A financial model does not think in terms of picks. It thinks in terms of edge, variance, and bankroll survival. This is the same mental shift that separates gamblers from traders. If you want to treat sports betting like a real business rather than entertainment, you need to build it like one.

Here is how that model actually comes together.

Made a summary video using blog2video.app

Start With Probability, Not Odds

The sportsbook’s odds already contain a probability estimate. Your job is not to read that number. Your job is to build your own number and compare the two.

If a sportsbook prices a team at -150 to win, that implies roughly a 60% chance of winning, once you strip out the vig. Your model needs to independently estimate that same probability using inputs the market may be underweighting: injury data, travel schedules, weather, referee tendencies, or simply a better statistical model of team strength.

The entire business rests on one question. Where does my probability diverge from the market’s probability, and by how much?

The Core Engine: A Rating System

Every serious sports betting model is built on top of a rating system. Elo is the simplest version. You assign every team a numerical rating, update it after each game based on the result and the strength of the opponent, and use the rating gap between two teams to generate a win probability.

More advanced versions layer in additional factors. Offensive and defensive ratings separately. Home field adjustments. Rest days. Pace of play. None of this needs to be exotic. The Elo framework from chess adapted cleanly to sports decades ago, and it still forms the backbone of most public and private models today.

The lesson from market history is relevant here. Simple models that are used consistently beat complex models that are used inconsistently. Renaissance Technologies did not win by having the most complicated math. They won by having a repeatable statistical edge and executing it with total discipline.

Convert Probability Into Expected Value

Once you have a probability, the next step is expected value math. This is the same calculation used in options pricing and insurance underwriting, just wearing a different jersey.

Expected value equals your probability of winning times the payout, minus your probability of losing times the stake. If your model says a team has a 55% chance to win and the sportsbook’s implied probability is only 50%, you have a positive expected value bet. The size of that gap is your edge.

This is the number that matters. Not whether you think a team will win. Whether your estimate diverges meaningfully from the market’s estimate.

Bankroll Management Is the Financial Model

Here is where most bettors fail even when their picks are good. They have no position sizing discipline.

The Kelly Criterion solves this. It tells you what fraction of your bankroll to stake based on your edge and the odds offered. The formula rewards bigger edges with bigger stakes and punishes overconfidence by shrinking stakes as uncertainty rises. Most professional bettors use a fraction of full Kelly, often a quarter or a half, because full Kelly sizing produces violent swings that are psychologically and practically difficult to survive.

This is identical to position sizing in a trading portfolio. A hedge fund does not bet the same size on every trade regardless of conviction. Neither should you.

Track Variance, Not Just Results

A financial model needs to separate skill from luck. Over a small sample, a good model can lose money, and a bad model can make money. This is the same problem value investors face when the market ignores fundamentals for years before eventually correcting.

Track your closing line value instead of just your win rate. Closing line value measures whether you got a better price than the market settled on by game time. If you consistently beat the closing line, you have a real edge, even if a short-term losing streak says otherwise. If you consistently lose to the closing line, your early results were luck, no matter how good they looked.

This single metric is the difference between a real quantitative operation and a hot streak that will eventually mean revert.

Build the Feedback Loop

A model that does not update itself is not a model. It is a fixed opinion wearing a spreadsheet’s clothing.

Every result needs to feed back into your ratings, your probability estimates, and your calibration. If your model says a team has a 70% chance to win, it needs to actually win about 70% of the time across a large sample of similar bets. If it wins 85% of the time, your model is underconfident and leaving value on the table. If it wins 55% of the time, your model is overconfident and will bleed money at scale.

This calibration check is the same discipline used in weather forecasting and epidemiology. A 70% chance of rain needs to mean something real, or the forecast is theater.

The Real Takeaway

Sports betting modeled correctly is a market efficiency problem, not a prediction problem. You are not trying to guess who wins. You are trying to find the small, recurring gaps between your probability estimates and the market’s, size your bets rationally against that edge, and survive the variance long enough for the law of large numbers to work in your favor.

That is not gambling. That is quantitative finance with a scoreboard.

At some point this stops feeling like reading. It starts feeling like a strategy. You have built a “portfolio” of information sources the same way you built a portfolio of stocks. The question worth asking is whether that portfolio actually behaves like one.

I made a summary video using blog2video.app

The Pitch Behind Every Newsletter

Almost every paid financial newsletter sells the same promise. Subscribe and you get something the crowd does not have yet. A macro read before it is consensus. A stock idea before it is priced in. A framework that lets you see around corners.

This pitch works because it flatters the reader. Nobody wants to think they are buying entertainment. They want to think they are buying alpha.

But alpha is scarce by definition. If ten thousand people pay thirty dollars a month for the same idea, the idea is not an edge anymore. It is already in the price by the time your inbox refreshes. The economics of a mass-market newsletter are the economics of a broadcast, not the economics of a hedge fund letter. The moment a signal is cheap enough and public enough to sell as a subscription, its value as a signal has already started to decay.

Diversification That Is Not Diversification

Investors know that adding a tenth uncorrelated asset to a portfolio reduces risk. Adding a tenth newsletter does not work the same way, because newsletters are not independent. Most of them are reading each other. Most of them are reacting to the same Fed meeting, the same earnings print, the same viral chart. During a strong trend, nearly every macro newsletter will sound bullish. During a drawdown, nearly all of them will suddenly discover reasons to be cautious.

This is herding dressed up as independent analysis. Subscribing to more of it does not diversify your information. It just increases the volume of the same signal arriving from different directions, which can trick you into thinking a view is more validated than it actually is. Ten people agreeing with each other is not ten independent data points. It is one data point repeated ten times.

The Track Record Problem

Newsletter writers advertise their wins. Nobody opens a pitch page that leads with the trade that lost forty percent. Survivorship bias is not just a phenomenon that happens to mutual funds. It happens inside every writer’s own memory of their own calls. The right way to judge a track record is to see every call, timestamped, including the ones that were wrong and quietly dropped. Almost no newsletter operates this way. Most operate more like a slot machine that only shows you the jackpots.

This does not mean every newsletter writer is dishonest. Most genuinely believe their own good calls are representative. That is exactly what makes it dangerous. Confirmation bias is more persuasive when it comes from someone who is not lying to you.

Where The Real Cost Sits

The subscription fee is the smallest cost. The larger cost is attention and behavior. A trader who reads eight sources before every decision is not gathering more information. He is gathering more permission to act on impulses he already had. Newsletters are extremely good at supplying a narrative that justifies a trade you were already tempted to make. That is not research. That is confirmation shopping, and it usually shows up later as excess turnover and worse entries.

There is also a subtler cost. Time spent reading opinions is time not spent building your own process. An investor who outsources their thinking to ten writers ends up with ten half-formed opinions and no framework of their own to reconcile them.

Where Newsletters Actually Earn Their Keep

None of this means newsletters are worthless. The honest value in a good newsletter is rarely the specific call. It is the framework behind the call. A writer who consistently explains how they think about position sizing, how they define risk, how they read a balance sheet, is teaching you a process you can reuse for the rest of your life. That has real compounding value, the same way a good textbook does.

The newsletters worth paying for are the ones you would still value even if every prediction in them turned out wrong, because the way of thinking survives the outcome. The ones not worth paying for are the ones that only have value if the next call happens to land.

A Simple Filter

Before adding another subscription, ask what you are actually buying. If the answer is a specific prediction, you are buying a lottery ticket with a subscription fee attached. If the answer is a repeatable way of analyzing markets, you are buying education, and education compounds in a way that hot tips never do.

Most people’s newsletter stack should be small, deliberately uncorrelated in perspective rather than topic, and judged on process rather than the last three calls. Treat it less like a portfolio of assets and more like a small, carefully chosen faculty. You do not need ten professors saying the same thing. You need two or three who disagree with each other for reasons you can actually follow.

This month I went looking for those writers. Here are five I think deserve more attention, plus one shameless plug at the end.

I made a summary video using blog2video.app

Run by an ex-investment banker who worked at Morgan Stanley and JPMorgan before stepping away to invest on her own terms. The bio reads “Mum, ex-investment banker, investor, amateur chef.” That combination matters. Most finance writing comes from people trying to prove they belong in the room. This comes from someone who already spent a decade in the room and left because she wanted to write for people outside it. If you want markets explained without the desk jargon, this is worth your inbox.

A Wall Street veteran writing about what it actually looks like to build and manage portfolios for wealthy clients. The subscriber count is small right now, which is exactly why it belongs on this list. The value here is access to thinking that normally stays behind closed doors at private banks and family offices. Not financial advice, as the author is careful to say, but a rare window into how money actually gets managed for people who have a lot of it.

- The Upside Brief

The newest and smallest publication on this list. I am including it because early-stage writers publishing in finance are worth watching before the crowd arrives. If you like being early to a newsletter the way you like being early to a stock, this is the one to bookmark and watch develop over the next few months.

Five years and over 2,500 daily market reports. That is the kind of consistency that almost nobody in this space actually pulls off. The framing is blunt: equities, crypto, macro, and the idea that the “black swan” event has quietly become the norm rather than the exception. If you want a daily habit that keeps you oriented across markets without the fluff, this is a strong candidate.

, Inside China Innovation

The one name on this list you may already recognize, just not in this format. Ashley is a ten-time bestselling author and a Thinkers50-recognized authority on China’s digital economy, with work that has been used by Alibaba, JD, ByteDance, and Tencent. Her Substack is new and still small, which means you get her level of access and expertise before the rest of the finance world catches up to the fact that she is writing here. If you care about China as an investing theme rather than a headline, subscribe now.

Bonus:

And since this is my newsletter, I will make my own case. I write about behavioral finance, market history, valuation, and the psychology behind bubbles and crashes. Recent pieces have covered LTCM, the financial parallels in Napoleon’s rise and fall, prediction markets, and why the FIRE movement’s math falls apart for most people who try it. No hot takes, no hype. Just an attempt to understand markets through history and human behavior. If that sounds like your kind of reading, come check it out.

video made using blog2video.app

The Growth Bet

A stock price is a claim on future cash flows, discounted back to today. A mature industrial firm grows revenue by a few percent a year. Its earnings today are a decent proxy for its earnings tomorrow. Value it on trailing profit and you are not far off.

A fast growing software company is different. Most of its value sits in cash flows that do not exist yet. They arrive five, ten, fifteen years out, if the growth holds. Small changes in the assumed growth rate or how long it lasts swing the fair value wildly. That sensitivity is why tech multiples look absurd against last year’s earnings. The market is not pricing last year. It is pricing a decade.

