QuantSeeker

Is There Alpha in the COT Report?

QuantSeeker — Fri, 12 Jun 2026 22:40:18 GMT

Every Friday at 3:30 PM Eastern, the CFTC publishes the Commitments of Traders report, a public record of how hedge funds, commercial hedgers, and other market participants are positioned in US futures markets. Traders have used this data for decades.

The obvious question is: Does it actually predict returns?

I discuss recent research on this topic and test a commodity signal that appears to predict returns. Importantly, I examine what happens when you account for publication timing and implement the strategy in a way that is actually tradable.

Weekly Research Recap

QuantSeeker — Tue, 09 Jun 2026 20:33:43 GMT

Each Tuesday, I share the most interesting market and investing insights I came across during the week, including new research papers, blogs, and podcasts. Links are included throughout for readers who want to explore the ideas in more detail.

Crypto

Price Path Continuity and the Cross-Section of Cryptocurrency Returns (Kim)

Crypto momentum is more nuanced than it appears. Cryptocurrencies whose gains are built through many small, persistent price increases significantly outperform those driven by a few large jumps. The effect is strongest outside the largest coins. Key takeaway: Smooth trends typically contain information that investors incorporate more slowly than explosive price jumps do, leading to greater predictability of returns.

The Election Anomaly in Bitcoin Returns (Shanaev, Maksikov, and Vasenin)

Investors often debate whether Bitcoin is digital gold. This paper points to a different role: A hedge against political uncertainty. Across 60 major elections in G20 democracies since 2010, Bitcoin gained roughly 2–3% on election days on average, regardless of which party won. Key takeaway: Bitcoin’s value may partly stem from its role as a politically neutral asset when uncertainty spikes.

Crypto has Fundamentals: A Seven-Factor Model for Digital Asset Returns (Babayev and Aliyev)

A common belief is that crypto is disconnected from fundamentals. This paper suggests otherwise. Across 90 cryptocurrencies, the strongest return predictor wasn’t size or momentum; it was on-chain quality, a measure of actual network usage. The authors also find short-term reversal, with recent losers outperforming recent winners. Key takeaway: Blockchain fundamentals seem to contain information about future returns that markets do not fully price.

Equities

Rejoicing, Regret and Stock Returns – US and International Evidence (So and Zhang)

This paper introduces a signal based on a stock’s volatility-adjusted performance relative to industry peers. Stocks that recently underperformed their peers outperformed recent industry winners by over 16% annually in the U.S. The signal resembles an industry-relative short-term reversal strategy but remains predictive after controlling for traditional reversal measures. Key takeaway: Recent industry laggards typically offer better opportunities than recent leaders.

The Power of Position: Display Salience in Specialized ETFs (Park)

Stocks displayed in a thematic ETF's top-10 holdings attract more retail attention, see larger increases in retail ownership, and outperform other constituents by about 1.6% around ETF launch. Most of the gain later reverses. Key takeaway: Visibility can create temporary price pressure, even when fundamentals haven't changed.

Option-Implied Market Risk Premium and Asset Pricing (Kang, Chen, and Gan)

For decades, investors have questioned whether higher-beta stocks actually earn higher returns. This paper suggests the answer depends on the market’s expected risk premium. Using a forward-looking measure derived from S&P 500 index options, the authors find that the classic beta-return relationship reappears when investors demand unusually high compensation for market risk. Key takeaway: Beta matters most when aggregate risk aversion is high.

Crowded Anomalies over the Business Cycle (Jung)

Many anomaly strategies look similar on average, yet behave very differently across the business cycle. Using a new measure called excess centrality, the author identifies which anomalies are most exposed to macroeconomic conditions. Key takeaway: Anomaly returns seem to depend as much on macroeconomic regimes as on the signals themselves.

Aggregation Consistency and Return Predictability: Evidence from CAPE Ratios (Ma, Marshall, Nguyen, and Visaltanachoti)

Traditional CAPE ratios implicitly weights firms by earnings, while the market itself is value-weighted. Rebuilding CAPE from the stock level and then value-weighting the components improves long-horizon return forecasts by more than 10%. Key takeaway: Sometimes alpha comes from measuring familiar signals more correctly, not from inventing new ones.

Machine Learning and Large Language Models

Dynamic Multi-Pair Trading Strategy in Cryptocurrency Markets with Deep Reinforcement Learning (Lebiedz and Slepaczuk)

Adding a Deep Reinforcement Learning execution layer on top of a traditional mean-reversion strategy improves out-of-sample performance in crypto pairs trading. Rather than generating signals, the model learns how to execute them while operating within predefined risk limits. Key takeaway: In systematic trading, execution can be just as important as signal generation.

The Implied Volatility Surface and the Cross-Section of Stock Returns: Evidence from Machine Learning (Ye, Xiong, Yang, Jiao)

Investors often compress option markets into a handful of indicators such as skew, smirk, or volatility spreads. This paper suggests that compression may discard valuable information. Machine-learning models that analyze the entire implied volatility surface generate stronger stock-selection signals than most traditional option metrics. Key takeaway: The predictive information in option markets seems to reside in the full volatility surface rather than in any single option metric.

