Large traders usually hide their intentions because revealing an order’s size and direction appears to invite front-running. Yet price impact reflects two different forces: information about an asset’s value and the temporary pressure created by a need to trade. When market makers cannot distinguish an informed trader from a benign trader, they must price every hidden order as a mixture of both. Hiding therefore protects genuine alpha, but it also forces uninformed traders to pay an adverse-selection premium.

Hyperliquid offers a rare test of the alternative. Its protocol-native TWAP orders publicly reveal their direction, total size and duration while they execute. Comparing 465,000 visible TWAPs with 4.3 million reconstructed hidden metaorders, Barone and Lillo (in Trading in the Sunshine or in the Shade: Market Impact and Adverse Selection on Hyperliqui) find that visible orders face lower execution costs, leave less permanent price impact and attract additional liquidity. Displayed depth increases while these orders run, and the order book tilts toward the side needed to absorb them, especially for larger programs.

The conclusion is not simply that transparency is always better. Transparency sorts order flow. Traders willing to make their demand predictable allow market makers to compete for relatively benign flow, while hidden orders become more likely to contain private information and bear more of the adverse-selection cost. Secrecy remains valuable when protecting real alpha, but for traders executing for mechanical or liquidity reasons, trading “in the sunshine” can produce better execution.

Market Making & Adverse Selection

Trades do not arrive labeled “benign inventory demand” and “toxic private information.” A resting limit order is an offer to trade against whoever arrives next, including someone who may or may not know more. Liquidity providers protect themselves by widening the spread, showing less size, or moving their quotes after suspicious flow. The uninformed trader ends up paying part of the cost created by informed traders because both are drawn from the same pool.

Put differently, adverse selection is partly a classification problem. The market maker is not only estimating how many dollars a buyer wants to trade. It is estimating what those dollars mean. Flow is toxic to a liquidity provider when trading against it predicts an adverse markout, whether the advantage comes from information, latency, or some other structural edge.

In 1991, Anat Admati and Paul Pfleiderer formalized a way for liquidity traders to separate themselves in Sunshine Trading and Financial Market Equilibrium. If part of future liquidity demand is announced, market makers no longer need to assign the same information premium to it. The remaining unannounced flow also becomes more informative because known liquidity demand has been removed. Counterintuitively, revealing some uninformed trading can improve both the announcer’s execution and the information content of prices.

Preannouncement can also change who supplies liquidity. A known buyer gives market makers time to arrange capital, hedge on another venue, and decide how much inventory they are willing to warehouse. In Admati and Pfleiderer’s costly-entry model, the benefit is strongest when the announced demand draws in suppliers who would otherwise stay out.

Of course, the announced order can also be front-run. In a model of stealth versus sunshine trading, Torsten Schöneborn and Alexander Schied show that disclosure can attract counterparties or predators depending on market liquidity, the number of competitors, and their trading horizons. The question is not whether someone can trade ahead of a known order. They can. The question is whether the extra cost from anticipation is larger or smaller than the benefit from classification and competing liquidity supply.

Predictable Trading Literature

Traditional markets make the question difficult to test. But there are two close precedents. Nasdaq once allowed “flash orders,” briefly showing an order to a subset of participants before routing it elsewhere. Skjeltorp, Sojli, and Tham find that Nasdaq liquidity and overall market quality improved after the functionality appeared and deteriorated when it was removed. Their interpretation is that less-informed traders advertised liquidity needs and liquidity providers responded.

The scheduled roll of the United States Oil Fund (and other similar predictable institutional flow) provides another example. The fund had to move a large crude-oil futures position each month on a predictable schedule. This sounds like ideal prey. Instead, Bessembinder, Carrion, Tuttle, and Venkataraman find narrower spreads, greater depth, better resiliency, and more accounts supplying liquidity on roll dates, without evidence of systematic predatory trading. When impact is partly temporary and the market is resilient, absorbing the order can be more profitable than racing ahead of it. Competition then improves the liquidator’s proceeds.

Neither setting is the same as Hyperliquid. Nasdaq’s disclosure was selective and lasted only briefly. The USO roll was inferred from a calendar rather than announced through a protocol. They nonetheless show why a predictable order need not be a defenseless one: predictable demand can recruit its own liquidity.

Hyperliquid TWAP

Hyperliquid runs the largest centralized limit order book onchain. During the sample studied in the paper, its native TWAP mechanism submitted an order every thirty seconds and targeted a constant execution rate through time. Each slice was subject to a 3% maximum-slippage constraint, and underfilled slices could be made up later at no more than three times the normal child-order size.

The important feature is that the parent order is visible while it is active. Once the TWAP begins, the market can observe its direction, total size, and horizon. This is more informative than simply watching a series of buys appear on the tape. The remaining scheduled demand is known rather than guessed.

Anyone can announce an intention to buy. A native TWAP immediately begins carrying out the schedule under protocol rules. It can underfill if later slices fail to catch up, and the user can cancel it, so the displayed size is not a guarantee of completed volume. While the program is active, however, its parameters and progress are publicly verifiable on the Hyperliquid blockchain.

Barone and Lillo compare these visible programs with what they call statistical metaorders: sequences of same-address, same-market, same-direction market orders separated by less than thirty minutes. The statistical parent order is not directly observed. It is reconstructed from the child orders after the fact and becomes only gradually inferable while it is running.

Execution Schedule Features

Native TWAPs execute nearly linearly through time, as the mechanism intends. Their average price path is smooth and concave: impact rises quickly at first and then more slowly, reasonably close to the square-root trajectory produced by uniform trading under a transient-impact model.

Statistical metaorders are U-shaped. During the first tenth of their execution window, they trade at roughly 1.5 to 1.7 times the uniform rate. They slow to 0.6 to 0.8 times uniform in the middle, then accelerate to around 1.1 to 1.4 times uniform near completion. Their price trajectories inherit the same shape: impact bends twice, peaks sharply around completion, and decays afterward.

Price Impact Features

The two populations are not described by a common aggregate impact curve. Native TWAP orders display an approximately linear relation in log-log scale over the displayed range. Statistical metaorders display a markedly different shape. Their binned impact is fitted by the nonlinear arctangent specification.

Plotted in signed basis points on the right vertical axis, the TWAP trajectories display the typical concave, square-root-like profile documented in the empirical literature, rising as execution proceeds and well fitted by a power law.

Statistical metaorders display a different dynamic pattern. Their trajectories are markedly non-concave, with a double-inflection shape, a sharp peak around completion, and a more pronounced post-trade decay. This suggests that the difference between statistical metaorders and native TWAPs is not only a difference in impact levels, but also in the way impact accumulates during execution and relaxes afterwards. Such patterns are consistent with optimal execution under transient-impact models, which prescribes a U-shape in the trading velocity.

Announcer Execution Discount

If preannouncement separates visible liquidity demand from potentially informed order flow, or if it attracts additional liquidity supply when entry margins are important, announced orders should face lower execution costs than comparable latent orders. We test this prediction by comparing the volatility-normalized temporary impact of protocol-native TWAPs and statistical metaorders.

On common regions of participation and volume-time duration, statistical metaorders have greater temporary impact than native TWAPs in 81% of populated cells. Weighting each cell by the smaller sample increases the figure to 88%.

Among cells where both group means are positive, the median ratio of statistical to TWAP impact is about 2.3. The difference is concentrated among shorter and intermediate executions. The median statistical-to-TWAP impact ratio is about 5.4 for the shortest volume-time group and 3.1 for the intermediate group.

Among the longest executions, the surfaces nearly converge and the sign becomes unstable. The visible-versus-latent gap is not equally large everywhere. At the median execution profile and median daily volatility, a native TWAP has about 8.9 basis points less temporary impact than a comparable statistical metaorder.

Announcer Permanent Impact

The paper also studies the price after execution. Mechanical impact should decay as market makers redistribute inventory. Informational impact should persist because the trade has changed the market’s estimate of fair value. The amount of signed displacement left after completion is a useful proxy.

The authors measure prices half, one, two, and three execution durations after completion, not at an infinite horizon. At those four points, native TWAPs are associated with roughly 5.0 to 5.5 basis points less signed price displacement than statistical metaorders after controlling for relative size, participation, duration, volatility, market, hour, and day.

Impact on Unannounced Flow

We now focus on hidden traders who execute while visible TWAP flow is active. The key question is whether the presence of announced order flow changes the cost of latent metaorders. If visible TWAPs are interpreted by the market as relatively uninformed liquidity demand, then hidden same-side trading may become a more adversely selected residual category and therefore face higher execution costs. At the same time, visible TWAP activity may also attract liquidity supply, in which case hidden traders could benefit rather than suffer.

The data shows that a ten-percentage-point increase in already-visible same-side TWAP dominance is associated with roughly 0.84 to 0.92 basis points more signed impact for the statistical metaorder, from completion through three order durations afterward. Two simultaneous buyers will mechanically push in the same direction, so part of the relationship could simply be shared price pressure.

The authors add a square-root-law proxy for the mechanical pressure of overlapping flow. At completion, the already-visible same-side coefficient falls from about 0.86 to 0.64 basis points per ten percentage points of dominance, but remains positive. The adjusted coefficient is reported for this horizon only.

Impact on Liquidity

During active TWAP windows, the quoted spread is around 0.28 basis points wider. If liquidity is defined only as the best bid-ask spread, it does not degrades. However, displayed depth on the absorbing side, measured within five quoted spreads of the midprice, increases by about $4,200. The book tilts in the same direction: asks accumulate during a buy and bids during a sell. The cost of immediately sweeping $10,000 falls by approximately 0.025 basis points.

One interpretation is that an announced buyer creates predictable inventory risk: market makers need not place all additional size at the best ask, but can add it behind the touch at prices that compensate them for holding inventory.

Transparency, Competition, and Market Efficiency

Competition does not mean that nobody can trade ahead of a public order, but it does mean that a repeatable opportunity attracts capital. A market maker who sees a future buyer can acquire or hedge inventory before later slices arrive, then quote more size. If the price offered is too high, another maker can undercut it.

Anticipation may move part of the price response earlier, but competition determines how much of the opportunity remains as rent and how much comes back to the trader as better execution.

The paper observes only the net outcome: visible orders have lower measured impact in the conditional comparisons, and active TWAP windows have more displayed depth. It cannot separately quantify predation or tell us how the same orders would have performed if hidden.

This only works if competition is real. If access is controlled by a small set of firms, predictable flow can become a toll road. Public information helps the trader when multiple counterparties can act on it and compete away part of the rent. Hyperliquid has shown that it has sufficient competition on its platform such that announcing larger order is indeed optimal for execution in expectancy.

System-level transparency on blockchains makes the same information available to firms competing to fill the order. This stands in contracts to traditional centralized markets where a broker, venue operator, dark-pool participant, or fast observer often sees information first first. That can be the worst equilibrium: enough information to create a rent, but not enough access to compete it away.

We discussed the broader version of this issue in Market Fragmentation. Competition among venues is not the same as competition for an order. Dark pools, internalizers, and OTC dealers can give particular customers excellent execution while removing uninformed flow from the central book that supplies the reference price. The customer-level result may be good even as the public market inherits a more informed residual pool.

Hyperliquid presents a different arrangement. The order book and execution signal are shared. The native TWAP functions like a public request for liquidity rather than a request sent to one dealer. Any market maker with the required infrastructure can compete for the flow, and the address history is not locked inside one broker’s database. In principle, repeated fills let market makers estimate whether an account’s trades predict later prices and whether its flow is usually mechanical or opportunistic.

Traditional dealers build the same histories privately. An onchain market can make the economically relevant part of that repeated game public without tying the address to a legal identity, although the paper does not test whether makers use those histories this way.

CLOB vs AMM Announced Execution

At first the results in the paper seems inconsistent with our post on order routing, where we argued for keeping large swaps out of public mempools and using private relays or resolver auctions. It is not. A visible future schedule and a signed transaction waiting to be ordered are different objects.

A native TWAP announces a bounded schedule of future trading in an open order book. Many market makers can respond by posting liquidity, and the trader retains protections on how each child order executes.

A signed swap broadcast into a public mempool has already been authorized by the user. A block builder or searcher with control over transaction sequence can insert trades immediately before and after it. In Flash Boys 2.0, Philip Daian and his coauthors document bots competing for exactly this kind of transaction priority on early decentralized exchanges.

In the first design, visibility expands the set of potential counterparties. In the second, it can give an ordering actor a nearly free option on a trade the user has already authorized. Competition among sandwich bots may dissipate their profits into gas or builder payments without returning the value to the user whose execution deteriorated. Competition only helps the trader if the market structure points it toward the trader’s price.

Private does not necessarily mean uncompetitive. A resolver auction can conceal intent from the public mempool while making several solvers compete over a guaranteed outcome. The bad structure is selective information combined with weak competition or discretionary ordering, whether the venue calls itself public or private.

Conclusion

The lesson is not that traders should always reveal their hand. Secrecy remains valuable when an order contains genuine information or when the market structure allows others to exploit it without meaningful competition. But hiding is not free: when market makers cannot distinguish informed flow from mechanical demand, even uninformed traders pay an adverse-selection premium.

Hyperliquid shows what can happen when trading intentions are public and liquidity providers compete on equal terms. A visible order can become an open invitation to supply liquidity rather than merely an opportunity to front-run. The practical rule is simple: hide genuine alpha, but when the need to trade is mechanical, letting the market compete for the order may produce better execution.

Disclaimer: The information provided on TheLogbook (the “Substack”) is strictly for informational and educational purposes only and should not be considered as investment or financial advice. The author is not a licensed financial advisor or tax professional and is not offering any professional services through this Substack. Investing in financial markets involves substantial risk, including possible loss of principal. Past performance is not indicative of future results. The author makes no representations or warranties about the completeness, accuracy, reliability, suitability, or availability of the information provided._

This Substack may contain links to external websites not affiliated with the author, and the accuracy of information on these sites is not guaranteed. Nothing contained in this Substack constitutes a solicitation, recommendation, endorsement, or offer to buy or sell any securities or other financial instruments. Always seek the advice of a qualified financial advisor before making any investment decisions.

For the 100th post, we mined the archive for the pieces readers engaged with most. This isn't necessarily a list of our best work, or even our favorite. We finish with two personal favorites that narrowly missed the top spots. Both required substantial work and contain results we think are genuinely beautiful.

1) The Effects of Spot-Vol Correlation

If spot goes up or down, how should the volatility surface move with it? How should the observable dK/dσ inform our estimate of dS/dσ? Sticky strike and sticky delta are both approximations to this question. Sticky strike says that the volatility surface is fixed for every strike, no matter where spot is. Sticky delta says that the volatility surface moves along with spot such that options with the same delta have roughly equal vol. Both are useful approximations.

A skew trade is therefore is a view about the future relationship between spot and volatility being different than the present one encoded in options. The post also discusses the connection between vanna and gamma as well as the shadow delta.

We think the most fascinating take from this post is that it provides a useful way to think about the volatility surface as evolving over time. The shape the surface takes today matters, but so does the surface expected to appear once spot has moved.

2) Professional Punters S1E1

The first episode of Professional Punters features Evan Semet, a prediction-market trader and quant who has built real edge in some of the strangest and most inefficient corners of modern markets. Evan made low five-digits a day market-making fifteen-minute crypto binaries, with Sharpe ratios in the 40-50 range. He also found major edges in election arbitrage, AI model leaderboard markets, and insider-tracking.

We discuss trades Evan made in the following markets

• 2024 Election (and other elections)

• 538 Presidential-Approval Ratings

• AI Leaderboards (what model is the best)

• Mention Markets (words mentioned in speeches)

• Bitcoin (and other crypto) 15-Minute up/down

• Discretionary Book in Political Topics

• Resolution Arbitrage

3) Market Probabilities are NOT Real Probabilities

A prediction contract trading at 67 cents is often interpreted as the market pricing a 67% probability of the event occurring. The post argues that this interpretation is not always strictly correct.

The post begins with a simple market populated by Kelly bettors, showing that a market populated by rational agents ends up with a price different from the 67% probability forecast. Even when participants are rational, have equal bankrolls, and bet according to their honest beliefs, the market-clearing price does not generally equal the average of those beliefs. The nonlinear relationship between prices, odds, and optimal position sizes means that some beliefs receive more influence than others.

Once prediction contracts are connected to the rest of a trader’s portfolio, the interpretation becomes even less literal. A contract’s price can reflect risk aversion, existing exposure to stocks or rates, perceived correlation with other events, funding costs, or the opportunity cost of locking up capital.

The most valuable takeaway from this post might be the recognition of the difference between market efficiency and probabilistic literalness. A market can be difficult to beat while still not pricing outcomes as independent probabilities. The number might be informative, but it is not a reliable poll. This is an example of a concept that seems somewhat technical but has significant implications for almost every prediction-market plot or chart.

4) Options Implied Distributions are NOT Real-World Distributions

By combining option prices across strikes, we can construct an implied distribution for the future price of an asset. It is tempting to read that distribution as the market’s collective forecast; this post argues that is wrong.

The risk-neutral distribution is not a real-world distribution, despite incorporating all market information. By construction, risk-neutral probabilities discard views on expected returns, assuming all assets have the same return. One way to interpret the risk-neutral distribution is that it considers the value of money in different states of the world. During a recession, cash is more valuable than during an expansion.

An option’s price reflects these risk premiums in addition to the probability of each state occurring. The value of cash during a recession is higher than during a period of expansion, so crash insurance is more valuable than it would seem at first glance. A similar argument applies to long-dated options and futures. If investors expect higher returns from risky assets than the risk-free rate, the real-world distribution will have a higher mean than the risk-neutral one.

This nuance has crucial implications for understanding options markets. An option trade can have positive expected value to buy for a trader who is willing to take on additional risk while also being good to sell for a market maker who is able to manufacture it via dynamic hedging.

5) Decomposing Volatility Skew

The volatility skew of a single stock is not just the skew of the stock itself. It can also incorporate macro skew, sector skew, and the skew of particular factors. The stock’s sensitivity to the overall market, sector, and themes can be extracted and analyzed separately from the unique characteristics of the security.

A market downturn will increase the volatility of a stock, but the reason for the increase is crucial. If the decline is due to a broad market sell-off, implied volatility will rise because the stock’s beta is magnifying the impact of the market move. If the drop is caused by a decrease in the stock’s unique risk, implied volatility might actually decrease depending on the shape of the idiosyncratic skew.

6) In Defense of Leveraged ETFs

Leveraged ETFs are frequently considered dangerous by default. The argument is that a leveraged position is inherently riskier. The post argues that this perspective is deeply incorrect. 1x is not risk-free, but it is simply one point on the risk spectrum.

The post was not intended to defend leveraged ETFs in all circumstances, but it highlights the importance of understanding leverage as a tool for constructing a portfolio with desirable risk characteristics. Leverage makes risk exposure a continuous variable rather than a binary choice between 1x long and 1x short.

A young investor with stable future earnings, for example, may have much of their total wealth tied up in relatively bond-like human capital. Viewed across the entire balance sheet, a conventional portfolio may leave them underexposed to market risk.

Markowitz, CAPM, constant-relative-risk-aversion utility, and the Kelly criterion all point toward a similar framework: first construct the best risky portfolio available, then scale that portfolio up or down to match the investor’s risk tolerance.

7) Implied Correlation and Index Skew

An index’s volatility skew reflects not only the volatility of its components but also the degree to which its constituent stocks move together. This skew can be used to infer the market‘s expectation of correlation between the securities in the index.

Dispersion trading simultaneously expresses a view on volatility and correlation risk. Long dispersion sells correlation and volatility, while short dispersion buys correlation and volatility.

An additional result is that the diversification lost when shifting from completely uncorrelated stocks to mildly correlated ones is substantial compared to the loss of diversification when shifting from highly correlated to moderately correlated stocks.

8) Market Impact

Price impact grows approximately with the square root of an order’s size relative to normal trading volume. An order equal to 1% of daily volume can therefore move the price by something on the order of 10% of its daily volatility. The post attempts to reproduce this law using dimensional analysis. Then, it discusses the order-book dynamics that can generate this relationship.

This market-impact model is interesting in how it connects to the discussion of volatility. If we consider volatility to be a measure of how much trading pressure a market can absorb before rebalancing, the discussion of market impact helps explain some of the most confounding aspects of equity market structure: volatility clustering, effects of crowd positions, and increased correlations during selloffs.

High volatility can be a consequence of both increased information arrival and a thinning out of liquidity providers, reducing the market‘s ability to dampen price changes. There are many ways to analyze markets, but we find this framework helpful in connecting ideas that are often treated in isolation.

9) Markets are Efficient if and only if P = NP

Two of the posts we worked hardest on fell outside the most-read list, and we’re still a little sad about that, so we‘re using our editorial cartographic privilege to put them here anyway. Both contain results we think are genuinely beautiful.

Market efficiency is usually discussed in terms of information. If there is a pattern in asset prices, traders will profit by finding it, and the pattern will stop existing.

Patterns that persist despite being known are rationalized as the result of different levels of information among traders, inattentiveness, institutional restrictions, market frictions, etc... What that formulation misses is that publicly available information is not necessarily information that is readily incorporated into security prices.

A profitable trading strategy might involve information that is easy to understand and apply after being found, but extremely challenging to find in the first place.

Under a certain stylized formalization, the weak-form market efficiency is equivalent to P=NP. In the first direction, identifying a consistent, budget-constrained technical trading strategy is equivalent to the NP-Hard Knapsack problem. In the second direction, a sufficiently efficient market with a suitable order syntax can be used to implement an algorithm that solves a 3-SAT formula given its clauses.

A market that can “discover” the implications of its own history immediately and consistently would be able to perform computations as hard as any NP-complete problem. Market efficiency is therefore a claim not simply about the presence or absence of information and incentives but about computational feasibility and representational capacity.

10) How to Extract Max Value from Your Edge?

A trader with an edge has a probability distribution that differs from the market‘s. This post provides an exact quantitative answer to the question: how much does this edge actually help the trader, and what position realizes that help?’

The result is technical but combines derivative pricing, Kelly betting, utility theory, information theory, portfolio selection, arbitrage, and presents them as a single, coherent whole.

Disclaimer: The information provided on TheLogbook (the “Substack”) is strictly for informational and educational purposes only and should not be considered as investment or financial advice. The author is not a licensed financial advisor or tax professional and is not offering any professional services through this Substack. Investing in financial markets involves substantial risk, including possible loss of principal. Past performance is not indicative of future results. The author makes no representations or warranties about the completeness, accuracy, reliability, suitability, or availability of the information provided._

This Substack may contain links to external websites not affiliated with the author, and the accuracy of information on these sites is not guaranteed. Nothing contained in this Substack constitutes a solicitation, recommendation, endorsement, or offer to buy or sell any securities or other financial instruments. Always seek the advice of a qualified financial advisor before making any investment decisions.

Typical High-Beta Observation

Suppose a semiconductor stock is commonly described as “four beta.” On ordinary days a 1% move in the index is often accompanied by something like a 4% move in the stock. It is tempting to extend the same line and say that a 3% index decline should produce a 12% stock decline in expectation.

Then we observe multiple occasions where index falls 3% and the stock falls only around 8%. The usual reaction is to search for an idiosyncratic story. But there is a simpler possibility that the stock may genuinely be four beta near spot while having a much lower beta over a large move.

The Central Equation

The central idea is not that the original beta estimate was wrong but that a beta estimated around one state does not have to describe the entire path to another state.

The ordinary definition of beta is the covariance of the stock with the market divided by the variance of the market:

Writing covariance as correlation multiplied by the volatilities of the two assets:

This equation separates two questions that are often blurred together. Correlation asks how tightly the stock and index move together. The volatility ratio asks how large the stock’s typical movement is relative to the index’s movement. A stock can be highly correlated with the index but have only a modest beta if its volatility is close to index volatility. A stock can also have a high beta with less-than-perfect correlation if its own volatility is several times index volatility.

The direction of each effect is immediate. Higher correlation raises beta. Higher single-stock volatility raises beta. Higher index volatility lowers beta, because index volatility sits in the denominator. In proportional-change form:

In a selloff, stock volatility usually rises and correlation usually rises, but index volatility can rise even faster. The stock can therefore become more volatile, more correlated, and yet lower beta at the same time. Beta compression does NOT require correlation compression.

Options-Implied Beta Curve

Options provide a forward-looking beta curve via the stock-volatility surface, the index-volatility surface, and an implied correlation surface.

The superscript Q is a reminder that option prices live under the risk-neutral measure. This is not automatically the same as the beta expected under real-world probabilities. But the decomposition is still valuable because it tells us where the option market is embedding the exposure.

Single-stock volatility and index volatility are directly visible in their respective option markets. Correlation is not quoted with the same completeness. Instead, average implied correlation is backed out from the variance identity for a weighted basket which gives the index:

Constituent options tell us how much standalone variance is priced stock by stock. Index options tell us how much total basket variance is priced. The difference is the covariance the market must be pricing. Under an approximation:

Concluding Notes

A four-beta label does not mean that for every 1% market move, regardless of size or state, in expectation produce a 4% stock move. It means that around some reference state, the stock’s covariance with the market is about four times market variance.

As the market moves, the three ingredients of beta move with it. In the stylized semiconductor example, stock volatility rises from 80% to 84%, correlation rises from 0.70 to 0.88, and index volatility rises from 14% to 30%. The stock becomes substantially more macro-driven, but beta falls from 4.00 to 2.46 because the index-volatility denominator reprices fastest.

The beta of the complete 3% index decline is not the spot beta and not the endpoint beta. It is the average local beta traversed along the path. Here that average is 2.68, which turns a 3% index decline into roughly an 8% stock decline.

Historical returns show how that curve has realized. Single-stock and index options reveal the two volatility surfaces. The gap between index variance and constituent variance reveals the market’s aggregate dependence pricing.

Once those pieces are put in the correct order, the apparent puzzle disappears. A four-beta stock falling only 8% when the index falls 3% is not necessarily an anomaly. It may be exactly what the volatility and correlation surfaces imply.

Disclaimer: The information provided on TheLogbook (the “Substack”) is strictly for informational and educational purposes only and should not be considered as investment or financial advice. The author is not a licensed financial advisor or tax professional and is not offering any professional services through this Substack. Investing in financial markets involves substantial risk, including possible loss of principal. Past performance is not indicative of future results. The author makes no representations or warranties about the completeness, accuracy, reliability, suitability, or availability of the information provided._

This Substack may contain links to external websites not affiliated with the author, and the accuracy of information on these sites is not guaranteed. Nothing contained in this Substack constitutes a solicitation, recommendation, endorsement, or offer to buy or sell any securities or other financial instruments. Always seek the advice of a qualified financial advisor before making any investment decisions.

Daniel traded fixed income at a top investment bank and analyzed public and private investments at a tech-focused, single-manager hedge fund. He was also forward-deployed at Palantir and has a research background in development economics. He is now a startup founder in prediction markets, building low-latency trading infrastructure to sports markets.

This conversation is about what actually separates a bank from a hedge fund, the esoteric risks bank traders are paid to warehouse, how the AI data-center buildout is being financed, and where the edge really lives in prediction markets, from Strait of Hormuz contracts to correlated parlays and courtside live betting.

