In their August 2026, Academic Alpha: A (Re)Introduction to Style Premia Investing, Thomas Maloney and Tobias Moskowitz argue that a class of well-researched, transparent, and systematic strategies still deserves a place in modern portfolios. These strategies, often grouped under alternative risk premia or style premia, seek returns from persistent patterns such as value, momentum, carry, and defensive investing. The ideas are public. The implementation is where the edge lives.

This distinction matters because style premia suffered a reputational collapse during the Quant Winter of 2018 to 2020. Several products lost money, investor patience evaporated, and parts of the industry closed. Yet the strategies that survived later recovered strongly. The episode exposed a deeper truth: investors often abandon diversification precisely when it begins behaving differently from everything else they own.

Academic alpha is therefore more than a factor story. It is a test of whether investors can separate a valid long-run premium from a short-run period of pain.

Read more

Japan has just conducted one of the most aggressive currency defence campaigns in modern financial history. Between July 30 and August 26, Japanese authorities spent ¥15.4 trillion, equivalent to roughly $96.5 billion, buying yen in the foreign exchange market. The intervention included rare coordination with the United States and briefly lifted the currency from almost ¥164 per dollar to around ¥155.20.

Less than a month later, USD/JPY was back near 160.

On August 28, the exchange rate reached approximately 159.97, rising 0.42 percent in a single session. The yen remains around 8.9 percent weaker than one year ago. More than half of the exchange rate improvement created by the intervention has already disappeared. Tokyo deployed a record amount of capital, shocked speculative traders and secured international support, yet the market gradually returned to the same fundamental calculation that had weakened the yen in the first place.

Read more

That distinction does little to soften the commercial verdict. OpenBB had one of the strongest brands in open-source finance. Its main GitHub repository accumulated more than 72,000 stars, 7,500 forks, and nearly 6,900 commits. The company says its products reached millions of people. It built a Python data platform, a web workspace, an AI copilot, an Excel integration, a Snowflake application, an app marketplace, and support for MCP before agentic finance became fashionable. (Check OpenBB on GitHub)

Yet admiration did not become enough revenue. OpenBB’s story therefore deserves more than the familiar explanation that startups often run out of cash. It exposes a deeper problem in financial software: a product can be technically excellent, widely used, and directionally correct while occupying the least profitable layer of the value chain.

The Viral Beginning Created the Wrong Commercial Signal

Read more

The model is presented as a state-event generative foundation model for market microstructure. It converts order events into tokens, conditions generation on the limit order book, and passes every generated order through a deterministic matching engine. Rather than producing one price forecast, it can generate many executable future trajectories.

This shift matters because the most valuable questions in trading are often counterfactual. What happens to the spread if liquidity suddenly vanishes? How does a large execution program alter the orders that follow it? How wide is the distribution of possible prices over the next five minutes? Historical data contain only the path that actually occurred. A credible market world model could let researchers examine the paths that did not.

Read more

In a wide-ranging conversation with Nicolai Tangen, Jensen describes three forces reshaping global markets. The first is the rise of modern mercantilism, as economic policy becomes increasingly organised around national security and industrial power. The second is an AI investment cycle moving from software into power, chips and physical infrastructure. The third is the extraordinary concentration of global wealth in US equities and illiquid assets.

Together, these forces point toward a more fragmented, capital-hungry and inflation-sensitive world. The market may still look familiar on the surface, but the machinery underneath it is changing rapidly.

Read more

This is the central argument of Intelligent AI Delegation, a paper by Google DeepMind. The authors contend that current multi-agent systems depend heavily on hard-coded workflows, brittle prompts, and simplistic routing heuristics. These methods may work inside controlled demonstrations, but they become dangerous when agents interact across companies, financial systems, supply chains, and public infrastructure.

Read more

Google DeepMind’s ICLR 2025 study, “Generative Verifiers: Reward Modeling as Next-Token Prediction,” proposes an elegant solution. Instead of training a verifier to compress its judgment into an opaque numerical score, the researchers train it to generate a critique and predict whether the solution is correct through an ordinary Yes or No token.

This apparently small redesign produces striking results. On algorithmic reasoning tasks, Best-of-32 performance rises from 5.0% to 45.3%. On GSM8K grade-school mathematics, it increases from 73.0% to 93.4%. A verifier trained only on grade-school mathematics also generalises to harder competition problems, raising MATH500 performance from 28.0% to 44.6%.

The deeper lesson is clear. Reasoning systems may improve dramatically when evaluation becomes a form of reasoning itself.

Read more

A new paper from researchers affiliated with Princeton, Stanford, and Ant Group confronts that danger. AQuA contains two independent research systems. One discovers crypto factors. The other develops models for US equities. Each remembers what worked, what failed, and why, then uses that evidence to shape the next experiment.

The results are striking. The crypto system reaches a combined information coefficient of roughly 0.190. The equity model produces a per-stock IC of 0.0843 and a held-out Sharpe as high as 2.50 after costs. Yet the deeper idea is architectural: if an AI researcher cannot reliably avoid leakage, remove its ability to create leakage.

Read more

By that measure, something serious has happened at Man Group. In a recent Bloomberg Odd Lots conversation, Chief Technology Officer Gary Collier and Head of Data and AI Tushara Fernando disclosed that internal token usage has grown roughly 86 times since January of this year. That is not a technology team experimenting at the margins. Compounded over the period, the growth rate sits somewhere around 50 to 65 percent per month, sustained without pause. Curves like that do not come from a pilot project. They come from a workflow that has become load bearing.