Macro Economists in the Machine: A Multi-Agent LLM Framework for Commodity-Related ETF Portfolio Construction (Wang, Dai, Ma, and Geng)

Investors often assume macro investing is a forecasting problem. This paper suggests it may be an interpretation problem. Using the same economic data and the same portfolio rules, LLM agents modestly outperformed a traditional macro model by handling conflicting signals more flexibly. Key takeaway: Better decisions can come from combining existing information more intelligently rather than collecting more of it.

Blogs

UFC - Ultimate Fitting Championships (Evaluating and calibrating portfolio optimisation methods with random data) (Rob Carver)

Forecasting statistical estimates when data gets real (Rob Carver)

The crossword puzzle of fitting - why across and then down? (Rob Carver)

Resourcing a Triangulated Stat Arb Operation as a Solo Trader (Robot Wealth)

The Non-Linear Costs of Trading (Concretum Group)

Podcasts

Jack Schwager and George Coyle Interview with Michael Covel on Trend Following Radio (Michael Covel)

Ben Carlson: Investing at All-Time Highs (Rational Reminder)

Social Media & Industry Research

Don’t Look Down: Reflections on Cross-Asset Drawdowns (Man Group)

A pessimistic take on optimistic growth forecasts (Acadian Asset Management)

Reading Between the Lines: Natural Language Processing for Long-Horizon Factor Investing (Research Affiliates)

Last Week’s Most Popular Links

Harvesting Factor Premia Across Regimes: An Anchor-Stabilized Hidden Markov Framework for Multifactor Portfolios (Yang)

Capital Market Assumptions And Strategic Asset Allocation Using Multi Asset Tradable Factors (Sepp, Hansen, and Kastenholz)

Recession Detection Using Real Time GDP Data (Sikand and Zhang)

Disclaimer: This newsletter is for informational and educational purposes only and should not be construed as investment advice. The author does not endorse or recommend any specific securities or investments. While information is gathered from sources believed to be reliable, there is no guarantee of its accuracy, completeness, or correctness.

This content does not constitute personalized financial, legal, or investment advice and may not be suitable for your individual circumstances. Investing carries risks, and past performance does not guarantee future results. The author and affiliates may hold positions in securities discussed, and these holdings may change at any time without prior notification.

The author is not affiliated with, sponsored by, or endorsed by any of the companies, organizations, or entities mentioned in this newsletter. Any references to specific companies or entities are for informational purposes only.

The brief summaries and descriptions of research papers and articles provided in this newsletter should not be considered definitive or comprehensive representations of the original works. Readers are encouraged to refer to the original sources for complete and authoritative information.

This newsletter may contain links to external websites and resources. The inclusion of these links does not imply endorsement of the content, products, services, or views expressed on these third-party sites. The author is not responsible for the accuracy, legality, or content of external sites or for that of any subsequent links. Users access these links at their own risk.

The author assumes no liability for losses or damages arising from the use of this content. By accessing, reading, or using this newsletter, you acknowledge and agree to the terms outlined in this disclaimer.

Paid subscriptions may not be available in all jurisdictions and may change without notice.

Weekly Research Recap

QuantSeeker — Tue, 02 Jun 2026 19:07:46 GMT

Every Tuesday, I highlight the best investing research, market insights, podcasts, and ideas worth your attention, all in one curated weekly digest.

Asset Allocation

Harvesting Factor Premia Across Regimes: An Anchor-Stabilized Hidden Markov Framework for Multifactor Portfolios (Yang)

This paper tests a factor timing model that uses a Hidden Markov Model to identify market regimes and dynamically rotate across equity factors. From 2013 to 2023, the strategy achieved Sharpe ratios above 1.2 and drawdowns below 6%, substantially lower than those of broad equity benchmarks. Key takeaway: Rather than forecasting factor returns directly, investors can improve factor allocation by conditioning on latent market regimes.

Capital Market Assumptions And Strategic Asset Allocation Using Multi Asset Tradable Factors (Sepp, Hansen, and Kastenholz)

Most asset allocation models estimate expected returns on an asset-by-asset basis, creating substantial estimation noise. This paper instead estimates risk premia for a handful of tradable factors and derives expected asset returns from their factor exposures. The result is a more stable efficient frontier and roughly 50% less estimation uncertainty. Key takeaway: Estimating expected returns through a small set of factor risk premia is more robust than estimating them asset by asset.

Equities

The co-pricing factor zoo (Dickerson, Julliard, and Mueller)

After analyzing 18 quadrillion stock and bond factor models, the authors find that many factors appear to be different expressions of the same underlying risks. No small factor set can jointly explain stock and corporate bond returns. The strongest results come from combining information across dozens of factors. Key takeaway: Alpha may come less from discovering new factors and more from identifying the common risks they share.