Banks Move Risk, Hedge Funds Take It

The cleanest way to understand the two institutions is by their relationship to risk. A bank, to some degree, is an institution that does not really try to take risk itself (partially due to regulatory constraints post 2008). It is in a sense a market maker: it wants to buy something, sell something, and collect a fee for standing in between.

A [discretionary] hedge fund is structurally built to take risk. It has a fundamental view on how a company or an asset will perform, and its investors give it money precisely to make that bet. At the fund where Daniel sat (long-biased, longer-duration) the work looked closer to traditional value investing: is this company going to beat earnings, is it actually in structural decline because of AI, or is the story overblown.

Banks Also Warehouse Esoteric Risks

In an ideal world, a trader at a market-making firm goes home with close to zero exposure. In practice that is almost impossible for a dealer in anything even slightly exotic or unlisted. So a lot of bank traders are essentially paid to warehouse esoteric risks, on an OTC or bilateral basis.

The best example is the deal-contingent hedge. Investment banks take companies public and finance M&A, where lenders borrow vast sums to buy a company that may be listed on public markets. Both prices move minute to minute: the equity through the stock market, and the cost of borrowing through the capital and money markets.

A macro shock may send the funding rate through the roof; the economics of the deal change, the buyer wants to pull out or the seller wants to force it through. A deal that takes six-plus months to close carries real timeline risk the whole way. The bank gets called in to hedge all of it with over-the-counter swaps across both the equity and bond markets. It is genuinely exotic risk, and it is warehoused rather than offloaded, because it is very hard to package into a product.

The Blurry Regulatory Line

So where is the line between market making and prop betting? Daniel’s answer: capital markets facilitation. Where you are directly supporting the investment banking practice, is something very few regulators take issue with. The rough rule is that speculation is not okay, but risk taken on in the facilitation of genuine economic activity or capital formation is. In addition, banks may use intermediaries to distance themselves from risk to be extra safe.

Mortgages are the cleanest illustration of how the machine got rebuilt around that rule. Banks mostly do not structure mortgage-backed securities themselves anymore. Instead they provide private credit financing to the firms that do (though these private credit firms are at times highly leveraged).

A mortgage-backed security, scary as the phrase sounds after 2008, is at some level just a giant pool of mortgages and a relatively safe investment. Lending to the structurer instead of structuring is safer, since there is another level of protection.

While the bank do still stay fundamentally exposed to some mortgage risk, regulators allow for this structure as banks are the ONLY financial intermediaries with sufficient capital to finance the US housing market.

Securitizing the AI Buildout

The most interesting thing happening in credit right now is the data-center buildout. If you pay close attention to Bloomberg you see such deals constantly, such as the one that Meta just closed with Blue Owl.

The logic is that very few capital providers, even the Blackstones of the world, are willing to take down $50 billion of data-center risk by themselves. So consortiums of banks lend alongside private credit providers, facilitating the creation of bonds backed by the data centers, the land, the revenue, the chips (i.e. GPU-backed loans), and ancillary equipment.

Is there a risk that new chips make today’s GPUs depreciate faster than the pricing assumes? Always. But Daniel’s more cynical observation is that much of private equity already faces exactly that risk elsewhere: B2B SaaS was underwritten as bulletproof recurring revenue, sometimes with loans literally made against ARR, and the advent of AI has put a lot of those prices into question.

The primary concern right now is simpler. Google, Anthropic, and OpenAI want to build as many data centers as possible, and the system’s job is figuring out how to finance that.

“Bonding” in Prediction Markets

Daniel trades prediction markets personally. His framework is that there are a few ways to make money, and they are the same in financial markets and prediction markets. One such way is being paid to take on risk.

At the time of recording there was a contract on whether the Strait of Hormuz returns to its pre-war import and export state by mid-July, with the market around 10 to 20%. Take the other side and $100 can make you roughly $120 in a month. But a high rate of return is not the same thing as a good bet. There is a legitimate chance the strait does not reopen, and you get hosed. Eight times out of ten you make money. The other two, you lose everything.

Nonetheless, some markets fundamentally inefficient. There is a contract on whether the Shah’s heir returns as Iran’s next political leader within the year. There is legitimately a chance it happens, but the market pays around 5% by year-end, roughly 10% annualized, way more than the risk-free rate of the US Treasury market.

That price is less a reflection of the actual probability than of market structure: some people really want to make that bet, and market makers do not want to warehouse long-duration risk for a limited gain.

This is why you see people making 100 to 200% a year on these markets, and a whole class of freebies yielding 5 to 15% from selling very-low-probability events that are still overpriced. The edge persists because there are so many other edges competing for attention, and diversified across enough contracts the positions are close to uncorrelated. The caveat worth flagging: the trade only works if your cost of capital is below that 5 to 15%, which is not always true for a lot of people.

Sports Betting Combo Market Structure

Prediction markets found their product-market fit in sports trading. On Kalshi it is still over 90% of volume one way or another, either directly through sports contracts or through combos, which are parlays with subtle differences (every sportsbook resolves parlays in its own way).

The structural difference is that combos on Kalshi and Polymarket are quoted through an RFQ system. You structure the bet yourself, say the Seahawks, the Ravens, and the Commanders across games, put the request out to the world, and market makers quote you a price.

Intrinsically that is a little more adverse; the market maker is trading a unique esoteric bet rather than a liquid contract, and only gets to trade it if other market-makers do not. In addition, the user keep the last look as they do not have to trade after the RFQ comes back. Nonetheless, on average the pricing produced by this system is still significantly better for users than what they get on sportsbooks.

The quoting side is more nascent than you would guess. Most combos on Kalshi today are quoted by much smaller independent traders, not large integrated firms, partly because the concept is only about three months old.

On collateral, smaller firms post close to full collateral while large market makers get preferential terms, and Kalshi now pays interest on balances at roughly 25 to 50 basis points below the risk-free rate. Your edge on the parlay quote has to overcome your firm’s cost of capital minus that rate, though this is probably not hard as most parlays expire within a week anyway.

Edge in Implied Correlations

Part of the reason dealers are compensated so much for quoting parlays is that they take on an extra dimension of correlation risk (two 50% legs do not make a 25% parlay when the legs are correlated).

Daniel talks about a case where, back in the early days of DraftKings, bet the under on every single player in a basketball game and the over on the team total would have made money. Those legs are obviously highly [anti]correlated: it is essentially impossible for the team to hit the over if every individual player hits the under.

However, the parlay was completely mispriced, treated as if it had a literally zero percent chance of happening (instead of merely very small), so on the rare occasions the combination did hit, the payout was something like 1000x at a time. In this case, the allowed betting size would have let a user make on the order magnitude of hundreds of thousands of dollars in total.

That is less an indictment of one book than of the category. Sportsbooks are known for being pretty bad at trading in general. Most of the company is focused on distribution and getting users onto the platform, and the business model is taking a cut of retail bets, not surviving a competitive landscape where shark hedge funds chase down mispricing.

Courtside Beats the Couch

Around six months ago, a big phenomenon in the development of sports betting on Kalshi and Polymarket was people going to games in person and live betting with literal knowledge of what was happening on the floor. Watch Steph Curry hit the three in front of you and you are trading ahead of the market. In the early stages of prediction markets, simply being at the game made you one of the fastest traders in the market.

The flip side: betting off your TV is a bad bet. Depending on where you are in the US, the broadcast is 20 to 30 seconds delayed, so the person you are trading against already knows what happened. There is still a very large volume of people who bet from what they see on TV anyway.

Market makers adapted quickly. Small size arriving very early in a move is probably someone sitting in the arena trading on their personal calculus. Some market makers pay for professional data feeds, and a whole industry has been built around the collection of sports data. If you trade within one or two seconds, the market probably has not fully priced it in. Wait five and you are late. Even DraftKings, which might not know exactly what is happening in the game, knows enough to anticipate volatility, lengthen its lines, and prevent you from trading.

Be Smarter, Be Faster, or Cheat

When the short-duration crypto markets first came out on Polymarket, the 5-minute and 15-minute Bitcoin contracts were at one point literally based off a single Binance price. A large individual whale could buy one side of the “Bitcoin up” contract, manipulate the market to push the indicator down for a while, accumulate even more shares, and then fire a huge buy order on Binance at the very last minute so the contract resolved their way. This happened with some regularity in the earlier Bitcoin markets, and people made quite a bit of money. It is also definitely illegal in any type of regulated financial market.

Which is the note the conversation ends on. As the line from Margin Call goes: be smarter, be faster, or cheat. There has been no new idea in trading in the last twenty years, just the same concepts implemented in new markets, and prediction markets are simply the newest place the old game is being played.

Professional Punters is presented by Freeport Markets: Freeport lets users trade 24/7, weekends included, with no KYC and up to 200x leverage across stocks, indices, commodities, crypto, rates, and more, with access to pre-IPO exposure in companies like OpenAI & Anthropic. For traders who already follow sharp sources across Twitter, Substack, hedge fund filings, corporate insider activity, and political trading disclosures, Freeport uses AI to read the sources they trust, connect the dots across markets, and surface trade ideas as narratives develop.

About the Host: Lihong Wang is the founder and CEO of Freeport Markets. Before starting Freeport, he traded discretionary semiconductor names at quantitative market-making firms including Jane Street and IMC Trading, building both systematic and discretionary strategies. Lihong graduated from Duke with a degree in mathematics and statistics, and has raised over $2.5M to build Freeport from investors including Y Combinator, Alliance DAO, and Informed Ventures. For more of his long-form research, visit freeportlogbook.substack.com.

Disclaimer: The information provided on TheLogbook (the “Substack”) is strictly for informational and educational purposes only and should not be considered as investment or financial advice. The author is not a licensed financial advisor or tax professional and is not offering any professional services through this Substack. Investing in financial markets involves substantial risk, including possible loss of principal. Past performance is not indicative of future results. The author makes no representations or warranties about the completeness, accuracy, reliability, suitability, or availability of the information provided._

This Substack may contain links to external websites not affiliated with the author, and the accuracy of information on these sites is not guaranteed. Nothing contained in this Substack constitutes a solicitation, recommendation, endorsement, or offer to buy or sell any securities or other financial instruments. Always seek the advice of a qualified financial advisor before making any investment decisions.

From Bond to Perpetuity

A bond promises a fixed schedule of nominal cash flows. Its Macaulay duration is the present-value-weighted average time until you receive them:

The weights above are each cash flow's share of present value, so they sum to one. Duration is a genuine probability-weighted mean of times, measured in years. A zero-coupon bond maturing in ten years has a duration of exactly ten.

Separately, and apparently unrelatedly, you can ask how the price P responds to a change in yield. That is modified duration, a sensitivity with some connection to time:

Differentiate the price with respect to yield and the algebra hands you the weighted average time, divided by one plus the yield. Each cash flow's price elasticity happens to be proportional to its own maturity, and elasticities aggregate linearly in present-value weights, so "average waiting time" and "price sensitivity" turn out to be the same number wearing two labels. Under continuous compounding the divisor disappears and they are literally identical.

Perpetuity provide the bridge to equities. An instrument that pays a fixed coupon forever has no maturity at all, yet its duration is finite and completely well behaved; in particular, the duration of an infinitely-lived claim is just the reciprocal of its yield.

An equity is a perpetuity. The only complication is that its coupon grows. In addition, bond duration rests on four other assumptions that equities violate.

Bond cash flows are contractual; equity cash flows are forecasts. Bond cash flows are nominal; equity cash flows are broadly real. Bond cash flows are fixed; equity cash flows grow at a rate that is itself estimated. And most importantly, bond cash flows are exogenous to the discount rate: when the Treasury curve moves, the coupons on a Treasury do not move with it. Corporate cash flows absolutely do.

From Perpetuity to Stock

The Gordon growth model prices a perpetuity whose cash flow Dn grows at a constant rate g and is discounted at r (i.e. how a stock hypothetically behaves). Price is

Differentiating gives the duration

Showing that the modified duration of an equity is the forward price-to-dividend ratio, assuming all shareholder returns are returned via dividends and not buybacks.

Equity duration is therefore a restatement of valuation. Any sentence of the form "growth stocks are long duration" is, under this model, the same sentence as "growth stocks are expensive," and carries exactly as much information.

Note also that duration is a rectangular hyperbola in the gap between the discount rate and the growth rate, which produces the explosive behaviour,

Halving the gap doubles the duration. Long-duration valuations are not merely sensitive to assumptions, they are unbounded in them. Now differentiate with respect to growth instead:

Equal in magnitude, opposite in sign: r and g enter the pricing function only through their difference, so the two partials must be negatives of each other everywhere. The asset does not know, and cannot know, whether the spread narrowed because the discount rate fell or because the growth rate rose. Write the total differential

shows that price responds only to the change in the spread, amplified by the multiple. If growth and discount rates rise together one for one, price is exactly unchanged, at any multiple, forever. Take the second derivative and something else falls out:

Convexity scales with the square of the forward multiple. And duration falls with the square of itself as rates rise, which means the asset shortens as it gets cheaper and lengthens as it gets more expensive. Duration is a state variable, not a constant. A regression estimated over a long sample spanning very different multiple regimes is estimating a moving target, and will report an average that describes no actual period.

Duration Interpretation

When Citrini writes that

Investors are always either discounting the promise of the future or the reality of the present. And they are never equally weighting them.

During the early part of a hype cycle, leading up to and directly following a technological advancement, investors are typically discounting the future while focusing on the present. A good example for this is Nvidia at the end of 2022: investors were solely focused on the headwinds presented by the crypto GPU glut, the anemic gaming PC market and the recent rise in rates causing fears about a near term recession.

Then, as the cycle begins, investors begin to shift to incorporate the future - they stop focusing so much on the present and see the promise. They move out in terms of valuing away from last twelve months current price / current earnings to next twelve months. Then, as price climbs and the technology becomes more exciting, their imagination takes hold. At a certain point they begin discounting the present much more heavily and the future becomes the only thing that matters. Valuation metrics over the next twelve months become useless in favor of 2, 3 or 5 years forward.

At the peak, the present is not considered at all, it is 100% driven by an imagined future (even when that imagination doesn’t necessarily align with a bullish outcome for the stocks driving the rally). Analysts aggressively raise estimates in ways that, at the time, seem fundamentally justifiable (if you take the assumptions at face value - for example, “everyone in the world will have two cell phones” was a good one from the mobile phone hype cycle). Capital is sucked in which ultimately forces performance chasing and crowds stocks with money that doesn’t really believe in the thesis. “A twilight period where people continue to play the game, but no longer believe in the rules” emerges, as Soros put it.

The valuation of SaaS stocks in mid-2021 is a great example of what happens when the future is overvalued relative to the present - nobody cared about climbing inflation, that rates had nowhere to go but up, that these companies were reliant on ZIRP or that software could become more competitive.

Then, a negative catalyst occurs - this can but doesn’t have to be related to the technology, macro, credit, underwhelming earnings. The estimates start to seem unattainable, and the present begins to matter more when the future seems more uncertain. That exact mechanism that drove future optimism to unsustainable heights mechanically reverses, everyone needs out. The future begins to be discounted until it results in a sense of disillusionment with not just the stocks but the technology itself. This overshoots to the downside, investors eventually become disillusioned and seemingly allergic to anything having to do with the technology. This happens in a very asymmetric manner to the climb (“stairs up, elevator down”).

This is the crucible in markets for truly transformative tech. If advancements persist, another opportunity to get long presents itself before capital once again begins flowing into the companies (the internet, for example). If they don’t - not necessarily “the tech goes away” but rather that it ceases to advance once the capital isn’t free or plateaus or the economics prove to be unfavorable - the cycle will still start again, just with a new technology.

He is restating the point that early in a hype cycle, investors still anchor valuations to current earnings and near-term risks, so the stock behaves like a relatively short-duration asset; as enthusiasm builds, analysts raise long-term growth estimates, investors push the valuation horizon from the next twelve months to three, five, or more years out, and an increasing share of the stock’s value comes from distant expected cash flows.

The stock therefore becomes longer duration, and highly sensitive to changes in interest rates, discount rates, and long-run growth assumptions, but less sensitive to present-day earnings. At the extreme, a quarterly earnings report may become almost a non-event unless it changes the perceived long-term trajectory, because investors are valuing the imagined future rather than current fundamentals.

When a negative catalyst undermines those distant growth assumptions or raises the discount rate, duration compresses abruptly, the present matters again, and the same mechanism that drove the stock upward reverses, often producing the familiar “stairs up, elevator down” dynamic.

The Impact of Inflation

Let inflation be π, let the nominal discount rate be r + π, and let nominal cash-flow growth absorb inflation with a pass-through coefficient λ:

At full pass-through, where companies in aggregate raise prices with inflation exactly one-to-one, λ = 1 and equity has exactly zero duration to expected inflation. At λ = 0 it behaves like a long nominal bond. So the popular framing "equity is a very long nominal bond" gets the inflation term precisely backwards.

The better analogy is that equity is a perpetual inflation-linked bond: real-rate exposure yes, inflation exposure no. That also implies the correct benchmark for equity's discount-rate exposure is the TIPS curve, not the nominal curve, which turns out to matter a great deal when you actually run the regressions.

Setting λ = 0 when it should 1 is called the Modigliani-Cohn money illusion: capitalising real cash flows at nominal discount rates. It is also the error embedded in the Fed model, comparing a real number (the earnings yield) to a nominal number (the Treasury yield), which is the objection Cliff Asness raised in 2003.

Decomposing the S&P 500 dividend yield into a rational growth forecast, a subjective risk premium, and a residual mispricing term, shows that inflation explains close to 80 percent of the time-series variation in the mispricing component, with high inflation associated with underpricing in the broad equity markets.

The Impact of Convexity

Consider two companies, valued with an explicit ten-year forecast plus a terminal perpetuity, with the duration computed as a present-value-weighted average of times. The terminal block is not a lump sum arriving at year ten but a perpetuity beginning at year eleven, and its own Macaulay duration measured from today is:

The value stock: Payout ratio 60 percent, growth 2 percent, discount rate 8 percent. Normalizing earnings to 1, this prices at 10.2 times earnings. Of its total present value, 43.5 percent sits in the first ten years of dividends and 56.5 percent in the terminal block, whose own duration is 28.0 years. The weighted average comes to a Macaulay duration of 18.00 years, or 16.67 modified.

The growth stock. Payout 10 percent, growth 12 percent for a decade then fading to 3 percent, same 8 percent discount rate. This prices at 19.0 times earnings, and 93.5 percent of its value sits in the terminal block. Macaulay duration: 29.93 years, or 27.72 modified. A summary of their cashflows is shown below

Equity duration is violently convex. The growth stock loses 22.8 percent on a 100 basis point rise and gains 35.1 percent on a 100 basis point fall. Any hedge ratio set from a point-estimate duration is wrong in both directions, and wrong asymmetrically (rate cuts makes stocks go up more than rate hikes).

Cashflow Duration of S&P 500

Gordon growth model inverted says the forward payout yield is the discount-rate gap, so duration is simply the reciprocal of the payout yield.

Moving from the dividend yield to the total payout yield collapses the S&P duration from 77 years to 30, a factor of 2.5, before any economics enters at all. Since the late 1990s the majority of US corporate cash distribution has run through buybacks rather than dividends. So the economically correct denominator in P/D is net payout.

30 years still seems long against empirical estimates that are far shorter. How much shorter is genuinely contested. An unanticipated 25 basis point cut in the fed funds target is associated with roughly a 1 percent rise in broad stock indexes, which implies something on the order of 4 percent per 100 basis points. Generally a figure of “2 to 6 years” to fed funds circulates widely in practitioner writing.

Of course, a fed funds surprise is NOT a move in the long real rate that actually discounts the cash flows, and a 25bp funds surprise moves the ten-year by considerably less than 25bp. Scaled against the right denominator the implied duration would be meaningfully higher.

Growth and Rates

Additional gap between empirical equity duration and the duration derived from cashflows can be explained by the positive long term correlation between interest rates and growth; let g' = dg/dr then

Real interest rates and real GDP growth are often positively correlated because stronger growth raises the demand for capital: firms want to invest more, households consume more, and the central bank may keep policy tighter to prevent overheating. At the same time, higher expected productivity growth increases the return on investment, so borrowers are willing to pay a higher real rate. Conversely, weak growth reduces investment demand and usually leads to monetary easing, pushing real rates lower. The relationship is therefore partly fundamental (higher growth raises the equilibrium real rate) and partly policy-driven.

In the neoclassical theory, the equilibrium real rate is anchored to trend growth by household patience regarding future versus present consumption

The Ramsey rule describes the steady state of the consumption Euler equation: where the real rate is expressed as the rate of time preference (rho) plus the curvature of utility (sigma) (risk aversion = inverse elasticity of inter-temporal substitution) times trend consumption growth (g). A fast-growing economy needs a high real rate to persuade households to defer consumption; a stagnant one cannot sustain it. Empirically, the relationship between real rates and growth are well-observed.

Model duration is therefore a partial derivative: the response to a pure discount-rate shock with cash flows held fixed. Empirical duration is a total derivative: the response to whatever actually moves yields in the sample, which is mostly news about nominal growth. They are different objects, both correct, and the ratio between them is a direct estimate of how much of a rate move the market reads as growth news.

Assuming a 0.8 (or indeed, any reasonable number) beta between rates and growth means that 0.8 of any move in the ten-year yield is matched by a move in expected perpetual nominal cash-flow growth, which reduces the effective duration of the market by 80%. Notice also that the same formula explains negative duration. Whenever cash-flow expectations rise more than one for one with the discount rate, empirical duration goes negative and the asset appreciates when rates rise.

Correlations between average real rates and average growth are empirically noisy, and other factors such as demographics, safe-asset demand, regulation, play a dominant role in average real rates. Nonetheless, native estimates of equity duration that does not take into account the relationship between rates and idiosyncratic, sector, and macroeconomic growth typically overshoots. See example below from Campbell

A 100bp surprise rate change does not typically cause a 20% drawdown in the S&P; but the graph does correctly note that growth assets have longer durations.

Stock-Bond Correlation

Over long horizons, real interest rates and real economic growth should generally be positively related. Higher trend productivity and growth increase the expected return on capital. Slow-growing economies normally have less demand for capital and lower equilibrium real rates. That long-run relationship does not mean rates and growth must move together over every shorter period. Their short-run correlation depends on the type of shock driving the economy.

In a demand- or productivity-driven regime, growth expectations and rates usually move in the same direction. Stronger demand or productivity raises expected corporate earnings, but it also increases investment demand and may cause monetary policy to become tighter, pushing rates upward. If the earnings effect dominates for equities, stocks rise while higher yields cause bond prices to fall. Stocks and bonds therefore move in opposite directions, producing a negative stock–bond correlation. In a conventional recession, the process reverses: growth expectations and rates fall together, stocks decline and bond prices rise.

In an inflation-, supply- or policy-driven regime, rates and growth can move in opposite directions. A negative supply shock raises inflation while reducing real output, and the central bank may respond by raising rates despite deteriorating growth. An independent monetary tightening can produce the same combination: higher real rates and weaker expected growth. Stocks are hurt by both weaker cash flows and higher discount rates, while bonds fall because yields rise. Stocks and bonds therefore move together, producing a positive stock–bond correlation.

The reverse can also occur. A favorable supply or disinflation shock can improve expected growth while allowing rates to fall. Stocks then benefit from stronger earnings and lower discount rates, while bonds rise as yields decline. Again, stock and bond prices move together, so their correlation is positive.

Factor Interpretations

Value stocks are often described as short duration because more of their valuation is supported by current earnings, while growth stocks depend more heavily on distant cash flows. But this relationship is partly mechanical: expensive stocks are assigned longer duration because more of their market value lies beyond current cash flows. Saying that growth is expensive and that growth is long duration can therefore amount to stating the same fact twice.

Profitable companies are generally shorter duration because they already produce substantial earnings relative to their valuation. Less of their value must be justified by distant and uncertain growth. High-payout firms distribute cash through dividends or repurchases rather than retaining it for future projects. Investors receive more of the company’s value today and depend less on management successfully reinvesting capital for many years. Low-volatility stocks heavily correlates with value, profitability, and high-payout factors.

Aggressive investment reduces current free cash flow in exchange for the possibility of higher future cash flow. High-investment companies are therefore longer duration, while low-investment companies receive more of their economic value from existing assets and nearer cash flows. The investment factor is correlated with growth.

Momentum has no permanent duration exposure. It owns whatever has recently performed well. During a speculative technology rally, momentum may become heavily long duration. During a rally in banks, energy or other cyclical value sectors, it may become much shorter duration. Its rate sensitivity is inherited from the current winners rather than being a stable feature of the factor.

Small-cap rate sensitivity is primarily a credit and refinancing story rather than a pure duration story. Smaller companies often depend more heavily on floating-rate debt, bank financing and frequent refinancing. Higher rates therefore reduce their actual earnings and cash flows. This is a numerator effect, not merely a higher discount rate being applied to unchanged future cash flows.

Equity Term-Structure

The persistent and well-documented premium of the value, quality, and profitability factors over the growth and investment factors suggest that the equity term structure is persistently downward sloping. Short-maturity dividend claims and short-duration stocks have earn higher returns than distant cash flows and long-duration stocks. This is the opposite of the standard bond intuition that investors should receive a larger return for waiting longer.

One explanation is that investors systematically pay too much for distant cash flows. Faraway earnings allow greater freedom to assume enormous markets, persistent competitive advantages and decades of compounding. Because those assumptions cannot be tested immediately, narratives can support valuations for long periods. Near-term cash flows leave less room for imagination. They are visible, measurable and constrained by present operating reality. Investors may therefore overpay for distant growth and require higher expected returns to own less exciting, less sexy, nearer cash flows.

This unifies several persistent factor premiums. Value, profitability, low investment and high payout all tend to select companies whose valuations are supported by current or near-term cash flows. They may not represent completely separate anomalies, but rather be different expressions of the same tendency to overvalue long-duration equities.

Sector Durations

We use publicly available table of empirical rate betas, from D.E. Shaw, estimated on daily data from 2010 to 2021 by regressing sector returns on ten-year Treasury returns and the equity market, so the rate beta is orthogonal to equity beta. A positive Treasury beta means positive equity duration: the sector falls when rates rise.

Technology has by a wide margin the longest cash-flow duration in the index: back-loaded earnings, minimal current payout, value concentrated in a distant terminal block. On the cash-flow-timing prediction it should top this table. It ranks fourth, barely above health care, at around a quarter of Utilities’ beta.

Technology cash flows are pro-cyclical, and so are interest rates. A rate rise carries two pieces of news at once: a higher discount rate, which is bad for long-dated cash flows, and stronger growth, which is good for cyclical ones. For technology the second partially cancels the first.

The tell is visible in the data: technology’s Treasury beta only spiked from the onset of COVID, when rate moves became purely about policy rather than about the cycle. Corroborating this, cross-sectional work finds that firms in cyclical sectors (energy, materials, industrials) exhibit higher measured rate sensitivities than non-cyclical sectors (staples, health care), which is the inverse of the pure cash-flow prediction.

Another possible explanation is the substitution effect: yield-starved investors move between bonds and high-payout stocks (the chase for yield). This effect has opposite implications for cash-flow duration. The mechanism is flows: when rates fall, money moves into income-oriented equity funds, whose managers overweight high-payout names, and the trade unwinds when rates normalize. Investors following a “live off the income” rule reaching for yield in exactly this way. The duration is therefore manufactured by the marginal buyer, not by the cash flows.