Zero Marginal Cost, Total Domination

Software scales differently than steel or retail. Serving one more customer costs almost nothing. Add network effects, where every new user makes the product better for existing users, and you get markets that tip toward a single winner or a small handful of winners.

Investors know this. So they do not price a company on its current market share. They price it on the probability that it becomes the eventual monopolist or duopolist. That optionality carries a real premium, long before the company has proven it will win.

Narrow Themes, Wide Capital

Every cycle produces a small number of dominant narratives. AI today. Cloud a decade ago. EVs before that. Capital searching for exposure to the theme is large. The number of companies plausibly tied to that theme is small. Simple supply and demand pushes prices up, independent of what the underlying business earns this quarter.

The Discount Rate Effect

Valuation models discount future cash flows back to the present. The discount rate is the denominator. When interest rates are low, cash flows far in the future lose less value when brought back to today. This mechanically inflates the price of long duration assets, and tech earnings are about as long duration as it gets.

This is also why tech got crushed in 2022. Rates rose fast. The same future cash flows, discounted at a higher rate, were worth less today. Nothing changed about the businesses. The denominator changed.

Reflexivity Does the Rest

Rising prices attract more buyers. More buyers bring media coverage, index fund inflows, and founder or employee wealth that gets recycled into more buying. This feeds on itself until sentiment turns. It is not unique to tech, but tech’s retail following and media attention amplify it more than most sectors.

Accounting Understates the Asset

R&D and software development get expensed immediately under standard accounting rules rather than capitalized like a factory or a machine. That makes a tech company’s book value look artificially small relative to what it has actually built. Price to book, a ratio built for industrial companies, breaks down when applied to a company whose real assets are algorithms, data, and a decade of accumulated code.

The Honest Conclusion

None of this proves tech valuations are justified. Some of it reflects genuinely superior economics. Some of it is speculative excess that eventually corrects, the way it did after 2000 and again in 2022. The uncomfortable truth is that in real time, nobody can cleanly separate the two. History suggests both are usually present at once.

The data backs up the vibe. This isn’t just a feeling. It’s a measurable shift in who owns stocks and how much of their money sits in them.

I made a summary video of this post using blog2video.app

The Numbers Tell the Story

US households held $55.15 trillion in equities and fund shares in the first quarter of 2026. That’s up from $51.24 trillion just a year earlier. Equities now make up 45.76% of household financial assets, near a record high.

Stock ownership in the US sits around 62% of adults. That number hasn’t been this high since before the 2008 crash.

This isn’t a US-only story. In the UK, 58% of adults have invested as of 2026. That’s up from 54% in 2025 and 51% in 2024. Nearly two-thirds of Gen Z has made an investment. So has two-thirds of Millennials. Even half of Baby Boomers are in the market now.

What Actually Changed

Three things made this possible.

The friction disappeared. Commission-free trading apps removed the last real barrier to entry. You used to need a broker, a phone call, and a fee. Now you need a smartphone and five minutes.

The information moved to social media. Financial education content and online communities pulled millions of new participants in. People learn about investing from creators now, not from a bank teller or a financial advisor.

Retail became a real force in daily trading. Roughly one-fifth of US equity volume now comes from retail flows. Payment for order flow, the mechanism that makes commission-free trading possible, rose 51% year over year in the third quarter of 2025. The number of brokers routing trades this way has doubled since the start of 2024.

The Part Everyone Skips

Here’s where the story gets less clean.

More people participating does not mean wealth is spreading evenly. The top 1% of households still hold 50.2% of all equities and fund shares in America. The bottom 50% hold just 1.1%.

So two things are true at once. Participation is genuinely rising. Ownership is still brutally concentrated. A rising tide of retail investors does not mean a rising tide of retail wealth. It mostly means more people now have skin in a game the wealthy were already winning.

There’s also a behavioral wrinkle worth sitting with. Retail money tends to chase momentum. Retail portfolios lean heavily toward stocks that already ranked in the top of recent price trends. That’s not a criticism, it’s just how newer, less experienced capital tends to behave. It also means retail flows can amplify moves that are already underway rather than cause them.

Why This Matters for You

If nearly half of the average household’s financial assets now sit in stocks, your own allocation is worth checking against that number, not against a vague sense of “I should probably invest more.”

A high participation rate across the population also raises the stakes of the next real drawdown. When everyone is in, everyone feels the correction. Two households with identical $200,000 portfolios can look the same in a good year. One holding 90% in stocks and one holding 55% in stocks will not feel the same during a 30% drop. The first household absorbs roughly twice the pain.

More people are playing the market because playing it got easy, and because a decade of low rates and easy access pushed money out of cash and into equities. That doesn’t make the game safer. It just means more people are now exposed to the same risks that used to sit mostly with institutions and the wealthy.

Know your number. Everyone else joining the market doesn’t change what you should be holding.

Hedge funds are not supposed to beat the market in a straight bull run. That was never the job.

Here is a summary video made via blog2video.app

The Comparison Is Wrong From the Start

Comparing a hedge fund’s return to the S&P 500 assumes both are trying to do the same thing. They are not.

The S&P 500 is fully invested in equities, all the time, with no hedging and no cash buffer. A hedge fund holds cash, shorts positions, hedges tail risk, and often runs at a fraction of the market’s volatility. Judging both by raw return ignores what each portfolio was built to survive.

A fund running at half the volatility of the index should be judged against half the return, adjusted for risk. Most critics skip that step entirely.

What Hedge Funds Actually Sell

Hedge funds sell three things the market itself does not offer.

Downside protection. A fund that loses 8% in a year the S&P 500 loses 35% did its job, even with a lower long-term average return. Capital preservation in a crash is worth more to a retiree than an extra 3% in a bull year.

Uncorrelated returns. A strategy that makes money when equities fall is valuable specifically because it does not track the index. If it moved in lockstep with stocks, there would be no diversification benefit to paying for.

Access to strategies retail investors cannot replicate. Merger arbitrage, distressed debt, market-neutral long-short, structured credit. These require capital, leverage, and infrastructure an individual investor does not have.

None of these show up in a simple return comparison against the S&P 500.

Who Actually Uses Hedge Funds

Pension funds, endowments, and family offices are not chasing the highest possible return. They are managing liabilities.

A pension fund needs to pay retirees on a fixed schedule regardless of what the market does in a given year. A 2008-style drawdown is not an inconvenience for that fund. It can mean missed obligations. Smoothing returns and cutting drawdowns matters more than maximizing upside.

This is portfolio construction, not stock picking. A 10% allocation to a low-correlation strategy can lower the volatility of an entire portfolio even if that sleeve underperforms equities on its own. The math works at the portfolio level, not the line-item level.

The Fee Argument Has Some Truth to It

The 2-and-20 model deserves real criticism. Paying 2% management fees for closet indexing is indefensible, and plenty of funds do exactly that.

But the fee critique and the underperformance critique are different arguments. A fund can charge too much and still deliver exactly what it promised: lower volatility, smaller drawdowns, uncorrelated returns. High fees are a reason to negotiate or walk away. They are not proof the strategy itself failed.

Where This Breaks Down

The industry does not get a full pass. Two failure modes are common and both are real.

Funds that quietly drift toward long-only equity exposure while still charging hedge fund fees are indefensible. If a fund’s returns correlate 0.9 with the S&P 500, it is not hedging anything and should be priced like a mutual fund.

Funds that market themselves on absolute return promises and then underperform on a risk-adjusted basis too are also failing at their own stated job. Risk-adjusted underperformance is a real problem. Nominal underperformance against an unhedged equity index is not.

The Right Question

Stop asking whether a hedge fund beat the S&P 500. Ask whether it delivered on what it was actually sold to do: lower volatility, smaller drawdowns, and returns that do not move in lockstep with equities.

For a 25-year-old with a 40-year time horizon, none of that matters much. Full equity exposure is the correct call.

For an institution managing near-term liabilities, it is the entire point.

Made a summary video of this post using blog2video.app

The pitch: cutting Wall Street out of the room

Larry Page and Sergey Brin had watched the late-1990s IPO circus up close, with underpriced shares handed to favored institutional clients who flipped them for quick profits while retail investors got left behind. Their answer was a Dutch auction. Instead of investment banks setting a price and allocating shares to their best customers, any investor could submit a bid stating a price and quantity, and the market would set the clearing price itself.

The framing was explicitly anti-establishment. Google’s prospectus took direct jabs at Wall Street’s obsession with smoothed quarterly earnings, and the founders were open about wanting a structure that protected the company from short-term pressure. That posture did not sit well with the banks running the deal. Merrill Lynch dropped out of the underwriting syndicate entirely, reportedly because the fee economics of a Dutch auction were worse than a traditional book-built offering. One venture capitalist quoted at the time called Google’s approach toward Wall Street “incredible arrogance,” conceding the company could get away with it only because it was printing money.

Confusion, not conviction, defined the run-up

The retail investors the auction was designed for were, by most accounts, baffled by it. Because Dutch auction IPOs were so rare, ordinary bidders had little idea how to actually participate, and the “quiet period” rules meant Google and its underwriters could not clarify much publicly. An NPR segment from the time captured the mood well: skepticism was mounting and small investors were finding it genuinely hard to take part, not because they didn’t want to, but because the mechanics were opaque.

There was also open academic uncertainty. Auction theorists interviewed by the Wall Street Journal called the deal “intriguing” but flagged real kinks in the design, with one Carnegie Mellon economist labeling it a three billion dollar experiment. That is not the language of a sure thing.

Two other distractions fed the negative narrative. Page and Brin sat for a Playboy interview during the quiet period, which forced Google to attach the piece as an amendment to its SEC filing and raised legitimate questions about disclosure compliance. And Google kept adding risk factors to its prospectus, including concerns about Gmail’s privacy implications, which gave skeptics more ammunition.

The price cut and the muted debut

As demand data came in, Google slashed its range down to $85 to $95 and shrank the offering size, which the market read as confirmation that the original ask had been too aggressive. Analyst reactions ranged from unimpressed to openly bearish. One analyst at Marquis Investment Research said he would not be stunned if Google closed down on its first day. At $85, Google was valued around $23 billion, well below Yahoo’s roughly $39 billion market cap, and several analysts argued that gap was justified because Yahoo was “more diversified.”

