Twenty-seven books and papers on finding exploitable patterns in financial markets, condensed into one essential read
Snuffling for Truffles: A Master Read on Pattern-Seeking in Markets to Generate Alpha. First Edition. Published May 2026.
Compiled and authored by Tony Woodhams & Dmitry Shibaev, Co-authors, Alphaxis Partners. This document synthesises publicly available academic papers, books, and market research. All sources are cited.
Intended for sophisticated financial practitioners. Does not constitute investment advice. Past performance of strategies described herein does not guarantee future results.
© 2026 Alphaxis Partners. All rights reserved. Internal distribution only.
In This ReadWhat all twenty-seven works are really trying to answer
Markets contain patterns. The question is whether those patterns are real, exploitable, and durable, or whether they are mirages conjured by selective memory, wishful mathematics, and the relentless urge of intelligent people to find order in noise.
That question has occupied the best quantitative minds in finance for over six decades. Jegadeesh and Titman showed in 1993 that past winners keep winning. But devastating critiques suggest that the vast majority of "discovered" patterns are statistical artefacts: the inevitable output of testing 400 hypotheses on the same dataset.
Both things can be true simultaneously. Some patterns are real. Most aren't. The job of a serious quantitative practitioner is to tell the difference, and then act on the real ones before they decay.
"The truffle is there. But most of what looks like a truffle isn't."
Every book and paper in this reading list is trying, in one way or another, to answer a version of the same question: where is the alpha hidden, and why hasn't it been arbitraged away? The answers cluster into three broad camps.
Markets misprice assets systematically because human beings are irrational in predictable ways. We overreact to recent news, underreact to earnings surprises, hold losers too long, sell winners too early. Kahneman, Shefrin, Thaler, De Bondt, and Shiller are the central voices.
Certain structural characteristics, size, value, momentum, profitability, low volatility, carry persistent return premia. Whether risk or behavioural mispricings, they show up reliably. Fama, French, Asness, Ilmanen, Berkin, and Swedroe lead this group.
Be very careful. Harvey, Liu, and Zhu documented 316 published factors. Hou, Xue, and Zhang tried to replicate 452 anomalies: 65% vanished under proper testing. McLean and Pontiff showed anomalies lose 58% of their returns post-publication.
"Simons and his mathematicians discovered that markets contain faint, recurring patterns across thousands of instruments, and that the edge in each is tiny but real. The Medallion Fund compounded at 66% annually from 1988 to 2018. That is not luck."
Evidence spans two centuries and 40+ countries. Buying past winners and shorting losers generates ~1% per month. Asness et al. (2013) proved it works in bonds, currencies, and commodities too. The crash risk is real: momentum can lose 30-40% in weeks when volatility spikes.
The value premium has equivalent statistical power to momentum but requires 3-5 year holding periods and psychological fortitude. It can underperform for a decade. Value combined with profitability (Novy-Marx) is the sweet spot.
Harvey, Liu & Zhu: 316 published factors. The correct significance threshold should be t > 3.0, not 2.0. Hou et al.: 65% of 452 anomalies fail to replicate properly. Trading frictions literature: 96% failure rate.
Kahneman's two-system model is the psychological substrate beneath every anomaly. Disposition effect, earnings underreaction (PEAD), post-announcement drift all trace back to predictable cognitive biases. Limits of arbitrage (Shleifer and Vishny, 1997) explain why they persist: real arbitrageurs face redemption risk.
Gu, Kelly & Xiu (2020): neural networks and tree-based methods doubled out-of-sample Sharpe ratios. But the most important predictive signals were variations on momentum, liquidity, and volatility, as conventional quants had already identified. ML finds better interactions between known patterns, not fundamentally new ones.
The January effect persists because tax-law mechanics create predictable flows. Options expiry dynamics create systematic price magnetism around large open interest strikes. Structural anomalies, rooted in regulation, market mechanics, or institutional constraints, are more durable than behavioural ones.
McLean and Pontiff (2016): 58% of anomaly alpha disappears after publication. Diversification across many uncorrelated signals is the only sustainable defence.
Narang: failure at any one of five layers (alpha model, risk model, transaction cost model, portfolio construction, execution) destroys the edge from the others. Carver: how much to trade is often more important than what to trade. Volatility targeting is empirically superior to fixed position sizing.