Banks & Financials

Banks and financials deserve their own treatment. Banks and similar entities borrow short and lend long (maturity transformation). Aggregate bank assets have an estimated duration of about a bit less than 4 years against less than half a year for liabilities, a gap of 3.5 years. At roughly 10x leverage on the balance sheet (10% reserves), a 100 basis point rate surprise should destroy about 34% of bank equity.

However, measured around FOMC meetings, the realized sensitivity is about 4.2 percent per 100 basis points (instead of 34%), against 3.7 percent for the market itself. Banks sit close to the median of sectors, and are no more rate-exposed than non-financial firms generally.

The resolution is within the business model. Banks pay high fixed operating costs (branches and salaries and marketing and technology) to buy market power over retail depositors, which lets them keep deposit rates low and insensitive. Economically, the deposit franchise is an interest rate swap: the bank pays the fixed leg (operating costs) and receives the floating leg (the deposit spread). It therefore has negative duration and rises in value when rates rise. Banks hold long-duration fixed-rate assets not despite this but because of it. Maturity transformation is a hedge, not a risk.

The deposit franchise is not a contract but a behavioral asset that exists only as long as depositors stay. Deposit betas averaged about 0.06 before the hiking cycle and rose above 0.81 from mid-2022 through 2023. The rate sensitivity of bank equity reversed, from +0.31 before the increase to −8.29 over mid-2022 through 2023. Marking bank assets to market for the 2022 to 2023 rate rise leaves the US banking system roughly $2.2 trillion below book, an average asset decline near 10 percent. When the depositors leave, the hedge evaporates.

Insurers are the mirror image. US life insurers' duration gap turned negative after the financial crisis, with liability duration now exceeding asset duration, so a parallel upward shift in the curve raises the value of their equity. Regress life-insurer stock returns on Treasury bond returns alongside the market, and the bond-return coefficient sits near zero before 2008 and near minus one after 2012, which at 2015 yield levels implied roughly an 8.8 percent equity decline per percentage-point fall in the ten-year (hence why insurer stocks did well during the 2022-2023 rate-hike cycle).

Closing Notes

Duration is a real property of an equity, correctly derived, and worth measuring. The mistakes come from importing bond intuition wholesale. Duration explains the cross-section of cash-flow timing. It also explains the time series of factor returns. It works cleanly where cash flows are genuinely bond-like, in utilities, REITs and staples.

Nonetheless, “equity duration” is a ceteris paribus statement being applied to a world in which ceteris is never paribus. That is not a reason to discard it. It is a reason to hold it the way you would hold any partial derivative: exactly correct about one channel, and silent about the others that usually dominate.

Disclaimer

The information provided on TheLogbook (the “Substack”) is strictly for informational and educational purposes only and should not be considered as investment or financial advice. The author is not a licensed financial advisor or tax professional and is not offering any professional services through this Substack. Investing in financial markets involves substantial risk, including possible loss of principal. Past performance is not indicative of future results. The author makes no representations or warranties about the completeness, accuracy, reliability, suitability, or availability of the information provided.

This Substack may contain links to external websites not affiliated with the author, and the accuracy of information on these sites is not guaranteed. Nothing contained in this Substack constitutes a solicitation, recommendation, endorsement, or offer to buy or sell any securities or other financial instruments. Always seek the advice of a qualified financial advisor before making any investment decisions.

Noah is the founder and co-CEO of Venari Asset Management, a multi-strategy hedge fund focused on public markets whose flagship fund has run above-market returns net of fees since inception. Before that, he co-founded Sage Tech Capital, an equities and options fund, and built TradeChoice.io, where he created day-trading indicators for retail investors and grew a community of more than 1,500 members.

This conversation is about how someone builds and runs an actual fund this early, what the strategies look like, from the AI memory trade and Bitcoin miners turned data centers, and what he has learned managing real outside capital along the way.

What’s On The Book

Venari runs long/short equities as the core, which at the time of recording meant AI infrastructure: stocks exposed to the data-center buildout on Earth and now in space with the SpaceX IPO.

The big theme last two year has been memory: LLMs need it, and the Microns, Samsungs, and SKs of the world have enormous amounts of money. However, one ought to not be overly concentrated, so the question becomes how to find other businesses that support the same thesis without the same idiosyncratic risk.

One answer Noah likes: Bitcoin mining companies transitioning into the data-center businesses. The energy problems and the GPU infrastructure a Bitcoin miner has to solve scale pretty proportionately to what data centers need, since mining is constant computation. Hut 8, a previous position, has done very well on exactly this pivot, since miners eventually realize that crypto is cyclical. The general theme Noah keeps coming back to: companies that are foundational experts in one field with the ability to take that expertise somewhere new.

It is not just memory and it is not just data centers. There is a huge energy issue, and it is not only the amount of energy needed but the efficiency of it. Nuclear is the hot-button hope, but it might be five, eight, or ten years out; nobody knows, which makes that bet a speculative asset.

So what is actually energy efficient right now? Noah argues for exposure in solar (Elon Musk has argued that if you build space data centers, solar panels plus batteries beaming data back down is the way to do it), hydro (the Hoover Dam already powers a city of millions), as well as domestic oil and gas.

Mom, HCA, and the First Trade

Noah’s mother has been in finance her whole career: Wall Street, then consulting, then CFO and C-suite roles across multiple health systems. The first trade is a story worth retelling. At the start of COVID, he was watching the market crash and asking his mom whether the family was going to be okay.

Her response was a framework, not a reassurance: what is the foundation of COVID? Healthcare. What is the largest health system in America? HCA. They bought it around $80 a share and rode it to the $400s over the next couple of years. The lesson that stuck: ask whether a company is a foundational business to society with a good leadership team. If the answer is yes, they will probably figure it out.

Boring Is the New Foundational

What still counts as foundational in an AI-driven economy? SaaS revenue was once treated as highly durable, with companies borrowing directly against ARR, but AI has made even long-term software contracts look less certain. Noah’s answer was that the next foundational businesses may be the ones that look the least exciting.

On the risk-on side, AI could be most transformative for old-line value companies that historically struggled to scale because of high operating costs, labor intensity, and thin margins. These businesses were considered “value” for a reason. Value itself is often a risk factor, reflecting structural inefficiency or limited growth.

But if AI can automate administrative work, customer service, scheduling, compliance, underwriting, and other labor-heavy functions, relatively small cost reductions can create enormous operating leverage. A boring business with stable demand and compressed margins can suddenly become a much better business when technology removes the cost structure that was holding it back.

On the defensive side, there are industries whose underlying demand is unlikely to disappear regardless of what happens in software. Compliance, healthcare, utilities, plumbing, housing, and physical infrastructure will continue to be needed. AI may improve these businesses, but it is less likely to eliminate them. Until robots can repair pipes, maintain buildings, install electrical systems, and care for patients at scale, the physical and unglamorous parts of the economy remain foundational.

The literal plumbing and electrical infrastructure supporting data centers represents a major investment theme for Noah. Eaton, or ETN, fits the electrical infrastructure side of the thesis. The same logic extends to real estate and private credit. Some of the strongest lenders, including groups such as Peachtree Group in Atlanta, have focused on lending against foundations rather than projections. Multifamily housing in the middle of a housing shortage fits that framework.

Rates, Oil, and the Macro Outlook

The macro backdrop was much less comfortable. Recorded the day before an FOMC meeting, the conversation came after a very hot CPI print, with oil emerging as the central inflation risk following the closure of the Strait of Hormuz.

At $120 a barrel, oil does not remain isolated to the energy sector. It flows through transportation, manufacturing, packaging, agriculture, construction, and nearly every physical supply chain. When a major input becomes dramatically more expensive, even temporarily, the cost does not simply disappear when prices normalize. It compounds through contracts, inventories, wages, and refinancing decisions.

That creates a serious problem for highly leveraged companies. Private credit lenders are beginning to recognize that borrowers cannot indefinitely absorb higher rates while lenders continue expecting the same returns and low default rates. Noah’s base case was that rates may need to remain flat. Cuts are difficult while conflict with Iran keeps energy prices elevated, and another hike cannot be ruled out.

The investment conclusion is not necessarily to abandon the market. It is to stay long the structural beneficiaries, particularly the boring businesses that either gain operating leverage from AI or provide services the economy cannot function without, while hedging the broader macro risk.

SpaceX, Anthropic, & OpenAI

Plenty of people argued that SpaceX was overvalued at $1.75 trillion, and it went up from there (at the time of recording it hit $3 trillion, before coming down). This naturally leads to dot-com comparisons, which Noah pushes back on. Part of why that bubble was such a disaster back in the 2000s was that those companies either had no revenue or were fraudulent in revenue (e.g. Enron).

Additionally, one may wonder if SpaceX at $3 trillion is a fair comp, what should Anthropic and OpenAI be worth? Both trade around $1 trillion in private markets, which arguably looks low: if SpaceX at 1.5 made Anthropic at 1 fair, then SpaceX at 3 implies something like 1.5 or higher.

Noah’s answer is about talent and transformation rather than multiples. These companies have changed the way we live, and they have allowed the average person to build a business worth millions. The market does not know how to price historic companies, the same way nobody knew how to price a train company when crossing the country suddenly stopped taking months. Because they are historic, they will probably be priced high, and they probably deserve to be.

AI Value Capture

Nonetheless, there is question regarding value capture. A company can create value X and capture only Y percent of it. A decentralized blockchain creates enormous value and captures almost none of it in fees; a bank pays pennies on your checking account and captures nearly all of it.

For the frontier labs and hyperscalers, the bear case runs through China: a distilled model that is 90% as good running the majority of inference tasks, backed by an energy infrastructure that is genuinely excellent and arguably over-invested, while the US is power constrained after years of underinvestment.

As an addendum, the thesis has begun to partially play out in the interval between recording and publication. U.S. semiconductors, neo-clouds, optics, photonics, and connectivity stocks have drawn down roughly 20% to 40% from highs as Moonshot AI, the Chinese frontier lab behind Kimi, released a model that reportedly outperformed Anthropic’s flagship model Fable across nearly all the benchmarks.

The market is reflecting growing concern that increasingly capable and efficient Chinese models could weaken the value-capture assumptions embedded in American AI valuations. Only time will tell whether Jevons paradox proves powerful enough, that falling inference costs cause global token consumption to expand so dramatically that today’s lofty valuations are ultimately justified, or whether AI becomes so strategically important that the required capital expenditure is partially underwritten, explicitly or implicitly, by the American taxpayer.

Either outcome could sustain enormous investment in AI infrastructure, but neither guarantees that today’s frontier labs and hyperscalers will capture returns proportionate to the value they create.

Professional Punters is presented by Freeport Markets: Freeport lets users trade 24/7, weekends included, with no KYC and up to 200x leverage across stocks, indices, commodities, crypto, rates, and more, with access to pre-IPO exposure in companies like OpenAI & Anthropic. For traders who already follow sharp sources across Twitter, Substack, hedge fund filings, corporate insider activity, and political trading disclosures, Freeport uses AI to read the sources they trust, connect the dots across markets, and surface trade ideas as narratives develop.

About the Host: Lihong Wang is the founder and CEO of Freeport Markets. Before starting Freeport, he traded discretionary semiconductor names at quantitative market-making firms including Jane Street and IMC Trading, building both systematic and discretionary strategies. Lihong graduated from Duke with a degree in mathematics and statistics, and has raised over $2.5M to build Freeport from investors including Y Combinator, Alliance DAO, and Informed Ventures. For more of his long-form research, visit freeportlogbook.substack.com.

Disclaimer: The information provided on TheLogbook (the “Substack”) is strictly for informational and educational purposes only and should not be considered as investment or financial advice. The author is not a licensed financial advisor or tax professional and is not offering any professional services through this Substack. Investing in financial markets involves substantial risk, including possible loss of principal. Past performance is not indicative of future results. The author makes no representations or warranties about the completeness, accuracy, reliability, suitability, or availability of the information provided._

This Substack may contain links to external websites not affiliated with the author, and the accuracy of information on these sites is not guaranteed. Nothing contained in this Substack constitutes a solicitation, recommendation, endorsement, or offer to buy or sell any securities or other financial instruments. Always seek the advice of a qualified financial advisor before making any investment decisions.

Tim studied computer science and mathematics at Stanford. He’s had experience at Weiss Asset Management, Jane Street, and Five Rings, moving from a hedge fund hunting statistical edge to two of the sharpest market-making shops in the business.

A little over a year ago he left trading entirely to become a founding machine-learning engineer at Coolant, a climate startup that raised their seed round from General Catalyst and Floodgate to build spatial intelligence and 3D vision; flying drones over forests and reconstructing every tree at centimeter resolution.

This conversation is about what separates a hedge fund from a market maker, where edge really lives in relatively efficient financial markets, and how that trader way of thinking informs one on how to run their own life.

Hedge Funds vs Trading Firms

Weiss, Jane Street, and Five Rings aren’t three versions of the same job. What separates them isn’t really “hedge fund versus market maker” but whose money they run, and how much risk each one is built to carry.

Weiss is a single-manager hedge fund, which means it runs outside capital. A fund like that lives inside a mandate( the strategy it raised money to run) and if it’s branded as, say, a high-Sharpe stat-arb shop (Sharpe being the ratio of return to the volatility you take on to get it; stat-arb being systematic, statistically-driven trading), it’s broadly held to that.

You can innovate within the mandate, and good desks do it constantly, but you can’t wander off it: LP money comes with real regulatory and investor constraints, so it’s innovation on a leash, not the wild west. When Tim interned there it had a single quant desk and it did everything statistically, because that’s what the mandate was.

Jane Street and Five Rings are the other kind of firm: proprietary trading shops, running their own pools of capital. That allows for freedom because there’s no outside LP to answer to, no mandate to justify, no particular return profile you need have; isntead, you make money for the firm, however the firm has decided is smart.

Both are nominally market makers (they quote a two-sided price and collect the spread) and both operate under market-maker privileges from exchanges (reduced fees, extended locate requirements for shorts, etc…).

In practice, Jane Street is far more willing than Five Rings to lean into a directional view and warehouse the variance that comes with it; Five Rings sits closer to the traditional market-maker archetype with less appetite for swings.

The Dumbest Idea in Quant

Tim previously said that he’s heard some dumb ideas at his previous employers, so I asked him for the dumbest idea he’d heard at these firms, half-expecting a story about a blown-up trade. His answer was better, because it’s a mistake a lot of shops are still making: treating AI as though it has no place in markets.

The objection used to sound sophisticated. Signal-to-noise in markets is too low; the distribution shifts too fast to isolate anything; there are too many players; it’s too game-theoretic; etc... Tim (and I) heard every version of this early in his quant career, and every version has aged badly.

Look at the scoreboard: HRT’s profit-per-head now rivals or exceeds Jane Street’s, and Jane Street is out building data centers, which tells you exactly what kind of method they’ve decided to bet on, because you do not need a data center to run classical inferential statistics. The asymmetry is stark. The cost of exploring deep learning is low: hire a handful of cracked PhDs, pay them a few dozen millions a year, hand them compute, let them dig. The cost of not exploring it is that you eventually get quietly competed away by the shops that did.

In high competitive industries, the risk of underinvestment is often greater than the risks of over-investment; an observation that also applies to frontier generative AI.

The Use of Modeling Volume & Volatility

For those outside of trading, modeling volume does not appear immediately useful since it can’t be directly monetized. However, for a market maker, volume informs how much slippage a trade will cause, how much inventory risk you should be willing to warehouse, how deep to quote and where. A retail trader can safely ignore all of it; a market maker lives on it. The lesson generalizes past volume, very often the edge is in modeling the regularities the market treats as boring and secondary.

Another regularity is the pattern of volatility realized. The observation there is that realized volatility doesn’t arrive linearly through the day. It clusters at the open and again into the close, and it’s asset- and clock-specific: a stock with heavy Chinese ownership bleeds more of its volatility during Chinese trading hours than during the US session, so you decay its vol differently depending on the clock (and set your price differently at the US opening auction). The same logic extends to stocks which may have other regional or commodity exposures.

Why Jane Street

Of the three firms, Tim’s clear favorite on culture is Jane Street. Tim’s observation is that these firms are extraordinarily good at attracting talent and, increasingly, less good at retaining it, because the competitor for their best people is no longer another prop shop, but the frontier AI labs.

What sets Jane apart is that the people at the top genuinely think about the employee experience, and that this is harder than it sounds, because the higher you sit the worse your information gets. Ask your own people how they’re doing and they’ll tell you it’s incredible; the honest signal never reaches you.

Most firms under-allocate to fixing that precisely because they can’t see the problem. Jane, Tim thinks, corrects for it better than most.

Look Past The Comp

On the topic of trading and Jane Street, it is impossible to avoid noting that pay numbers are enormous ($850K Total Comp for New Grads); however, it is noted that you should weight this number very little.

In the grand scheme of life they aren’t actually that big, and optimizing the starting figure is optimizing the one thing everyone can see: the price, not the value. Above all, price what different opportunities teaches you and which doors it opens, because that’s the thing compounding underneath the salary.

By his accounting there are exactly two good reasons to go into quant: you want to make money for five years and coast (high value placed on not working), or you genuinely love trading (high value placed on trading). Absent those, Tim notes that “if I could go back, I wouldn’t have,” because of how much is happening everywhere else right now, especially in AI.

Variance Adds in Square Roots

Tim only half-jokes when he calls this one of the most important rules of life: Expected value adds linearly but variance adds in square roots. Over a long enough horizon, twenty, thirty years, the square-root term shrinks relative to the linear one, which is the formal version of the folk wisdom that people usually end up roughly where they should over the long run.

Two consequences fall out of that. First, you should take many independent bets and stack them (i.e. try now things, take risks when yu are young), because independence is what makes the variance term behave. Second, and less obvious, is that your trading portfolio should be wired to the risk in your life.

If you’re already the kind of person taking startup risk with your career, and you’re simultaneously holding all your retirement in the S&P (with no leverage), there’s a contradiction sitting in plain sight: levering up, especially since those bets aren’t actually that correlated for most people. The covariance is real but small next to the enormous idiosyncratic variance of any individual life.

The Case for Leverage

His view on optimal leverage is simple: take the thing with high enough expected value and a long enough horizon (the S&P, a global stock basket) and lever it.

And the tooling to do this responsibly is now trivial. You can one-shot a Monte Carlo simulation in Claude Code that shows you the full distribution of outcomes at each leverage level, so you can stare at every percentile and decide how you actually feel about the bad ones before you take the position.

The compounding math is startling. QQQ is up roughly 16x since 2010 (from ~44 to ~720). TQQQ, the 3x-levered version, is up not 3x that but around 350x over the same window, from about 0.22 to 77, because leverage compounds.

The most under appreciated form of leverage, though, is on your future income. When you’re young, your current net worth is a rounding error against your expected lifetime net worth, so 3x leverage on 1% of your wealth is still only 3% of what you’ll eventually be worth, which is not a crazy decision at all.

The right posture, Tim argues, is to live as though you hold an infinite call option on yourself: a standing bet that you can always get back to a high-earning seat if you need to. And once you internalize that, it changes not just how you invest but how you think about money, because you stop being afraid of the downside that was never actually going to end you.

Trading to Trees

A little over a year ago Tim joined Coolant, then a seed-stage startup, because his heart had always been in climate and in the geospatial corner of computer vision.

Coolant scan trees. A drone carrying ordinary RGB imagery flies a forest, and the pipeline reconstructs it in 3D (i.e. a digital twin of the natural world) from which vision models read off the metrics clients pay for.

The timber use case makes the value obvious. If you run a timber operation you manage enormous tracts of land and know shockingly little about them, because the incumbent method is to send a person out with a tape measure. It’s slow, it’s expensive, and it samples a few acres of a vast holding.

This is a small-area estimation problem, and the square-root law again: your sampling error is brutal when the thing you’re sampling has huge local variation. Disease is localized. Species competition is localized. You specifically want to catch the localized thing, and a sparse tape-measure sample is blind to it.

It’s also just inaccurate; you can’t easily measure the height of a tree with a tape, a laser pointer, and some trigonometry. Coolant’s reconstructions are centimeter-resolution: thickness and height to within an inch, every tree, from a single drone covering hundreds of acres in the time it takes a person to measure a handful.

The use cases fan out well past timber: carbon-market verification (proving the trees you planted for credits actually exist and are growing), wildfire management, coastline monitoring. And underneath all of it: a large and growing proprietary dataset of the physical world, which in this era is the moat.

Lifetime Net-Worth Swaps

The most fun idea in the conversation is one Tim actually trades with friends, mostly his more risk-seeking ones. It’s a lifetime PNL swap. You pick an expiry (usually something like each other’s 40th birthday) and agree to exchange five basis points of the difference in your net worths at that date.

The logic is pure variance reduction. In the founder-heavy corner of SF Tim lives in, everyone has a high expected net worth and a much lower median one (a lottery-ticket distribution, empirically) and a swap like this is a clearly positive-utility trade whenever it moves money from the friend who hit to the friend who didn’t.

It’s a bet on your friends making it, priced better than their cap tables; blindly buying YC companies has historically returned something like a 20% IRR, but do you actually want to buy your friend’s startup at a $40M valuation? The swap sidesteps the entry price entirely and aligns incentives better (a founder can get acqui-hired into a $100M package while the company nominally dies and the equity holders get nothing).

It works best early in life, when there’s the least information about who will win and pricing is therefore most uniform; by forty the risks have largely realized and there’s little left to hedge, which is why you’d never write the contract then. And it’s especially good for lower-risk people, because the floor is zero; the worst case is that your friend goes broke and you give up five basis points of your net worth.

Tim’s half-serious coda: once on-chain identity (a World Coin-style proof of personhood) makes these enforceable, a swap like this could become significantly more popular, as well as replace a good chunk of what insurance companies do.

Professional Punters is presented by Freeport Markets: Freeport lets users trade 24/7, weekends included, with no KYC and up to 200x leverage across stocks, indices, commodities, crypto, rates, and more, with access to pre-IPO exposure in companies like OpenAI & Anthropic. For traders who already follow sharp sources across Twitter, Substack, hedge fund filings, corporate insider activity, and political trading disclosures, Freeport uses AI to read the sources they trust, connect the dots across markets, and surface trade ideas as narratives develop.

About the Host: Lihong Wang is the founder and CEO of Freeport Markets. Before starting Freeport, he traded discretionary semiconductor names at quantitative market-making firms including Jane Street and IMC Trading, building both systematic and discretionary strategies. Lihong graduated from Duke with a degree in mathematics and statistics, and has raised over $2.5M to build Freeport from investors including Y Combinator, Alliance DAO, and Informed Ventures. For more of his long-form research, visit freeportlogbook.substack.com.

Disclaimer: The information provided on TheLogbook (the “Substack”) is strictly for informational and educational purposes only and should not be considered as investment or financial advice. The author is not a licensed financial advisor or tax professional and is not offering any professional services through this Substack. Investing in financial markets involves substantial risk, including possible loss of principal. Past performance is not indicative of future results. The author makes no representations or warranties about the completeness, accuracy, reliability, suitability, or availability of the information provided._

This Substack may contain links to external websites not affiliated with the author, and the accuracy of information on these sites is not guaranteed. Nothing contained in this Substack constitutes a solicitation, recommendation, endorsement, or offer to buy or sell any securities or other financial instruments. Always seek the advice of a qualified financial advisor before making any investment decisions.

Jude Rizzo studied math, physics, and machine learning at Cornell and is a Stuyvesant alum. He co-founded Carousel Technologies (YC W24), which used AI to build financial models, and was acquired by the market-intelligence platform AlphaSense in October 2025. Before Carousel, he had a brief experience as a quant at Jump Trading.

Sam Rosenstrauch earned his BS in Math from Carnegie Mellon, where he placed top 200 nationally on the Putnam, and qualified for the USA Math Olympiad in high school. He traded options at Susquehanna (SIG) and worked as a Quantitative Research Analyst at Goldman Sachs. He’s now an engineer at Espresso AI, a frontier neo-lab building ML-driven compute optimization.

This conversation is about where edge actually comes from in sports betting markets: the NFL-drive latency trade that turned a small stake into 20x in a couple of months; the half-second window they exploited on hockey goals; the cat-and-mouse of staying under a book’s limits; why Pinnacle welcomes smart money when everyone else bans it; books you bet against versus exchanges like Kalshi and Polymarket; and how life eventually led them out of trading.

The Drives: 20x on a Half-Season

Jude is a lifelong Jets fan, which mostly means he has spent years watching his team punt. During one of those punts he pulled up the sportsbook odds out of curiosity: what’s the real probability the Jets turn this drive into a touchdown? (Low, always.) But looking across books with Sam, something else stood out: during a live game, the price of a given outcome didn’t line up across DraftKings, FanDuel, and the rest.

The mechanic was latency. On live betting, one book moved first, usually DraftKings, then FanDuel, then everyone else followed, and PointsBet in particular would sometimes just not move in time. So you could bet on something that had effectively already happened and still get stale odds.

Consider an example: a drive opens with the punt at −200, the field goal at +300, the touchdown at +200. As the drive develops you watch DraftKings reprice; the touchdown drifts from +200 to +100, which after the vig (the cut the book bakes in to tilt every line slightly in its favor) implies roughly 45%. Meanwhile PointsBet still has the touchdown at +200, about 33% implied. You’re buying something worth ~45% (if you believe that the line moved for good reasons) for ~33%.

Two things made it more than a one-off. First, drives happen many times a game and there are many games a week, so the edge was testable and repeatable enough to size into without betting the account on a single outcome and losing to variance even when you’re right. Second, a drive outcome is a side market. The book concentrates its attention on the money line (a straight bet on who wins) and the point spread (a bet on the margin of victory), where the volume is. The side markets are softer, so the usual assumption that “the book is right” is weakest exactly where you’re trading.

They sized by feel using the Kelly criterion, the formula for the size that maximized your rate of return over repeated bets. (Bet everything and you eventually bust even with an edge; bet too little and you leave money on the table; Kelly is the mathematical optimal). In practice they couldn’t be precise due to the time-sensitive nature of the opportunity, so they kept a small table of common American-odds cases and guesstimated live.

It compounded fast. At ~20% edge pre-vig, closer to 10% post-vig, and pressing hardest whenever the value gap was widest, the stake 20x’d in a couple of months: “basically one half of a season.” Then PointsBet caught on. Which, as it turned out, was the pattern for everything that followed.

Building OddsJam for Themselves

The drives play was one instance of the real engine. Their bread and butter was a giant sheet of every book’s odds across every market: “we basically built OddsJam for ourselves [before OddsJam]” scanning for arbitrage and, more usefully, near-arbitrage. A true arb is when two books disagree enough that hedging both sides locks in a profit; a near-arb is the more common and more interesting signal, because the very existence of the gap makes it more likely that at least one side carries edge.

The hard part is deciding which side is wrong. They started with consensus as a prior, if every book has a team at 40% and one suddenly prints 50%, the crowd is probably closer to the truth than the outlier.

Over time they learned book-specific tendencies: for a stretch you’d find softer underdog odds on FanDuel; BetMGM was consistently off from everyone else on basketball quarter point-spreads. Different books, different algorithms, different blind spots shown to the user to exploit.