Excess returns on idiosyncratic profitability: Evidence from a hedge portfolio strategy (Han, Jackson, and Monroe)

Most profitability signals treat earnings as one number. This paper suggests investors should look deeper. Stocks with high idiosyncratic profitability, profits unexplained by market or industry factors, earned significant abnormal returns even after controlling for standard factors. Key takeaway: Investors underreact to firm-specific profitability.

Pervasive Styles (Wang and Yu)

Investors spend enormous effort searching for the next great factor. This paper suggests the bigger edge may come from understanding how investors chase styles. Across 312 stock characteristics, over 84% of style-based signals predict future returns, even when the underlying characteristic itself have little predictive power. Key takeaway: Investor flows into styles are a more pervasive source of return predictability than the characteristics themselves.

Buying the Dip in Retail Trading (Yin and Zou)

Retail investors are often portrayed as momentum chasers. This paper finds the opposite. Using data covering 80% of U.S. retail trading, the authors show that retail investors systematically buy falling stocks, but don’t symmetrically sell rising ones. Key takeaway: Retail investors are helping stabilize markets by absorbing institutional selling pressure during drawdowns.

Passive Flows and the Limits to Arbitrage (Deng and Sammon)

Passive investing may be quietly eroding classic factor profits. Large passive inflows sharply compress returns to accounting-based anomalies by pushing up the stocks arbitrageurs are shorting more than the stocks they’re buying. Key takeaway: The growth of passive investing is creating a new limit to arbitrage by making the short side of anomaly trades increasingly difficult to exploit.

Conditional Equity Factor Risk Premium (Kim)

Most investors treat factor premiums as static. This paper suggests many aren't. Using a Kalman filter to estimate time-varying factor risk premia, the author finds that dynamically adjusting exposure to factors such as Value, Profitability, BAB, and Long-Term Reversal meaningfully improves risk-adjusted returns. Key takeaway: Factor investing may be as much about timing factor premiums as identifying them.

Machine Learning and Large Language Models

Daily Market Return Prediction with Transformer (Han, Huang, Huang, and Zhou)

A Transformer trained only on recent daily market returns generates Sharpe ratios above 1 and outperform simple moving-average signals. The edge comes from uncovering nonlinear patterns hidden in past returns. Key takeaway: Market returns contain more short-term information than traditional linear models can extract.

Regime-Based Portfolio Allocation Using Hidden Markov Models and Reinforcement Learning (Verma, Putri, and Lesupi)

Using a Hidden Markov Model to detect volatility regimes and Reinforcement Learning to allocate across stocks, bonds, and gold, the authors outperform static allocation rules out of sample. Key takeaway: Market regimes can be easier to predict than returns, and that information can improve asset allocation.

Macro

Recession Detection Using Real Time GDP Data (Sikand and Zhang)

Investors obsess over yield curves, payrolls, and dozens of other indicators to identify recessions. This paper suggests a much simpler signal may work. Using only real-time GDP releases available at the time, the authors build recession classifiers that identify all 12 U.S. recessions since 1947 without a single false signal in-sample. Key takeaway: Recession detection is less about finding new indicators and more about extracting the information already embedded in GDP data.

Blogs

A Simpler Way to Rotate Across Sectors (Quantseeker)

When Is a Mispricing Not a Mispricing? (Robot Wealth)

Tech Stock Singularity (John H. Cochrane)

The Sharpe stability ratio of trading strategies (Macrosynergy)

Building Reliable Algo Trading Systems (Concretum Group)

Podcasts

Ex-WorldQuant Head of Data Strategy: Quants “Don’t Care About the Stock Market” (Odds on Open)

The AI Bubble Might Be Exactly What We Need (William Goetzmann Explains) (Meb Faber)

Fmr FI Head Man Group: AHL Made a Billion When Everyone Thought the World Was Ending | Rob Carver (Personable)

Social Media & Industry Research

Trend-Following Filters – Part 10 (Alpha Architect)

The Inflation Wobble (Man Group)

Last Week’s Most Popular Links

A hidden Markov model for statistical arbitrage in international crude oil futures markets (Fanelli, Fontana, and Rotondi)

Factoring in the Low-Volatility Factor (Soebhag, Baltussen, and van Vliet)

Total Portfolio Approach: A Quant Lens (AQR)

Paid subscriptions may not be available in all jurisdictions and may change without notice.

A Simpler Way to Rotate Across Sectors

QuantSeeker — Mon, 01 Jun 2026 23:39:12 GMT

Sector rotation is one of the oldest ideas in investing. The challenge is figuring out which sectors are likely to lead next.

Some approaches rely on business-cycle forecasts: Overweight cyclicals during expansions and defensives during recessions. The problem is that economic regimes are often only obvious in hindsight, and macro indicators tend to lag. By the time investors recognize the regime, much of the market move has already taken place.