The stock opened and closed its first day up about 18%, ending near $100. Respectable, but for a company that had been billed as the hottest tech offering since the dot-com era, it was not the blowout everyone expected. A trailing P/E in the low 40s at the close looked rich to a market still scarred by 2000, even though it looked cheap next to what came later.

What the skepticism actually tells you

The interesting part isn’t that Google was doubted. Almost every consequential company gets doubted at the point of maximum uncertainty. What’s interesting is what the doubt was actually about. Very little of it concerned the underlying business, a company with roughly 37% search market share, a genuinely novel ad auction model, and a clear path to $1 billion in revenue that year. Almost all of it was about mechanism and behavior: an unfamiliar auction structure, a founder team seen as arrogant toward the institutions that usually control IPO pricing, a disclosure misstep with the Playboy interview, and a valuation that looked stretched relative to a still-recent bubble.

That is a pattern worth sitting with. Markets are often quite good at pricing structural and behavioral uncertainty in the short run, and quite bad at pricing the durability of a genuinely dominant business model over a long run. The Dutch auction itself never caught on. Wall Street’s conclusion, articulated years later by CNBC’s Bob Pisani, was that leaving pricing entirely to the crowd was a bad idea, and the industry went right back to book-built offerings for nearly every large IPO since. Google was right about search. It was arguably wrong, or at least ahead of a market that wasn’t ready, about how to sell shares in a search company.

If you’d bought at the $100.34 close and never touched it, the position would be worth over 65 times your money today. The lesson isn’t “ignore skepticism.” It’s that skepticism about process and optics is a different signal than skepticism about the business, and conflating the two is one of the more common and expensive mistakes investors make around high-profile listings.

But “data center” is not one business. It is at least three, and they make money in fundamentally different ways, with different risk profiles, different customers, and different reasons an investor might want to own a piece of them. Understanding the distinction is the difference between analyzing this boom and just repeating the word “AI” while pointing at a chart.

Watch this summary video made using blog2video.app

The three layers

Think of the industry as a stack, from smallest and most artisanal at the bottom to largest and most industrial at the top.

Retail colocation is the oldest model and the one most people picture when they hear “data center.” A customer rents a rack, a cage, or a small suite inside a shared facility. They bring their own servers. The operator supplies power, cooling, physical security, and network connectivity. Pricing runs per kilowatt or per rack, and the real money is in the add-ons: cross-connect fees for linking to other tenants or carriers, remote-hands support, managed services. This is the highest-margin layer per square foot because it is the most operationally intensive. It rewards operators who run dense, sticky ecosystems where tenants want to be physically close to other tenants (think financial exchanges, ad-tech networks, content delivery).

Wholesale colocation sits above that. Instead of a rack, the customer leases an entire suite or a dedicated power block, often multiple megawatts. Leases run five to fifteen years with credit-quality tenants. This is a bond-like business dressed up as real estate: long duration, contracted escalators, low churn. It is the layer that institutional infrastructure funds and data center REITs love, because the cash flow looks like a utility with better growth.

Hyperscale leasing is the newest and now the largest layer by capital deployed. A hyperscaler (Amazon, Microsoft, Google, Meta, or increasingly a “neocloud” like CoreWeave) either builds its own campus or pre-leases an entire facility from a developer before a single shovel hits the ground. That pre-lease is what makes the project financeable in the first place. Sovereign wealth funds, pension funds, and project-finance lenders will fund a multi-billion-dollar campus because the anchor tenant has already signed for the capacity, sometimes for a decade.

A useful mental model from the industry: retail colocation drives yield, wholesale colocation anchors stability, and hyperscale leasing unlocks scale and financing leverage. Each layer is really a different point on the risk-return curve, and the “data center stock” you are buying determines which curve you are on.

Where the actual revenue sits

The colocation market alone generated roughly $39 billion in tracked annual revenue in 2025, and that figure excludes the pass-through cost of power. Multiple forecasters expect the broader colocation market to keep compounding at 14-15% a year through the early 2030s. Hyperscale capacity, meanwhile, is growing even faster and is expected to control roughly two-thirds of global data center capacity by the early 2030s, up from a minority share not long ago.

The pricing power tells you something important about where we are in the cycle. Colocation rates in the US rose sharply from 2020 through 2023, reversing years of steady decline that had made the business feel almost commoditized. That reversal was not really about AI demand at the retail level. It was about power. Grid interconnection queues now stretch four to five years in many markets, and data centers are driving over half of incremental US electricity demand growth. When the bottleneck shifts from capital to physical infrastructure like transformers and substations, the operators who locked in power contracts early get to charge for scarcity rather than for square footage.

That is the single most important thing to understand about this business right now: the moat is not the building. It is the power interconnect. Land is not scarce. Steel and concrete are not scarce, though tariffs have made them pricier. Electrons delivered reliably at gigawatt scale, on a known timeline, are scarce. That scarcity is what has turned data center development from a real estate business into something closer to a utility monopoly with a five-year lead time.

The bear case nobody wants to say out loud

Every capital cycle this large eventually runs into the question of returns on the capital deployed. The build cost per megawatt has risen from about $7.7 million in 2020 to roughly $11 million in 2026, and tenant fit-out for AI-specific infrastructure can add another $25 million per megawatt on top of the shell. That is an enormous amount of capital chasing workloads whose economics are still being figured out in real time, by companies that are themselves not yet reliably profitable on the model-training side.

The structural question echoes the fiber buildout of the late 1990s. Back then the bandwidth thesis was directionally correct and the capital allocation was still wildly overdone, and a large share of the fiber laid in that boom eventually went to a handful of survivors at pennies on the dollar. The AI infrastructure buildout could rhyme with that pattern, but with one meaningful difference: hyperscale leases are signed with a handful of the most creditworthy companies on the planet, pre-committed before construction, rather than sold speculatively into an undefined future customer base. That does not eliminate the overbuild risk. It changes who eats the loss if utilization disappoints, and it is a large part of why the debt behind these projects has been rated the way it has.

What this means if you are looking at it as an investor

The layer you choose to own matters more than “data centers” as a category. Retail and wholesale colocation REITs behave like income real estate with a growth kicker, and their risk is mostly about churn and re-leasing spreads. Pure hyperscale developers and the project-finance vehicles behind them behave more like infrastructure credit, with concentrated counterparty risk to whichever hyperscaler signed the anchor lease. And the picks-and-shovels layer underneath all of it, power equipment, cooling, grid interconnection, is arguably the least discussed but most structurally advantaged position, because it gets paid regardless of which hyperscaler wins the AI race.

The building is not the business. The power contract is the business. Everything else is just concrete.

The bigger the thread gets, the more reach everyone in it gets. That is the whole point.

I write TheFinanceGuy here. Long form pieces on behavioral finance, market history, valuation, and right now a lot on energy investing and how weather patterns feed into gas demand forecasting. Contrarian takes, short sentences, no fluff. Drop a comment and I will check out your work too.

Additional Growth Tools

If you are a newsletter writer looking to grow beyond Substack itself, two tools worth a look:

Blog2Video (blog2video.app) turns your written posts into narrated videos automatically, using your actual content rather than generic AI video templates. Useful for repurposing long form essays into something shareable on video platforms.

Convert Your Substack into Video

BlogHub (bloghub.app) is a community-curated blog discovery platform. It helps readers find blogs and newsletters outside the usual algorithm-driven feeds, which is exactly the kind of discovery this comment thread is trying to create.

Drop your Substack below. Let’s build some reach together.

Showcase your substack

Natural gas is the clearest example. Unlike oil, gas is hard to store and hard to ship across oceans. Most of it gets burned close to where it comes out of the ground. That means supply and demand balance locally, in near real time, and the biggest swing factor in that balance is temperature.

A summary video made by using blog2video.app

The Demand Curve Is a Thermometer

Heating and cooling account for the majority of short term swings in gas demand. A few degrees of deviation from seasonal norms can move demand by billions of cubic feet a day. Power generators need gas to run turbines when air conditioning load spikes in July. Utilities need it to heat homes when a polar vortex drops into Texas in February.

This is why the weekly EIA storage report moves prices so violently. The report itself is backward looking. What actually drives the number is the weather that already happened, and what drives the market reaction is the weather forecast for the next two weeks.

Professional gas traders do not read the same weather report you get on your phone. They subscribe to services that run ensemble models like the GFS and the European ECMWF, and they pay close attention to when those two models diverge. A forecast that shows the American and European models disagreeing on whether a cold snap arrives on day ten versus day fourteen is worth more to a trader than most fundamental research.

Power Markets Compound the Problem

Electricity adds another layer. Power cannot be stored at scale. It has to be generated the instant it is consumed. So a heat wave does not just raise gas demand for cooling, it also strains the grid directly, which can spike spot power prices by an order of magnitude in a matter of hours.

Texas in February 2021 is the textbook case. A winter storm that meteorologists had flagged more than a week in advance still managed to freeze wellheads, spike gas demand, and knock out power generation simultaneously. Spot power prices went from around 20 dollars per megawatt hour to the market cap of 9,000 dollars. Traders who had the right weather read, and the right hedges in place, made careers. Traders and utilities who did not made headlines for the wrong reasons.

Renewables Made This Worse, Not Better

Wind and solar were supposed to reduce dependence on weather driven fuel demand. In practice they added a second weather variable on the supply side. Now a trader has to forecast not just how much gas or power will be demanded, but how much wind and solar will actually generate.

Grid operators in Texas, California, and Germany have all learned to watch wind speed forecasts the way gas traders watch temperature forecasts. A calm, high pressure system sitting over the Great Plains means low wind output, which means gas or coal has to fill the gap, which means prices move even if nothing about heating or cooling demand has changed. Solar has the same problem with cloud cover, though on a shorter and more local timescale.

Why This Matters for Investors, Not Just Traders

You do not need to trade daily gas futures to care about this. If you own utility stocks, midstream pipeline companies, or independent power producers, your earnings are indirectly a weather derivative. A mild winter can blow a hole in a gas utility’s quarterly numbers even if every other part of the business executed well. An unusually hot summer can be a windfall for a merchant generator with uncontracted capacity.

Analysts building models for these companies increasingly build in weather normalization, an adjustment that tries to strip out the noise from an unusually hot or cold period so you can see the underlying trend. That adjustment is itself an admission of how much weather distorts the raw numbers.

The Takeaway

Energy is one of the few sectors where a meteorologist’s forecast has more short term predictive power over earnings and prices than a Wall Street analyst’s model. That is not a knock on the analysts. It is a structural fact about a commodity that cannot be stored cheaply and a demand base that is fundamentally about staying warm or staying cool.