Medallion Fund: 66% average annual returns before fees, 30+ years, all market conditions. The edge in each individual pattern was tiny, faint signals that would not survive casual statistical scrutiny. Medallion never published its findings. The academic literature is a gift to those who find patterns early and a tax on those who follow published research.
A small number of factors, momentum, value, profitability, carry, low volatility, have delivered persistent risk-adjusted returns across multiple geographies. These are the foundation. Around them, structural anomalies can be layered carefully. At the frontier, micro-edges found by ML are small, fast-decaying, and require constant renewal.
The founding document of momentum research. ~1% per month excess return, consistent across all sub-periods, unexplained by known risk models. The 12-1-3 formation-skip-hold specification remains the canonical benchmark.
The most complete single reference in quantitative finance. Every major return premium covered: value, momentum, carry, low volatility, illiquidity. Honest treatment of when each style works and fails. Central finding: combining uncorrelated premia produces substantially better risk-adjusted returns.
The essential methodology text. Identifies specific failure modes: data leakage, triple barrier labelling, purged k-fold CV, Probability of Backtest Overfitting. Most published "edges" are curve-fits to historical noise.
The closest window into Renaissance Technologies. 66% pre-fee annual returns, 1988-2018. The edge comes not from one brilliant insight but from a vast, constantly refreshed library of tiny statistical patterns. Medallion's methodology: identify, validate, explain.
Value and momentum generate statistically significant excess returns in every single asset class tested across 40 years and 8 markets. They are negatively correlated (~-0.50), meaning combining them substantially reduces portfolio volatility without sacrificing expected return.
Tested 15 ML methods on 94 firm characteristics (1957-2017). Neural networks doubled out-of-sample predictive R2. The top predictors were momentum, short-term reversal, and volatility, exactly what conventional quants had already found. ML improves combination logic, not discovery of new patterns.
Each asset's own past 12-month return predicts its own future return. Tested on 58 liquid futures across equities, bonds, currencies, and commodities. Sharpe ~1.28 combined. Documents the "time series momentum crash", brief, severe reversals correlated with funding liquidity crises.
Reviewed 316 published factors. With this many tests, the correct significance threshold is t > 3.0, not 2.0. Even a t-stat of 4.0 may indicate a factor with less than even odds of being real under multiple-testing correction.
Five empirical criteria applied to every factor: persistent, pervasive, robust, investable, intuitive. Only five factors pass all five: market beta, size, value, profitability, and momentum. A useful triage framework for evaluating strategy ideas before investing research time.
The anatomy of a quant fund: five interacting layers (alpha model, risk model, transaction cost model, portfolio construction, execution). These layers are multiplicative: excellent alpha generation + naive execution = live results far below backtest. Every failure mode is documented.
Smooth momentum (high return achieved through a steady upward path) significantly outperforms jagged momentum. The "Frog-in-the-Pan" hypothesis: steady incremental price increases attract less attention, causing less crowding and more durable trends. Volatility regime filter reduces momentum crash risk.
System 1 (fast, emotional, pattern-matching) vs System 2 (slow, rational, expensive) cognition. Loss aversion, anchoring, availability heuristic, disposition effect, all create predictable mispricings. Hindsight bias and planning fallacy also corrupt the research process itself.
Volatility targeting: scale each position so its expected daily dollar volatility is constant. Naturally reduces exposure when markets become more volatile, a built-in risk management feature. "Democracy of instruments": every additional uncorrelated instrument improves Sharpe, with diminishing returns.
Attempted replication of 452 published anomalies with consistent methodology. Result: 65% produce statistically insignificant alphas. Under multiple-testing correction: 82% fail. Trading frictions literature: 96% eliminated. Surviving anomalies concentrated in momentum, profitability, and investment factors.
97 published predictors tracked across three periods. Post-publication decay: -26% before widespread adoption, then -58% as investors trade the anomaly. Predictors with higher trading volume and more academic citations decay fastest.
The paper that launched behavioural finance. Prior-3-year loser portfolios outperformed prior-3-year winners by ~25% in the following 3 years. Initial underreaction creates momentum; eventual overreaction creates the subsequent reversal. This explains why 12-month lookbacks work while 3-5 year lookbacks produce the opposite signal.