Rather than trying to forecast the economy, a different approach is to ask a simpler historical question:

When the market looked like this in the past, what did each sector do next?

No macro forecast or recession call is required.

In this article, I discuss research built around that idea, test it on sector ETFs, and assess whether it adds anything beyond simply holding SPY or an equal-weighted sector portfolio.

Weekly Research Recap

QuantSeeker — Tue, 26 May 2026 17:35:10 GMT

Every Tuesday, I curate a selection of market insights, research papers, podcasts, and investing ideas that stood out during the week, along with links for further reading.

Commodities

A hidden Markov model for statistical arbitrage in international crude oil futures markets (Fanelli, Fontana, and Rotondi)

By combining Brent, WTI, and the newer Shanghai crude oil futures in a hidden Markov regime-switching framework, the authors show that temporary dislocations between global oil benchmarks remain surprisingly tradable. The strongest models delivered roughly 6–8% annualized excess returns after transaction costs, with Sharpe ratios around 1 to 1.3 out of sample. Key takeaway: New markets may create fresh inefficiencies before arbitrage capital fully compresses them.

Crypto

Cryptocurrency Return Forecasting Using Technical Analysis: Out-of-Sample Evidence, Economic Value, and Macro-Conditional Predictability (Magner, Hardy, Lavin, and Sanhueza)

Testing 84 technical signals across the 10 largest cryptocurrencies from 2017 to 2026, this paper finds that trend and momentum signals in BTC, BNB, SOL, and XRP delivered the strongest out-of-sample performance after costs, with surviving strategies showing net Sharpe ratios around 0.8 to 1.1. Predictability strengthens during high BTC volatility, crypto sentiment shifts, and gold inflows. Key takeaway: Alpha in crypto is regime-dependent.

Equities

Factoring in the Low-Volatility Factor (Soebhag, Baltussen, and van Vliet)

Low-volatility investing manages sizeable AUM, yet most asset pricing models still ignore it. This paper argues that the disconnect stems from unrealistic assumptions about frictionless long-short investing. Once trading costs and factor asymmetry are considered, low-vol materially improves standard factor models because its returns survive implementation frictions better than many competing factors. Key takeaway: Low-vol investing becomes even more valuable once real-world frictions are considered.

Continuous Timing Signals for Growth-Defensive Style Allocation: Factor Attribution, Risk Matching, and Out-of-Sample Evidence (Xiong)

Many “growth vs defensive” rotation models rely on rigid market regimes. Instead, this paper adjusts exposure between growth/tech and defensive/value ETFs using continuous signals tied to rates, VIX stress, drawdowns, and growth crowding. From 2017 to 2026, the smooth-allocation model achieved a 1.01 Sharpe ratio with materially smaller drawdowns than 100% growth exposure. Key takeaway: Style timing may work better as a continuous risk-management process than a binary regime switch.

Machine Learning and Large Language Models

Battle of transformers: Adversarial attacks on financial sentiment models (Can Turetken and Leippold)

Financial sentiment models increasingly drive trading, yet this paper shows how fragile they remain. GPT-4o-generated paraphrases fooled FinBERT and FinGPT in 20–54% of cases, cutting accuracy by up to 26 percentage points. The main weakness? Overreliance on trigger words and weak numerical reasoning. Key takeaway: Many financial NLP models still pattern-match language better than they understand finance, making them vulnerable to manipulation.

The Volatility Premium of Machine Learning: Decomposing Signal from Mechanism in Directional Trading Strategies (Gonzalez Maiz Jimenez, Lopez-Herrera, and Reyes-Santiago)

Using 216k+ out-of-sample forecasts across 48 U.S. stocks, this paper shows that, under perfect fills, random signals produce Sharpe ratios nearly identical to those of ML models. But as fills became more realistic, ML adds substantial value, especially during high-volatility regimes associated with stronger investor overreaction. Key takeaway: Execution quality may matter as much as the model itself.

Options

Navigating IV: Options Trading Strategy in Earnings Season (Lu)

Across 5,000+ option trades from 2020 to 2025 on 30 high-volume tech stocks, selling 5–10% OTM strangles around earnings produced strong risk-adjusted performance, while long-volatility strategies were broadly unprofitable. Key takeaway: Systematically harvesting post-earnings IV crush appears more effective than betting on volatility expansion.

Retail Option Imbalance and the Cross-Section of Stock Returns (Liu)

Sorting stocks by retail call-option buying pressure generates a long-short strategy with a Sharpe ratio of 1.76 (before costs) and a daily alpha of roughly 27 bps. The effect peaks after 2 trading days, then gradually reverses. Key takeaway: Retail option flow may create short-lived price overshooting.

Vanilla Options: A Unified Mathematical Framework for Pricing, Hedging, and Volatility Modeling -A Comprehensive Review (Rahaman)

This is a comprehensive review paper, walking through option pricing, Greeks, hedging, volatility modeling, and American option valuation, while also highlighting where theory breaks down in practice through transaction costs, discrete hedging, and volatility dynamics.