If you invest in this space, the weather forecast is not a curiosity to check before your commute. It is part of the research process.

This asymmetry is called loss aversion. It explains almost everything irrational about how people handle money.

You can check out the summary video here: https://blog2video.app

Why You Hold the Losers

You buy a stock. It drops 20%. Every instinct tells you to hold. Selling makes the loss real. Holding keeps the story alive, the story where you were right all along and the market just needs more time.

This is the disposition effect. Investors sell winners too early and hold losers too long. The winner gets sold because you want to feel the win. The loser gets held because selling forces you to admit defeat.

The stock does not know your purchase price. It does not care what you paid. But your brain anchors hard to that number. Below it, you are “down.” Above it, you are “up.” The stock has no opinion on this. Only you do.

The Sunk Cost Trap

Money already spent is gone. It cannot be recovered by spending more. Rationally, every decision should be made fresh, based only on future expected value.

Nobody does this.

You put $10,000 into a position. It falls to $6,000. Now you have two options. Cut the loss and move on, or add more to “average down” and lower your cost basis. The second option feels smart. It often is not. It is frequently just the sunk cost fallacy wearing a strategy costume.

The test is simple. If you did not already own this position, would you buy it today at this price with this thesis? If the answer is no, the fact that you already own it changes nothing about the future. Your past decision is irrelevant to your next one. It only feels relevant.

Pain Changes How You See Risk

After a loss, most people do one of two things. They freeze, or they gamble.

Freezing looks like moving everything to cash after a drawdown, right before the recovery. Gambling looks like doubling position sizes to “make it back,” which is how one bad trade becomes three bad trades.

Both responses come from the same place. A loss activates threat circuitry, not analytical circuitry. You are briefly a worse decision maker than you were yesterday. This is not weakness. It is biology. The amygdala does not care about your Sharpe ratio.

The professionals who survive long careers in markets are not the ones who never feel this. They are the ones who built process specifically because they knew they would feel this. Stop losses, position sizing rules, and pre-committed exit plans exist because the person who sets the rule in a calm moment makes better decisions than the person who is inside the loss.

Comparison Is the Real Killer

A 10% loss in a falling market that everyone else lost 20% in feels like a win. A 5% loss in a market that was flat feels like a disaster. The number on your statement is identical to the number in an alternate universe where context did not matter. But context always matters to your nervous system.

This is why bear markets are psychologically easier in some ways than sideways markets. In a crash, everyone is losing together. The pain is shared and normalized. In a choppy, directionless market, every small loss feels personal, like the market singled you out.

It did not. Markets do not know you exist.

What Actually Helps

Understanding loss aversion does not make it disappear. You cannot think your way out of biology. But you can build systems that route around it.

Decide your exit before you enter, not after. The version of you that has not lost money yet is a better risk manager than the version of you staring at a drawdown.

Separate the decision from the outcome. A good decision can lose money. A bad decision can make money. Judging yourself purely on results teaches you the wrong lessons, usually the lesson to take more risk after a lucky win and less risk after an unlucky loss, which is backwards.

Write down your reasoning before you act. Not after. After, memory rewrites itself to make you look smarter than you were. A dated note is honest in a way your recollection never will be.

The market does not remember what you paid. Neither should you.

Made a summary video of this post using blog2video.app

1.

If you want to understand why Nvidia, TSMC, and the entire AI buildout move the way they do, this is the publication institutional investors quietly forward to each other. SemiAnalysis has crossed 180,000 subscribers and built out a full research operation, with proprietary models covering accelerators, datacenter power, wafer fabs, and AI networking. Ben Thompson of Stratechery has called it one of his most cited resources, and that endorsement tells you what kind of publication this is. It is not a newsletter with opinions about chips. It is closer to a sell side research desk that happens to publish on Substack.

The catch is depth. Some of the best material sits behind institutional pricing, and even the free posts assume you already know what HBM and CoWoS mean. Read this one for the picture behind the picture: the actual industrial mechanics driving the AI trade that most financial media only summarizes secondhand.

2.

Written by a France-born, Silicon Valley-based investor with a background at PwC and Bandai Namco, this is the cleanest breakdown of big tech earnings you will find anywhere online. Over 320,000 subscribers read it, and it currently ranks in the top 50 finance publications on Substack. The format is consistent. Each week takes an earnings report from Apple, Amazon, Meta, or a handful of others, and turns dense 10-Ks into charts a normal investor can actually use.

What makes this one worth a subscription rather than a skim is the discipline. There is no macro speculation, no hot takes about the Fed. Just company by company fundamentals, quarter after quarter, until you start recognizing patterns in how these businesses actually compound. If SemiAnalysis is the supply chain view, this is the income statement view of the same AI-driven tech economy.

3.

This one is aimed less at retail investors and more at the people who sit across the table from them. Its stated purpose is helping CFOs and finance leaders turn numbers into stories, and it has built an audience in the hundreds of thousands doing it. The angle is unusual for a finance newsletter. Instead of asking how to read a balance sheet, it asks how to present one so that a board, a bank, or a market actually believes the story behind it.

Worth following if you sit anywhere near financial communication, whether that is investor relations, corporate finance, or just building a habit of thinking about numbers as a narrative device rather than a spreadsheet exercise. It is a useful complement to writers who only cover the analysis side and never touch the persuasion side of finance.

4.

Josh Bobrowsky writes at the intersection the other four publications on this list only visit occasionally. AI, finance, and life, in his own words, and the informality is the point. This is not an institutional research shop. It reads more like a sharp friend thinking out loud about where markets and technology are heading, with the personal essay instincts that a lot of bigger finance newsletters have edited out of themselves.

At a few thousand subscribers, this is the smallest publication on the list, and that is exactly why it belongs here. Bobrowsky can take positions and follow tangents that a 300,000-subscriber operation cannot risk. Read the big three for information. Read this one for a point of view.

5.

Francis Henry runs this one closer to a personal notebook than a media brand, and it shows in a good way. It sits at a smaller scale than the others on this list, but the value is in the focus. Where the AI and finance intersection can drown you in noise, from hype cycles to earnings jargon to chip war geopolitics, this newsletter narrows in on the practical question most readers actually have. What does the AI shift mean for how money gets made, allocated, and analyzed.

It will not replace SemiAnalysis for supply chain depth or App Economy Insights for earnings rigor. But as a low commitment way to keep a pulse on the AI-finance conversation without subscribing to five different research operations, it earns its spot.

The pattern across all five. None of these writers started as journalists. A gaming industry analyst, a semiconductor researcher, a CFO communications specialist, and two independent thinkers building in public. That is not a coincidence. The best writing about where finance and technology meet tends to come from people who worked inside one of those two worlds first, then started explaining the other one to themselves in public. Substack just gave them a printing press.

If you only subscribe to one, make it SemiAnalysis. If you only subscribe to two, add App Economy Insights. The rest depends on whether you want the communications angle, the personal essay angle, or the low-commitment daily pulse.

A marketer calculates CAC. A finance person calculates cost of capital. Both are asking the same question: how much do I have to spend to acquire something worth more than what I spent?

I made summary video using https://blog2video.app

Customer Acquisition Cost Is a Discount Rate in Disguise

Every marketing dashboard has a CAC number. Every finance model has a discount rate. They exist for the same reason.

A discount rate tells you what future cash flows are worth today, adjusted for risk and the cost of waiting. CAC tells you what a future customer is worth today, adjusted for the risk that they churn and the cost of getting them in the door.

When a marketer says “our CAC payback period is 14 months,” they are running a discounted cash flow model without calling it one. LTV is the sum of discounted future cash flows. CAC is the initial investment. Payback period is the point where cumulative discounted cash flow crosses zero. This is corporate finance 101, just relabeled for a growth team.

The reason this matters: marketers who don’t know they’re doing DCF analysis make the same mistakes finance people spent decades learning to avoid. They ignore churn risk the way a bad analyst ignores default risk. They discount nothing, which is the same as assuming a 0% cost of capital, which is the same as saying money today and money in three years are worth the same. They are not.

Marketing Budgets Are Capital Allocation Decisions

A CFO allocating capital across business units asks: where does the next dollar generate the highest risk-adjusted return? A CMO allocating budget across channels asks the identical question, just about ad spend instead of factories or acquisitions.

Portfolio theory applies directly. A channel with high variance in return (paid social, where performance swings wildly week to week) should be sized differently than a channel with low variance and steady, boring returns (SEO, once it’s compounding). Diversification isn’t just a finance concept. A marketing budget concentrated entirely in one channel carries the same fragility as a portfolio concentrated in one stock.

Most marketing teams don’t think this way. They chase whichever channel had the best last-click number this month, which is the marketing equivalent of chasing last year’s best-performing stock. It works until it doesn’t.

Attribution Is Accounting, and Accounting Has the Same Fights

Attribution models exist to answer: which activity gets credit for a result? First-touch, last-touch, multi-touch, data-driven. This is the exact argument accountants have had for a century about how to allocate shared costs across departments. Overhead allocation is attribution modeling with a different name.

And just like accounting, the choice of attribution model changes the story without changing the underlying business. A company can make its blended CAC look fantastic by switching to last-touch attribution, the same way a company can make earnings look better by switching depreciation methods. Nothing about the business changed. Only the lens changed.

This is why sophisticated operators, and sophisticated investors evaluating a company’s growth marketing, ask for raw cohort data instead of attributed dashboards. The dashboard is opinion. The cohort table is fact.

The Real Overlap: Both Fields Are About Time

Strip away the jargon and finance is the discipline of comparing money across time. Marketing analytics, done properly, is the same discipline applied to customers instead of dollars.

A finance person asks: is a dollar today worth more than a dollar in five years, and by how much?

A marketer asks: is a customer acquired today worth more than a customer acquired in five years, and by how much?

Same question. Same math. Different noun.

Why This Matters More Than It Sounds

I’ve watched companies raise money on marketing metrics that would never survive a finance team’s scrutiny, and I’ve watched finance teams reject marketing spend using logic that ignores compounding brand value the same way an amateur investor ignores compounding returns.

The companies that get this right treat their CMO’s dashboard and their CFO’s model as the same document, viewed from two angles. The companies that get it wrong let marketing optimize for a metric finance doesn’t trust, and let finance cut budgets using a model marketing doesn’t understand. Both sides are right about the math and wrong about talking to each other.