Read more

Its Applied AI team recently revealed the architecture behind PAT, the Pocket Analyst Tool. According to Bridgewater, PAT has been deployed internally for several months and is used by hundreds of investment professionals. It can complete certain exploratory research tasks in minutes that would previously have occupied a specialist analyst for hours.

The speed is impressive, but it is not the most important part of the story.

Bridgewater is building PAT around a fundamentally different idea from the universal AI assistant. Instead of asking one general agent to search, calculate, code, validate and explain everything, the firm decomposes investment research into specialized workflows. Different agents handle data discovery, research planning, code generation, execution, validation and institutional learning.

Together, these capabilities point toward a much larger ambition: an Artificial Investor capable of performing the full research cycle currently completed by human investors.

Read more

Then you actually run one. After three or four days of continuous iteration, something strange shows up in the logs. The type of error keeps changing. The total volume of errors barely moves.

Three Days of Errors That Never Actually Shrink

In the early rounds the failures are directional. Momentum, reversal, volatility, the whole canon of classical paradigms marches through one after another, backtest IC means hover near the floor, and the pass rate stays miserable. In the middle rounds the failures become operational. Field names are wrong, data types do not match, and the backtest engine simply throws and returns. In the late rounds the system starts making strategic adjustments on its own, running a correlation self check before submission and voluntarily discarding any candidate that sits too close to the existing pool.

Read that arc quickly and it looks like maturation. Rookie mistakes, then mechanical competence, then strategic self discipline. That is roughly the career path of a human researcher, compressed into seventy two hours.

The analogy is seductive. The systems we have actually run give a different answer.

Read more

Across the last several years of enterprise AI deployment, a consistent pattern has emerged in project reviews. Of roughly fifty self-described “enterprise AI agent” builds examined in recent months, the overwhelming majority, on the order of nine in ten, reduce on inspection to a chat window wrapped around a system prompt. A user types a question, a model returns text, and a thin layer of role-play sits between them. Such systems demo beautifully. Placed in front of real business traffic, they typically expose their limits within days.

One retail deployment illustrates the failure mode precisely. A customer asked the agent to cancel an order. The agent responded with a well written, entirely useless walkthrough of how order cancellation works. The model had classified a command as a knowledge question, because nothing in the architecture gave it a path to execute anything. Model capability was never the constraint. The architecture was wrong before the first token was generated.

The distinction that matters is a shift in job description. A chatbot answers questions. An enterprise agent interprets a task, selects the right internal capability, executes against real systems, and returns a result that changes the state of the business. That transition demands four things at minimum: business intent understanding, invocation of internal capabilities, control over permission and data boundaries, and orchestration of multi-step work. None of those four can be solved by tuning a prompt.

Anatomy of One Request

Read more

A team has published a framework that attacks this directly. The study Self-Play Meets Skill Evolution, introduces SESA, short for Self-Evolving Skill-Augmented Agent. The core move is deceptively simple: take the failures that self-play normally discards, write them down as human-readable strategy cards, and feed them back into the next round of training. The results across seven question-answering benchmarks and seven model backbones are consistent, and the ablation study contains a finding that should change how practitioners think about memory-augmented agents.

Read more

The paper is called Toward Efficient Agents: A Survey of Memory, Tool Use, and Planning, by seventeen researchers spanning nine institutions. It runs to 63 pages and 244 references, and it tabulates 117 distinct methods. What makes it worth your attention has nothing to do with its size. The survey asks a question the industry has been quietly avoiding while it celebrates capability: when an agent gets smarter, who pays, and how much?

Read more

In May, Tencent Cloud went overseas with WorkBuddy, pitched not as a chatbot that answers questions but as an AI office partner that hands back finished work. Fewer than sixty days later, in July, Andrew Ng released OpenWorker with a tagline that reads like an echo: an agent that “doesn’t just chat with you, but delivers finished work.” Open source, local first, model agnostic.

When two teams on opposite sides of the Pacific converge on the same sentence within a single quarter, the sentence stops being marketing. It becomes a market signal. The interesting question is no longer whether AI coworkers are coming, but which layer of the stack captures the value once they arrive.

What Actually Ships When You Install It

Read more

Step 1: Pick your connection method

There are two ways to connect, depending on where your client runs.

Opt…

Read more

1. How deep is the wound

For the first half of 2026, momentum was the undisputed king of global equities. On Goldman Sachs’ construction, its momentum factor index returned roughly +57% in H1, not a single stock, but a long/short hedged factor portfolio.

Then, starting from the June 22 high, it gave back 22% in five weeks.

The same disclosure from Tony Pasquariello, head of Goldman’s hedge fund business, came with two other details: sentiment indicators sitting in the 98th percentile over five years, and speculative leverage at multi-year highs.

Read more

Prices sit on one screen, company filings on another, macroeconomic releases arrive through a calendar, institutional holdings hide inside 13F reports, and prediction markets move while most of the world is asleep. The real challenge is converting these sc…

Read more

What makes the paper worth an hour of your attention has little to do with the deep learning fashion of the moment. The provocation sits deeper. FinGAT argues that the relationships between listed companies, the invisible web of supply chains, rivalries and shared exposure that every experienced investor senses but nobody can document, can be learned directly from price movements alone. No knowledge graph. No industry database. No analyst telling the model that TSMC feeds Apple. Just prices, and a network that infers who is listening to whom.

Why Knowing the Price Is Not the Same as Knowing the Trade

Read more

The framework below comes from Steve Hanly of Engineered Portfolio, who treats a financial life the way an engineer treats a bridge: define the load, choose materials that survive stress, and model the failure points before you ever pour concrete. What follows translates his five-part system into plain language, with the actual numbers that make each move worth your attention.

Start With the Finish Line, Not the First Step

Read more