Uncovering the asymmetric information content of high-frequency options (Alexiou, Bevilacqua, Hizmeri)

The sign of high-frequency option returns matters. Negative jumps in OTM calls and positive jumps in OTM puts strongly predict future volatility, variance risk premia, and even equity returns. In out-of-sample volatility timing tests, these signals generate economic gains worth up to 206 bps annually. Key takeaway: Measures of downside risk in options contain information that aggregate volatility measures miss.

Blogs

Timing VX Futures with the Front-End VIX Curve (Quantseeker)

Active Dual Momentum GTAA Strategy (Quantpedia)

The Metamorphosis (Robot Wealth)

Warsh’s Challenges: Monetary Policy (John H. Cochrane)

Podcasts

The Only 4-Sharpe Crypto Fund — Leigh Drogen, Starkiller Capital (Odds on Open)

He Studied 100 Years of Bubbles. He Exposed Private Equity’s Volatility Illusion | The Weekly Wrap (Excess Returns)

Patrick Welton’s Journey from Stanford Oncologist to one of Trend Following’s Quiet Legends (RCM Alternatives)

Social Media & Industry Research

When Everyone Trades the Same Factor Playbook (Alpha Architect)

Total Portfolio Approach: A Quant Lens (AQR)

The Hidden Macro Bets in Quant Portfolios (Man Group)

Last Week’s Most Popular Links

Agentic Workflows for Alpha Research (Jonathan Kinlay)

ENSO Signals and Out-of-Sample Predictability in Soft Commodity Futures (Apte)

Dynamic Momentum Trading via Deep Q-Networks: An Intelligent Execution Framework for Portfolio Management (Deng, Xu, Li, Ji, and Xu)

Paid subscriptions may not be available in all jurisdictions and may change without notice.

Timing VX Futures with the Front-End VIX Curve

QuantSeeker — Mon, 25 May 2026 19:07:32 GMT

Last week, I showed that the front end of the VIX term structure is a strong predictor of near-term realized volatility, even when controlling for the VIX level itself and standard HAR persistence effects.

The natural question is whether it translates into better trading.

In a previous post, I discussed a VX futures strategy utilizing bond market implied volatility as a cross-asset predictor of VIX. This week, I replace bond vol with the front-end VIX slope, run both models side by side on the same out-of-sample period, and test what happens when the two signals are combined.

The signals reflect different dimensions of market stress. Bond vol captures macro uncertainty spilling over from bond markets into equity volatility, while the front-end VIX slope reflects concentrated short-term fear embedded in equity options.

Combining them proves to be more effective than using either one alone.

Weekly Research Recap

QuantSeeker — Tue, 19 May 2026 22:30:07 GMT

Every Tuesday, I share the most interesting investing and market insights I found during the week, including new research papers, blog posts, and podcasts. I’ve included links throughout for readers who want to explore the ideas further.

Commodities

ENSO Signals and Out-of-Sample Predictability in Soft Commodity Futures (Apte)

Using NOAA El Niño data alongside price-based features for coffee, cocoa, sugar, cotton, and orange juice futures, a simple Ridge model achieves an out-of-sample Sharpe ratio above 1 after transaction costs. Climate variables alone have little standalone predictive power, but improve risk-adjusted performance by roughly 10% when combined with market signals. Key takeaway: Public climate data may still be only partially reflected in commodity prices.

Crypto

Are Day-of-the-Week Effects in Cryptocurrencies Real? Intraday Evidence from Active and Less Active Cryptocurrencies (Aalipour, Mehdian, and Rezvanian)

Most crypto “calendar anomalies” may be statistical illusions created by daily data aggregation. This paper utilizes hourly data across 12 cryptocurrencies and finds that apparent day-of-week effects are typically driven by just a few isolated intraday hours, rather than persistent daily behavior. Bitcoin’s “Monday effect,” for example, comes largely from only two specific hours. Key takeaway: Many crypto inefficiencies are short-lived microstructure effects rather than durable alpha signals.

What Do Crypto Options Tell Us? Risk Premia Implied by BTC Option Prices (Atanasova, Miao, Segarra, and Willeboordse)

Bitcoin options may contain a genuine risk-premia structure, not just speculative flow. This paper finds that option-implied factors, especially volatility-of-volatility, help predict future BTC excess returns, while crypto variance risk premia remain highly persistent even without strong institutional hedging demand. Key takeaway: Crypto derivatives are evolving into a genuine risk-transfer market with their own priced risk factors.

Arbitrage trading between decentral and central cryptocurrency exchanges (Schwertfeger and Vogt)

This paper analyzes live high-frequency arbitrage between decentralized exchanges (DEXs) and centralized exchanges (CEXs). Four arbitrage bots generate monthly returns between 6% and 33%, but realized profits are heavily constrained by slippage, liquidity limits, and latency. Key takeaway: The edge in crypto arbitrage is increasingly coming from execution infrastructure.