If you run a business, the fix isn’t complicated. Sit your growth lead and your finance lead in the same room and make them use the same discount rate. Most of the tension disappears once everyone agrees on how to value a dollar over time. That agreement, more than any dashboard, is the actual foundation of both disciplines.

Made a video of this post using blog2video.app

Loss Aversion

People feel the pain of a loss roughly twice as strongly as the pleasure of an equivalent gain. This is why investors hold losing stocks far too long, hoping to “get back to even,” while selling winners too early to lock in the good feeling.

The fix is mechanical, not emotional. Set exit rules before you buy, not after the position moves against you. A rule made in advance is not vulnerable to the fear you will feel in the moment.

Anchoring

Once a number enters your head, it becomes a reference point you cannot fully escape. If a stock was $150 and is now $80, that $150 feels like the “real” price, even though it was set by a market that no longer exists. This is why investors wait for a stock to “come back” to a price that has no actual bearing on future value.

The anchor is always the enemy of a fresh look. Ask what you would pay for the business today, with no memory of where the price used to be.

Recency Bias

The last few years feel like the permanent state of the world. After a decade of low rates, investors assumed low rates were the default setting of the universe. After a crash, investors assume crashes are always around the corner. Neither is true. Markets move in regimes, and regimes end.

History is the antidote. Read about the 1970s, the 1990s, and the 2008 crash side by side, and the “this time it is different” feeling gets a lot weaker.

Herd Behavior

Humans are wired to feel safer in a crowd. In markets, this shows up as buying because everyone else is buying, and panic selling because everyone else is selling. The crowd is often right during calm periods and badly wrong at the extremes, which is exactly when it matters most.

The useful habit is to ask a simple question before following the crowd: would I want this asset if the price had not just gone up?

Confirmation Bias

Once you own a stock, you start noticing every bullish article and ignoring every bearish one. Your brain is not trying to find the truth. It is trying to protect the decision you already made.

A good discipline is to actively seek out the best bear case for anything you own. If you cannot argue against your own position convincingly, you do not understand it well enough.

Overconfidence

A few good trades in a row and the brain quietly rewrites the story. Luck becomes skill. Skill becomes destiny. This is when position sizes creep up and risk management gets quietly abandoned.

The market has no memory of your last win. Every new decision deserves the same process as your first one, not a looser one earned by a recent streak.

Mental Accounting

Money is fungible, but the brain treats it as if it is not. A $5,000 inheritance feels different from $5,000 earned through a bonus, even though both spend the same. Investors will happily gamble with “house money” from a big gain while being extremely cautious with their original capital, despite it being the exact same dollars.

Treat all capital in a portfolio as one pool, not separate buckets with separate rules.

The Sunk Cost Fallacy

Money already spent should not affect a decision about what to do next. Yet investors keep adding to a losing position purely because they have already put so much in, as if the size of the mistake justifies making it bigger.

The only relevant question is what you would do today, starting from zero, with the money you have left.

Why This Matters More Than Most Analysis

You can build the best discounted cash flow model in the world and still lose money if you cannot manage your own reactions to volatility. The market is a mechanism for transferring wealth from the impatient to the patient, and patience is a psychological skill, not a financial one.

Understanding these biases will not make you immune to them. Nobody is fully immune to their own wiring. But naming the bias while it is happening is often enough to slow down the reaction long enough to make a better decision.

A better approach treats every investment claim as a probability problem. You are not asking “does this feel shady.” You are asking “given everything I know about base rates, incentives, and verifiability, what is the likelihood this is what it claims to be.” That question survives contact with a smart con artist. Pattern matching does not.

Here is the framework I actually use.

I made a summary video using blog2video.app

Start with the base rate, not the pitch

Before you evaluate a single detail of the investment in front of you, ask what fraction of similar-sounding opportunities turn out to be fraudulent. This is the reference class question, and almost nobody asks it.

If someone offers you 20 percent annual returns with low volatility, the first move is not to scrutinize their strategy. The first move is to ask how many people in history have sustainably delivered 20 percent with low volatility, and how many people who claimed to have done so were lying. The base rate for the claim, independent of the specific pitch, should dominate your prior. Bernie Madoff ran for decades on a return stream that looked steady because steady is what sells, not because steady is what real strategies produce. Real strategies are lumpy. Fraud is smooth, because fraud is manufactured to be reassuring rather than to be true.

This single habit, checking the reference class before checking the details, filters out a huge share of scams before you even need clever reasoning.

Extraordinary claims need proportionally extraordinary evidence

This is an old principle and it still does all the work. The size of the claim should set the size of the evidence bar. A fund claiming 8 percent annual returns needs ordinary evidence: audited statements, a custodian you can call, a track record you can verify independently. A fund claiming 40 percent annual returns with no down years needs extraordinary evidence, and if the evidence offered is the same as what the 8 percent fund offers, that mismatch is itself the signal.

Fraudsters understand this instinct exists, so they try to manufacture proportional-looking evidence. They show you PDFs. They show you a dashboard. They show you other investors who are happy. None of that is evidence in the sense that matters, because none of it is independently verifiable. A statement generated by the same entity making the claim is not evidence for the claim. It is just the claim, restated with more formatting.

The test is simple. Can a party with no financial stake in the outcome confirm this independently. If the only confirmation available comes from the promoter, their employees, or their existing investors, you have zero independent evidence, no matter how much paperwork exists.

Ask who benefits if this is false

This is inversion applied to trust. Instead of asking “is this person lying,” ask “if this were a lie, who gains, and does that person’s behavior match what a liar in that position would do.”

A person running a real, profitable strategy has no reason to rush you. Their edge exists whether you invest ten dollars or ten million. Urgency is manufactured because time works against fraud. Every day you spend independently verifying is a day the fraudster is not collecting your money and is at risk of exposure. So urgency, “this closes Friday,” “the allocation is almost full,” is not just an emotional manipulation tactic. It is a logical tell. It reveals that the promoter’s payoff depends on you skipping the verification step, which tells you the verification step would probably fail.

Apply the same test to complexity. A real strategy is usually explainable at a level of abstraction, even if the details are proprietary. You can say what the edge is, even without giving away exactly how it’s captured. A fraudulent strategy often hides behind complexity that explains nothing. If someone cannot describe, in one paragraph, why the strategy should work, and instead retreats into jargon whenever asked, ask who benefits from you not understanding. Usually it’s them, because your confusion is what stops you from noticing there was never a real mechanism underneath.

Check whether the claim is falsifiable

A claim that cannot be wrong is not a claim, it’s a story. Before trusting any investment thesis, ask what evidence would prove it false, and whether that evidence is available to you.

Legitimate strategies produce falsifiable numbers. Audited returns, verifiable trade records, a custodian holding assets separately from the manager. You can check these against something external. Fraudulent schemes tend to produce numbers that are internally consistent but externally unfalsifiable. The only source of truth is the promoter’s own reporting. There is no outside system that could ever contradict them, which also means there is no outside system that could ever confirm them. Unfalsifiability isn’t neutral. It’s a structural feature that makes fraud durable, because nothing you could ever check would flip your belief. If you notice that no possible piece of evidence would change your mind about an investment, that’s not conviction. That’s the absence of a real test.

Separate custody from management

This is less about probability and more about pure logic, but it belongs here because it collapses most fraud cases on its own. Ask one question: who holds the actual assets, and is that party different from and independent of the person managing the strategy.

Nearly every major investment fraud in history shares this structural feature: the person telling you how much money you have is the same person who controls the money. When custody and management are combined in one party, the reported numbers become a story that party tells about itself, and there is no external check on that story. When custody sits with an independent, regulated third party, a fraudulent manager cannot simply announce whatever return they like, because the assets and their valuation are visible outside their control.

This single structural check, independent of anything about strategy or personality, eliminates the majority of Ponzi-style fraud. It does not require you to be a forensic accountant. It requires you to ask one factual question and verify the answer with the custodian directly, not with the manager’s description of the custodian.

Weight the asymmetry of the bet

Even when you cannot fully resolve whether something is fraud, you can still reason correctly about the decision using expected value and asymmetry.

If you are wrong in one direction, you skip a legitimate opportunity, the cost is an opportunity cost, real but bounded. If you are wrong in the other direction, you invest in a fraud, the cost can be your entire principal, and in leveraged or borrowed-money cases, more than your principal. These outcomes are not symmetric, so your evidence bar should not be symmetric either. When the downside of being wrong is capped and the downside of being fooled is catastrophic and irreversible, the correct probability threshold for saying yes is much higher than 50 percent confidence. This is why professional allocators demand so much verification before committing capital that ordinary investors skip. It’s not caution for its own sake. It’s the correct response to an asymmetric payoff structure.

Watch for the smoothness that shouldn’t exist

One more logical tell worth naming on its own, because it comes up constantly and fools smart people specifically because it looks like skill. Real markets produce volatility. Real strategies, even good ones, have losing months, losing quarters, sometimes losing years. Volatility is not a flaw in a track record, it’s evidence that the track record was actually generated by markets rather than manufactured by a person with a keyboard.

So when a return stream is suspiciously smooth, consistently positive, low drawdown, month after month, for years, the correct response is heightened suspicion, not admiration. Smoothness like that is either a genuinely rare structural edge, which is possible but should be treated as a base-rate long shot, or it is fabricated. Given the base rate of each explanation, fabrication should be your leading hypothesis until proven otherwise, not your last resort after everything else fails to explain it.

The actual checklist, restated as questions

Pull all of this together and the process becomes a short sequence of questions you ask before you ask anything about the specific pitch in front of you. What is the base rate for claims like this one. Is the evidence offered proportional to the size of the claim, or is it just paperwork. Who benefits if this turns out to be false, and does the promoter’s behavior match that incentive. What would have to be true for this claim to be provably wrong, and can I actually check it. Is the person reporting my returns the same person holding my money. Given the size of what I could lose versus what I could gain, how much confidence do I actually need before saying yes. Does the smoothness of this track record make sense given how markets actually behave.

None of these questions require special expertise. They require the discipline to ask them before you get emotionally invested in the story, because once you want something to be true, your ability to evaluate it honestly drops fast. The fraud isn’t usually cleverer than you. It’s just faster than your skepticism, if you let it be.

It is a good chart. It is also a bad explanation.

Nobody at the Federal Reserve bought a single share of Apple. No newly created dollar was ever wired into a brokerage account by a government official. The money supply and asset prices move together, but the mechanism connecting them is not the one most people describe. And if you get the mechanism wrong, you will get the next cycle wrong too. You will be waiting for a crash that does not come, or standing in front of one that does.