Size Limits From Sportsbooks

The first thing to understand about beating a sportsbook is that the book is built to win, and it has defenses. It won’t usually ban a winner outright, but it will limit you. The algo flag your account and suddenly you can only put $8 on a given line. There’s a livable zone in between, smart enough to win but small enough not to trip the wire, and the pair lived there for a while before getting throttled.

The exception is Pinnacle, the sharp offshore book that does the opposite of everyone else: it takes smart money, even publishes an API for its odds, and uses that flow as signal to keep its spreads tighter than anyone’s. As Sam put it, it’s a skill issue.

Books vs. Exchanges

Underneath all of this is a structural split. One kind of book: DraftKings, FanDuel, the US incumbents that were basically the only legal option before Kalshi, has you betting directly against the house. The book doesn’t want you to win; it’s incentivized against you, and it makes its money on the vig. The other kind is an exchange; now Kalshi and Polymarket, where you’re betting against another person and the venue genuinely doesn’t care who wins.

The against-the-house model appears to mostly a regulatory artifact. Everywhere else in finance the two jobs are split: exchanges aggregate liquidity (and the liquidity is the moat; everyone wants to trade the S&P where everyone else trades it), while front ends like Robinhood, Schwab, Interactive Brokers and E-Trade compete for users on top.

User acquisition and liquidity provision are different businesses, run by different companies, for a reason. Sportsbooks fuse them only because the law made them, and now that exchanges like Kalshi and Polymarket exist, they’re pulling flow off the books (and rightfully so, as they are more open and transparent venues).

Hence the irony of DraftKings running ads about how Kalshi isn’t a “real” financial contract, and a founder complaining on Twitter that sizing up on Kalshi walks him through the order book, coming from a business that limits anyone who wins. On an exchange both sides of a trade are happy at the moment it prints; a book that throttles winners while lecturing the exchanges is hard to take seriously. It also set up the next trade; because Kalshi, unlike the books, doesn’t lock its market during a game.

The Half-Second Edge: Hockey Goals on Kalshi

That Kalshi keeps its book live during a game was the whole opening. Some traders leave resting orders up while a major event happens on the ice, because they haven’t seen it on the broadcast yet. That gap is the trade.

To find out how big the gap was, they ran an experiment instead of guessing. They pulled up a Rangers game, coded a scraper that read Kalshi’s WebSocket messages to watch the order book tick by tick, and grabbed a stopwatch app. The moment they could visually react to a goal, they hit the button to timestamp their own reaction; afterward, they lined up their click times against when the market actually moved. The answer: about half a second of wiggle room.

Hockey wasn’t a random choice; it’s close to the optimal sport for this trade. The edge is a latency advantage, and a latency advantage is worth the most when the event it front-runs swings the odds the most and arrives unambiguously. A hockey goal does both. There’s essentially only one event that moves the line (a goal), it moves it a lot, and there aren’t so many goals that any single one is diluted.

Basketball scores constantly, and the game is long enough that no single basket moves much; baseball events (a double, a triple) are harder to interpret in real time. Hocky is nice because there’s no skill in judging whether the puck crossed the line, unlike a jump shot still in the air. That matters because Kalshi’s taker fees (what you pay to cross the spread and hit a resting order) top out around 7% on a sliding scale, so you need a lot of clean edge to clear them.

The ceiling is set by depth, not skill. Pregame, the book is millions of dollars deep; the instant the game starts, that depth collapses. You might get $1–2k down on a goal, maybe $5k on a big game if you work it, but there’s no putting a million to work. Jude’s estimate for what the whole hockey program could earn was mere a few hundred thousand dollars a year. Whether it still works he isn’t sure; the arb may barely exist now, though he’d bet it’s there for anyone willing to grind it.

The 50-Second Mirage

Through a friend, Daniel, who sells data infrastructure to trading firms, the host had come to see that some large firms, Jump among them, only had access to the same broadcast as everyone else which is running about 50 seconds behind live (on select games). Graph the point-spread orders around a goal and you can see it: a tight cluster of small orders over roughly five seconds (people reacting live) then the same cluster again 50 seconds later, bigger, from people on delayed feeds who don’t know they’re behind. Sam confirmed it matched what they saw on Kalshi exactly.

The microstructure that falls out of this:

1. The first ~5-second cluster is live reactors

2. The second cluster, ~50 seconds later, is delayed-broadcast money trading a goal that’s already old news.

3. The smart money realizes the move is in, it fades the second cluster; the late buyers pay a premium and the level evens back out.

Once everyone knows to buy when the goal lands, the market overcorrects, and the natural move is to sell into the overcorrection. Instead of crossing the spread and eating the taker fee, provide liquidity when your live information is better than the market maker’s. Basketball is the clean case: someone watching live knows whether a late basket was a two or a three, while a market maker on a delayed feed does not.

Why There’s Edge on the Table

The recurring question underneath every one of these trades is why the edge survives at all; why the biggest, best-capitalized firms don’t simply take it? Mostly, they can’t be bothered. Imagine you’re Jump: do you care about $1,500 on a hockey game when your trading bot makes a million dollars a day? The compute tokens and trader salaries (upwards of millions a year) aren’t worth it.

Comparison to Options Trading

Scale the same thinking about benign flow and dislocations that it causes, and you arrive at the options market, where the players are larger but the game is identical.

The same instinct, work out exactly who’s on the other side and why, is what professional desks run on, and the guests were candid that the trades there are often less sophisticated than outsiders imagine.

Start with what you can see. After you trade an option, depending on which of the roughly thirteen OCC options exchanges you’re on, you can often see your counterparty, and if it’s a known market maker (Jane Street, Citadel Securities, SIG, Optiver, IMC), you get out of the way, because you assume they know something.

The simplest money came from watching a maker telegraph an institutional order: buy a level up in small clips, then a giant block prints where they sold that high level to a pension fund; or sell it down in twos and fives, then a block prints where they bought from an institution. (To the CFTC this isn’t manipulation; the maker is merely pre-positioning the impact the order would have had.)

Then there’s forced flow, the cleanest edge of all, because it’s on a calendar. Covered-call ETFs, funds that systematically sell call options against their holdings, have a mandate to sell a fixed number every week; come Friday, if a fund has only sold ~60% of its quota, market makers push implied vol down to buy those calls cheaply, knowing the seller has to show up.

For funds that sell calls in enough size it gets blunter: instead of moving vol you can move spot. And this isn’t a small-cap curiosity; it happens on the S&P 500 and ES futures (the S&P’s flagship futures contract). A large block of calls hits in the morning, the S&P ticks down half a percent, certain makers absorb them, and it ticks right back up. Index rebalances are the same idea; enormous, scheduled, entirely predictable flow, and, as the host noted, the biggest provider of that trade may not be the market makers at all but a particular team at a multi-manager (a fund running many independent trading pods under one roof).

The same predictability, the host added, is why companies stay private longer: index funds are forced to buy at IPO, so the move is to grow as large as possible in private markets and then list at a rich valuation the index has no choice but to fund.

From Trading to Startups

Both of guests eventually stopped trading and started building. The skills carried over more directly than you’d expect.

Sam describes his work at Espresso AI: optimizing customers’ compute, as feeling like trading a live game. You’re forced to make fast routing decisions, and you can make them decently good by layering ML on top of interpretable heuristics. The edges even feel like trading. Snowflake bills on a 60-second minimum, so a warehouse switched on for one second and idle for the next 59 still bills the full minute, and batching compute tasks allows cloud bills to therefore be lowered by an order of magnitude. This sort of attention to detail and the fine print is exactly the kind of structural quirk they used to hunt on a sportsbook; now hunted on a cloud bill.

Jude’s company, Carousel, puts AI into Excel. A banker-quality spreadsheet holds far more context than it appears to using formatting that is both load-bearing and ruinous to token budgets, so reading every cell naively doesn’t work. They ended up reading “bricks” of a file (connected units of meaning, not just formulas) and building a dependency graph the hard way, since you can see what a cell references but not what references it until you’ve walked the whole sheet once. Again, the skills of attention to detail and systematic thinking came into good use.

The One Rule to Take Away

Step back, and every edge in the conversation has the same shape. A puck crosses the line and for half a second the order book hasn’t caught up. A broadcast runs fifty seconds behind and a second wave of money trades a goal that’s already history. A covered-call fund has to sell by Friday, and the price moves before the flow arrives. None of it was about predicting the game or the price better than anyone else: a hockey goal is unambiguous, a rebalance date is on a calendar. The edge was being the one person standing in the gap between a fact and its price. The catch is that the gap always closes over time. PointsBet caught on and eventually quant money poured into Kalshi.

Jude now builds at AlphaSense and Sam designs core algorithms at Espresso AI. They're two of the sharpest people I've had the pleasure of knowing, and also two of the most genuinely enjoyable to talk and work with.

Professional Punters is presented by Freeport Markets: Freeport lets users trade 24/7, weekends included, with no KYC and up to 200x leverage across stocks, indices, commodities, crypto, rates, and more, with access to pre-IPO exposure in companies like OpenAI & Anthropic. For traders who already follow sharp sources across Twitter, Substack, hedge fund filings, corporate insider activity, and political trading disclosures, Freeport uses AI to read the sources they trust, connect the dots across markets, and surface trade ideas as narratives develop.

About the Host: Lihong Wang is the founder and CEO of Freeport Markets. Before starting Freeport, he traded discretionary semiconductor names at quantitative market-making firms including Jane Street and IMC Trading, building both systematic and discretionary strategies. Lihong graduated from Duke with a degree in mathematics and statistics, and has raised over $2.5M to build Freeport from investors including Y Combinator, Alliance DAO, and Informed Ventures. For more of his long-form research, visit freeportlogbook.substack.com.

Disclaimer: The information provided on TheLogbook (the “Substack”) is strictly for informational and educational purposes only and should not be considered as investment or financial advice. The author is not a licensed financial advisor or tax professional and is not offering any professional services through this Substack. Investing in financial markets involves substantial risk, including possible loss of principal. Past performance is not indicative of future results. The author makes no representations or warranties about the completeness, accuracy, reliability, suitability, or availability of the information provided._

This Substack may contain links to external websites not affiliated with the author, and the accuracy of information on these sites is not guaranteed. Nothing contained in this Substack constitutes a solicitation, recommendation, endorsement, or offer to buy or sell any securities or other financial instruments. Always seek the advice of a qualified financial advisor before making any investment decisions.

Evan studied computer science, mathematics, and economics at Vanderbilt before working across options trading, quantitative research, and systematic execution at firms including Belvedere, the Cutler Group, J.P. Morgan, GTS, and a family office.

He eventually left a traditional quant seat to trade prediction markets full time, building his own statistical-modeling and execution stack to price markets across Kalshi and Polymarket. He was recent recruited by a major market marker to run their prediction markets operation.

Evan made low five-digits a day market-making fifteen-minute crypto binaries, with Sharpe ratios in the 40-50 range. He also found major edges in election arbitrage, AI model leaderboard markets, and insider-looking prediction-market flow.

This conversation is about what it actually takes to find edge in prediction markets: the modeling, the execution, the weird rules, the toxic flow, the oracle disputes, the bankroll constraints, the trades that worked, the trades that stopped working, and the losses that teach you what the rules really mean.

2024 Election Markets

Kalshi and Polymarket both ran “who wins” contracts that settle to a dollar, but Polymarket carried a persistent two-to-three-cent premium on Trump; partly a user-base story (Polymarket’s crypto & Trump-leaning base vs. Kalshi’s). Buy the dollar for 97–98¢ across the two venues and you collect ~2% a week with only exchange risk.

The interesting part isn’t the spread, but why it survived. To trade Kalshi you needed a U.S. entity; to trade Polymarket you needed a non-U.S. one. Almost no one could put on both legs cleanly. Susquehanna potentially could of (Dublin office for the Polymarket leg, a U.S. office for Kalshi, settle internally), but most U.S. prop shops couldn’t capitalize. On a Sharpe basis a 2%/week with minimal vol is almost nonsensical, with around 280% annualized returns.

538 Presidential-Approval Markets

Evan’s first real loss came on the 538 presidential-approval markets. The play looked clean: once the week’s number prints, buy the winning bracket up to 99¢ and sell the others. He’d “bonded” 99¢ for something he was sure was worth a dollar (bonding here refers to the practice of buying high probability events and collecting “safe” payments like one would with a bond). The problem was that he assumed 538 published one value per day whereas they revised, sometimes up to three times in a day. He’d loaded up on shares at 99¢ that were worth zero.

Why didn’t the market warn him with pricing that was significantly below 99¢? Because he was the market. His own 99¢ bids were the only thing holding the price there. Someone chipped away at him periodically, but he wrote it off as a participant freeing up capital, which is a perfectly normal thing to see in these venues because the opportunity cost of capital in prediction markets is enormous (if your book returns 200% a year, you’ll gladly pay 1% to free capital two weeks early). So a steady stream of people hitting your 99¢ bid looks like business as usual, not a red flag.

Latency Edge: Scraping the AI Leaderboard

The first systematic winner was the “top AI model” market on Kalshi, priced off LMArena’s Elo leaderboard (the human head-to-head voting site). The structural quirk exploited was that LMArena pushed updated results to Hugging Face before their own front-end refreshed, sometimes a 15-minute head start. Scrape Hugging Face and waiting for a surprise would have been profitable because other market participants didn’t realize this quirk. The biggest was Gemini’s debut topping the board when nobody expected Google to be competitive (their prior model, Bard, was bad). One book-clearing print on a model update was a ~$30k click; the whole edge added up to ~$50–60k.

Mention Markets & Fine Prints

Evan doesn’t trade mention markets much, but he flagged a trader who’s made a lot of money the rules-are-the-product principle. The example given was a Bernie Sanders town hall that wasn’t carried on a qualifying broadcast, so under the literal rules none of the words said could count. If you’d read the resolution criteria, you knew the market was mispriced. Everyone arguing on the Kalshi Discord hadn’t. His view on settlement: even when an outcome feels against the spirit of a market, the exchange has to honor the written rules, which is exactly why reading them is an edge.

Bitcoin 15-Minute Binaries

The recurring 15-minute (and 5-minute) crypto markets have a single strike, wherever the previous window closed, so it’s a pure up-or-down binary over the next window.

While technically options, these are not volatility products. As time-to-expiry goes to zero, Vega goes to zero (the price barely responds to implied vol), but it’s enormously sensitive to the underlying. And for a binary that’s exactly at-the-money, Vega is literally zero. So your only edge at the money is modeling short-horizon returns; it behaves like a Delta-one product, not an options trade.

Then why not just trade spot directly? Because spot is a different sport. Market-making BTC/USDT perps on Binance puts you against Jump Crypto, Tower, Radix, which is an infrastructure arms race you will lose. The binary venues, by contrast, were still tractable with an AWS EC2 box and good Python, because the heavyweights weren’t there and the flow was price-insensitive.

The microstructure also presents some opportunities not present in spot markets”

1. Quote to the very end: In the final minute, many makers widen out or stop quoting entirely. Spreads blow out but Kalshi has price-insensitive takers who will cross a 20–30¢ spread in the last few seconds. That’s where the edge is fattest. But don’t quote so tight that another maker picks you off.

2. Adverse selection is detectable, after the fact. In post-trade analysis you can see it: if you’re marked 10–20ms after a fill, you got picked off. You can use this data to improve your modeling by looking at correlated statistics.

3. Toxicity isn’t linear in size. You’d think bigger fills mean higher conviction (toxic). Past a point this principle inverts; a huge fill is often just a clueless market order punching through levels because someone can’t read a book.

4. Multi-level quoting is a real tradeoff. Resting deep in the book catches the punch-through orders, but the back of the book is also where the most adverse fills live. We discussed an old equities project measuring slippage across dozens of quoted layers, where getting run through all of them was extremely toxic. (Caveat: equities size, hundreds of millions, isn’t five-minute Bitcoin size.)

The results: at peak, $10–12k a day. Per-coin Sharpes in the 30–50 range; aggregated across the ~7 coins he quoted, some days hit 50–90. The kind of return graph where, in his words, there’s “no intraday drawdown, it’s just a straight line up.”

Discretionary Book: Trading Against Identity

Evan’s systematic strategies were most of his volume, but he spent a disproportionate share of time on discretionary trades, because they’re more interesting. He thinks sports and elections are where people are most irrational. People are personally attached to candidates the way fans overrate their home team, so no-hope candidates always trade at some non-zero probability. And as results come in, you get a flood of flow from people buying whoever’s “currently leading” without understanding the county-by-county or mail-in breakdown.

Recent wins came from reading exactly that: Peru (Fujimori was going to take the uncounted expat vote even as her lead narrowed on screen, while Polymarket’s comment section panicked), the LA mayoral race (he just bought “no” on a candidate he saw no credible path for), and Hungary (the market implied a ~25% chance Orbán would refuse to leave / contest the result, which he found absurd, inflated partly by U.S. observers feeling sudden skin in the game after JD Vance’s association).

His noted there are only a handful of genuine election sharps on Polymarket; if they trade against you, worry, but most other flow is benign, and some whale accounts are just spectacularly bad (one down ~$5M YTD that he was happy to fade).

Unfair Resolutions

The big loss is a governance story. A market on whether streamer Clavicular would be announced to have gotten someone pregnant said “jokes don’t count”; the streamer (who’s said he’s sterile) joked about it on stream, and UMA’s resolvers called it serious and resolved YES. Evan was long ~50k “no” shares and lost ~$30–35k.

The deeper point: Polymarket settlement runs through UMA’s “decentralized” oracle, but in practice voting power is hoarded, outcomes often come down to a couple of large holders (UMA Rocks and a big whale account). Polymarket can override but rarely does, sometimes letting UMA take the heat. Oracle governance is a real, priceable risk, not a footnote.

Solo vs. Firm

A trading firm eventually recruited him off his Bitcoin work, through Discord connections, not a cold recruiter email. The biggest difference he noticed wasn’t appetite for trade risk (firms have the bankroll to size up, so a given position is relatively less risky for them), it’s their intolerance for system risk. No “vibe code looks good enough to me” on a quoter; PRs get reviewed by multiple people; compliance scrutinizes everything down to whether Perplexity is allowed on a work machine. Solo, you set your own hours and answer to no HR, some days are 30 minutes, some are till 4 a.m., but you also own all the system risk yourself.

The One Rule to Take Away

His closing advice doubles as the thesis of the whole episode. Don’t treat prediction markets as free money, there are millions of recreational bettors, but the sharps are real. When you see a price far from your fair value, first try to explain why the market disagrees with you. With a no-hope candidate, the answer might be “there’s a fan base that buys him at any price,” fine, fade it. But if you can’t explain the gap, you don’t understand the market well enough to be trading it.

That’s the through-line. The money isn’t in predicting outcomes better than everyone else. It’s in knowing exactly who’s on the other side, why they’re there, and whether the structure of the venue lets that mispricing persist.

Evan writes up more of his trades and thinking on his Substack, if you trade these markets, it’s worth the subscribe. He’s on X at @evan_semet, and his discretionary book is public on Polymarket.

Professional Punters is presented by Freeport Markets: Freeport lets users trade 24/7, weekends included, with no KYC and up to 200x leverage across stocks, indices, commodities, crypto, rates, and more, with access to pre-IPO exposure in companies like OpenAI & Anthropic. For traders who already follow sharp sources across Twitter, Substack, hedge fund filings, corporate insider activity, and political trading disclosures, Freeport uses AI to read the sources they trust, connect the dots across markets, and surface trade ideas as narratives develop.

About the Host: Lihong Wang is the founder and CEO of Freeport Markets. Before starting Freeport, he traded discretionary semiconductor names at quantitative market-making firms including Jane Street and IMC Trading, building both systematic and discretionary strategies across equities. Lihong graduated from Duke with a degree in mathematics and statistics, and has raised over $2.5M to build Freeport from investors including Y Combinator, Alliance DAO, and Informed Ventures. For more of his long-form research, visit freeportlogbook.substack.com.

Disclaimer: The information provided on TheLogbook (the “Substack”) is strictly for informational and educational purposes only and should not be considered as investment or financial advice. The author is not a licensed financial advisor or tax professional and is not offering any professional services through this Substack. Investing in financial markets involves substantial risk, including possible loss of principal. Past performance is not indicative of future results. The author makes no representations or warranties about the completeness, accuracy, reliability, suitability, or availability of the information provided.

This Substack may contain links to external websites not affiliated with the author, and the accuracy of information on these sites is not guaranteed. Nothing contained in this Substack constitutes a solicitation, recommendation, endorsement, or offer to buy or sell any securities or other financial instruments. Always seek the advice of a qualified financial advisor before making any investment decisions.

Introduction

In 1886, a depositor at a New York commercial bank paid the financial system about $0.019 per year for every dollar of intermediated assets, between savings stored, credit extended, equity issued, and payments cleared. The same dollar, routed through the 2012 financial system, with index funds, securitization, repo, high-frequency trading, securitized credit cards, and interest rate swap dealers, cost about $0.018. A century and a quarter of compounding technological progress in one of the most competitive industries in the world has NOT brought down the net cost of intermediation.

Per the author’s own words in the abstract for Has the U.S. Finance Industry Become Less Efficient? the core empirical results show that:

A quantitative investigation of financial intermediation in the U.S. over the past 130 years yields the following results : (i) the finance industry’s share of GDP is high in the 1920s, low in the 1960s, and high again after 1980; (ii) most of these variations can be explained by corresponding changes in the quantity of intermediated assets (equity, household and corporate debt, liquidity); (iii) intermediation has constant returns to scale and an annual cost of 1.5% to 2% of intermediated assets; (iv) secular changes in the characteristics of firms and households are quantitatively important.

Defining The Cost of Financial Intermediation

Imagine the entire financial sector as a black box. Money goes in from savers; money comes out to borrowers and equity issuers. The black box collects a toll for its trouble. That toll is the income of the finance industry: every fee, every spread, every wage, every bonus, every trading profit. From the outside, you don’t need to know whether the toll was collected by a bank net interest margin, an asset-management fee, an IPO underwriting spread, or a swap-desk markup. You just need the total.

In equilibrium, then, the user cost of external finance to a borrower or investor decomposes cleanly:

The borrower’s cost of capital is what the saver demands, plus a wedge the intermediary keeps. The wedge ψ can be estimated as a ratio:

The numerator is finance-industry value added (labor income + capital profits). The denominator is the dollar volume the system services, measured on the balance sheets of end users who are not banks, so that interbank lending doesn’t double-count.

The quantity in the denominator is the sum of the three economically meaningful things the financial system actually produces for non-financial agents:

Add up what the system charges and divide by what it services gets us the price per unit of finance, comparable across centuries. Every other measure such as bid-ask spreads, mutual-fund expense ratios, and lowering brokerage commissions, only captures one product at one moment and misses the rest of the iceberg.

Core Empirical Results

The two relevant long-run series from the BEA are finance value added which goes back to 1880., and intermediated assets which goes back to 1886.

The income series tells the story that laypeople already know: finance is huge again. From the post-war trough of 3% of GDP to a 2007 peak of 8.2%. Doubled relative to the rest of the economy in a single human lifetime. But notice that the red line, intermediated assets, moves in lockstep with the green line. Finance got big because the things finance services got big: corporate equity values, household debt, repo markets. Per dollar serviced, the system charged about the same.

The unit cost refuses to fall over the long run. It rises in the 1920s, falls back to ~1.4% in the 1960s, then climbs again from 1980 onward. The quality-adjusted line (red), which corrects for the fact that the modern system services riskier borrowers (more young firms, more low-income households) per dollar of credit, is flatter still.

In short, the unit cost of intermediation is about as high today as it was at the turn of the 20th century. Improvements in information technologies do not appear to have led to a significant decrease in the unit cost of intermediation. Explaining this puzzle is an active area of research.

Almost every other industry that absorbed comparable IT investment got dramatically cheaper. The chart below puts ψ alongside two industries the BEA tracks where the price decline is so steep you have to plot it in logs.

Explanations for Persistent Intermediation Costs

The literature has accumulated, roughly, five serious hypotheses for why intermediation costs refuses to fall. Each is probably partially right but incomplete.

1. The asset-management mix shift. Individual mutual-fund and ETF fees have fallen; allocations shifted to higher-fee active and alternative managers such as hedge funds, venture capital, and private equity and credit. Average fee per dollar AUM stays flat. (Greenwood & Scharfstein, 2013.)

2. Trading-cost decline, volume explosion. Spreads have dropped over 90%. But volumes rose more to compensate. Active management still extracts ~0.67% of AUM attempting to beat the market (French, 2008).

3. Arms-race investment in expertise. Agents over-invest in expertise (think HFT infrastructure) to defend against counterparties’ opportunism. Privately rational, socially wasteful (Glode et al, 2012). Pagnotta-Philippon also notes that when exchanges compete on latency, faster venues charge more, and the rent persists.

4. Trust as the scarce resource. Gennaioli, Shleifer & Vishny notes that investors may be paying intermediaries for the confidence to bear risk. Trust doesn’t scale.

5. Information production is invisible in q. Bai et al notes that maybe finance got more efficient at producing informative prices (think the expansion of prediction markets) but that doesn’t show up in dollar quantities (i.e. financial intermediation has positive externalities).

Conclusions

The unit cost of finance appears to be one of the most stable macroeconomic constants we have. This may be because some of its proposed causes appear to be fundamental facts about human behavior.

For example, the first two causes (asset-management mix to and trading volume) both come from the desire of investors to beat the market. This does not compete away because the supply of people who believe they are above average appears to be inexhaustible. As long as there are prices, some fraction of humanity will pay to bet against them.

We should hesitate before calling this pure waste though. Since the presence of noise traders is necessary for the engine of price discovery. Measured ψ counts their fees but not the use of the information that those fees produce i.e. a positive externality the denominator q never sees.

The third hypothesis is the one place we may be hopeful, because its waste is a choice. The arms race of optimized optimal latency and microwave tower is privately rational but socially wasteful. Speed to trade redistributes who captures the information rather than producing more of it. But the size of that waste is not a constant of human nature but is rather encoded in market structure.

The same technology built for a head start under a continuous limit-order book contributes nothing extra under a frequent batch auction. Greed and self-interest are facts of human nature, but under different rules, carry wildly different social cost. The job of regulations is never to suppress competition but to shape the arena so competitive energy is spent producing information instead of shaving microseconds.

And finally, just as there will always be people who pay to beat the market, there will always be people who pay to forget it exists: to outsource the worry, to buy the ability to sleep through a drawdown. The advisor sells confidence, trust, and temperance, not insight or skill. This scrambles the tidy line between rent extraction and value creation. The wealth manager who trails the index by exactly his own fee is, on one ledger, pure rent; on another, he is selling a service his client values, i.e. the service of not having to care. Both are true at once. And the service does not get cheaper when chips do, because trust is built by reputation, personality, and time.

I nonetheless remain an optimist about the ability for the financial services industry to provide value. The deepest inefficiency in the real economy is not the two cents charged per dollar by banks and trading firms, bur rather the dollars that they never tough: the risks that no one trades because its not been securitized and the claims that cannot be collateralized because no market for it exists.