Equities

The decision-usefulness of stock recommendations in analyst reports: Evidence from long-window excess returns (Li)

Using 440k+ analyst reports from China’s A-share market, this paper finds that “Buy” and “Strong Buy” recommendations outperform comparable stocks by roughly 2.7 to 2.9% over the next 240 trading days. But much of the move starts before publication. The real edge may come from analysts identifying improving earnings and undervalued firms before the market fully reprices them. Key takeaway: Sell-side research still contain slow-moving information in less efficient markets.

Do Insider Trading Profits Result from the Superior Processing of Public Information? Evidence from Subtle Peer Firms’ Earnings Announcements (Campbell, Raleigh, and Zhao)

Most investors assume insider trading profits come from private information. This paper suggests another edge: Superior processing of subtle public signals. The authors show insiders buy their own company’s stock after positive earnings surprises from linked firms. CEOs appear especially skilled, and these trades earn annualized alphas of roughly 14 to 28%. Key takeaway: Some insider alpha come from spotting information spillovers before the market does.

Amplifying Long-Horizon Equity Returns: Evidence from Long-Term Equity Anticipation Securities (LEAPS) (Hegde)

Over 30 years of rolling 2-year windows, at-the-money SPY LEAPS produced average returns above 100%, versus roughly 18% for SPY itself, with realized amplification near 5x. But the tradeoff is real: Some LEAPS expired worthless, volatility was extreme, and risk-adjusted performance was not materially better than simply owning SPY. Key takeaway: LEAPS may offer convex exposure to long-run equity growth.

Machine Learning and Large Language Models

Dynamic Momentum Trading via Deep Q-Networks: An Intelligent Execution Framework for Portfolio Management (Deng, Xu, Li, Ji, and Xu)

This paper combines momentum signals, LSTM forecasts, portfolio optimization, and reinforcement learning that dynamically decides when to enter or exit trades. Across U.S. and Chinese equities, the framework improves Sharpe ratios and reduces momentum-crash drawdowns versus static strategies. Key takeaway: Adaptive execution matter as much as the signal itself in momentum investing.

Portfolio Construction

Portfolio Risk Parity with different Risk Aversion Degrees (Piantoni and Lozza)

Using S&P 500 stocks, the authors show that classic risk parity increasingly converges toward 1/N allocation as the number of holdings rises. But a Gini-based version, designed to emphasize downside and tail-risk protection rather than overall volatility, delivers stronger drawdown control and, in several tests, superior risk-adjusted performance during stressed markets. Key takeaway: The real innovation in risk parity may come from explicitly targeting tail risk, not just equalizing volatility.

Volatility

Maturity Alignment Matters: The Predictive Power of the Variance Risk Premium (Plihal and Mampouya)

Using SPY options from 2013 to 2025, the authors show that variance risk premia forecast realized volatility far better when option expiries closely match the prediction horizon. Forecast gains became strongest after SPY options moved to daily expirations. Key takeaway: Volatility signals lose information when the option maturity and forecast window are misaligned.

Blogs

How to Manage an Intraday Trend Trade (Concretum Group)

What the Front End of the VIX Curve Knows (Quantseeker)

How to build macro-aware equity indices (Macrosynergy)

Agentic Workflows for Alpha Research (Jonathan Kinlay)

Podcasts

“Market Crashes Are Good for My Strategy” - One-Man Hedge Fund PM George Livadas (Odds on Open)

When Crisis Alpha Hides in Plain Sight ft. Yoav Git & Rob Croce (Top Traders Unplugged)

He Invested Through Five Bubbles. He Wrote the Book on Them | What We Learned This Week (Excess Returns)

Last Week’s Most Popular Links

Systematic Tactical Allocation in Emerging Markets vs. U.S.: A Momentum-Based Approach (Vojtko and Dujava)

Dual Momentum Allocation Between Physical Gold and Bitcoin (Digital Gold) (Vojtko and Dujava)

Volatility Scaling in Multi-Asset Portfolios: Evidence from a Systematic Risk-Targeting Strategy (Almeida and Farias)

Paid subscriptions may not be available in all jurisdictions and may change without notice.

What the Front End of the VIX Curve Knows

QuantSeeker — Mon, 18 May 2026 19:12:33 GMT

It has long been known that the shape of the VIX term structure contains information beyond the VIX level itself. When short-dated implied volatility spikes above longer-dated implied volatility, the market may be signaling not only higher uncertainty but also an imminent volatility shock.

Researchers have long used the slope of the VIX curve as a regime indicator and tactical filter for equity exposure. In a previous post, I explored a related idea by using bond market implied volatility as a cross-asset signal for forecasting shifts in the VIX complex.

But a more interesting question is whether the magnitude of the inversion matters. Not just whether the curve has inverted, but how sharply, and whether the very front end of the curve contains information that the conventional short-to-medium slope misses.