So let me explain what actually happens.

Money Is Credit, and Credit Needs Collateral

Start with where money comes from. Not the central bank. Commercial banks.

When a bank makes a loan, it does not reach into a vault. It writes two entries. A loan on the asset side, a deposit on the liability side. That deposit is money. It did not exist ten seconds earlier. This is not a fringe view. The Bank of England published a paper saying exactly this in 2014, which was mildly embarrassing for everyone who had spent a decade teaching the money multiplier.

Now ask the question that matters. What did the bank lend against?

This is the whole game. New money enters the economy attached to a specific piece of collateral. If a bank lends to a manufacturer building a new factory, the new deposits chase newly produced goods and labour. Supply expands alongside money. If a bank lends against an existing house, or existing shares posted as margin, or an existing company being taken private in a leveraged buyout, the new deposits chase an asset whose supply cannot expand. There is only one of that house. There is a fixed float of that stock.

Same money supply growth. Completely different price outcome.

Adair Turner spent a book making this point and it still has not landed properly. In most developed economies, the majority of bank lending is not funding new production. It is funding the transfer of existing assets at higher and higher prices. That is not a moral failing of bankers. It is a rational response to collateral quality. A house can be repossessed. An idea cannot.

So the sentence “money supply inflates asset prices” is incomplete. The accurate version is that money created against existing assets inflates the price of existing assets. The composition of credit matters more than the quantity.

Cantillon Was Here First

Richard Cantillon worked this out before anyone had a central bank to blame.

He was an Irish banker in Paris in the 1710s, and he wrote a book in the 1730s about what happens when new money enters an economy. His observation was simple and nobody has improved on it. New money does not arrive everywhere at once. It arrives somewhere specific, in someone’s hands, and that person spends it on the things they buy. Prices rise in a sequence, not in a level.

Cantillon used the example of gold mines. The mine owner gets rich first. He hires servants and buys land. The servants and landowners get rich second. Prices rise for the goods those people want, in that order, over years.

Translate this to the present. When credit expands through mortgage markets, house prices move first. When it expands through corporate bond markets at low spreads, buybacks and takeout multiples move first. When it expands through margin and derivatives, index levels move first. The money is not neutral. It has an address.

This is why the same monetary expansion can produce a housing bubble in one decade and an equity bubble in the next. The plumbing changed. The water is the same.

The Purest Experiment Ever Run

If you want to see the mechanism naked, look at John Law.

Law took control of French finance in 1717. He founded a bank that issued paper notes. He also controlled the Mississippi Company, a colonial trading venture whose shares he was selling to the public. Then he did the thing that makes economic historians wince. He allowed his bank to issue notes that people used to buy his company’s shares, and he accepted the shares as backing for more notes.

The loop was explicit. Print, buy, revalue, print against the new value, buy more.

Mississippi shares went from a few hundred livres to around ten thousand in roughly a year. Paris invented the word millionaire to describe the people it created. Then in 1720 the loop ran in reverse, which is what loops do, and it took the French financial system with it. France did not have a proper central bank again for eighty years.

Law was not stupid. He understood something most of his critics did not, which is that money supply and asset prices are not two separate variables. When assets serve as collateral for money creation, and money creation bids up assets, they are one variable feeding itself. That is the actual mechanism. Not a printer spraying cash at the stock market. A reflexive loop between the price of collateral and the willingness to lend against it.

Every serious asset bubble since has some version of this loop. American broker call loans in 1928. Japanese land collateral in 1988, where rising land values supported bank lending that bought more land. American mortgage-backed collateral in 2006. Crypto lending against crypto in 2021. The asset backs the loan, the loan buys the asset.

Why the Supermarket Does Not Notice

The obvious objection. If money supply grew forty percent between 2020 and 2022, why did stocks triple off the low while eggs did not?

Because assets and consumer goods have different supply curves and different buyers.

You cannot eat more than a few thousand calories a day no matter how rich you get. Consumer demand saturates. Financial assets have no saturation point, because their demand is not consumption demand, it is savings demand. A person with an unexpected windfall buys some extra goods and then puts the rest somewhere. That somewhere is an asset.

Add the supply side. Manufacturers respond to demand by making more units. The float of existing equities does not expand when prices rise. It usually shrinks, because buybacks accelerate in exactly the conditions where credit is cheap. So money grows, share count falls, and the arithmetic does the rest.

Then add the discount rate, which is the part that makes the swings violent. An asset is a claim on future cash flows priced against a rate. Cut the rate that money is created at, and every long-duration cash flow reprices upward, hard. A company earning nothing today but a lot in 2035 is nearly worthless at six percent and quite valuable at one percent. This is why the most speculative end of the market moves the most in monetary expansions. It is duration, not sentiment. The sentiment is downstream.

Look at the Denominator

Here is the exercise that changes how people see their own portfolio.

Take an index and divide it by the money supply instead of quoting it in dollars. Do it for the S&P against M2 since 2008. The chart does not look like a triumph. It looks like a long flat range with some noise.

That does not mean returns were fake. Dividends and earnings growth were real. But a meaningful chunk of what investors experienced as “the market went up” was actually “the unit of account went down.” Priced in money, assets rose. Priced in assets, money fell. Those are the same event described from two ends.

This is the honest version of the money supply argument, and it is more uncomfortable than the conspiratorial one. The people who owned assets did not get richer relative to each other. They got richer relative to everyone holding cash and wages. Which is most people. That is the real distributional story, and it happened through arithmetic, not through a plot.

The Counterexample That Saves the Theory

Now let me undercut everything I just said, because the version of this argument that goes around social media is falsifiable and it has been falsified.

Japan. The Bank of Japan has been expanding its balance sheet aggressively since 2001, more aggressively than any other major central bank relative to the size of its economy. The Nikkei peaked at 38,915 in December 1989 and did not durably exceed that level for over thirty years. Enormous monetary expansion. No asset inflation for a generation.

Why? Because money creation requires two consenting parties. A central bank can supply reserves. It cannot make a bank want to lend or a borrower want to borrow. After 1990 Japanese balance sheets were impaired and the private sector spent two decades paying down debt rather than taking it on. Reserves piled up in the banking system and went nowhere, because nobody wanted the loan that would have turned them into deposits.

The United States between 2010 and 2015 was a milder version of the same thing. Large quantitative easing programmes, weak bank credit growth, and consumer price inflation that stubbornly undershot target while people wrote books about hyperinflation.

So the conclusion is narrower than the meme, and more useful.

Money supply growth is a permission slip, not a cause. It sets the outer limit on how much nominal asset appreciation is possible. Whether the limit gets used depends on whether the private sector wants leverage, and on which collateral it wants to post. When both conditions are present, money creation and collateral values reinforce each other and you get 1719, 1928, 1988, 2006, 2021. When the balance sheet impulse is absent, you can print for twenty years into a flat market.

What to Actually Watch

Stop watching the central bank balance sheet. It is the least informative series in this whole story.

Watch private credit growth, because that is where money is actually created. Watch what it is being created against, because collateral composition tells you which asset class gets the Cantillon effect. Watch whether the marginal buyer is levered, because leverage is what turns a repricing into a loop. And watch the denominator, because a portfolio that keeps pace with money growth has preserved purchasing power and nothing more, whatever the nominal number says.

The uncomfortable part is that none of this gives you a timing signal. Law’s loop ran for three years. Japan’s ran for six. The 2010s ran for eleven. Knowing the mechanism tells you what kind of animal you are looking at. It does not tell you when it will turn.

But it does tell you the thing worth knowing, which is that in a monetary expansion the safest-looking asset in the world, cash, is the one quietly guaranteed to lose. That is not a reason to buy anything in particular. It is a reason to understand what you are being paid in.

That is not how it works. The market does not pay you for being correct. It pays you in proportion to how much you had on when you were correct, and how much you had on when you were wrong. Those are two different skills, and only one of them gets talked about at dinner parties.

You can be right most of the time and still go broke. You can be wrong most of the time and get rich. The variable that decides which one happens to you is size.

The arithmetic nobody wants to look at

Start with a coin that is rigged in your favor. It pays even money and lands heads 60 percent of the time. This is a spectacular edge. No real market gives you anything close to it.

Bet your entire net worth on every flip and you will be wiped out. Not maybe. Almost certainly. The chance of never hitting a tail across a long run of flips is effectively zero, and one tail ends the game. Your edge was real. Your sizing killed you anyway.

Now bet 5 percent of your capital each time. You compound steadily and end up wealthy. Same coin, same forecast accuracy, same information. Completely different outcome. The only thing that changed was the number in the size box.

This is the part that breaks people’s intuition. They believe the edge produces the return. The edge only produces the return if the sizing lets you survive long enough to collect it. Ruin is not a bad outcome on the distribution. Ruin removes you from the distribution.

Losses are not symmetric with gains

A 50 percent loss requires a 100 percent gain to get back to even. A 20 percent loss requires 25 percent. An 80 percent loss requires 400 percent.

This asymmetry is the reason large positions are so dangerous even when they are correct in the long run. Drawdowns do not just hurt your feelings. They damage the base you compound from. Two people with identical average annual returns can finish decades apart if one took a single catastrophic hit along the way, because compounding runs on the geometric mean and the geometric mean punishes volatility.

The practical translation is simple. Your job is not to maximize how often you are right. Your job is to maximize the amount you compound over time, and that means protecting the base.

Being right is cheap

Consider two investors.

The first has a 70 percent hit rate. She is right on seven trades out of ten. She makes 1 percent on her winners and loses 3 percent on her losers. Her expected value per trade is negative. Seven times one is seven. Three times three is nine. She is right constantly and losing money constantly.

The second has a 35 percent hit rate. He is wrong nearly two thirds of the time. But his winners return 5 percent and his losers cost 1 percent. Thirty five times five is one hundred seventy five. Sixty five times one is sixty five. He is wrong most days and rich most decades.

The second investor is essentially every successful trend following fund in existence. Managed futures firms have built entire businesses on hit rates in the thirties and forties. They lose small over and over and occasionally catch a move that pays for years of losses. Nobody who runs one of those funds would describe their skill as prediction. Their skill is sizing and exiting.

Meanwhile the first investor sounds brilliant at parties. She can list her calls. She was right about the Fed, right about oil, right about the election. Right is a story. Size is a number. Only one of them shows up in the account statement.