Securitize and financialize the rest of the economy and you do not lower ψ but rather enlarge q. A real economy in which more of its claims can be priced, hedged, pledged, and exchanged is a more efficient economy, even if the intermediary still keeps his cut. The per unit cost of intermediation may simply be permanent, but the territory it is levied on can still grow by orders of magnitude. That expansion may be the prize actually worth chasing. The next great financial institution will not be the one that reduces costs for investors, but rather one that create new markets of securitized assets for investors to trade.

Disclaimer

The information provided on TheLogbook (the “Substack”) is strictly for informational and educational purposes only and should not be considered as investment or financial advice. The author is not a licensed financial advisor or tax professional and is not offering any professional services through this Substack. Investing in financial markets involves substantial risk, including possible loss of principal. Past performance is not indicative of future results. The author makes no representations or warranties about the completeness, accuracy, reliability, suitability, or availability of the information provided.

This Substack may contain links to external websites not affiliated with the author, and the accuracy of information on these sites is not guaranteed. Nothing contained in this Substack constitutes a solicitation, recommendation, endorsement, or offer to buy or sell any securities or other financial instruments. Always seek the advice of a qualified financial advisor before making any investment decisions.

The Relationship Between Rates & Growth

The link between “expected to be rich in the future” and “high interest rates today” is the one most counterintuitive piece of the argument for non-economists, so let us build it up before formalizing.

Suppose you learn today, with perfect certainty, that you will inherit $100 million in five years. What do you do?

Most people asked the question directly say something like: take out a loan now, buy a house, take a sabbatical, start a business. The thing all those answers have in common is that they involve borrowing against the future windfall. You do not need to save for retirement anymore, because retirement is taken care of. You do not need to maintain a precautionary buffer, because the buffer is on the way. The marginal dollar five years from now is worth much less to you than the marginal dollar today, because in five years you will have so many dollars that one more makes very little difference to your life, while today every dollar still has a noticeable effect.

You would also be willing to pay quite a lot of interest on that loan. If a bank offered you a five-year loan at 15% annualized, you would probably take it, because 15% on a small loan today is paid back from a much larger pile in five years. The high rate does not deter you. The rate that does deter you is the rate that exceeds the implicit “exchange rate” between your current dollar-poor self and your future dollar-rich self, which is a high rate indeed.

Now imagine the same thing, but not just for you. Imagine everyone learns today that AGI will arrive in five years and they will all be much richer than they currently are. What does the loan market look like?

Demand for loans is high (everyone wants to borrow against the future). Supply of loans is low (why save when you are about to be rich anyway). With high demand chasing low supply, the price of loans, which is the interest rate, rises. It rises a lot. The market clears at whatever rate equates the smaller pool of people willing to delay consumption against the larger pool of people who want to consume now and pay back later. This is the whole mechanism. Expected future wealth raises interest rates because it shifts both sides of the loanable-funds market in the same direction.

Households have a strong preference to smooth consumption across time. A consumption path of $100 today, $100 next year, $100 the year after that feels much better than a path of $50 today, $200 next year, $50 the year after that, even though both have the same total dollar value. The reason is diminishing marginal utility: the second hundred dollars in a single year is less valuable than the first hundred.

Take this preference seriously and combine it with an expectation that future consumption will be much higher. The household sees a future of $1000K a year (post-AGI prosperity) and a present of $100K a year. The smoothing preference says this is a bad allocation. The household would rather have, say, $400K a year for both present and future. To get from $100K today to $400K today, it has to consume more than it earns today, which means borrowing.

It is willing to pay quite a bit to do so, because the $400K it wants to spend today is worth much more in utility terms than the $400K of future income it gives up. The interest rate it is willing to pay is bounded by how much it values the smoothing. Aggregate this across all households, and the same conclusion follows. The economy-wide interest rate rises because everybody wants to do the same kind of borrowing.

The Ramsey Rule Formalized

With the intuition in hand the formal version is a single equation away. Start with the canonical consumption-Euler equation. A household with time-separable utility u(C), subjective discount factor β ∈ (0, 1], and survival probability δ ∈ (0, 1] solves an intertemporal allocation problem whose first-order condition is

The survival probability δ represents a chance that the household might not be around tomorrow to enjoy future consumption, so it discounts the future more. Under isoelastic utility u(C) = C^(1 − 1/σ) / (1 − 1/σ) with elasticity of intertemporal substitution σ, perfect foresight, and constant consumption growth g, this collapses to the Ramsey rule

where ρ bundles time preference and survival into a discount parameter (ρ ≈ −ln(βδ)). Each term has an interpretation:

• ρ is the rate at which households discount the future for impatience and mortality reasons alone, even at constant consumption.

• g is the expected growth rate of consumption.

• 1/σ measures how much the household dislikes intertemporal consumption inequality. Higher 1/σ (lower σ) means a stronger desire to smooth, which translates into a stronger willingness to pay for borrowing when the future is expected to be richer.

Add uncertainty in growth. With log-normally distributed consumption growth (mean g, variance Var(g)) and isoelastic utility, the rule becomes

The new term is precautionary savings: uncertainty about future consumption raises desired saving, which pushes rates down. In the limit of certain growth, this term vanishes. In the limit of very uncertain growth, this term can be large enough to offset most or all of the growth effect.

Empirical Evidence

Real rates predict realized GDP growth: take inflation-linked bond yields (TIPS in the US since 1999, indexed gilts in the UK since 1985, Australian and Canadian linkers since the mid-1990s), and plot them against subsequent realized GDP growth over 10-, 15-, and 20-year horizons. If the Ramsey rule is doing real work, higher real yields should predict higher subsequent growth, because the real yield is the market’s pricing of expected growth on average.

The relationship is noisy because long-horizon predictions involve a lot of unrealized future variance, but the sign is right and the slope is meaningful. This is the sanity check on the equivalent claim in the other direction: when markets expect higher growth, they bid up real yields. The same dataset that lets the paper estimate the EIS at 0.74 also says the rate-growth link is empirically real and not an artefact of theory.

The deeper test is a panel regression with Consensus Economics survey-based growth forecasts on the right-hand side and inflation-linked sovereign yields on the left-hand side, across 59 countries from approximately 1990 to today. The dependent variable is the 5-year-forward 5-year real rate (the rate from year 5 to year 10 implied by current bond prices). The key right-hand-side variable is the Consensus Economics 5-year-forward 5-year GDP growth forecast. Controls include the standard deviation of the forecast, the 0-to-5-year growth forecast, and 10-year sovereign CDS spread. Standard errors are Driscoll-Kraay to handle cross-country correlation.

The labeled country scatter is striking. Across-country variation alone produces a strong positive slope. The clear anomaly is China and Vietnam, which combine high growth expectations with low real rates. The paper interprets this as evidence of capital controls preventing the rate-growth link from binding in countries with restricted external capital flows. Strip those out and the relationship is even cleaner.

The empirical conclusion is that the Ramsey-rule slope holds in the data. Real rates do in fact move with expected growth, by roughly the amount the theory predicts. This validates using observed rate movements to back out implied growth-expectation changes, and therefore validates using observed rate movements to back out implied probability shifts on regime-change scenarios like AGI.

Aligned & Unaligned AI

The Euler equation has two parameters that move with AI scenarios. Aligned AI moves g (expected consumption growth) up substantially. Unaligned AI moves δ (survival probability) down substantially. Both push the equilibrium real rate up.

The growth channel is what we previously covered. The survival channel is worth developing on its own. Take the Euler equation and solve for the implicit relationship between r and the discount parameters

If δ falls (survival probability drops), the term −ln(βδ) rises. A 1% per year drop in survival probability translates roughly to a 1 percentage point rise in real rates. The intuition: if there is a non-trivial chance you will not be alive next year, you discount future consumption more heavily, which is the same thing as demanding a higher return to delay current consumption.

This is empirically grounded. We see it in several historical episodes:

• AIDS antiretroviral rollout in Malawi (Baranov-Kohler 2018): when aid arrived and life expectancy rose by roughly 13 years, household savings rates rose measurably and investment in human capital (schooling) increased. The survival-probability shock was visible in aggregate decisions.

• Sri Lankan mortality drop, 1946-53 (Jayachandran-Lleras-Muney 2009): every additional year of life expectancy translated to 0.11 additional years of schooling. The mechanism: parents invest more in children’s human capital when those children are likelier to live long enough to use it.

• Huntington’s disease testing (Oster-Shoulson-Dorsey 2013): people who learn they are positive for Huntington’s complete college at rates roughly 30 percentage points lower than those who learn they are negative. Survival probability shocks change life-cycle decisions on time horizons that match the savings response.

These episodes confirm the survival-probability channel runs in the right direction at the household level. The unaligned-AI scenario in Halperin et al. has people putting nontrivial probability on existential risk over the next several decades. Even modest probability estimates, say 1% per year of extinction-level catastrophe, translate to roughly a 1 percentage point rise in real rates through the survival channel.

The important point is that both AI scenarios push rates up. Aligned AI raises g, which raises r via the growth channel. Unaligned AI lowers δ, which raises r via the survival channel.

The Recent Path of Rates

US 10-year TIPS real yields bottomed at roughly −1.0% in late 2021. They have risen steadily since, and long-dated yields across the developed world have continued to push higher: 30-year sovereigns in the US, UK, France, and Japan are now trading at nominal yields in the 5–6% range across the long end. That is roughly a 450 to 600 basis point move on the long end from the 2021 trough.

The popular explanation is fiscal dominance: unsustainable deficits, debt-to-GDP grinding higher, and term premium re-emerging as the marginal buyer of long-dated paper questions whether the issuer can credibly stabilize the path of debt. There is a real case for this. Deficits are large and structural, and it is not clear that necessary cuts to popular entitlement programs are politically viable.

But it is not the only story consistent with what we are observing. The Ramsey rule under uncertainty is linear in probability

solving for the implied probability p given an observed r gives

The point of the framework is that anything which shifts expected future consumption growth, fiscal-driven supply pressure, demographic re-pricing, or rising probability of transformative AI, feeds into the same real-rate observable. Fiscal dominance and AGI repricing are observationally adjacent on the rate channel. The first does not falsify the second; they would coexist in the same long-yield path.

Plug today’s roughly 3% real long-end rate into the calibration with ρ = 1%, g_TAI = 30%: the implied probability runs ≈5% at σ = 0.74, ≈7% under log utility (σ = 1), and ≈10% under an asset-pricing-prior σ = 1.5.

The Forecasting Research Institute’s elicitation of 69 economists puts the median probability of a rapid-AI scenario by 2030 at 10%, with an interquartile range of 6.7 to 20%. The implied rate-based probabilities sit roughly inside this range. The market is not “strongly rejecting” near-term AGI under any defensible reading of the long end; it is pricing a single-digit-to-low-double-digit probability, which is approximately what professional forecasters say.

Plug today’s roughly 5% real long-end rate with ρ = 1%, g_AGI = 30%: the implied probability runs ≈10% at σ = 0.74, ≈13% under log utility (σ = 1), and ≈20% under an asset-pricing-prior σ = 1.5.

The Forecasting Research Institute’s elicitation of 69 economists puts the median probability of a rapid-AI scenario by 2030 at 10%, with an interquartile range of 6.7 to 20%. The rate-implied probabilities sit on the FRI median and span its full interquartile range. The market is not just consistent with single-digit AGI pricing; it is consistent with the central tendency of professional forecasters, and at the high-EIS end with the upper quartile.

A nuance about which “AGI growth rate” to use in the calibration. The 30% benchmark is at the high end; FRI economists who put 10% probability on a “rapid AI scenario” by 2030 expect, conditional on that scenario, GDP growth of only 3.3 to 3.5%, an order of magnitude lower.

If the market is pricing a milder rapid-AI scenario (say, 5% growth instead of 30%), the implied probability from the observed rate rise climbs to 60% or higher. The framework is sensitive to the assumed conditional growth rate; pick a lower g_AGI and the implied probability is much higher for the same observed rate.

The Joint Move with Stocks

Take the Gordon equation in the simplest form

A pure rates-up shock with g_D unchanged drops P unambiguously. A pure growth-up shock with r unchanged raises P unambiguously. A simultaneous shock where both r and g_D rise leaves the sign of P’s change ambiguous. The relative sizes matter.

Aligned AGI raises both the numerator (profits) and the denominator (the discount rate that prices those profits), and the net sign depends on the elasticity of intertemporal substitution. The condition turns out to be:

• σ > 1: the growth effect dominates and equities rise on a AGI shock.

• σ < 1: the discount effect dominates and equities fall on a AGI shock.

• σ = 1: the two effects approximately cancel and equities are roughly unchanged.

A quick reminder on what σ, the EIS, is actually saying. EIS measures how much a household will shift its consumption-growth path in response to a change in rates.

Low EIS (σ around 0.5) means households are sticky about smoothing: they want roughly the same standard of living year to year and need large rate moves to budge their savings decisions. High EIS (σ around 2) means households are flexible: a higher rate is a great deal, so they happily cut spending now to consume more later. Log utility (σ = 1) is the textbook middle case.

Macro consumption-Euler estimates of the EIS in the broader literature cluster at σ between 0.2 and 1. Under this prior, AGI being priced in implies equities fall, not rise. The current pattern of both rates rising and equities rising is in that case inconsistent with the AGI hypothesis.

But the asset-pricing literature reaches the opposite conclusion. Long-run-risks models, recursive utility calibrations consistent with the equity premium, and the macro-finance term-structure literature all calibrate σ at values above 1, typically 1.5 to 2. Under this prior, AGI raises equities.

There is no settled answer to which σ is right. The empirical consumption-Euler literature has historically said low, the asset-pricing literature has historically said high, and the gap between them has not closed in 40 years.

If you believe the macro-Euler σ, the joint move is evidence against AGI being priced. If you believe the asset-pricing σ, it is evidence for. Given the asset-pricing prior, current markets are consistent with low-double-digit AGI probability being priced. Given the macro-Euler prior, they are not.

Stocks are a Bad Signal for AGI Risk

Setting aside the EIS-sign-ambiguity problem, there are three other independent reasons stocks are a worse direct signal of AGI than rates are.

1. Alignment risk zeroes everything. In the unaligned scenario, equities, bonds, and currency all go to zero. So a meaningful portion of the probability mass on transformative AI sends the equity price to zero regardless of which companies were positioned to benefit. The equity level is already adjusted for that wipeout probability, and disentangling the alignment piece from the underlying-business piece is not identified. Rates do not have this problem: the survival channel pushes them up under unalignment, the growth channel pushes them up under alignment, and the observed rate carries a clean signal of total AI probability.

2. Profit capture is uncertain even conditional on aligned AI. If 30%/year GDP growth materializes, who captures the surplus? There are several reasons public-equity holders may not: explicit profit caps (OpenAI’s original 100x cap), Windfall Clause-style precommitments to redistribute, government expropriation or nationalization, and simple competitive dispersion that splits the surplus across thousands of firms so that no single one captures much. Each of these is a path on which “aligned AGI arrives” coexists with “current public-equity valuations were not justified by the AGI thesis.”

3. The most AGI-relevant firms are private. The auto industry was transformative for the 20th-century economy, but most equity investors who tried to capture the auto boom in the 1900s and 1910s lost money to bankruptcies and dilution; the surviving giants mostly were not public during the formative period. Today’s analog is direct: the leading labs (OpenAI, Anthropic, xAI) are predominantly private. Public-equity investors have to bet on adjacent firms, NVDA, MSFT, semi equipment, rather than the labs themselves, and if value capture happens at the lab level it is invisible to public-equity prices.

Conclusions

None of the above arguments show that the AGI hypothesis is indeed what is driving rates and stocks up together. There are many candidate stories: fiscal dominance, structural shifts in the savings-investment balance, demographic re-pricing, an inflation regime change, term-premium normalization. Most of them have real evidence behind them, and on a simple weighting of priors the AGI explanation is not the largest single contributor to what we are seeing.

Nonetheless, the AGI hypothesis is one explanation that is consistent with the full joint pattern rather than just the rates side. A world where the market is gradually shifting weight onto a non-trivial near-term AGI probability looks like exactly what we have: earnings expanding rapidly as firms capture AI-driven productivity, equity prices climbing the wall of worry over higher rates and stickier inflation, and earnings pulling ahead of price fast enough to absorb the discount-rate move.

Climbing rates do not have to kill the equity story if the growth story is moving faster. So the calibrated read is not “this is a bubble” and not “AGI is priced in.” It is that AGI risk is one out of several plausible drivers, that under a defensible-but-not-settled choice of prior it could be a meaningful share of the joint move, and that on the rate-implied calibration the market is already pricing a probability that sits at the central tendency of professional forecasters.

Disclaimer

The information provided on TheLogbook (the “Substack”) is strictly for informational and educational purposes only and should not be considered as investment or financial advice. The author is not a licensed financial advisor or tax professional and is not offering any professional services through this Substack. Investing in financial markets involves substantial risk, including possible loss of principal. Past performance is not indicative of future results. The author makes no representations or warranties about the completeness, accuracy, reliability, suitability, or availability of the information provided.

This Substack may contain links to external websites not affiliated with the author, and the accuracy of information on these sites is not guaranteed. Nothing contained in this Substack constitutes a solicitation, recommendation, endorsement, or offer to buy or sell any securities or other financial instruments. Always seek the advice of a qualified financial advisor before making any investment decisions.

All good advice, but not the whole game.

There is another side of markets: the side where people take concentrated views, search for mispriced risk, size up conviction, and put real capital behind ideas before the rest of the market catches on. That is where Professional Punters live.

Professional Punters is a new podcast about the traders, investors, founders, and risk-takers finding edge across modern markets. We will be covering DeFi, prediction markets, sports betting, memecoins, perpetual futures, equities, options, and whatever other strange corner of the market happens to attract serious risk-taking.

This is not a podcast about playing it safe. It is not going to be limited to perfectly diversified portfolios, efficient frontiers, or textbook asset allocation. It is about how people actually find edge, build conviction, manage risk, and decide when a trade is worth taking.

For our first episode, we are starting with someone who embodies that spirit.

Our guest began trading prediction markets independently and got so good that he was recruited by a professional trading firm. He became one of the largest liquidity providers and takers on BTC five-minute markets, operating in one of the fastest, strangest, and most competitive corners of modern trading.

In the episode, we talk through how he found edge, what the market looked like when the opportunity was still underappreciated, the strategies that used to work but no longer do, and the ones that may still have real edge today.

I left the conversation extremely impressed.

This is exactly the kind of person we built Professional Punters to learn from: someone who did not just theorize about markets, but found a live edge, scaled it, and paid close attention as the edge evolved.

Follow our podcast below to get notified when we release the episode:

For more on our long-form research visit freeportlogbook.substack.com

Disclaimer

The information provided on TheLogbook (the “Substack”) is strictly for informational and educational purposes only and should not be considered as investment or financial advice. The author is not a licensed financial advisor or tax professional and is not offering any professional services through this Substack. Investing in financial markets involves substantial risk, including possible loss of principal. Past performance is not indicative of future results. The author makes no representations or warranties about the completeness, accuracy, reliability, suitability, or availability of the information provided.

This Substack may contain links to external websites not affiliated with the author, and the accuracy of information on these sites is not guaranteed. Nothing contained in this Substack constitutes a solicitation, recommendation, endorsement, or offer to buy or sell any securities or other financial instruments. Always seek the advice of a qualified financial advisor before making any investment decisions.

1 · Challenger and Morton Thiokol, January 28, 1986

The space shuttle Challenger lifted off from Kennedy Space Center at 11:38 a.m. Eastern Standard Time on Tuesday, January 28, 1986. Seventy-three seconds later it broke apart. All seven crew members died. The public watched live on television.

Four publicly traded prime contractors had built the shuttle. Lockheed had built the orbiter’s flight software. Rockwell International had built the orbiter itself. Martin Marietta had built the external fuel tank. Morton Thiokol had built the two solid-rocket boosters. By the time the New York Stock Exchange closed that afternoon, three of those four stocks had fallen between two and three percent.

The fourth, Morton Thiokol, had fallen 11.86%. Trading in Thiokol had effectively dried up within minutes of the explosion. In the half-hour following the launch, the tape showed only six prints in the stock.

The Rogers Commission, the official investigation, would not report for another five months. When it did, it identified the failure mode as an O-ring seal in the right solid rocket booster that had lost flexibility in cold weather and allowed hot exhaust gases to breach the seal. The booster was Morton Thiokol’s. The O-ring was Morton Thiokol’s. Engineers inside the company had warned the night before the launch that cold temperatures would compromise the seal; the warning had been overridden.

The market did not need to read the Rogers Commission report to know any of this. By the close on January 28, hours after the failure, the equity market had decisively concentrated the loss on the responsible firm. Michael Maloney and J. Harold Mulherin published the canonical study of this event in 2003 in the Journal of Corporate Finance (“The Complexity of Price Discovery in an Efficient Market: The Stock Market Reaction to the Challenger Crash”).

They considered every alternative explanation. Could insider trading by Thiokol employees have produced the pattern? No: option flows and SEC filings showed no unusual activity. Could selective news leaks have done it? No: contemporaneous wire reports did not single out Thiokol that day. Could the market have just been moving on the largest contractor, by coincidence the responsible one? No: by size, Lockheed and Rockwell were larger. The market had aggregated the diffuse engineering knowledge of thousands of analysts, suppliers, ex-employees, and tracking specialists into a price that named the culprit hours after the event and months before the official investigators.

2 · Three Mile Island and nuclear utilities, March 28, 1979

At 4 a.m. on Wednesday, March 28, 1979, the Unit 2 reactor at Three Mile Island Generating Station near Middletown, Pennsylvania, lost feedwater to its steam generators. The cascade of failures over the next several hours produced a partial meltdown of the reactor core, the most severe accident in the history of US commercial nuclear power. General Public Utilities Corporation (GPU), the utility that owned the plant, traded on the New York Stock Exchange.

GPU equity opened down sharply on Wednesday morning and continued to fall over the following two weeks. By April 11, the stock had lost approximately 40% of its pre-accident value, erasing roughly $230 million in market capitalization. The cleanup and decommissioning would eventually cost approximately $1 billion in 1979 dollars, spread over fourteen years.

The interesting feature for the cross-sectional thesis is the contagion. Bowen, Castanias, and Daley published an event study in the 1983 Journal of Financial and Quantitative Analysis tracking the abnormal returns of all publicly traded electric utilities with nuclear exposure in the two weeks after Three Mile Island. Utilities with operating nuclear plants earned negative abnormal returns of approximately 9% in the two weeks following the accident. Utilities with planned but unbuilt nuclear capacity earned negative returns of approximately 6%. Pure non-nuclear utilities earned essentially zero abnormal returns. The market priced the regulatory and operating cost of a generic “nuclear future” before the Kemeny Commission’s October 1979 report and before the multi-year regulatory tightening that would actually follow.

Hill and Schneeweis (1983, Journal of Finance) document the same cross-sectional pattern in the bond market: utility bonds with greater nuclear exposure widened in spread by amounts proportional to the share of nuclear capacity in their issuer’s generation mix. The accident on the morning of March 28 was processed across both equity and bond markets within days, with the cross-section of exposure correctly identified before any official investigation.

Three Mile Island is the closest US analogue to the Challenger story for industrial disasters. The pricing was cleaner because the responsible firm was a single corporate entity rather than a cross-section of contractors. The contagion to similarly exposed firms was visible within weeks. The decades-long shadow on US nuclear construction (no new ground-up reactor was ordered in the United States between 1977 and 2012) was implicit in the price the market wrote in April 1979.

3 · Lithium-6 and the hydrogen bomb, 1954

The United States detonated its first deployable hydrogen bomb on March 1, 1954, in the Castle Bravo test at Bikini Atoll. The bomb yielded 15 megatons, two and a half times the predicted yield, contaminating a Japanese fishing boat and producing the first international fallout incident. The design of the bomb was classified at the highest level. Specifically classified was the choice of fusion fuel: not liquid deuterium, as the earlier Ivy Mike test had used, but solid lithium-6 deuteride, which made the bomb small enough to be dropped from an aircraft.

That a chemical lithium compound was the fuel of the hydrogen bomb would not be public knowledge for decades. The United States government continued to classify the design choice through the late 1970s.

In 1954, the equity of the publicly traded lithium-producing firms rose 461%. The Dow Jones Industrial Average rose 50% in the same year. Other rare-earth and radioactive-metal producers (uranium, thorium, beryllium) stayed roughly flat. Lithium Corporation of America alone earned an abnormal return of approximately 28% within one month of Castle Bravo, settling into the 461% annual return as the year progressed.

This story is preserved because Armen Alchian, working at RAND, ran what was probably the world’s first formal event study on it. Alchian observed the lithium-stock spike, inferred from the magnitude and timing that lithium had to be the fusion fuel, and wrote a short paper describing the inference. RAND’s security officers read the paper, recognized that Alchian had reconstructed a classified design choice from public price data, and ordered the paper destroyed.

The story would have been lost entirely if Joseph Newhard had not reconstructed it from CRSP-era data in his 2014 Journal of Corporate Finance paper, “The Stock Market Speaks: How Dr. Alchian Learned to Build the Bomb.” Newhard’s chain of inference is the same as Alchian’s: there were only a handful of publicly traded lithium producers, only one Teller-Ulam-design choice plausibly explained the spike, and the timing aligned to within weeks of the Bravo shot. The market had priced a state secret.

The interesting epistemic point is that nobody who bought lithium stock in 1954 needed to know what lithium was being used for. They needed to know that demand for lithium was spiking, that the buyers were not commercial chemists or battery manufacturers (lithium-ion batteries were 30 years away), and that the most plausible large-volume buyer was the federal government. The chain “government is buying lots of lithium” plus “the government is at war for nuclear primacy” plus “lithium has plausible nuclear-relevant chemistry” was available to anyone with a chemistry textbook and a procurement contact. The market did the rest.

4 · The Cuban Missile Crisis and city-level geography of risk, 1962

For thirteen days in October 1962, the United States and the Soviet Union came closer to nuclear war than at any other point in the Cold War. Soviet missiles had been emplaced in Cuba. The Kennedy administration debated invading; the Soviets debated launching; analysts on both sides put their probability of nuclear exchange between 10% and 50%.

The aggregate equity market response to this was, by reputation, modest. The S&P 500 fell roughly 3% over the crisis window, less than other 1962 drawdowns. For decades, financial economists treated this as a puzzle: surely the largest existential-risk event of the postwar era should have moved the market more than a routine recession.

David Andrew Finer’s 2022 working paper “No Shock Waves through Wall Street?” reframed the question. The aggregate index is the wrong place to look. The cross-section is the right place. Finer geocoded the headquarters cities of every NYSE-listed firm, overlaid them with declassified Cold War expert maps of expected Soviet-strike probability per US city (which weighted by economic importance, military bases, and political symbolism), and ran a panel regression of firm-level abnormal returns during October 16 to October 28, 1962. Firms headquartered in cities estimated by intelligence analysts as higher-priority Soviet targets earned significantly lower returns during the crisis window, controlling for industry, size, and aggregate risk.

The market was, on a daily basis through the crisis, implicitly computing a city-level actuarial estimate of where the missiles would land if they were launched. The pricing was not the gross expected value of nuclear war (which would have wiped out everything roughly equally, since US economic activity would not survive a full exchange). It was the conditional pricing: given that we get a strike, where does the strike land, and what fraction of the surviving economic activity does each firm represent? That information was distributed across thousands of analysts, military commentators, journalists, and informed amateurs. The market aggregated it into a usable cross-section.