Recent research suggests the answer is yes. The magnitude of front-end VIX curve inversion appears to contain incremental information about future realized volatility, even after controlling for the VIX level itself and standard HAR-style models of volatility persistence.

In this post, I replicate the latest research findings on independent data and discuss what they imply for anyone using the VIX term structure as a volatility or risk signal.

Weekly Research Recap

QuantSeeker — Tue, 12 May 2026 20:42:52 GMT

Asset Allocation

Systematic Tactical Allocation in Emerging Markets vs. U.S.: A Momentum-Based Approach (Vojtko and Dujava)

Simple 6 to 12 month trend signals between EM and U.S. equities help transform a historically weak EM-vs-US spread into a profitable long-short allocation strategy with Sharpe ratios above 0.5. The strongest performance is achieved by blending multiple trend horizons rather than optimizing a single parameter. Key takeaway: Global allocation may depend less on forecasting macro turning points and more on adapting to persistent relative-performance trends.

Commodities

On the Comovement of Contango and Backwardation Across Futures Commodity Markets (Luisi, Roccazzella, and Triantafyllou)

Contango and backwardation tend to move together across energy, metals, and agricultural futures markets, especially during crises. Gold is the outlier: When other commodity curves become more contangoed, gold often moves the opposite way. Key takeaway: Macro shocks increasingly dominate commodity term structures, meaning diversification across commodities may be far weaker than expected.

Dual Momentum Allocation Between Physical Gold and Bitcoin (Digital Gold) (Vojtko and Dujava)

A simple momentum strategy rotating weekly between GLD and Bitcoin based on relative and absolute strength improves risk-adjusted performance versus buy-and-hold. The strongest variant deliver 80% annualized returns with a Sharpe of 1.64. Key takeaway: There is potential alpha in systematically adapting to when investors prefer digital versus physical stores of value.

Equities

The Crash Risk in Individual Stocks Embedded in Skewness Swap Returns (Pederzoli)

Investors appear willing to pay heavily for protection against crashes in individual stocks. Skewness-based option strategies earn roughly 20% monthly average returns with Sharpe ratios near 0.5, though punctuated by rare but severe losses during market stress. The effect strengthened materially after 2008 as deep OTM put protection became more expensive. Key takeaway: Stock-specific tail risk appears to command a large positive risk premium in option markets.

Economic Uncertainty and the Beta Anomaly in G10 Countries (Atilgan, Demirtas, Gunaydin, and Tosun)

Across G10 equity markets, low-beta stocks outperform high-beta stocks primarily during periods of low economic uncertainty. In calmer environments, the performance gap between low- and high-beta portfolios is often around 1% per month. During high-uncertainty periods, the relationship weakens substantially. Key takeaway: The beta anomaly appears highly regime dependent.

The Difficulty of Market Timing: Proximity Matters More Than You Think (Lars N. Kestner)

Getting the direction right isn’t enough in market timing; precision matters. Using 22 years of SPY data, this paper shows that even investors who correctly identify every major turning point lose most of the edge if trades are early or late by more than a few weeks. Timing errors of 8 to 16 weeks quickly converge toward buy-and-hold performance, while only trades within roughly 4 weeks retain meaningful outperformance. Key takeaway: Market timing is less about being right and more about being precisely right; small timing errors can eliminate most of the edge.

Why the Standard Sharpe Ratio Misleads for Market-timing Strategies with Many Zero Return Days (Lars N. Kestner)

Market-timing strategies often look worse on paper than they actually are. This paper shows that the standard Sharpe ratio mechanically declines as time out of the market increases, even when in-market performance is unchanged. Zero-return days create a mixture distribution that depresses Sharpe and increases kurtosis. But when measured only on invested days, Sharpe remains stable. Key takeaway: Standard Sharpe ratios can penalize low-exposure strategies for being inactive, not for having worse returns.

Machine Learning and Large Language Models

On the predictability of ETF returns with technical predictors (Gong and Muller)

A random forest model trained only on global stock technical indicators generates long-short ETF return spreads of 0.76% per month (t=2.76), with volatility and momentum signals carrying most of the predictive power. The edge is strongest in less efficient, lower-liquidity markets like China and fade quickly at longer horizons. Key takeaway: Cross-sectional patterns in individual stocks appear to carry over into ETF returns.

A Comparative Analysis of Regressor Machine Learning Models in Forecasting SPDR S&P 500 ETF Trust (SPY) Movements (Lee and Haldankar)

This paper tests five machine-learning regressors on short-term SPY prediction and finds that adding technical indicators consistently lifts directional accuracy from roughly random levels (46 to 49%) to about 54%. Random Forest benefited the most from added features, while Prophet achieved the highest hit rate at 54.6%. Key takeaway: In short-term market forecasting, feature engineering may matter more than model complexity.