WorldQuant's model: treat each insight as an "alpha" (a function mapping from data to position), combine thousands of weakly correlated alphas. Quality metric: Information Coefficient (IC) and its stability (ICIR). An alpha with IC = 0.05 and ICIR = 1.5 is more valuable than IC = 0.15 with ICIR = 0.4.
Stocks reporting positive earnings surprises continue to outperform by 2-5% over the following 60 days. Not risk-based; present in all quarters; increases with surprise magnitude; persists for multiple quarters. The drift is most pronounced for smaller companies with lower analyst coverage.
Rational arbitrageurs (who manage other people's capital) face constraints: redemption risk, margin calls, career risk. They cannot hold positions indefinitely against temporary adverse price movements. This explains why anomalies are largest in the least liquid segments of the market, not because sophisticated investors don't see them, but because the constraints are too high.
Gross profitability (revenue minus COGS / assets) predicts returns as well as book-to-market value, and is essentially uncorrelated with it. Cheap and profitable is the sweet spot, a natural filter against classic value traps.
~50 strategies tested on US stocks 1964-2009. Price momentum is the single most powerful predictor. "Trending Value" (composite value score + 6-month momentum) produced the highest risk-adjusted returns over the full 45-year period.
Addresses ten commonly cited objections to momentum: large caps, transaction costs, risk exposure, US-specific, ML replacement, crashes, value superiority, turnover, short-term reversal, current markets. Each refuted with data from multiple geographies and time periods.
Only objective (mechanically defined, reproducible) technical analysis can be scientifically tested. Subjective pattern recognition cannot. Bootstrap and permutation testing as corrections for the data-mining problem in technical analysis.
CAPE above 25-30x predicts low subsequent 10-year real returns. Not a tactical timing tool but a macro regime context for sizing aggregate equity risk. Published exactly at the NASDAQ peak in March 2000. The argument that market bubbles are social phenomena driven by narratives is relevant to understanding crypto cycles.
The definitive reference on market microstructure. Economics of market making, adverse selection, order types. Key insight: bid-ask spread is not merely a cost to minimise; it is a signal. Wide spreads indicate high uncertainty and frequently predict near-term volatility.
More philosophical argument than technical manual. Aggregates long-run track records of Winton, Man AHL, Campbell, and Millburn. The key message: systematic, rules-based trend following has produced the best risk-adjusted returns through 2008 and 2001-2002. Follow rules without override, accept extended underperformance without treating it as evidence against the approach.
Makes the multiple-testing problem viscerally intuitive via simulation. If 200 researchers each test 200 candidates, the expected number of factors with t > 4.0 purely by chance is approximately 1. T-statistics cannot be interpreted in isolation without knowing how many tests were run to find the factor.
If time allows only three texts; read these, in this order
The most complete single text in the field. Every major factor and strategy style, when each works and fails, how to combine styles in a portfolio. If you only read one book about why certain patterns generate alpha, this is it. Allow 8-10 hours. Read value, carry, momentum, and the section on combining styles first. Skip fixed income deep dives on first pass.
The most important methodology book for anyone running systematic strategies today. Read chapters on financial data structures, labelling, cross-validation, and backtesting. Even without ML, the framework for data leakage and overfitting applies to every backtest we run. The chapter on combinatorial purged cross-validation is dense; read it twice. Allow 6-8 hours on relevant chapters.
Download the PDF from SSRN. 60 pages. Read it as a corrective discipline, a reminder that the bar for claiming a real pattern is much higher than it feels when a backtest looks good. The table of 316 factors is genuinely humbling. Allow 3-4 hours.
This trio takes you from "what are the real patterns" (Ilmanen) through "how do I avoid fooling myself" (Lopez de Prado) to "just how much fooling myself is possible" (Harvey et al.). Everything else in the list builds on these three.
What the best practitioners in this literature would say if they reviewed our current strategy stack
We run systematic, rules-based strategies, not discretionary trading. This is the single most important methodological choice in the field. Carver, Covel, Chan, and virtually every serious practitioner agree: systematic outperforms discretionary because it eliminates the behavioural biases Kahneman identified. The Ron review gate, requiring independent validation before capital deployment, is exactly the institutional discipline that separates research-grade from deployment-grade claims.