The corollary that Finer’s paper makes most clearly: aggregate-level event studies can fail to detect aggregation that is happening cleanly at the cross-sectional level. The Cuban Missile Crisis “puzzle” of modest aggregate response dissolves once you look at the right unit of analysis.

5 · Orange juice futures and the National Weather Service, 1984

Richard Roll’s “Orange Juice and Weather,” published in the American Economic Review in December 1984, is the foundational academic paper on market-as-information-aggregator. The setup is mundane and the finding is elegant.

Frozen-concentrate orange juice futures trade on what was then the New York Cotton Exchange. The largest US orange-juice-producing region is Central Florida. Cold weather in Central Florida damages the orange crop and pushes up futures prices. The National Weather Service issues forecasts of overnight low temperatures in the Central Florida region. Roll asked the obvious question: do OJ futures contain information about the weather beyond what NWS forecasts already contain?

The answer was yes. Statistically significant. If OJ closed above its open, the NWS forecast for that night was systematically too high (too warm) by an amount that could not be explained away by reporting noise. The market knew, by close of trading, things about the night’s weather that the NWS would only catch up to in its next-day revision.

The mechanism is not magic. Physical traders watch citrus groves directly, monitor freight rates for trucks moving to harvest areas, observe grower hedging volumes, and read agricultural-economist newsletters that the NWS forecaster does not. Each individual data source is small. Aggregated through the futures price, the data sources add up to a more accurate temperature forecast than the federal government produces.

The Roll paper became the touchstone of the 1980s “wisdom of crowds” literature and the inspiration for the climactic scene of the film Trading Places. It remains the cleanest case of a financial market visibly outperforming a competing federal information aggregator on the federal aggregator’s home turf.

6 · Pre-announcement runups in M&A targets

The single most quantified version of “everything is priced in” is the corporate-control literature. Whenever a public company is bought by another, the target’s stock price rises sharply on the day the deal is publicly announced. The interesting question is: how much of that rise has already happened in the weeks before the announcement?

Keown and Pinkerton, in 1981 in the Journal of Finance, ran the first careful study. They looked at 194 mergers between 1975 and 1978. Across the 25 trading days before the public announcement, the target’s stock price drifted up by roughly 40 to 50 percent of the total takeover premium it would eventually earn. The drift was concentrated in the final week before announcement, but it started visible at least a month earlier.

This is leakage. The corporate decision to acquire is known to a small set of executives, lawyers, bankers, and consultants, none of whom is supposed to trade on it. Some fraction of them do. The drift is the trace of that trading in the public tape.

Lisa Meulbroek, in 1992, looked at the question from the inside out. She used SEC enforcement records of insider-trading prosecutions to identify the specific days on which insiders had illegally traded ahead of merger announcements. On those days, the target stock earned an abnormal return of approximately 3% on average. The total pre-announcement runup of roughly 30 to 50% of the eventual premium could be substantially accounted for by insider activity that the SEC later detected. Two implications. First, illegal trading is detectable by both regulators (through enforcement) and by the market (through the runup). Second, the market processes the information about an impending deal in essentially real time, even when the information is being introduced through illegal channels.

The M&A runup literature is the most replicated finding in this entire collection. It is one of the strongest pieces of evidence that public markets price information that is, in the most literal sense, private at the moment of pricing.

7 · The Iowa Electronic Markets and the wisdom of small-money political crowds

The Iowa Electronic Markets opened in 1988 as a research project at the University of Iowa. Real-money political-event contracts (with strict per-trader position limits, around $500 in the early years) traded on US presidential elections and many other races.

Across five presidential election cycles from 1988 to 2004, Berg, Nelson, and Rietz documented (in International Journal of Forecasting, 2008) that the IEM vote-share contract was closer to the eventual outcome than the contemporaneous Gallup-style poll on 74% of polling days. The advantage grew at longer forecast horizons: at 100+ days out, the IEM beat polls by a wider margin than at 1 day out.

The result was striking because the IEM had no obvious informational advantage over polls. The traders were a few hundred academics and graduate students. The position limits made it impossible for a single well-informed trader to dominate the price. And yet the aggregate of these small bets, across a small market, outperformed the most sophisticated polling apparatuses of major newspapers.

The mechanism is the wisdom-of-crowds story plus a selection effect. Polls survey a random sample of voters and ask them to predict their own behavior; the answer is noisy because voters are bad at predicting their own behavior. The IEM surveyed a self-selected sample of analytical hobbyists who were betting on the outcome they actually expected; the answer was less noisy because their incentive was accuracy.

The two methods are not measuring the same thing. The IEM was measuring belief, weighted by the strength of belief plus the willingness to risk capital. That turned out to be a better predictor than self-reported voter intent. The IEM was the proof of concept for the larger prediction-market industry that would emerge two decades later.

8 · DARPA’s Policy Analysis Market, 2003

The Defense Advanced Research Projects Agency, between 2001 and 2003, designed a real-money prediction market called the Policy Analysis Market. The market was to trade contracts on Middle East political and military events: civil unrest in specific countries, regime stability, conflict probabilities. The intended use was as an alternative information aggregator for the intelligence community. The hypothesis, owing largely to Robin Hanson, was that aggregated bets from a wide pool of informed traders would outperform analyst panels.

On July 28, 2003, Senators Ron Wyden and Byron Dorgan held a press conference labeling PAM an “assassination market” and accusing the Pentagon of incentivizing terrorism. The Secretary of Defense cancelled the program the next morning. The hypothesis was never tested at scale.

This case is in the essay because of what it implies rather than what it demonstrated. The Pentagon’s own R&D arm had concluded that the prediction-market mechanism was the most promising alternative to analyst-led intelligence aggregation. The CIA’s Studies in Intelligence journal published a favorable retrospective in 2006. The premise that markets aggregate intelligence-relevant information better than analyst panels was, inside the relevant institutions, taken seriously enough to fund.

The implementation was killed for political optics, not because the underlying claim had been disproved. The current generation of real-money political prediction markets (Polymarket, Kalshi, others) is the descendant of the IEM and PAM.

9 · Trump and the peso, election night 2016

US polls began closing on Tuesday, November 8, 2016. Pre-election polls had shown Hillary Clinton with a roughly 4-percentage-point national lead and approximately 70 to 90% probability of victory across the major poll aggregators. Equity futures and the Mexican peso were positioned for a Clinton win.

The Mexican peso had been trading as a proxy for the election outcome for months. Donald Trump’s anti-trade and immigration positions implied a structurally weaker peso, and the peso had been rallying or selling off in approximate inverse correlation to Trump’s daily polling odds. As state-level results began arriving Tuesday evening, the peso began collapsing. By 10:30 p.m. Eastern, with Florida and North Carolina trending Republican, USD/MXN had moved from approximately 18.5 to 20.8, a 12% peso decline in roughly three hours. By the overnight low, the peso had fallen approximately 13% intraday, its sharpest move in over twenty years.

US Treasury yields moved in parallel. The 10-year Treasury yield rose from approximately 1.86% at the 4 p.m. cash close to over 2.05% by Asia morning, an overnight move of nearly 20 basis points. Equity futures fell sharply through the night and then reversed into a rally that would carry the S&P 500 to new highs over the following weeks.

The interesting feature of the night for the everything-is-priced-in thesis is the speed. AP did not call the race for Trump until 2:29 a.m. Eastern, November 9. The peso, the bond market, and the equity futures had effectively priced the Trump-win outcome by midnight, two and a half hours earlier. The signal was being read off state-level returns and county-level precinct data, parsed in real time by traders watching the AP feeds and the New York Times’ “needle” updates, and impressed onto deep, liquid global markets before the formal media call.

Polymarket did not exist as a serious market in 2016 (it would launch in 2020). The Iowa Electronic Markets did, and its Trump contract crossed 50% around the same time the peso was crossing 20.5. The aggregate of FX, rates, and small prediction markets identified the eventual outcome hours before the formal media infrastructure caught up. The 2016 election night is the proof-of-concept for the 2024 election night, scaled down by an order of magnitude in market depth and lengthened by an order of magnitude in lag.

10 · Deepwater Horizon, April 2010

The Deepwater Horizon offshore drilling rig exploded on April 20, 2010, killing 11 workers and beginning the largest marine oil spill in US history.

Three publicly traded companies were directly involved. BP owned the leased well and was the operator. Transocean owned the rig itself. Halliburton had cemented the wellbore. Over the next two months, equity in all three fell sharply, with cross-sectional dispersion that tracked specific assignments of fault as they emerged.

BP fell from approximately $59.48 on April 19 to $27 on June 25, a loss of approximately 55%, erasing approximately $90 billion in market capitalization. Options volume on BP increased twentyfold during the response period. Credit-default-swap spreads on BP debt widened by hundreds of basis points. Transocean fell less sharply but visibly. Halliburton’s response was muted, anticipating the eventual finding that its cementing work had been compliant with industry practice.

The market priced the eventual liability allocation in approximate real time. The final civil and criminal penalties against BP, settled over the following years, totaled approximately $60 billion. The market’s June 2010 low for BP reflected a present-value estimate of this exposure within roughly the right order of magnitude, computed within two months of the spill and well before any formal liability assignment.

The interesting cross-sectional pattern is that Transocean and Halliburton equity moved by amounts proportional to their eventual liability shares, while BP took the brunt. This is the same mechanism Maloney and Mulherin documented for Challenger: the market resolves the cross-section of responsibility correctly while the official investigation is still organizing.

Conclusion · What the price knows

Prices in deep markets routinely identify causes, allocate blame, forecast weather, and call elections before the institutions formally tasked with these jobs have finished assembling their conference rooms. The market’s edge is not omniscience. It is aggregation: it pools the small, partial, often unconscious knowledge of thousands of marginally informed people, weights each contribution by the willingness of its holder to risk capital, and prints a number. The number is almost always closer to the truth than any single contributor could be, and frequently closer than the formal expert apparatus mobilized to answer the same question.

The standard objection is that markets are “just speculation,” that derivatives are gambling, that prediction markets monetize tragedy. The objection has the structure of an aesthetic complaint rather than an empirical one. Speculation is the name we give to forecasting when we disapprove of the forecaster. The trader who shorted Morton Thiokol within minutes of the Challenger explosion was not desecrating the dead; he was, in the act of trading, telling the rest of the market who had built the failed component. Risk has to be priced by someone.

The choice is not between a world with speculation and one without it. The choice is between a world that prices risk through people with skin in the game, and a world that prices it through people without.

A society that took this seriously would be quicker to legalize the instruments that aggregate information and slower to ban them. It would treat loud calls to outlaw prediction markets the way it treats calls to outlaw thermometers: as the complaints of people who dislike what the instrument reads. None of this requires believing markets are sacred. It requires only noticing what they have repeatedly done, and asking, with humility, what we might be missing when we insist on doing it ourselves.

Disclaimer

The information provided on TheLogbook (the “Substack”) is strictly for informational and educational purposes only and should not be considered as investment or financial advice. The author is not a licensed financial advisor or tax professional and is not offering any professional services through this Substack. Investing in financial markets involves substantial risk, including possible loss of principal. Past performance is not indicative of future results. The author makes no representations or warranties about the completeness, accuracy, reliability, suitability, or availability of the information provided.

This Substack may contain links to external websites not affiliated with the author, and the accuracy of information on these sites is not guaranteed. Nothing contained in this Substack constitutes a solicitation, recommendation, endorsement, or offer to buy or sell any securities or other financial instruments. Always seek the advice of a qualified financial advisor before making any investment decisions.

What is Volatility Skew

Briefly, the implied volatility of an option is the volatility you would have to plug into Black-Scholes to make the model price match the observed market price. Different strikes on the same expiry generally trade at different implied vols, and a plot of implied vol against strike is called the volatility skew (or smile, or smirk, depending on its shape). For equity indices and most single names, the curve slopes down, downside puts trade at higher implied vols than upside calls, which is what practitioners mean when they say “equity skew is negative / put-skewed.”

Where does that asymmetry come from? The classical answer is spot-vol correlation.

If realized volatility tends to rise when the spot falls, which it overwhelmingly does in equities, then the paths that finish below an OTM put’s strike are paths along which vol has on average been higher than usual, so the strike vol on the put has to compensate and prints rich.

Symmetrically, the paths that finish above an OTM call’s strike are paths along which vol has been compressing, and the strike vol on the call prints cheap. This is the dynamic, or local-volatility, view of skew: skew exists because the underlying’s vol moves opposite to its price.

So far, so standard. The thing the standard story leaves implicit is that “the underlying’s vol moves opposite to its price” is an empirical statement about a single, undifferentiated random variable: the stock. The stock, however, is not a primitive variable. It is a function of underlying drivers, and the vol response to those drivers does not have to point in the same direction.

Decomposing the Stock

When we talk about a stock’s daily P&L, we are perfectly comfortable saying “today NVDA was up three percent, but index was up two so that was expected.” That is a casual decomposition of return into a market-beta component and an idiosyncratic component. We do not think of the stock’s daily return as a primitive; we routinely partition it.

What naturally follows is that we ought to partition is the stock’s vol response to those moves. A naive local vol model treats spot-vol correlation as a single function of the single variable “stock price.” But if the return can be decomposed into a beta piece and an idio piece, the spot-vol relationship can be decomposed too. The vol of the stock responds differently to a one percent move that is the index’s fault than it does to a one percent move that is purely the name’s own, and those two responses, importantly, can point in opposite directions.

Consider a meme stock with a beta to the broad market, sitting in some otherwise normal macro regime. Hold all else equal and consider each of the following moves.

Case 1) Stock up with beta, slow and orderly. The S&P drifts up half a percent over the morning, the stock follows it up about a percent in line with its beta, nothing unusual is happening. Implied vols on the index and the stock decline modestly.

Case 2) Stock up violently, while the market is flat. The vol response here is positive. The meme is signaling that it has not yet died, and option markets respond by paying up for upside.

Case 3) Stock down with the market. The S&P sells off two percent on a macro headline; the stock sells off three in line with beta. Here we are back in the textbook. Vol goes up. VIX goes up. Skew steepens.

Case 4) Stock down, while the market is flat. This is the case that most clearly exposes the limits of the single-factor view. The market is fine. The stock idio drops ten percent as the meme deflates into nothing. The vol response here is negative. The story is over. The convexity is gone. Forward realized vol on the name collapses because the source of variance itself (the meme) has evaporated.

Two Skews, One Curve

The same sign of spot move gives the opposite vol response depending on whether the move was macro-driven or idio-driven. The realized correlation between the stock’s price and its vol is therefore not a single number but a blend of two underlying numbers, with the weights set by which channel is currently doing the moving.

If we adopt the local-volatility view that skew is the option market’s pricing of spot-vol correlation, then the skew on a single name must, by the same logic, be a blend of two skew curves:

• A macro skew, inherited from the stock’s beta exposure to the broad market. This is the equity-fear curve with negative spot-vol correlation, OTM puts richer than OTM calls, the classic SPX-style smirk on a smaller scale.

• An idio skew, reflecting the stock’s spot-vol relationship at the residual level. The shape of this curve depends on what the name actually is.

The skew you observe on the option chain is some weighted combination of those two underlying curves, weighted by roughly the relative contribution of macro and idio variance to the name’s total variance (technically not true since the macro skew also takes into account the correlation between spot and index-wide correlation but for the purposes of this post let us not consider additional effects).

This isn’t strictly additive. Spot-vol relationships aren’t linear; beta itself isn’t constant; and the conditional distribution of an idio move given a macro move is messy. The decomposition is best read as a heuristic, not an identity.

Further Decompositions

Classical local volatility writes the instantaneous vol of an asset as a function of its own spot price and time, σ = σ(S, t). The observed skew curve is the option market’s pricing of that function.

What we have just argued is that, for a single stock, the relevant local-vol function isn’t really σ(S_stock, t). It is closer to σ(S_stock, S_index, t) with two variables. And there is no principled reason to stop at two. You could extend to σ(S_stock, S_index, S_sector, t), or further.

A name in semis might want a separate factor for AI Capex. A bank might want a rates factor and a credit spread. A miner might want the commodity. Each new variable opens another channel through which spot moves can be sourced and through which vol can respond differently.

However, there is a calibration asymmetry that makes the macro/idio decomposition especially useful. We have deep, liquid options markets on the broad index such as SPX, NDX, and RUT. The macro skew is therefore directly observable. We can pull a clean macro spot-vol relationship off market prices any trading day of the week. We do not, in general, have clean factor option markets for the things that drive other aspects of vol.

But when we do, a linear multi-factor local vol model is the simplest possible generalization of single-asset local vol. Though as soon as we permit ourselves to do this, the usual problems of any multi-factor linear model show up as the inputs are correlated with each other, the loadings shift across regimes, and calibration is underdetermined.

You could also put flow into the model. You could put dealer options-positioning. You could put a sentiment index or a search-trend metric also. Each of those will, in principle, shift the conditional vol of a single name. The macro/idio decomposition is just the cleanest two-factor approximation to a much higher-dimensional problem; we pick it because we can always see the macro piece in liquid index options, not because it is the only piece that matters.

Closing Thoughts

None of what we’ve written here is a description of reality. Markets are not models. Markets are what models try to explain, and any model elegant enough to be useful is going to be wrong in interesting places. However, at times each new layer of does give us a sharper articulation of what the market is currently pricing, and by contrast it shows us the places where the market is doing something we don’t yet accommodate.

The places the model breaks are the places that matter, because those are the places where the market is telling us something we did not yet have a word for. When that happens, and it happens often, we get to make better models (and make a lot of money along the way). The two-skew view is one such step. There are many more.

Disclaimer

The information provided on TheLogbook (the “Substack”) is strictly for informational and educational purposes only and should not be considered as investment or financial advice. The author is not a licensed financial advisor or tax professional and is not offering any professional services through this Substack. Investing in financial markets involves substantial risk, including possible loss of principal. Past performance is not indicative of future results. The author makes no representations or warranties about the completeness, accuracy, reliability, suitability, or availability of the information provided.

This Substack may contain links to external websites not affiliated with the author, and the accuracy of information on these sites is not guaranteed. Nothing contained in this Substack constitutes a solicitation, recommendation, endorsement, or offer to buy or sell any securities or other financial instruments. Always seek the advice of a qualified financial advisor before making any investment decisions.

For years, Matt Levine has argued that “eventually, every possible trade will be packaged into an exchange-traded fund.” This past week, on the back of Robinhood’s launch of an agentic trading account, he revised himself:

But now I think that this argument was naïve, and the ETF, as a universal package for trade ideas, might already be outdated. “Every possible trade” will be an ETF? There are infinite possible trades, some of which ETF providers and brokers and wealth advisers have probably never thought of… What you want is a more general-purpose technology, a way to describe any imaginable trade in a few sentences and then execute it with a click of a button. And that, increasingly, exists.

He’s right. The agentic broker is the natural successor to the ETF. Any trade you can describe, you can now execute. Robinhood’s own examples make that concrete:

A thematic investor with conviction in AI or semiconductors can have an agent build an initial portfolio matching their criteria, monitor the space for new entrants and analyst upgrades, and rebalance toward the strongest opportunities at a regular cadence. An active trader can backtest a mean reversion strategy to see how it performed historically, and deploy it to automatically buy oversold stocks and sell when they revert to the mean.

Robinhood’s VP for brokerage, Abhishek Fatehpuria, told Bloomberg that “when we enable customers to connect their outside agents and outside tools, the possibilities are pretty vast.” That is exactly the right adjective. They are vast. And that vastness is the part most people are not yet thinking carefully about.

Because the question Levine’s framing leaves open is the one that actually matters: where does the trade idea come from? Levine himself is uneasy about the answer. In the same piece, he writes:

Stock trades are not really free, and anyone who works in professional quantitative trading was horrified by [the suggestion that they are]. Presumably these agents will be good for business at the electronic market makers who fill Robinhood’s orders and collect a spread on each trade. “Agentic trading,” to them, means something like “retail traders making lots of small predictable rule-based stock trades all the time,” which is what they want.

And then, the line that is really the heart of the matter:

One possibility is that widespread agentic retail investing would allow each individual investor to automate his or her idiosyncratic investing ideas, leading to, you know, a lot of random noise trading. Another possibility is that widespread agentic retail investing would lead to all the individual investors automating the same, like, 20 investing ideas, leading to herding and predictability.

The first possibility suggests that AI will have no edge at all in trading, but the second possibility is the one we should worry even more about, and it is, with overwhelming probability, the one that plays out by default.

Every general-purpose AI chatbot, asked some version of what should I buy, is going to lean on the same training distribution of financial commentary, the same surface of recent headlines, the same set of well-known names. The defaults are convergent. The agents on top of those defaults are not going to disagree with each other in any interesting way, because they were not given any interesting way to disagree.

You don’t get edge from running a customized strategy on top of a generic input pipeline; you get a slightly different rendering of the same twenty trades that every other agent is also lining up to make. That isn’t investing. It is a slower, politer version of the meme-stock cycle, automated and rebalanced quarterly, with a market-making industry happy to collect the spread on each lap.

The conversation has to move past the agent itself and back onto what it reads. And that begins with being honest about what today’s AI is, and is not, capable of.

Today’s AI is not yet capable of financial creativity. It cannot originate a thesis the way Citrini does in Let There Be Light, identifying an optical-interconnect market structure shift years before it reaches headlines. It cannot do what Druckenmiller does when he meets Milei in Davos and walks out thinking Argentina is a trade. Those acts are creative human judgments. They require taste, conviction, scar tissue, and time spent in the world. The model does not have them, and nothing about scaling current architectures suggests it is about to.

What AI does extraordinarily well is the work around the creative judgment. It can read everything. It can track which catalysts a writer noted and which have actually happened. It can cross-reference Substacks, earnings transcripts, sell-side notes, and live newswires, and surface the moments when two or more credible independent voices converge on the same name. It can take a four-thousand-word thesis and turn it into a specific tradeable expression with sizing, hedges, and a monitoring rule attached.

None of that is glamorous, but it is exactly the work that no individual retail account has the time or the budget to do manually across the full surface of credible analyst output.

It can also verify connections that the human author has flagged but that no individual reader is realistically going to chase down. When Irrational Analysis writes that Tower Semiconductor’s real incremental driver is its COUPE competitor, a best-in-class silicon-nitride photonic-integrated-circuit process bolted to a “meh” 65nm-class electronics process, and then tells the reader, in so many words, to do their own homework:

Check the loss (dB/cm) of Si vs SiN waveguides. Check the Q-factor of Si vs SiN ring structures. Check the system-level tradeoffs of a slow-and-wide NRZ link with clock forwarding versus traditional PAM4 224G SerDes driven optical link. Pay attention to laser power requirements to meet SNR.

…almost no retail investor is actually going to do that homework. An AI can. It can read the optics literature, walk through the Q-factor math, and come back with a written verdict on whether the analyst’s leap is supported by the underlying physics. The creative leap, look here, is human. The verification of whether the leap holds up is something AI can do.

This is the part of the picture the generic agent is missing, and it is why we believe, at Freeport, that the input matters as much as the output. A generic agent reading a generic feed will converge to a generic portfolio, which is approximately the consensus retail trade, which is approximately what the market-makers are already pricing for.

The way out of that loop is not a smarter agent on top of the same inputs. It is a customized pointer: a deliberate, curated surface of writers and analysts who are doing original work, fed into a system that can synthesize, verify, monitor, and act on it at the speed and scale that the writers themselves cannot. The agent is the easy part. Pointing it at something worth reading is the entire game.

Concretely, this is what Freeport is building. We continuously read the highest-signal writing in financial markets; independent analysts on Substack, the credible accounts on Twitter, the major financial journalists, the live newswire; and turn that surface into a filtered feed of trade-relevant ideas in something close to real time.

When a writer you’ve trusted before flags a setup, the notification finds you. When a thesis is quietly building across several independent sources, we surface that convergence before it becomes a consensus headline. And on top of the raw feed, we publish synthesis on the themes themselves: what is actually being talked about this week, which catalysts have already fired and which are still ahead, where the smart writers agree, and where the disagreement is doing real work. The pointer is curated. The processing is automated. The judgment that matters is still yours.

We believe that this is also where the durable reward is. Short-term, latency-driven arbitrage has been effectively monopolized by the firms with the fastest infrastructure; that contest is over, and it was over before most of us showed up.

The edge still available to traditional investors, hedge funds, and quant strategists is at the horizon of weeks, months, and years, where the constraint is not how fast you can act, but how well you can read the world. AI is enormously useful at that horizon, because it lowers the cost of reading the world. It does not, and for some time will not, lower the cost of seeing it differently.

That is the distinction Freeport is built on. AI does the reading. Humans, the best ones, still do the seeing. Our job is to make sure the readers are pointed at the right writers, and that what comes out the other side is something better than a rerun of everyone else’s trade.

Disclaimer

The information provided on TheLogbook (the “Substack”) is strictly for informational and educational purposes only and should not be considered as investment or financial advice. The author is not a licensed financial advisor or tax professional and is not offering any professional services through this Substack. Investing in financial markets involves substantial risk, including possible loss of principal. Past performance is not indicative of future results. The author makes no representations or warranties about the completeness, accuracy, reliability, suitability, or availability of the information provided.

This Substack may contain links to external websites not affiliated with the author, and the accuracy of information on these sites is not guaranteed. Nothing contained in this Substack constitutes a solicitation, recommendation, endorsement, or offer to buy or sell any securities or other financial instruments. Always seek the advice of a qualified financial advisor before making any investment decisions.

The Heston Model

The model starts by letting the asset price grow at the risk-free rate while its diffusion scale is the square root of a stochastic variance. The variance itself follows a square-root diffusion (CIR Process). Under the risk-neutral measure, the core dynamics are:

The terms in the above dynamic equations are interpreted as such:

The variance is mean-reverting with speed kappa toward a long-run level theta, it is buffeted by its own randomness with intensity sigma (vol-of-vol), and its shocks can be correlated with price shocks through rho.

Two immediate practical features fall out. First, the square-root term keeps variance nonnegative and, under a familiar condition below, away from zero. Second, by allowing correlation between price and variance shocks, the model can create realistic downside skew in equities.

Mean reversion is the backbone of the model’s intuition. If today’s variance starts below its long-run level, the drift term pushes it upward; if it starts above, the drift pulls it down. That story is made precise by the conditional expectation of the variance as time moves forward. This single formula already tells you how quickly the market “forgets” today’s variance regime and drifts toward the background climate:

Autocorrelation—volatility clustering—drops out immediately from the same mean-reverting structure. Two nearby points in time share much of the same history, and that correlation decays exponentially as you separate them. This is both a realistic statistical description and a key pricing ingredient, because it tells you how much today’s volatility regime bleeds into tomorrow’s:

Options, however, do not care about the instantaneous variance in isolation. What they “see” is the time-average of variance over their life. At a high level, the at-the-money implied variance for maturity T is well-approximated by the risk-neutral expectation of that time-average, with small corrections we will discuss shortly. This makes the integral of variance—the realized variance over the option’s life—the central object. The model gives a clean, closed-form expression for the expected integrated variance and, dividing by maturity, for the expected average variance:

This compact formula is the engine behind the term structure. The factor multiplying the initial deviation from the long-run level is a decreasing function

When today’s variance is below its long-run level, the initial deviation is negative. Multiplying a negative number by a function that decreases with maturity makes the overall correction term rise toward zero as maturity grows, which means the expected average variance rises toward the long-run level.