Right thesis, wrong size

The graveyard of finance is not filled with people who were wrong. It is filled with people who were right and sized like they could not be early.

Long Term Capital Management is the cleanest example. Their convergence trades were mostly sound. In the years after the fund collapsed in 1998, many of those spreads did in fact converge exactly as the models predicted. The thesis was fine. The problem was that they ran the fund at roughly twenty five to thirty times leverage with over a trillion dollars in notional exposure sitting on a few billion of equity. When Russia defaulted and correlations went to one, the position size that had produced years of smooth returns turned a temporary move against them into a permanent loss. They were right and they were gone.

Michael Burry is the other version of the same lesson. He was correct about subprime years before it happened. Because he was early, his positions bled premium and marked against him, his investors demanded their money back, and he had to gate the fund and fight his own clients to keep the trade alive. He won. He almost did not, and the reason had nothing to do with the accuracy of his analysis. It had to do with whether the position was small enough and financed well enough to survive the gap between being right and being paid.

Being early is indistinguishable from being wrong, and your sizing has to be built for that fact.

The other failure: right and too small

The opposite mistake gets far less attention because it is quiet. Nobody blows up from it. They just underearn forever.

In 1992 Stanley Druckenmiller had done the work on sterling and concluded the pound had to break its peg. He proposed putting the whole fund into the trade. George Soros told him that if the conviction was that high, the position was too small, and pushed him to go several times larger. The trade made around a billion dollars and defined both of their careers.

Druckenmiller’s analysis was already correct before that conversation. Soros added nothing to the forecast. He only changed the size. That is the entire value he contributed to the most famous macro trade in history.

John Paulson’s subprime trade is the same shape. Plenty of people saw the housing market was rotten. A handful acted on it. Almost none of them sized it in a way that produced fifteen billion dollars. Insight was widely distributed. Conviction expressed as capital was not.

Most retail investors have had at least one genuinely great idea. They put 2 percent in it, watched it triple, and made nothing that mattered. The idea was not the constraint.

What good sizing actually looks like

There is real math here. The Kelly criterion tells you the fraction of capital that maximizes long run growth given your edge and your odds. For an even money bet, the formula reduces to twice your win probability minus one. A 60 percent edge implies betting 20 percent of capital, not 100.

Almost nobody should bet full Kelly. Kelly assumes you know your true edge, and you do not. You have an estimate, and your estimate is optimistic, because everyone’s is. Overestimating your edge causes Kelly to overbet dramatically, and overbetting is far more destructive than underbetting. Most professionals run at a half or a quarter of Kelly for exactly this reason. You give up some growth in exchange for surviving the fact that your model of yourself is wrong.

Beyond the formula, a few principles do most of the work.

Size to your uncertainty, not to your excitement. The strength of a feeling is not evidence. Ask what would have to be true for you to be wrong, and how quickly you would find out. Positions where you learn you are wrong slowly deserve to be smaller than positions where you learn quickly.

Size so that being early does not force you out. If a position can be liquidated by a drawdown, by a margin call, by a redemption, or by your own nerves before the thesis has time to play out, it is too large regardless of how good the thesis is. Staying power is part of the position.

Size against correlation, not against names. Ten different positions that all express the same view on rates is one position wearing ten costumes. Correlations rise exactly when you need them not to, which is another way of saying your real position size is largest at the worst possible moment.

Cap what a single mistake can cost. Not because you expect to be wrong, but because the cost of being wrong about your own accuracy has to be bounded. This is the whole game. You are not sizing against the market. You are sizing against your own overconfidence.

The reframe

Stop asking whether you are right. Start asking what happens to your capital across the full range of outcomes where you are wrong, early, or right for reasons that stop being true.

Prediction is the part of investing that feels like intelligence. Sizing is the part that produces money. The people who last are rarely the best forecasters in the room. They are the ones who structured their bets so that a good call could pay for a decade of bad ones, and no single bad one could end them.

Being right is a claim about the world. Bet size is a claim about yourself. The second one is harder to get honest about, and it is the one that determines what you keep.

A paid Substack newsletter is typically worth 2x to 5x its annual revenue. A publication earning $100,000 a year usually sells somewhere between $200,000 and $500,000. Where you land inside that range is decided mostly by churn, then by growth rate, audience quality, and how much the publication depends on you personally.

That is the whole formula in one sentence. The rest of this piece is about the second half of it — because the gap between 2x and 5x is the difference between $200,000 and $500,000 on the same revenue, and that gap is where all the real money is.

Find your substack's worth

Why revenue multiples, and not subscriber counts

Writers tend to think in subscribers. Buyers think in revenue. A 100,000-subscriber free list with no paid tier is worth far less than a 900-subscriber paid list at $15 a month, because only one of them has proven that people will pay.

This is why per-subscriber rules of thumb are misleading. The number that matters is annual recurring revenue, and the multiple applied to it is a judgment about how durable that revenue is. Everything below is really a statement about durability.

Churn is the single biggest lever

Churn decides how long the revenue a buyer is purchasing actually lasts, which makes it the most heavily weighted input in almost any newsletter valuation.

The math is unforgiving. At 7% monthly churn, the average subscriber sticks around about 14 months. At 1.5%, they stay about 67 months — nearly five times longer. A buyer purchasing the second business is buying five years of a relationship; a buyer purchasing the first is buying just over one. They will not pay the same multiple, and they should not.

If you are planning to sell in the next couple of years, cutting churn is worth more than adding subscribers. It compounds into the multiple itself rather than just into this year’s revenue line.

• Below 2% monthly churn: exceptional, and priced as such.

• Around 3% to 4%: normal for a healthy paid newsletter.

• Above 6%: a buyer will discount hard, or walk.

Where your readers live changes what they are worth

Two newsletters with identical revenue are not identically valuable if one list is concentrated in high-income markets and the other is not.

This is not about the revenue you have today — a subscriber pays the same USD price wherever they live. It is about what happens next. A high-income list absorbs price increases better, commands higher sponsorship rates, and is less exposed to currency moves and discretionary-spending shocks. Buyers price that forward risk.

The honest version of this adjustment is a dampened one. Weighting your paid list by GDP per capita against a US benchmark gets you a directional signal, but applying it at full strength would be wrong, since it would discount revenue you are already reliably collecting. Our calculator dampens it heavily for exactly this reason, and lets you turn it down further if you disagree.

Niche, priced the way the market already prices it

An AI newsletter and a recipe newsletter with the same revenue do not fetch the same multiple. Rather than inventing a ranking of which topics are fashionable, the cleanest approach is to read it off the public markets, which price categories all day long.

Compare the price-to-sales ratio of the listed sector closest to your topic against the market as a whole. Information technology trades at a large premium to the index; consumer staples and energy trade at a discount. Finance sits modestly above. That relative spread is a reasonable proxy for how much appetite there is for exposure to your category.

As with geography, the raw spread is far too wide to apply directly — sectors can differ by 8x, and no one thinks a cooking newsletter is worth an eighth of an AI one. Dampen it, and let it tilt the number rather than decide it.

Find your substack's worth

Scale and growth: the two things that move the multiple up

Larger businesses sell for higher multiples. There is less key-person risk, more institutional buyers able to write the cheque, and more of an actual operation to acquire rather than one person’s habit. The effect is roughly logarithmic — each 10x of revenue adds a step, rather than scaling linearly.

Growth works the same way but faster. A buyer is purchasing next year’s revenue, not last year’s. A list still compounding at 3% a month is on a very different trajectory from a flat one at the same size, and gets paid for it.

These are the two factors most within your control on a 12-month horizon, and the two most worth optimising before a conversation with a buyer.

What the model cannot see

Every valuation model, including ours, is blind to the thing that most often decides a newsletter sale: whether the audience follows the writer or the publication.

If readers subscribed for your voice, the business does not fully transfer, and sophisticated buyers know it. Publications with a distinct editorial identity, a repeatable format, or multiple contributors survive the handover far better than a one-person column does. This is why some newsletters sell at 5x and structurally identical ones struggle to clear 2x.

Sponsorship concentration is the other blind spot. Revenue that depends on three advertiser relationships you personally maintain is worth less than the same revenue spread across a thousand subscribers, even though both show up identically in ARR.

Find your substack's worth

Estimate your own number

We built a calculator that runs this whole model: base revenue multiple, adjusted for audience GDP per capita, sector price-to-sales, churn, scale, and growth — with weight sliders on every factor so you can dial out any adjustment you disagree with.

It also shows the distribution rather than a single figure, because a valuation is a range and anyone who gives you one number is selling you something.

Find your substack's worth

This list is different. These are publications that are either small enough that you probably have not seen them, or strange enough that you skipped past them. Every one of them does something the others do not.

I read through the archives of each before writing this. Here is what actually sets them apart.

I made a summary video using blog2video.app

1.

Jib Hunt writes about one thing. Participating whole life insurance as a personal capital system, the strategy usually called infinite banking.

This is a category most finance writers refuse to touch. Say “whole life insurance” in a room of index fund people and you will get a reaction. The strategy has a bad reputation, and a lot of that reputation is earned, because it has been sold badly by people who did not understand it.

What sets Hunt apart is that he writes like someone who expects to be argued with. His posts open with compliance notes. He published a page on the risks. He built a free calculator specifically so readers could check the math without taking his word for it. His piece on why your CPA and financial advisor have never mentioned the strategy does not claim they are hiding something. It makes an incentives argument. A fee-only advisor is paid on assets under management, so any strategy that moves capital out of a managed account shrinks their compensation base. A CPA is usually not licensed to place insurance contracts. Neither of those is a conspiracy. It is just how the industry is structured.

He also sells policies. He says so at the top of every page. I would rather read a writer with a disclosed conflict than one with a hidden one.

Read it if you want to understand a strategy you have probably dismissed, whether or not you end up agreeing with it.

2.

Divyesh Dave writes for accountants. Not for investors who read financial statements, but for the people who produce them.

This is a genuinely underserved corner. Almost every finance Substack is written for someone allocating capital. Very few are written for the professional trying to move from staff accountant to controller to CFO. Dave’s archive is career infrastructure. A five-year roadmap from accountant to CFO. A study framework for CA, CPA, ACCA and CMA exams. A practical guide to finding accounting work in Dubai. A post on when to use SUMIF instead of SUMIFS.

His thesis is that technical competence is not the bottleneck. Most accountants know their work. What holds them back is the inability to explain it, so a lot of his writing is about communication. There is a companion podcast, The Accountant Show, with seventeen episodes.

