Why 15,000 Hedge Funds All Look the Same

Academic finance has been busy. Since Fama and French published their three-factor model in 1993, the research community has identified over 400 distinct "factors" that supposedly predict stock returns. Profitability. Investment patterns. Earnings surprises. Short interest. Analyst revisions. Seasonal effects. Lunar cycles. The list is long, growing, and increasingly absurd. John Cochrane, in his 2011 presidential address to the American Finance Association, called it "the factor zoo" — a menagerie of variables, most of which are either statistical noise or redundant expressions of the same underlying forces.

In practice, the zoo collapses. When you strip away the academic branding and look at what money actually does inside quant funds, nearly every systematic strategy reduces to some combination of four ideas: value (buy cheap assets, sell expensive ones), momentum (buy what's been going up, sell what's been going down), carry (collect premiums from assets that yield more than they cost to hold), and low volatility (own the boring stocks, avoid the exciting ones).

The specific implementation varies. One fund might measure value using price-to-book. Another uses enterprise-value-to-EBITDA. A third has a proprietary composite of seventeen valuation metrics weighted by a machine learning model trained on three decades of data. These are real differences in engineering. They are not real differences in exposure. When the value factor falls — when cheap stocks get cheaper and expensive stocks get more expensive, as happened during the tech bubble, the COVID recovery, and the AI rally — all three funds lose money. The proprietary composite of seventeen metrics provides exactly zero protection against the shared exposure to value.

This is not a controversial claim within the industry. Cliff Asness of AQR — one of the largest and most transparent quant firms — has written extensively about factor crowding. Marcos Lopez de Prado, formerly of Guggenheim Partners, has published research showing that most "new" factors are either repackaged versions of known factors or the product of data mining. The academic incentive is to publish novel findings. The practical reality is that novel findings in finance are extraordinarily rare.

An allocator choosing between quant Fund A and quant Fund B is, in most cases, choosing between two slightly different expressions of the same underlying bet. Fund A charges 2-and-20 for its version. Fund B charges 1.5-and-15 for its version. Both versions will make money when value and momentum work and lose money when they don't. The allocator, believing they have diversified by hiring two quant managers, has in fact doubled down.

On Monday, August 6, 2007, something unusual happened. Every major quantitative equity fund lost money. Not most of them. All of them. Simultaneously. Funds that had never spoken to each other, that operated in different cities with different teams using different models, posted nearly identical losses. Goldman Sachs Global Alpha, Renaissance Technologies, AQR, D.E. Shaw, Citadel — firms that competed fiercely for talent and guarded their strategies as proprietary secrets — all bled at the same time.

The event, later called the "quant meltdown," was a live demonstration of what happens when apparent diversification is actually concentration. These funds had developed their strategies independently. Their code was different. Their signal-generation processes were different. Their risk models were different. But the positions they held — the stocks they were long and the stocks they were short — were nearly identical. Because the factors are the same. Value, momentum, and statistical arbitrage, implemented independently by brilliant teams, produce convergent positions. The same stocks screen as cheap. The same stocks screen as having momentum. The same stocks show up as pairs trading candidates.

The August 2007 event lasted roughly a week. Losses for the most concentrated funds ran to 30% or more. The recovery was swift — most factors reverted within two weeks — but the lesson was permanent. Goldman Sachs Global Alpha, which had been the most profitable quant fund in the world, never recovered its footing. It closed in 2011. The fund didn't die because its models were wrong. It died because its models were right in exactly the same way as everyone else's models.

This dynamic has repeated, less dramatically but more pervasively, many times since. The January 2021 meme stock frenzy punished quant funds that were short the same basket of low-quality stocks. The November 2020 "value rotation" — when COVID vaccine announcements triggered a sharp move from growth to value — was profitable for value funds only if they survived the preceding nine months of losses. Most had already reduced their value exposure because the drawdown had been so painful. The timing of the recovery was random. The crowding of the drawdown was structural.

Investors in these funds were told they were diversified. They owned Fund A (value and momentum, U.S. equities), Fund B (statistical arbitrage, global equities), and Fund C (systematic macro, multi-asset). Three different strategies. Three different pitch decks. Three different risk management frameworks. When August 2007 arrived, all three lost money in the same week, for the same reason, because they were all holding the same positions in different wrappers.

The problem is not that factor investing doesn't work. Over long horizons, value and momentum have delivered real returns. The problem is that when 15,000 funds are all harvesting the same premiums, the premiums shrink, the crowding increases, and the crashes become correlated. Diversification across managers is not diversification across risk. It is the appearance of diversification, maintained until the moment it matters.