Interpretation of Parameters

What we described above is precisely an upward-sloping at-the-money variance term structure. Mirror the argument when today’s variance is above its long-run level and you get a downward-sloping term structure. If today equals the long-run level, the term structure is flat—nothing in the model needs to work harder than this basic mean-reversion arithmetic.

A useful first-order bridge between the variance path and option prices is the approximation that at-the-money implied variance equals the expected average variance, up to small corrections that vanish as vol-of-vol goes to zero.

However, refinements matter for practitioners. First, option prices are convex in variance. Randomizing variance while holding its mean fixed pushes option prices up compared with a constant-variance world at that mean; in implied-variance terms, this adds a small upward “convexity premium” proportional to the variance of the time-average of variance. In Heston, that variance is governed mainly by vol-of-vol and mean-reversion speed.

Second, the correlation between price and variance shocks—the leverage effect—tilts the surface, especially at short maturities, by making downside price moves coincide with increases in variance. These refinements shift levels and near-term curvature, but they do not overturn the central message: the sign of the at-the-money slope is determined by where today’s variance sits relative to the long-run level and by the monotone shape of the factor shown above.

You can summarize the “why upward?” logic in a crisp checklist when the market opens in a calm regime (today’s variance below its long-run level). One, the conditional mean of variance increases over time, so the path you expect to average lies above today’s level and points toward the long-run anchor. Two, the expected average variance is a decreasing transform of that initial gap, so it moves upward with maturity. Three, short maturities remain anchored to today’s regime due to autocorrelation, leaving room for longer maturities to blend in more future days closer to the long-run climate. Four, convexity and leverage sprinkle adjustments on top—raising levels slightly and shaping skew—but the basic slope logic holds. The mirror image applies in stressed regimes: a high starting variance leads to a downward slope that relaxes toward calmer long-run conditions.

Parameter roles fall neatly into place once you view the surface through this lens. The long-run level pins the far-end at-the-money variance (plus a modest convexity cushion), the mean-reversion speed controls how quickly the curve flattens and how fast volatility memory fades, the vol-of-vol deepens smiles and increases the convexity uplift, and the correlation sets the sign and intensity of skew.

The initial variance sets the short end and, relative to the long-run level, determines the sign of the slope. This parameter map is not just academically tidy; it is operational. For example, if you observe an unusually steep upward at-the-money term structure on a quiet morning, the model’s story would be: initial variance is below its long-run level, the gap is sizeable, and mean reversion is not so fast that the surface is already flat—hence the slope.

Analytic Solution

For completeness, we must mention that the model remains analytically tractable because the log price has an exponential-affine characteristic function, which lets you price vanillas via Fourier inversion. The details are standard, but the critical point for intuition is that tractability comes from the variance process being affine in its own state; all of the term-structure logic above survives regardless of whether you compute prices by quadrature or by an FFT.

Over very short maturities, a small-time expansion further clarifies the initial slope. The leading-order term is simply today’s variance; the next term contains the mean-reversion pull toward the long-run level and a leverage tweak. When today’s variance is below the long-run level, that next term is positive, which is exactly the upward initial slope you see on quiet days.

Taken together, the Heston model’s appeal is its minimalism with punch. By turning variance into a mean-reverting, autocorrelated, stochastic state, it reproduces the key term-structure patterns you see on trading screens. Today’s variance sets the short end; mean reversion and averaging set the long end; autocorrelation, convexity, and leverage draw the curve between them. And because the model is both interpretable and tractable, it gives you levers - long-run level, mean-reversion speed, vol-of-vol, and correlation - that map cleanly to surface features you care about in practice.

Disclaimer

The information provided on TheLogbook (the “Substack”) is strictly for informational and educational purposes only and should not be considered as investment or financial advice. The author is not a licensed financial advisor or tax professional and is not offering any professional services through this Substack. Investing in financial markets involves substantial risk, including possible loss of principal. Past performance is not indicative of future results. The author makes no representations or warranties about the completeness, accuracy, reliability, suitability, or availability of the information provided.

This Substack may contain links to external websites not affiliated with the author, and the accuracy of information on these sites is not guaranteed. Nothing contained in this Substack constitutes a solicitation, recommendation, endorsement, or offer to buy or sell any securities or other financial instruments. Always seek the advice of a qualified financial advisor before making any investment decisions.

The most convincing modern reading of the evidence is therefore an interaction story: information quality, investor processing, and market frictions jointly determine how quickly public news gets reflected in price.

Classic PEAD

In 1968, Ray Ball and Philip Brown published a An Empirical Evaluation of Accounting Income Numbers showing that after the announcement, prices kept moving in the direction of the surprise. Not a lot. Not instantly. Just a slow, persistent drift, upward for good news, downward for bad, that continued for at least 60 days. This was the first documentation of post-earnings-announcement drift (PEAD), and it became one of the most replicated anomalies in all of empirical finance.

Victor Bernard and Jacob Thomas sharpened the finding across two landmark papers in 1989 and 1990. Using data from 1974 to 1986, they showed that the spread in average returns between stocks in the top and bottom deciles of standardized unexpected earnings (SUE) was positive in 41 of the 48 quarters in their sample.

The magnitudes were large. Bernard and Thomas (1990) found that zero-investment portfolios formed on SUE deciles generated abnormal returns of approximately 8–9% per quarter, roughly 35% annualized before transaction costs. Later studies have documented the range of annual PEAD returns between roughly 8.76% and 43.08%, depending on sample period and methodology.

A particularly striking feature was that about 25–30% of the total post-announcement drift was concentrated in just the three-day windows surrounding the next quarterly earnings announcements, despite those windows representing only about 5% of total trading days, suggesting that the drift wasn’t random but instead had a structure.

They documented that quarterly earnings follow a systematic autocorrelation pattern. Seasonal differences in earnings (comparing the same quarter year-over-year) show positive autocorrelations for the first three lags that decline in magnitude , approximately 0.34 at lag one, 0.19 at lag two, and 0.06 at lag three. In other words, if a company beats earnings this quarter, it will probably beat the next three quarters too (by shrinking amounts).

Bernard and Thomas concluded that PEAD occurs because investors fail “to recognize fully the implications of current earnings for future earnings.”

This is important because it reframes PEAD not as a market plumbing problem but as a reading comprehension problem. The information is public. The pattern is stable. But parsing its implications in real time, across hundreds of firms, is hard cognitive work. And that work is where the drift comes from.

Complexity Increases Delayed Reaction

If the core issue is interpretation difficulty, then anything that makes interpretation harder should make the drift larger. The empirical evidence strongly supports this.

Attention Scarcity: Hirshleifer, Lim, and Teoh (2009) tested the investor distraction hypothesis directly. On days when many firms announce earnings simultaneously, the immediate price reaction to any single firm is weaker and the subsequent drift is stronger. Even purely public information gets systematically mispriced when there is simply too much of it arriving at once.

Delayed Disclosure: Li, Nekrasov, and Teoh (2020) found that when companies delay full financial-statement disclosure until their 10-Q filing (rather than including it in the initial earnings announcement), the drift increases dramatically. Moving from zero to full delayed disclosure corresponds to approximately a 37% decrease in the immediate earnings response and a 42% increase in PEAD.

Limits to Arbitrage: Research documents that PEAD is consistently more pronounced for stocks with higher transaction costs (wider bid-ask spreads), lower share prices and trading volume, lower institutional ownership, less analyst coverage, and higher idiosyncratic volatility. In other words: the harder a stock is to analyze and trade, the longer the drift persists.

Risk-Based Explanations

Not everyone agrees PEAD is purely behavioral. Risk-based explanations suggest the drift is compensation for bearing some form of underlying risk. In particular, it is true that the log price-dividend ratio is convex with respect to expected growth rates. This convexity implies that stock returns are more sensitively affected by changes in expected growth when expected growth is higher.

Classic PEAD Fades

The magnitude of classic numerical PEAD has declined substantially over time. The spread in average returns between high- and low-SUE portfolios fell from about 5% in the 1980s/1990s to 3% or lower by the late 2010s. Charles Martineau (2022), in a paper provocatively titled Rest in Peace Post-Earnings Announcement Drift, showed that PEAD began disappearing from non-microcap stocks around 2001 and was essentially zero for large-cap stocks by 2006.

The causes are structural: decimalization made electronic arbitrage more precise; Reg NMS in 2005 accelerated high-frequency trading; prices began fully adjusting on announcement day. The Sarbanes-Oxley Act (SOX) also played a role; post-SOX, market reactions show faster immediate adjustment and lower abnormal returns in the +2 to +60 day window, consistent with improved financial reporting quality, stronger audit committees, and CEO/CFO certification requirements.

Kettell, McInnis, and Zhao (2022) add an important nuance: declining earnings persistence is itself a key driver. The persistence of SUE has fallen over time, and after controlling for this, the declining trend in PEAD becomes statistically insignificant. In other words, part of the reason PEAD shrank is that earnings surprises became less predictive of future earnings.

Complex PEAD Remains

Meursault, Liang, Routledge, and Scanlon, in a Philadelphia Fed Working Paper titled PEAD.txt, constructed a purely text-based analogue of the earnings surprise. Their measure, SUE.txt, is derived entirely from earnings-call language, without using the reported earnings number at all.

The result was that SUE.txt generates a larger post-announcement drift than classic numerical PEAD at every calendar-quarter mark. The PEAD.txt portfolio generates larger alpha than the traditional PEAD portfolio within the Fama-French five-factor-plus-momentum framework, and remains economically significant in recent years when classic PEAD is close to zero.

The researchers note that text and numbers “compress primitive data in different ways and are neither completely orthogonal nor identical.” The meaningful distinction isn’t between different types of information, but rather it’s between structured and unstructured form.

The easy, numbers-based underreaction got competed away first. The harder, language-based underreaction survived. That’s exactly what you’d expect if the bottleneck is interpretation of messy information, not access to data.

From Quarterly Earnings to the Entire Internet

PEAD was the scheduled, quarterly, single-number version of a much more general phenomenon: markets are slow readers. The empirical record now tells a clear story:

1. Public information that is easy to parse (a single EPS number) eventually gets priced efficiently once markets mature, infrastructure improves, and arbitrageurs target it.

2. Public information that is hard to parse, unstructured text, fragmented signals, complex context, continues to generate persistent drift even in today’s markets.

3. Anything that increases interpretation difficulty, distraction, delayed disclosure, illiquidity, low analyst coverage, high idiosyncratic volatility, amplifies the drift.

4. AI systems can already extract predictive signal from both text and visual representations of earnings data, often outperforming human analysts.

Now we generalize. The world doesn’t hand traders a neat quarterly EPS surprise every 90 days. It hands them screenshots on X, rumors in Telegram groups, app-store rankings, GitHub commits, and offhand comments in podcasts. Each of these is an information event. Each one has a “surprise” component.

Onchain Markets & Delayed-Reaction Effects

The PEAD literature gives us a precise checklist of conditions that make drift worse:

Lower liquidity ✓ Lower institutional ownership ✓ Less analyst coverage ✓ Higher volatility ✓ Greater information asymmetry ✓

Liquidity is bad. Most prediction markets, equity perps, and tokens trade with a fraction of the depth of even a small-cap U.S. equity. Thin order books are the single strongest predictor of persistent post-announcement drift in the academic record.

Price reaction is delayed by design. Equity perp oracles must bridge the gap between TradFi trading sessions and 24/7 onchain availability. When external price feeds go dark, oracles fall back on internal smoothing mechanisms, exponentially weighted moving averages of their own order flow. That is a structural lag baked into the market.

There is virtually no institutional ownership. The largest allocators, such as pensions, endowments, sovereign wealth funds, the multi-strat hedge funds, are almost entirely absent from onchain markets. The PEAD literature is explicit that low institutional ownership is one of the strongest correlates of persistent drift These are retail-dominated venues.

Analyst coverage is near zero. No sell-side research desk publishes models on DeFi assets. The limits-to-arbitrage research shows that less analyst coverage means slower information incorporation and larger drift.

Idiosyncratic platform risk is high. Oracle failures, smart-contract exploits, governance attacks, depegs, bridge hacks; these are platform-specific risks with no analogue in traditional equity markets. High idiosyncratic risk is another condition the literature directly associates with stronger PEAD.

Information asymmetry is extreme. Many onchain assets have no mandatory disclosure, no standardized reporting, and no audited financials. The information that moves prices is fragmented, unstructured, and unevenly distributed.

These are not bugs to be fixed before onchain markets “mature.” They are the current reality, and they describe an environment where PEAD-style logic, slow reading, persistent mispricing, exploitable drift, should be at its most alive. These markets don’t need less drift. They need better readers.

The Application of AI

Historically, the ability to read raw information, form a view, and express it in a market belonged to hedge fund PMs, prop desks, and a small number of discretionary traders. The toolkit was expensive: Bloomberg terminals, expert networks, proprietary NLP pipelines, teams of sector analysts.

PEAD persisted for decades in public equities partly because not enough people could do the reading. AI changes the economics entirely. The natural product implication is to give every user the capabilities of a hedge fund PM: AI analysts that ingest information across every source, social feeds, onchain data, news, community sentiment, product signals, filings, synthesize it in real time, and surface actionable trades (incidentally, this is what we are doing at Freeport).

That doesn’t eliminate markets. It makes them better. More participants interpreting information means faster price discovery, better capital allocation, and the creation of entirely new market types around events, narratives, attention, and emerging realities that older financial rails are too slow and too rigid to support.

Disclaimer

The information provided on TheLogbook (the “Substack”) is strictly for informational and educational purposes only and should not be considered as investment or financial advice. The author is not a licensed financial advisor or tax professional and is not offering any professional services through this Substack. Investing in financial markets involves substantial risk, including possible loss of principal. Past performance is not indicative of future results. The author makes no representations or warranties about the completeness, accuracy, reliability, suitability, or availability of the information provided.

This Substack may contain links to external websites not affiliated with the author, and the accuracy of information on these sites is not guaranteed. Nothing contained in this Substack constitutes a solicitation, recommendation, endorsement, or offer to buy or sell any securities or other financial instruments. Always seek the advice of a qualified financial advisor before making any investment decisions.

We consider a realistic example of 10% drift in equity index with a 2% risk-free rate. The numerical ceiling for max compound growth rate is around 14.5% above cash. A continuum of long calls or a continuously rebalanced ~312%-leveraged stock position achieves this exactly; a single five-year ATM LEAPS captures 93% of it. Information theory’s measure of statistical distinguishability turns out to be, in finance, the measure of value of an edge.

The material for this post was originally my final presentation for a graduate information theory course at university. Most of the mathematical results are drawn from existing literature, except the framing of single-LEAPS selection as a constrained KL projection with, which is my original mathematical contribution.

Introduction

This post walks through the question in a few steps.

(1) We set up the disagreement: the market prices under one measure Q, the trader believes another P, and we ask what the maximum extractable rate is given that gap.

(2) We solve the Kelly problem in a complete market and find that the growth-optimal payoff is, up to a constant, the likelihood ratio dP/dQ

(3) Plugging that payoff back into the objective shows that max growth rate is given by r + KL(P‖Q)/T i.e. value of an edge is its KL divergence from the market.

(4) We consider the special case of Black–Scholes, where dP/dQ becomes a clean power function of the terminal stock price, and show three different ways to actually trade it: a Carr–Madan continuum of long calls, a continuously rebalanced leveraged stock position, and a single call option.

(5) We frame single-option selection as a constrained KL projection, derive a tail-matching first-order condition for the optimal strike, and show that one ATM five-year call captures about 93% of the theoretical ceiling. The closing comparison plots terminal wealth distributions across all five strategies side by side.

Risk Neutral versus Real Distributions

As explained in one of my previous posts in greater detail, markets price contingent claims under a measure Q called the risk-neutral or martingale measure. Under Q, discounted asset prices are martingales, which is just the statement that option prices admit no arbitrage. The trader has a different belief, the physical or subjective measure P. If P = Q there is no edge. If P does not equal Q, there is an edge, and the question of this essay is: what is the maximum compound growth rate that edge can deliver, and what payoff structure realizes it?

Consider the following example with one risky asset under Black–Scholes:

With risk-free rate r = 2% and trading horizon T = 5 years. The trader believes the drift is μ = 10%; the market prices options as if the drift were 2%. Volatility is agreed. The disagreement looks like this in the distribution of the asset S at time T:

Both densities are lognormal with the same log-variance. Only the means differ:

The trader’s distribution is shifted upward by (μ - r)T = 0.40 in log-units, putting more mass on outcomes above 126. Equivalently, the same Brownian motion produces different price paths under the two measures because the drift is different. Girsanov’s theorem makes this precise:

is a Brownian motion under Q, where

is the market price of risk. For our parameters theta = 1/2.

The two panels show six paths of the same underlying noise simulated under each drift. The shape of the noise is identical between the two, only the drift is different. Everything that follows can be read as: how much money is the trend gap worth, and what’s the cheapest way to extract it.

The Kelly Problem

Among all contingent claims W paying off at time T that cost W0 today, find the one that maximizes expected log-growth under the trader’s measure P:

The constraint is the no-arbitrage budget: every claim has a price, and that price is its Q-expected discounted value. The objective is the Kelly criterion (Kelly, 1956). Log utility is the right preference for compounding wealth because terminal wealth is multiplicative across reinvestment periods, so log is the unique utility for which one-period optimization gives the asymptotically growth-optimal strategy (Algoet and Cover, 1988). The market is assumed complete. We note that this is a constrained variational problem. Form the Lagrangian:

Write the expectations as integrals against densities p and q. The Lagrangian becomes

The first-order condition in W(s) (pointwise functional derivative) is

Plug into the budget constraint to solve for λ:

The optimal payoff is therefore

This is the central object of the entire post. The growth-optimal contingent claim is, up to a constant, the Radon–Nikodym derivative dP/dQ. Where the trader’s density exceeds the market’s, the optimal payoff is large; where it is smaller, the optimal payoff is small. The optimal payoff is the likelihood ratio. (This is the classical duality solution to utility maximization; see Karatzas and Shreve, 1998, Ch. 3) Visualized for our example, dP/dQ as a function of S at time T looks like this:

It is identically 1 at around 126 in our example (where P = Q), small below, and growing steeply above. The Kelly-optimal trade pays you a multiple of p/q at every state; implicitly it is telling us that we shoud load up where the market is underpricing and take nothing where it isn’t.

The Kullback-Leibler Divergence

Plug W* back into the objective:

The Kullback–Leibler divergence of P from Q is defined as

So the maximum expected log-growth, divided by time, is

This is a core result. The compound growth rate available to the optimal Kelly investor equals the risk-free rate plus the relative entropy rate between the trader’s view and the market’s. KL divergence, a purely information-theoretic object, is literally the value of the trader’s edge, denominated as a compounding rate.

Nothing about this part requires normality, continuous-time trading, or other Black–Scholes assumptions; it requires only a complete market and log utility. The original information-rate version of this result is in Kelly (1956); Cover and Thomas (2006, Ch. 16) develop the modern portfolio-theory restatement. The integrand tells you which stock outcomes contribute most to the divergence:

There is a negative region around spot, where p < q (the trader thinks these outcomes are less likely than the market), and a much larger positive region where p > q. The integral over the whole range is around 0.625 nats which gives around 14.5% max excess compounded annual returns attainable from the risk premium (note this is higher than the 8% risk premium from just buy and hold). We show this below.

Specializing to Black–Scholes

To compute the optimal payoff and growth rate explicitly in our example, we need dP/dQ as a function of S(T). Recall that the variance of these distributions are agreed-upon by both the trader and the market, and that only the means are different. Recall also that the density ratio of two same-variance normals is

for some normalizing constant c. The likelihood ratio in Black–Scholes is a power function in S(T). The exponent

is the squared-Sharpe-per-variance which will reappear in three more places before we are done. The KL divergence has a closed form and can be computed directly using the lognormal moments:

So the Black–Scholes Kelly–KL identity is

That is the hard ceiling. No strategy can beat it in expected growth rate. In our case

The Optimal Payoff

We now substitute

back into the optimal payoff formula so see that per dollar invested

where the normalization

ensures the contract is priced at $1 today. The lognormal moment

is closed-form, giving

for our β = 3.125. So the Kelly-optimal payoff per dollar is approximately

How does this compare to other strategies a trader might run? The chart below shows terminal wealth per dollar invested as a function of S at time T for five strategies:

Cash is flat at around 1.10. Stock buy-and-hold is linear. A single ATM call plus cash is a hockey stick. The static call strip and dynamic Kelly are the same smooth power curve, plotted on top of each other. The strip and Kelly curves dominate single-stock and single-LEAPS aggressively in the upper tail, where the trader’s edge lives.

The LEAPS Call Approximation

Plot the integrand p(s) log W(s) for both the optimal payoff and ATM call:

Both strategies generate negative contribution at low prices. Both also generate positive contribution in the upper tail. The differences is that the single call dips much further negative around 100 in the expired worthless region where the call paid nothing and you only have your cash residual. The optimal payoff is positive everywhere above roughly 130 and tracks the call closely from there up. There is however still a slight difference in the area between the curves. This is what the edge lost by suboptimal execution looks like geometrically.

Implementing W* with Vanilla Calls

The optimal payoff W* is an exotic and doesn’t trade. To realize it, the trader has to build it from instruments that do trade. The Carr–Madan static replication theorem says that for any twice-differentiable payoff H(S) with H(0) = H’(0) = 0, the identity

decomposes H as a portfolio of long calls. The proof is integration by parts:

when the boundary terms vanish. In our case

Differentiating twice shows that the Kelly-optimal portfolio holds calls at strike K with density

For our example

This shows that the further out of the money, the more contracts. (Each individual deep-OTM call is cheap; you can afford a lot of them per dollar.) The complete portfolio is a continuum of long calls; no stock, no puts, no cash. The replication identity is due to Carr and Madan (1998). We can observe convergence to the exact target as we increase the number of strikes:

The contract density keeps rising with the strike but individual call prices fall fast. So the capital density i.e. dollars deployed per unit strike is actually hump-shaped:

Most of the dollars go into a broad mid-strike region peaking around 60, well below spot, even though the contract count is heaviest in the wings.

Single LEAP as a Constrained Projection

In practice traders rarely buy a 30-strike strip. The realistic question is: given that you’re going to buy one call, which strike, and how much of your bankroll? This is a constrained version of the Kelly problem. Restrict the admissible payoffs to the two-parameter family generated by a call at strike K plus cash with fraction f in the call:

where

is the Black–Scholes call price. The normalized payoff

satisfies

implying that strategy W satisfies

Note that

implying

which gives

Define the call-induced measure tilt Q by

which we can show to be a valid probability measure (proof left as an exercise to the Substack reader). The expected log-growth under P is

Apply

to calculate the expectation as

Combining the above gives

In words, the result shows that choosing the strike and position size of a single call is exactly the problem of finding the call-generated probability measure closest to the trader’s true measure in KL divergence. The 2D landscape of G over the full (K, f)-space looks like this:

The maximum is in the upper-left region, with f approximately being 0.76 at strikes near spot, and decays smoothly in all directions. The optimum achieves roughly 13.55% excess returns a year. Cutting along the optimal f*(K) ridge gives the one-dimensional growth profile:

From the left, we see that the peak is around 100 numerically and the curve is broad such that anywhere from 90 to 110 delivers within 10 bps of optimal. The right panel shows two opposing recommendations under different objectives: linear expected return (orange) grows monotonically with K because deeper-OTM calls have higher payoff multiples, while the optimal Kelly fraction (purple) falls because log utility punishes the rising probability of finishing worthless.

The first-order condition for the optimal strike also has a clean form. Differentiating G in K (envelope theorem on the inner f-optimization) and using

gives, after simplification

where the right hand side is the physical measure tilted by

Thus, the optimal strike is the one where the risk-neutral digital probability of finishing in the money matches a Kelly-tilted physical digital probability. There is no closed-form solution; it must be solved numerically, and the answer is parameter-dependent. ATM is NOT generically optimal, it just happens to be near-optimal under “reasonable” parameters.

Why a Single Call Works

The reason a static long call does so well, capturing 93% of the dynamic Kelly edge with one trade, is convexity. Under Black–Scholes, the dynamics of any option price can be derived by Itô’s Lemma. Starting from the BS PDE

and applying Itô to C under P:

Dividing by C and writing

for the call’s elasticity (also called leverage or lambda) shows

Omega is equal to the Spot times Delta divided by the Call price and measures the option’s elasticity / leverage i.e. the percentage move in the option price per 1% move in the stock. If the stock rises 1%, an option with Omega = 5 rises 5%. ATM short-dated options typically have Omega in the 10–30 range; deep ITM options have Omega near 1; deep OTM options can have Omega in the hundreds.

Three observations. First, the option’s expected excess return is Omega times (mu − r) and its volatility is Omega times sigma, so the instantaneous Sharpe ratio is exactly (mu − r) / sigma which is the same as the stock. Second, Omega varies with the underlying: Omega = S · Delta / C and as the stock rises, Delta rises (due to positive Gamma everywhere). Third, in the continuous-rebalancing limit at Kelly fraction is (mu − r) / (Omega² · sigma²) for an option, the achievable growth rate r + (1/2) · (mu − r)² / sigma² is strike-independent.

For a static long call, the leverage profile Omega(S) does the work that dynamic rebalancing would have to do explicitly. As S rises the position naturally levers up; as S falls it de-levers. This is “synthetic momentum trading” baked into the contract. But the profile is fixed by the Black–Scholes formula and is concentrated near the strike. The Kelly-optimal target is a smooth power that grows without bound. A single call’s Delta(s) is a sigmoid that saturates at 1 once s is far ITM.

The orange call delta approaches 1 and saturates around 130, after which the call cannot lean any further into a continued rally. The blue Kelly target slope has no such ceiling. However, note that a single call gets ~93% of it because the orange and blue curves agree in the middle of the action around the 80 to 140 range where most of the distribution lives anyways, and only diverge in the deep tails.

The Dynamic Alternative

There is a third way to realize the optimal payoff, neither static-call-strip nor single-LEAP. Hold no options at all; instead, continuously rebalance a position in the underlying stock. I have written about this before. Let pi be the fraction of wealth in stock and 1 - pi in the risk-free asset, rebalanced continuously. Wealth dynamics are

Apply Itô to log W

Maximizing the drift over pi gives the Kelly fraction

The maximum drift is

the same expression as the KL bound. That is the third place beta appears: as the optimal leverage. This is the log-utility special case of the classical Merton fraction (Merton, 1969); see also Long (1990) for the numéraire-portfolio interpretation. To see what terminal wealth this strategy produces, integrate d log W

The Brownian term on the very right hand side is recoverable from the stock SDE

Substituting and simplifying

Equivalently

The dynamic Kelly strategy produces a terminal wealth that is a deterministic function of the terminal stock price under Black-Scholes assumptions. Note that even if the stock ends where it started, continuous Kelly loses money because holding leverage constant forces buying after rallies and selling after drawdowns, skimming wealth on every round trip through volatility. The fact that final wealth depends only on the terminal state (and not the path) is purely an artifact of the assumption that rebalancing is continuous and costless; in practice, even daily rebalancing introduces meaningful path dependence.