One honest note. Publishing has been sparse in 2026. If you subscribe, subscribe for the archive, which is substantial, and treat new posts as a bonus.

Read it if you work in accounting or finance operations and every newsletter you follow is written for portfolio managers.

3.

The name is doing something. It is a filter. If you scroll past it because it sounds unserious, you were not the target reader.

C.D. Lawrence writes long research reports at the intersection of alternative energy, biotech, longevity and emerging markets. Recent pieces cover Vietnam’s move from semiconductor assembly to sovereign industrial strategy, iron-air batteries and hundred-hour grid storage, tokenized treasuries replacing settlement infrastructure built in the 1970s, and the gut-brain axis as an investable frontier. The Bloom Energy deep dive runs twenty-six minutes.

What sets it apart is the pairing. Plenty of newsletters do speculative science reporting. Plenty do options strategy. Solar Kitties does both in the same post, so a piece on perovskite nuclear batteries ends with an actual spread structure rather than a shrug. The research is odd and the trade is specific. That combination is rare.

At twenty-six thousand subscribers it is the largest publication on this list, so calling it undiscovered would be a stretch. Calling it underrated is not.

Read it if you want research on things that are not yet consensus, with the trade attached.

4.

This is the one I want you to look at closest.

The publication was created in May 2026. It has fewer than three hundred subscribers. It used to be a semiconductor newsletter before pivoting entirely to precious metals miners.

And it publishes more than any of them.

There is a daily premarket note called Bullion Breakdown that walks through where gold, silver and the mining ETFs are before the open. There is a weekly Ore Report. There are single-name deep dives on Eldorado Gold, Snowline Gold, Dundee Precious Metals, Franco-Nevada and Allied Gold, several a week. There is a scored ranking of twelve junior miners, each graded out of a hundred, updated as the research expands.

That is the shape of a paid institutional mining research product. It is being run by one person for an audience the size of a lecture hall.

The other thing worth noting is that the writer is willing to be specific and wrong in public. The posts name levels, name setups, and then follow up the next morning on whether the setup held. Most newsletters at this size hedge everything into meaninglessness so they can never be scored. This one does the opposite.

Read it if you own miners, or if you want to watch a research operation being built in real time before the subscriber count catches up.

5.

Allied Investment: Market Insight has one format and never deviates from it. Every other week, roughly twenty charts with a sentence of commentary on each.

Vermulm is Chief Investment Officer at Allied Investment Advisors, so he has a day job that involves being right about this. The recent issues cover the widening gap between trailing and forward earnings multiples, semiconductors entering bear market territory after leading the spring rally, CAPE approaching its 1999 level, value stocks sitting in the fourth percentile of historical cheapness, and margin debt near record highs as a share of GDP.

Two things set it apart. First, he sources every chart. Koyfin, Deutsche Bank, GMO, FRED, CME Group, the ChartStorm. You can go check them. Almost nobody does this.

Second, he argues against his own charts. In one issue he shows an overlay comparing the current bull market to the late 1990s, and then immediately warns you to discount it, citing Darrell Huff on how easy it is to make two unrelated lines appear to move together. A writer who undercuts his own most persuasive chart is a writer who is trying to inform you rather than convince you.

There is also a running dry humor to it. He recorded, for posterity, that his Standard Oil of New Jersey joke got zero laughs.

Read it if you want to see the data without someone forcing a narrative onto it.

6.

Mine. Daily posts on markets, financial history and the psychology of investing. Recent pieces on the South Sea Bubble and AI stocks, why Nobel laureates still fail to generate consistent returns, what the FIRE movement gets wrong, and why prediction markets are gambling regardless of what you call them.

If you like the historical angle, you already know where to find me.

The pattern

Look at what these five have in common. None of them are trying to cover everything.

Hunt writes about one insurance strategy. Dave writes for one profession. Lawrence writes about energy and biology. Silver & Gold Alpha writes about metal in the ground. Vermulm writes charts.

The generalist market commentary Substack is finished. There are ten thousand of them and they are all writing the same post about the Fed. The publications worth your inbox now are the ones that picked a lane narrow enough that nobody else is standing in it.

That is the actual lesson here, whether you are reading or writing.

But look at the actual numbers. A modern five day forecast today is as accurate as a one day forecast was forty years ago. Rain forecasts for tomorrow are right more than eighty percent of the time. No other complex, chaotic system gets predicted this well. Not earthquakes. Not elections. Not markets.

This is strange, because weather and markets are built from the same raw material. Both are massive, interacting systems with millions of variables. Both are sensitive to small changes that cascade into large outcomes. Both resist any single clean equation. Yet meteorologists have gotten dramatically better at their job over the last fifty years, while most market forecasters have not.

The gap is not in the difficulty of the problem. It is in the method. If you understand how weather forecasting actually works, you already understand most of what it takes to think clearly about markets.

I made a summary video using blog2video.app

Climate Is Not Weather

A meteorologist never confuses climate with weather. Climate is the long run average. Weather is what happens on a given Tuesday. Knowing that a region has a wet climate tells you almost nothing about whether it will rain three weeks from now. Knowing the climate is still useful. It sets your baseline expectation, the range of outcomes you should not be surprised by.

Markets have the same split, and almost nobody respects it. Valuation is climate. Price action is weather. A cheap market can stay cheap for years. An expensive market can grind higher for a decade. Knowing valuation tells you almost nothing about what happens tomorrow, or next month, or even next year. It tells you the range of outcomes you should expect over a much longer window.

Most retail investors try to use climate data to predict weather. They see a low valuation and expect an immediate bounce. They see a high valuation and expect an immediate crash. Meteorologists do not make this mistake, because their discipline forces the distinction on them daily. Anyone trying to time markets should force the same distinction on themselves.

Ensembles Beat Single Answers

Modern weather forecasts are not single predictions. They are ensembles. A forecasting center runs the same model dozens of times, each with a tiny variation in starting conditions, because the atmosphere is chaotic and small differences compound fast. If ninety of the hundred runs show rain, the forecast says high chance of rain. If the runs are split fifty-fifty, the forecast says uncertain, and a good meteorologist tells you exactly that.

This is the opposite of how most people discuss markets. Financial media wants a single number. Where will the S&P be in December. What is the price target. Pundits deliver a confident point estimate because a range does not make for good television.

The skill worth building is the opposite instinct. Instead of asking what will happen, ask what the distribution of outcomes looks like, and how spread out that distribution is. A market with a tight distribution of likely outcomes is a very different bet than one with a wide distribution, even if the average outcome looks the same on paper. Traders who think in distributions size their positions correctly. Traders who think in single point predictions get wrecked when reality lands outside their one guess.

Confidence Has to Be Calibrated

When a forecaster says seventy percent chance of rain, that number has to mean something specific. Across a large sample of seventy percent days, it should actually rain about seventy percent of the time. If it rains ninety percent of the time on seventy percent days, the forecaster is underconfident. If it rains only forty percent of the time, the forecaster is overconfident and dangerous. Meteorology has decades of data to grade this calibration, and forecasters get held to it.

Nobody grades market pundits this way, which is exactly the problem. A commentator who says a crash is coming will eventually be right, the same way a stopped clock is right twice a day. Nobody tracks how many times they said it and were wrong. Nobody asks what their seventy percent calls actually resolved to.

You can build this discipline yourself even without an external grader. Write down your predictions with a number attached. Not “I think stocks go up” but “I think there is a sixty five percent chance stocks are higher in six months.” Then check yourself later. Most people who do this for the first time discover they are far less calibrated than they assumed. That discovery alone makes you a better investor than most of the people shouting on financial television.

Skill Lives at a Specific Time Horizon

Meteorologists are excellent at one to three day forecasts. They are decent at five to seven day forecasts. Beyond ten days, forecasts degrade toward the climate average, because chaos wins. No amount of computing power fixes this. It is a property of the system, not a limitation of current technology.

Markets have an almost identical shape, just inverted. Nobody has skill at predicting tomorrow’s price move. That horizon is closer to pure noise, dominated by order flow and sentiment shifts that are effectively unforecastable in advance. But skill reappears at longer horizons. Valuation, business quality, and capital allocation genuinely predict returns over five and ten year windows, the way climate genuinely predicts the range of temperatures next August.

The investor’s version of a meteorologist’s discipline is knowing which horizon you actually have edge on, and refusing to pretend you have edge on the other one. Day traders are trying to forecast tomorrow’s weather with climate data. Long term investors who panic sell on a bad quarter are doing the reverse, treating a single stormy day as proof the climate has changed.

Updating Without Losing the Thread

A forecaster who wakes up to new satellite data does not throw out yesterday’s model and start from zero. They update the existing forecast with the new information, shifting probabilities incrementally. The forecast this morning and the forecast last night are related, not unrelated events.

This is the market skill that separates good investors from bad ones under stress. New information should move your position size and your conviction, not force a total identity crisis about your thesis every single day. If you owned a stock because you believed in the five year earnings trajectory, a single disappointing data point should shift your confidence a little, not zero it out. The people who get whipsawed hardest in markets are the ones without a standing model to update. Every headline becomes a brand new forecast built from scratch, and a system with no memory is a system that panics constantly.

Respecting the Edge of the Model

Every meteorologist knows their models eventually break. Rare events like sudden stratospheric warming or unusual ocean current shifts sit outside what the standard models handle well. Good forecasters flag this uncertainty explicitly instead of pretending the model still applies.

The market equivalent is knowing when you are in a regime the historical data does not cover well. LTCM ran models built on decades of correlation data and got destroyed the moment correlations broke down in a way the model had never seen. The South Sea Bubble investors were pricing a company using logic that assumed markets always eventually reflect fundamentals, right up until the mania stopped caring about fundamentals entirely. In both cases, the failure was not bad math. It was trusting a model past the edge of its own validity, and not knowing where that edge was.

The Actual Takeaway

Forecasting weather well is not about eliminating uncertainty. It is about handling uncertainty honestly: separating climate from weather, thinking in distributions instead of single answers, calibrating confidence against real outcomes, matching your claims to the time horizon where you actually have skill, updating gradually instead of resetting to zero, and knowing where your model stops working.

None of that requires a supercomputer. It requires the discipline meteorologists were forced into by decades of being graded on their forecasts. Markets rarely grade anyone this honestly, which is exactly why the discipline has to be self imposed.

Learn to think like a forecaster, not a fortune teller. The weather never rewards certainty. Neither do markets.