Our strategy stack is heavily momentum-weighted (Flying High, Dual MA, Weekly Opening Range). The literature on value/momentum correlation (-0.50 to -0.60) shows that holding a contrarian or value-tilted strategy alongside momentum improves Sharpe without reducing expected return. Novy-Marx's gross profitability measure would be a natural complement.
WorldQuant's "Finding Alphas" describes a model where hundreds of individual signals are combined into a portfolio. Alphaxis is building full strategies, each with complex logic. This is not wrong, but our diversification is coarse. A signal-level framework, where each market insight is treated as an alpha and combined in a portfolio, would improve risk-adjusted returns and reduce concentration.
Every strategy must pass Harvey et al.'s implicit t-statistic threshold of 3.0 (not 2.0), must survive at least two different out-of-sample periods, must have a stated mechanism, and must show fee-adjusted performance with realistic transaction costs before it is considered live-ready. Not all tests that show a positive Sharpe are real.
There is a truffle buried somewhere in every market. It is real. It generates real returns. Medallion's 66% annual return, compounded across 30 years, proves it beyond reasonable doubt. Something is there.
But the undergrowth between the truffle hunter and the truffle is extraordinarily dense, and most of it is deceptive. The 452 anomalies that Hou, Xue, and Zhang tried to replicate looked, in their original papers, exactly like the real thing. And 82% were false discoveries. The academic finance industry has, for decades, been producing a kind of truffle-hunting map that mostly leads to empty holes.
"The biggest mistake in this business is to mistake luck for skill, noise for signal. The second biggest mistake is to give up because patterns decay and replace them with nothing."
What distinguishes the real from the fake? Real patterns share certain properties: they persist across very long histories; they appear in multiple asset classes without requiring asset-class-specific stories; they survive with proper transaction cost modelling; they decay slowly as awareness grows; and they have at least one plausible psychological or structural mechanism that explains why the inefficiency is not immediately arbitraged away.
The behaviouralists gave us the why. The factor hunters gave us the what. The sceptics gave us the discipline: none of this is as easy as it looks, most published results are noise, and the bar for conviction must be very high.
For Alphaxis, the practical synthesis is this. We are a truffle-hunting firm. Our competitive advantage is not access to superior data, nor to superior technology. It is methodology: the discipline to test rigorously, to demand multiple-period confirmation, to look for mechanisms rather than patterns, and to resist the temptation to declare a result real before it has earned that status through live performance.
The Medallion Fund's three-step methodology: identify, validate, explain. It is simple enough to memorise but hard enough to execute that most practitioners never manage it. Alphaxis is trying to execute it at a fraction of Medallion's scale, with a fraction of Medallion's resources, but with the same intellectual framework. That is the right aspiration.
"The truffles are there. The question is always: is what I'm smelling actually a truffle, or is it just the damp earth making its familiar promises?"
Stay disciplined. Stay sceptical. And when the real thing surfaces, you'll know it, because it survives everything you throw at it.
Tony Woodhams & Dmitry Shibaev · Co-authors, Alphaxis Partners · 12 May 2026
27 books and papers · 27,000 pages distilled · 18 minutes to read
This document is the first in the Alphaxis Research Series, a programme of internally commissioned research briefings covering topics relevant to the firm's investment programme and intellectual development.
The Research Series is compiled by the Alphaxis intelligence function and is not for distribution outside the firm. It draws on publicly available academic papers and published books. All source materials are identified by title, author, and publication year.
This synthesis does not constitute investment advice. It does not represent the views of Alphaxis Partners Ltd as a regulated entity. References to Alphaxis strategies describe internal research programmes and do not represent offers or performance guarantees.
The Alphaxis Research Series is published on an irregular basis as material of sufficient quality and relevance is identified.
Typeset in Raleway. Compiled May 2026. © 2026 Alphaxis Partners Ltd. All rights reserved.
Alphaxis Partners Ltd is incorporated in Gibraltar (Company No. 119454). Suite 4, 4 Giro's Passage, Gibraltar GX11 1AA.
The flipbook view is best on desktop. On mobile, please open this on a larger screen, or download the PDF version.