Conclusions

Comparing terminal wealth distributions across paths, with each strategy’s bin masses normalized so they sum to 100%

Cash is a delta function (the clipped 100% spike). Buy-and-hold stock gives a tighter lognormal that is moderately wider than cash but with much less upside than the option strategies. Single LEAP has a discrete probability mass near from expired worthless paths (~11% of the time) plus a smooth right-tail distribution that dominates buy-and-hold stock for most upside outcomes.

Strip and dynamic Kelly are visually identical broad smooth distributions, exactly as predicted by their being mathematically identical contracts. The Kelly/strip distributions are wider than the call in both directions, they pay much more on the upside but also leave you with less than the call when the stock falls hard, because the leverage embedded in the convex payoff cuts both ways.

Reading across the derivation, the key results are:

Note that the identity

is distribution-free. It does not require normality, continuous-time trading, or any specific market model. It says: in a complete market, the maximum compound growth rate equals the relative entropy rate between the trader’s measure and the market’s.

Whatever the source of the trader’s edge, fundamental research, alternative data, microstructure, behavioral views, its dollar value as a continuous compounding rate is the KL divergence. What you are paid for is distinguishability which is measurable from an information theory point of view.

The optimal payoff is always the likelihood ratio dP/dQ. Black–Scholes is the special case where the answer is a power function and a stack of calls reproduces it.

References

The Kelly criterion and the information-theoretic value of an edge

Kelly (1956), “A New Interpretation of Information Rate,” Bell System Technical Journal 35(4): 917–926.

The original paper. Introduces the log-optimal capital-growth rule and proves that for fair-odds gambling, the maximum exponential rate at which capital can compound equals Shannon’s information rate, the foundational form of the identity between an information-theoretic quantity (rate at which a signal carries information) and a financial quantity (rate at which wealth grows).

Cover & Thomas (2006), Elements of Information Theory, 2nd ed., Wiley, Ch. 16.

The standard modern textbook treatment. Generalizes Kelly’s result from horse-race gambling to log-optimal portfolios in arbitrary asset distributions, identifies the maximum compound growth rate with the relative entropy between the trader’s distribution and the market’s, and shows that the optimal portfolio weights are proportional to the trader’s probability over the market’s at each state.

Algoet & Cover (1988), “Asymptotic Optimality and Asymptotic Equipartition Properties of Log-Optimum Investment,” Annals of Probability 16(2): 876–898.

Proves that the log-optimal strategy is asymptotically optimal in a strong path-by-path sense, not just on average, no causal investment strategy can outperform it in the long run, regardless of the underlying return process. Justifies why log utility (rather than any other utility) is the privileged criterion when wealth compounds.

Bell & Cover (1980), “Competitive Optimality of Logarithmic Investment,” Mathematics of Operations Research 5(2): 161–166.

Establishes a separate optimality property of the log-optimal strategy: it cannot be beaten in head-to-head wealth comparisons more than half the time, against any opposing strategy. A third angle on why log utility is special.

Optimal payoff and the likelihood ratio (utility-maximization duality)

Karatzas & Shreve (1998), Methods of Mathematical Finance, Springer, Ch. 3.

The cleanest textbook derivation of the martingale/duality method: maximizing expected utility of terminal wealth subject to the no-arbitrage budget constraint reduces to a one-dimensional Lagrangian, with the optimal contingent claim equal to the inverse marginal utility evaluated at a multiple of the market’s likelihood ratio. For log utility this collapses to the optimal terminal wealth being proportional to the trader’s probability density divided by the market’s.

Kramkov & Schachermayer (1999), “The Asymptotic Elasticity of Utility Functions and Optimal Investment in Incomplete Markets,” Annals of Applied Probability 9(3): 904–950.

Generalizes the duality / likelihood-ratio characterization of optimal investment to general semimartingale markets, including incomplete ones, and establishes the conditions on utility functions (”asymptotic elasticity”) under which a clean duality solution exists.

Continuous-time portfolio choice (the Kelly fraction)

Merton (1969), “Lifetime Portfolio Selection under Uncertainty: The Continuous-Time Case,” Review of Economics and Statistics 51(3): 247–257.

Derives, for the first time in continuous time, the optimal constant fraction of wealth to hold in a risky asset. For a log-utility investor in a Black–Scholes economy the answer is the asset’s excess return divided by its variance.

Merton (1971), “Optimum Consumption and Portfolio Rules in a Continuous-Time Model,” Journal of Economic Theory 3(4): 373–413.

Extends to multiple risky assets, giving the optimal vector of asset weights as the inverse covariance matrix times the vector of excess returns over cash.

Change of measure (Girsanov) and the closed-form KL divergence in Black–Scholes

Karatzas & Shreve (1991), Brownian Motion and Stochastic Calculus, 2nd ed., Springer GTM 113, Section 3.5.

The rigorous probability-theory treatment of Girsanov’s theorem, the result that lets you change the drift of a Brownian motion by switching probability measures, with an explicit formula for the likelihood-ratio process between the two measures.

Static replication of nonlinear payoffs (Carr–Madan)

Carr & Madan (1998), “Towards a Theory of Volatility Trading,” in R. A. Jarrow, ed., Volatility: New Estimation Techniques for Pricing Derivatives, Risk Books, pp. 417–427.

Establishes the static replication identity that any twice-differentiable European payoff can be written as a weighted integral over strikes of vanilla call payoffs, with the weight at each strike equal to the second derivative of the target payoff.

The numéraire portfolio (the power-curve payoff)

Long (1990), “The Numéraire Portfolio,” Journal of Financial Economics 26(1): 29–69.

Introduces the numéraire portfolio: a self-financing trading strategy whose value, when used as the unit of account, makes the discounted prices of all other assets into martingales without any change of measure. In a Black–Scholes economy the numéraire portfolio’s value is proportional to the stock price raised to a power equal to the excess return divided by the variance.

Disclaimer

The information provided on TheLogbook (the “Substack”) is strictly for informational and educational purposes only and should not be considered as investment or financial advice. The author is not a licensed financial advisor or tax professional and is not offering any professional services through this Substack. Investing in financial markets involves substantial risk, including possible loss of principal. Past performance is not indicative of future results. The author makes no representations or warranties about the completeness, accuracy, reliability, suitability, or availability of the information provided.

This Substack may contain links to external websites not affiliated with the author, and the accuracy of information on these sites is not guaranteed. Nothing contained in this Substack constitutes a solicitation, recommendation, endorsement, or offer to buy or sell any securities or other financial instruments. Always seek the advice of a qualified financial advisor before making any investment decisions.

The piece below is drawn from a long conversation with one such user, who generously agreed to let us write it up. All names and personally identifying details have been changed; the piece has been stylistically edited for readability; and while we trust our source, we cannot independently verify every claim he makes in the course of a day’s conversation.

Where possible, we have grounded the narrative in publicly available onchain data, and all figures in this piece are produced from public data for illustrative purposes and does NOT constitute constitute investment, legal, tax, or any other form of professional advice.

Sea Cliff

Sea Cliff at quarter past eight on a Saturday. The fog has pulled down so low over the Pacific that the Golden Gate is more idea than bridge, you can’t see it, but you know it’s there. The hedges on El Camino del Mar are glistening with damp. In this light, the stucco houses behind them look briefly like something other than stucco; softer, almost fabric. A foghorn sounds, twice and then twice again.

Trader 12 had asked me over for coffee and pastries. He opens the door in a linen shirt, barefoot. Chemex on the counter, Arsicault croissants on a parchment-lined plate. His day job is at a venture-backed AI hedge fund downtown, the kind where the machines do the trading. Saturdays, though, are for the side gig. “On weekends, markets no longer sleep” he says, pouring two cups of coffee. “You just have to figure out what’s moving it.”

The Inference Problem

For a few dozen hours from Friday close to Monday open, Korea is dark, Japan is dark, Wall Street is dark. The only live tape on EWY, EWJ, and a handful of American chips and hyperscalers are perps onchain @HyperliquidX. Someone has to price those names to the Monday open.

“The trap most people fall into,” he says, “is to look at the S&P and then just multiply by a beta someone fit on a year of data, but that’s not how markets work.” (A beta is just a multiplier. If Broadcom’s beta to the S&P is 1.3, the rough rule says Broadcom moves 1.3% for every 1% move in the S&P. )

He sketches on a napkin. “Three prices are relatively liquid on perps over the weekend: SPX (US market), NDX (excludes financials and energy), and CL (crude oil). But those three numbers aren’t the real drivers. They’re summaries of something underneath, behind them sit what we call factors.”

“A factor is just a hidden force that moves a lot of things at once. You can’t trade it directly, you can only see it through the prices that respond to it. The reason factors are useful is that most of what happens in the world is actually explained by relatively few of them. Thousands of stocks bounce around every day. But if you take a few well-chosen forces and ask how each stock responds to each one, you can explain most of what you see. The rest is company-specific noise.”

“First factor is macroeconomic growth. How fast the world economy is actually expanding. When it’s down, everything bleeds together.”

“Second is real rates. Long-term interest rates after you subtract inflation. The companies that get hit hardest when real rates rise are the ones whose profits sit far in the future.”

“Then there’s stuff like country and sector, many smaller stories bundled together. Korean memory, Japanese autos, Taiwanese foundries, US hyperscalers. Each reacts to the same news differently. An oil spike crushes Asian manufacturers because they import almost all their energy. For US names it’s a shrug since shale made the U.S. an energy exporter.”

“There are a handful more I’m not naming. The three I just gave you make the point. The point is that there are more factors than there are liquid prices.”

“Nonetheless, you can use the liquid tape as a constraint. SPX, NDX, and WTI trade hundreds of millions of dollars a day over weekends. Treat those three as efficient. Whatever values the factors take, they have to reproduce the liquid returns. That doesn’t pin them to one answer, but it cuts the space from infinite to a finite envelope of combinations consistent with the liquid tape.”

“Then you project that envelope onto the thin names that only trade a few million to dozens of millions a day in volume. Each combination inside the envelope implies a price for Samsung, for Hynix, for EWY. Across the whole envelope you get a range. The implicit bet is that the thin names are not efficient: Korea is closed, the perps trade thin.”

“There seems to be just one problem,” I said. “More factors than prices means many combinations of factor values can produce the same SPX, NDX, and WTI returns; and each combination, projected onto less liquid targets, like Samsung or Hynix, gives a different answer. The same liquid prints can be consistent with Samsung up two or Samsung down five, depending on which factor story is doing the work.”

“How do you decide?” I ask.

“News tells you direction. The envelope gives you magnitude. Direction without magnitude is a tweet. Magnitude without direction is noise. Both together is a trade.”

“The news being.”

“What the headline points at. An OPEC decision points at growth via oil, with importers and exporters reacting opposite directions. An Nvidia blowout points at country and sector. A Fed surprise points at real rates. The news doesn’t narrow the envelope to one answer. It tells you which corner of it the market wakes up believing.”

The Trading Problem

“In practice, the way I express a view is mechanically simple: I’m saying some less liquid symbol, Samsung, Hynix, EWJ, should move on a higher or lower beta to the liquid tape than the perp is currently pricing.”

“But the claim underneath really is about the factor drivers behind this or that particular move being different from the ones averaging into its prior. When Hormuz is the story, the relevant driver is energy intensity, and Asian importers should load harder. When earnings are the story, the relevant driver is hyperscaler capex, and memory should decouple from oil entirely.”

“The different beta is just the shadow that a different factor mix casts on the liquid axis. You’re never really betting on a coefficient; you’re betting on which factors are doing the work.“

“Let me give you some concrete examples.”

“First. The weekend the war began. Tehran and US rhetoric had been escalating all week; a strike package widely reported as imminent. The cross-section already told you who was exposed. Korea and Japan imports almost all its energy through Hormuz, the U.S. is a net exporter.”

“Saturday the strike landed. Joint U.S. and Israeli operation. Oil was up onchain with SPX and NDX down sightly; but the interesting thing was that asian tickers like EWY EWJ Samsung and Hynix all traded inline with American symbols even though they had higher exposure to energy shocks.”

“The amount that the less liquid Asian symbols were down was inconsistent with how much oil was up and how much the US market was down, so I shorted all of them. On Monday, Samsung opened down nearly ten. Hynix down eleven and a half.”

“Then, earnings season started. Financials went first, the way they always do. Strong credit expansion, consumer spending holding up, capital markets activity firming. Financials are the leading indicator for the real economy because they sit at the front of every transaction.”

“The intraday correlation between the US market, especially AI-exposed names, and oil started breaking down. Memory and photonics specifically, weren’t tracking oil intraday anymore. They were trading their own factor tied to hyperscaler capex. The decorrelation persisted, day after day.”

“You’d think a correlation isn’t tradable, but it is because the way that weekend perp traders trade doesn’t seem to update as fast as the weekday market does. When bad macro news lands on a Saturday, oil rips, and the perp sells chip names down on the old beta-to-SPX, you fade it.”

“Long the chips at the perp lows. Cover Monday open. The weekday markets already knows those chips don’t load on oil the way they used to. The weekend perp traders gets there eventually. You just get there first.”

“The factors themselves don’t change. Macroeconomic growth, real rates, country and sector. What changes is how each single stock loads on each one, and that drifts faster than most people give it credit for as investor narrative shifts. The hard part is picking which factors actually matter. The principle is simple. The construction is almost all of the work.”

Land’s End

As we continued to walk, the morning fog cleared and the bridge swung into full view. He filled our coffee into red Solo cups before we left the house, the only cups he had. Pretty degenerate look for two young guys at nine in the morning. As we walk along the beach, the Pacific stretches below us. A jogger goes by. A woman with a stroller. Two old men in windbreakers. Dogs chasing each other in the surf.

“All these people,” he says. “None of them is predictable on their own. The woman with the stroller, why she’s out here and not the other side of town. The runners. The two old guys. Each one is its own thing. Each one would tell you a long story about why they’re here and not elsewhere.”

“Sure.”

“But step back. Watch a year of this trail. Watch ten years. The aggregate is structured.”

“Same as the factor model.”

“Same deal. Markets are millions of agents trading their own thing for their own reasons, and each trade is unforecastable on its own. But the covariance structure of returns, what moves with what, collapses onto five or six axes that explain most of the variance. You don’t need to know what any individual trader is doing. You just need to know how each name loads on those few axes. Same thing with people on the trail. Same thing with most aggregates of independent agents.”

“Why does it collapse like that.”

“Because the world is shaped that way. The number of underlying drivers is much smaller than the number of things they push around. The geometry is low-dimensional even when the surface looks impossibly high-dimensional. Most people only see the surface. The trade is in the geometry.”

The path curves out toward the point. We keep walking.

Disclaimer

This is an article about a conversation with a trader, not a recommendation that you do any of what he does. Nothing here constitutes investment, legal, tax, or any other professional advice. Perpetual futures carry risks that include, but are not limited to: total loss of principal, counterparty failure, protocol failure, illiquidity, and regulatory change. Past results, including the backtest figures in this piece, do not predict future outcomes.

Access to Freeport and to the venues referenced in this article may be restricted in certain jurisdictions; it is your responsibility to comply with all laws and regulations that apply to you, including those governing access to DeFi protocols and derivatives. Freeport is a non-custodial interface. We do not hold your assets, we do not take the other side of your trades, and we merely allow access to third-party venues (including Hyperliquid) which we do not operate, audit, or endorse. Every protocol surfaced through Freeport carries its own smart-contract, counterparty, oracle, bridge, and operational risk, and we make no representation as to the safety, solvency, or ongoing availability of any of them.

Any AI-generated content in the Freeport app, summaries, feeds, trade ideas, and buttons, are informational and not advisory. Surfaced actions are shortcuts, not recommendations; they reflect computed signals, not fiduciary advice, and should be treated as a starting point for your own research rather than an endpoint.

The piece below is drawn from a long conversation with one such user, who generously agreed to let us write it up. All names and personally identifying details have been changed; the piece has been stylistically edited for readability; and while we trust our source, we cannot independently verify every claim he makes in the course of a night’s conversation.

Where possible, we have grounded the narrative in publicly available onchain data, and all figures in this piece are produced from public data for illustrative purposes and does NOT constitute investment, legal, tax, or any other form of professional advice.

This post was originally published on our Twitter has has been edited for display on Substack.

A Portrait of Trader 7

Trader 7 picks a restaurant near 79th and Columbus because he can walk there from his apartment and because he says the Parisian sea bass is the only thing above 72nd he isn’t bored of. He is already at the table when I arrive. Gray fleece zip-up over a Patagonia sun hoodie, the haircut expensive in the way that’s supposed to look cheap. I’ve met him exactly once before and I am still getting used to the fact that his name is not actually Trader 7.

He found me a month ago through a LinkedIn post. I’d written something, half serious, about the similarities of running an AEPi chapter to running a venture-backed startup, and he DMed me twenty minutes later. He was also AEPi, at a midwest school that I will not name because he does not want me to, and he thought the post was funny enough to download my app. That is the entire origin story. He says the word serendipity without irony, which is something I notice about him.

He grew up on the Upper West Side, two blocks from where we’re sitting. He traded at a Chicago prop-shop for three years and then quit, which is a thing he tells people within the first three minutes of meeting them. He now trades@Kalshi and @Polymarket full-time, and also @ventuals perps and @PreStocks tokens and anything else onchain where he can find edge.

Dinner, 7:34 PM [Fading Large Moves on Thin Books]

“Can I tell you about Christmas?”

“Please.”

“I was at my parents’ apartment. Two blocks that way” - he points west.

“My family was trying to get me to play Catan. But I get a notification on Anthropic perps spiking. The book was, I’m telling you, two dozen bids, three dozen asks, total depth maybe a few hundred grand of notional. Some guy walks in with market orders for a mil. The mark prints minus twenty-eight percent in an hour.“

“And the oracle.”

“The oracle can’t move. The oracle is half offchain and half onchain and the offchain component can’t even move! So the mark is in another zip code and the oracle’s barely budged. The funding rate crosses negative two thousand percent annualized on the hour.”

“So I fade the move and go long. In size. No thinking. The discount has to fade, it’s math. Thirty minutes, I cover half. Ninety minutes, I cover the rest.”

“How much did you make.”

“Business-class ticket,” he says. “To somewhere nice.”

The bread arrives. He orders the sea bass and an IPA. I order lamb over rice and a diet coke and try to write this down in my head without making it obvious that I am writing it down in my head.

Between Courses [Oracle & Funding, Explained]

The thing about Trader 7, and this is the thing you notice about anyone who has traded for a living, is that he loves talking about trading. He keeps explaining until you tell him, in plain English, that you understand; if you nod without saying it he takes the nod as a request to go again from the top. I haven’t said the words yet, so he describes the trade again.

Here is what a perp is. It’s a contract on an exchange that tracks the price of something else, in this case, Anthropic, a private company that doesn’t itself list a share anywhere. The contract doesn’t expire. Two people are on either end of it, a long and a short, swapping exposure to whatever Anthropic is worth. Because the thing it tracks doesn’t trade on the exchange, the exchange needs a rule for what the contract is supposed to be worth. That rule is called the oracle.

The oracle is not a person, it’s an average. Ventuals takes two numbers and splits the difference: an offchain feed of recent secondaries transactions in the actual shares, and a moving-average for whatever the perp itself last traded at. Half and half.

The result publishes every few seconds. The funding rate is what keeps the two sides honest. If the perp drifts rich against the oracle, longs pay a fee to shorts every eight hours. If it drifts cheap, shorts pay longs. Small gap, small fee. Big gap, big fee. The point is to make sitting on the wrong side of the gap expensive, so somebody walks in and closes it.

Thin-book moments break the system. The mark jumps because the mark is whatever the last market order printed. The oracle moves less since half of it is the offchain data, which doesn’t update. For the first few minutes the mark is significantly below the oracle and funding prints more than two thousand percent annualized in favor of shorts.

And the gap has to close. Funding is continuous. You get paid every eight hours, every day, as long as you’re on the right side. As long as the underlying secondary markets isn’t going down, and Anthropic, which just raised at three hundred and eighty billion at the time, is really not about to be marked down.

Somebody with a clear head eventually notices the exchange is offering a king’s ransom to sit on the right side of a mispriced trade. They come in. They take the short. The mark drags back to the oracle.

Funding falls. Trade ends.

It is not a story about what Anthropic is worth. It is a story about how long prices can stay dislocated without anyone paying attention. Christmas morning, the answer was roughly a few hours.

Bar, 9:20 PM [Basis Carry / Mean-Reversion]

We walk over to a bar on Amsterdam because he says they made good negronis there. The bar is half-full and nobody looks up.

“The other trade is the basis.”

“Explain.”

“Same company, two different venues. PreStocks token on Solana. Vntl perp on Hyperliquid. Different flows, different liquidity, different crowds. The number on one side drifts away from the number on the other side. Sometimes it drifts a lot. And, in theory, it should come back.”

“In theory.”

He shrugs at the ceiling in that way you do when you have been losing arguments with yourself for a while. “Look. I think there’s edge here. I’m not sure there’s edge here. What I know is that when I am on this trade and funding is in my favor, I get paid to wait, and the waiting is most of the P&L.”

“How much of the P&L?”

“More than half of it. I make more on the funding than I do on the prices actually converging. If I stripped the funding out I’d be about half of what I’m up on the basis book overall.”

“Which is.”

“Call it low-forties. On prices alone, low-twenties. The rest is the funding carry. And the composition is bumpy, I’m currently down on SpaceX but my Anthropic position has made more than enough back”

Walking up Amsterdam, 11:40 p.m. [Buy & Hold]

After a few rounds of drinks, we pay and walk north on Amsterdam in search of additional food, because the sea bass, while excellent, was portioned for people who’ve never been hungry.

There is a Joe’s on 74th. We agree on a slice.

“Let me tell you what actually made me the most money.”

“I just held Anthropic and SpaceX spot. PreStocks. Bought in October, never sold. That’s the trade.”

“Not fading the move on Christmas?”

“That trade was great, had a Sharpe of four and a half. But it’s not every day that someone comes in and smashes through the orderbook. I’ve been long Anthropic and SpaceX spot every hour since October.”

Joe's Pizza [Trading Funding Round Announcements]

Around midnight we’ve finished our first slice and ordered a second because we are still hungry and because the first one went down too quickly to count. The second one is hotter and we eat it more carefully.

“The Microsoft-Nvidia thing. November 18th.”

“I was long perps and spot when it came out. Not just that one, but I went long on most positive funding and partnership headlines .”

“How did you know.”

He points at his phone.

“You have a group chat?”

“I have a group chat with some college friends, yes. But I also have” - he opens up my app - “Freeport on my phone. The SpaceX IPO headline was on my lock screen before it was on Bloomberg.”

“The rule,” he says, “is you are probably early enough if you long within a few minutes of the headline coming out, and you definitely want to sell within at most a few days. Flow keeps coming in days after the announcements from folks who FOMO and that’s your exit.”

“Do you ever miss on these?”

“Some of these are real. Sometimes the rumor dies. Sometimes the round gets pushed. Sometimes the price just doesn’t move. But the point is that it works more often than it doesn’t, and your risk is asymmetric”

Joe’s, More Slices [Launch-Day Pumps]

It is now something like 12:40 a.m. and we are the only people in Joe’s.

“Tell me about the launch-pumps.”

“OK. New PreStocks listings. Whenever @XavierEkkel adds a name, a new private company tokenized on Solana, I buy in the first minute of listing.”

“Kalshi listed. Spot went plus twenty-something percent in the first few hours. Neuralink listed. Same thing, maybe forty. Both of those were basically free money if you were willing to buy at the first print.”

“I think I tried the same thing, but it didn’t work on Polymarket,” I said.

“Polymarket did not pump. But two of three in the small sample is enough conviction for me. Long the launch and get out in a day regardless of whether it’s red or green, rinse and repeat.”

“This doesn’t seem scalable.”

“No. A new listing is maybe once every few months, and the size I can put through the first thirty minutes is limited by the AMM’s depth. But it’s free money, who cares if it is scalable?”

Waiting for Uber, 1:08 AM [Long SpaceX perps into the IPO]

It’s late. He suggests that I call an Uber. The app shows that the driver is a few minutes away.

“Do you mind if I ask? All-in, five months in, everything: how much.”

He laughs.

“Enough to retire anywhere in the world. Except New York City.”

“At least let me know what the percentage return is.”

“I’d say I’ve roughly 3x my money.”

I had one more thing to ask him.

“Give me a trade rec. One real one. Right now.”

He nods like he’s been waiting for me to ask.

“SpaceX. The IPO is coming. Here is the trade. You long the SpaceX vntl perp, TWAP it over the weeks leading into the IPO. Scale in. Don’t buy it all at one print.”

“Why perps and not spot.”

“At IPO, Ventuals swaps the oracle from the offchain market valuations feed to the real stock; the perp has to price to the public-market tape. And the public-market tape at IPO is not at the IPO price, because IPOs pop. The float is tiny. The insiders are locked up for six months. The people who actually own the company can’t supply. You are going to have every retail brokerage in the world piling in at once to buy, and no one who can sell.”

“So the perp has to mark up.?”

“The perp has to mark up toward wherever the real stock prints, and the real stock prints above the IPO price. Fifteen to twenty-five percent above the IPO price is my conservative estimate. That is what I am long for.”

“What’s the downside?”

“Cost of carry. If funding gets to a few thousand percent annualized while you’re in the trade, you’re bleeding on the long. If funding runs hot for a few weeks without the IPO timeline tightening up, you get out. The trade works only if the IPO actually happens on the rumored horizon. If it slips by a quarter, the carry eats into your PnL.”

My phone beeps. The Uber is here.

“I’m writing a post, can I put your name in?”

He laughs. “I’d prefer not.”

“The trades?”

“The trades you can put on whatever you want. That I don’t mind.”

We split the tab and walk out into the cold. He goes north up Columbus. I step into the car. The driver takes me south on Amsterdam, and I watch him get smaller in the side mirror, already on his phone, already back at it.

Disclaimer

This is an article about a conversation with a trader, not a recommendation that you do any of what he does. Nothing here constitutes investment, legal, tax, or any other professional advice. Pre-IPO tokens and perpetual futures on private companies carry risks that include, but are not limited to: total loss of principal, counterparty failure, protocol failure, illiquidity, and regulatory change. Past results, including the backtest figures in this piece, do not predict future outcomes.

Access to Freeport and to the venues referenced in this article may be restricted in certain jurisdictions; it is your responsibility to comply with all laws and regulations that apply to you, including those governing access to DeFi protocols, derivatives, and tokenized private-company claims. Freeport is a non-custodial interface. We do not hold your assets, we do not take the other side of your trades, and we merely allow access to third-party venues (including Hyperliquid and Solana AMMs) which we do not operate, audit, or endorse. Every protocol surfaced through Freeport carries its own smart-contract, counterparty, oracle, bridge, and operational risk, and we make no representation as to the safety, solvency, or ongoing availability of any of them.

Any AI-generated content in the Freeport app, summaries, feeds, trade ideas, and buttons, are informational and not advisory. Surfaced actions are shortcuts, not recommendations; they reflect computed signals, not fiduciary advice, and should be treated as a starting point for your own research rather than an endpoint.