Genuine diversification requires more than different implementations of the same idea. It requires different ideas, operating in different markets, driven by different sources of return. This sounds obvious. In practice, it is extraordinarily rare, because most fund managers operate within a single domain — equities, or fixed income, or macro — and diversify within that domain rather than across domains.

Prism runs eleven engines. This is not eleven variations of the same strategy applied to different stock universes. It is eleven structurally distinct approaches to generating returns, spanning genuinely different asset classes and genuinely different market dynamics.

A regime rotation engine reads the macro environment and shifts equity exposure between value and growth depending on the current state of volatility, credit spreads, and yield curve shape. A crypto funding arbitrage engine collects the premium that leveraged long traders pay to maintain their positions — a structural feature of crypto derivatives markets that has no relationship to equity factor premia. An NLP-based scoring engine reads corporate communications — not for sentiment, but for the linguistic patterns that distinguish adaptive management teams from rigid ones — and generates scores that predict forward returns with near-zero correlation to traditional factors.

Mean reversion engines activate during dislocations — when panic drives assets below fair value and the snap-back creates short-term returns. Trend-following engines ride sustained moves in commodity and currency futures, profiting from the behavioral tendency of institutional investors to under-react to new information and then over-react once the trend is established. These two engines are structurally opposed: mean reversion profits from reversals, trend following profits from continuation. Owning both is not a contradiction. It is the architecture of genuine independence. When one is losing, the other is gaining, by design, not by hope.

The correlation between these engines is not low by accident. It is low because the return drivers are fundamentally different. Crypto funding rates are determined by leverage demand in perpetual swap markets — a mechanic that has nothing to do with equity valuations or bond yields. NLP-based scoring captures managerial quality signals embedded in language — a source of information that has no structural relationship to momentum or value. Commodity trend following profits from supply-demand imbalances in physical markets — forces driven by weather, geopolitics, and production cycles, not by the stock market.

This is what structural independence means. Not different implementations of the same factor. Different sources of return from different markets driven by different forces. When the quant equity world has its next August 2007 — and the conditions that caused it have only intensified — Prism's equity engines may lose alongside them. But the other eight or nine engines will be doing something entirely different, because they always are.

Every portfolio looks diversified during a bull market. Correlations are low when everything is going up. The true test of independence is not how engines behave when conditions are benign. It is how they behave when conditions are extreme — when fear is high, when liquidity evaporates, when the selling is indiscriminate.

During the COVID crash of February-March 2020, Prism's equity-exposed engines lost money. The regime rotation engine, despite shifting allocation toward defensive positioning, could not avoid all equity losses when the market fell 34% in twenty-three days. The direct equity loss was approximately -8%. This is honest and expected. Any portfolio with equity exposure will lose money when equities lose a third of their value in a month. The question is what happens everywhere else.

The mean reversion engine gained approximately 18% during the same period. As correlations spiked to one and securities fell indiscriminately, the dislocation created exactly the conditions mean reversion was designed to exploit. Oversold equities, mispriced relative values, panic-driven gaps between fundamentally related securities — these are the fuel that mean reversion engines burn. The worse the crash, the more fuel is available.

The regime rotation engine, reading the shift in real-time through its ten-signal composite, increased allocation to the strategies that thrive in crisis conditions and reduced allocation to those that don't. This isn't prediction. It's responsive reading of the current environment, executed without the attachment to yesterday's positioning that cripples human decision-making. The regime engine contributed approximately +6%.

Other engines were approximately flat. The crypto-linked strategies saw reduced activity as markets locked up. The trend-following components had modest gains from short equity momentum that had begun building before the crash accelerated. The net across all eleven engines was +2.8%. Not because every engine made money. Because the engines that lost money and the engines that made money had nothing to do with each other. The losses didn't compound. The gains didn't cancel. The independence was real.

Compare this to the experience of allocating to three or four quant equity funds that all run value-momentum strategies. When COVID hit, those funds all lost money simultaneously. The "diversification" across managers provided zero protection because the underlying exposure was identical. An investor holding all four funds experienced roughly the same drawdown as an investor holding any one of them. The fees were four times higher. The protection was zero times better.

The 2022 experience sharpened the point. When stocks and bonds fell together, the traditional 60/40 portfolio lost 16%. Quant equity funds, which are benchmarked against equity markets, lost money alongside the market they were supposed to beat. Prism's non-equity engines — crypto carry, commodity signals, regime rotation into cash and alternatives — operated in a different universe. The equity engines lost. The portfolio didn't. Because the engines are genuinely independent, not diversified in name while correlated in practice.