A Review of Large Language Models for Stock Price Forecasting from a Hedge-Fund Perspective (Zhang and Zhang)

Most investors think LLMs will easily unlock alpha in markets. The reality may be far messier. This review paper argues that many impressive LLM trading results may overstate real-world performance due to data leakage, short sample periods, and illiquidity. In some cases, strategies with extremely high reported Sharpe ratios deteriorate sharply once realistic trading frictions are considered. Key takeaway: The biggest challenge for AI in investing may not be prediction; it’s surviving contact with real markets.

Portfolio Construction

Volatility Scaling in Multi-Asset Portfolios: Evidence from a Systematic Risk-Targeting Strategy (Almeida and Farias)

A simple multi-asset vol-scaled portfolio nearly doubles the Sharpe ratio of a 60/40 benchmark during calm market regimes (0.75 vs 0.41) by mechanically increasing exposure when realized volatility is low. But the tradeoff is severe during crises: De-leveraging leads to underperformance during sharp V-shaped recoveries like 2020. Key takeaway: Volatility scaling works best when markets stay calm but struggles during violent rebounds.

Blogs

Don’t Be Too Smart About History (Quantseeker)

Reinforcement Learning for Optimal Execution (Jonathan Kinlay)

Bad Month for Your Strategy? Should You Change It? (Alvarez Quant Trading)

The Anatomy of a Successful Trend Program (Concretum Group)

Book Review: Wrong Number (CFA Institute)

Podcasts

John Gu – Crypto Market Making & The Cold Start Problem (Flirting with Models)

Martyn Tinsley - 1 of 2 - Building Robust Trading Strategies - The Masterclass (The Algorithmic Advantage)

“If it is easy and obvious, there is no edge in it” - TD Quant Matt Schrager (Odds on Open)

Crisis Alpha, Cocoa Trends, and Correlated Trendlessness: Inside Aspect’s Strategy with Chris Reeve (RCM Alternatives)

Social Media & Industry Research

Why Momentum Investing Has Been Struggling—And What Volatility Has to Do With It (Alpha Architect)

Reads You May Have Missed (Man Group)

Last Week’s Most Popular Links

Variance and Skewness Risk Premium and Expected Equity Returns (Ito)

Multi-Asset Commodities Volatility Portfolio (Dottin)

Getting the Target Right in Return Prediction (Cakici and Zaremba)

Paid subscriptions may not be available in all jurisdictions and may change without notice.

Don't Be Too Smart About History

QuantSeeker — Sat, 09 May 2026 22:47:40 GMT

Statistical models used to predict returns typically make a quiet assumption: That the past is uniformly informative about the future. A standard predictive regression assigns the same weight to turbulent periods, such as October 2008, as to calm periods. The data goes in, the estimate comes out, and the differences between market regimes are averaged away.

The natural fix is to condition on relevance, to identify which historical episodes most resemble today and place greater weight on those observations when forming forecasts. It is an intuitive idea with an elegant mathematical foundation, and it has attracted interest in both academic research and applied quantitative investing.

But there is a deeper question beneath the intuition:

Does conditioning on “relevant” history actually improve forecasts, or does it simply amplify noise precisely when markets become most unstable?

Weekly Research Recap

QuantSeeker — Tue, 05 May 2026 17:47:52 GMT

This week’s Tuesday Roundup brings together the most useful investing ideas I’ve found over the past week, from fresh academic papers and market commentary to standout blog posts. You’ll find links throughout if you want to explore further.

Commodities

Multi-Asset Commodities Volatility Portfolio (Dottin)

Commodities alpha may be less about direction and more about volatility. By selling overpriced insurance (delta-hedged straddles), buying tail hedges, and scaling exposure with inventory signals, this strategy harvests the volatility risk premium while adapting to different regimes, delivering 8% returns at 9% volatility (Sharpe ratio of 0.9). Key takeaway: In commodities, systematically harvesting volatility, conditioned on fundamentals, may be more robust alpha than betting on price direction.

Crypto

Arbitrage trading between decentral and central cryptocurrency exchanges (Schwertfeger and Vogt)

Crypto arbitrage looks easy until you actually execute. Evidence from live trading bots shows that cross-exchange strategies can post very high gross returns (roughly 25 to 33% per month in strong periods), but most of that gets eaten by slippage, fees, limited liquidity, and execution risk. Key takeaway: The challenge isn’t finding arbitrage spreads; it’s capturing them at the price you expect.

Equity

Variance and Skewness Risk Premium and Expected Equity Returns (Ito)

Markets price downside risk. This paper shows that the downside variance risk premium is the key signal, strongly predicting returns at 3 to 6 month horizons, while upside volatility carries little information. At longer horizons, the skewness risk premium dominates with more negative skew (higher tail-risk pricing) predicting higher future returns. Key takeaway: Markets don’t reward volatility equally; they mainly reward bearing downside and tail risk.

Foreign Eyes on Wall Street: Investor Attention and U.S. Stock Reactions (Fan, Nikolsko-Rzhevskyy, and Talavera)

Foreign attention can move U.S. stocks more than most investors realize. When investors outside the U.S. actively search for company filings, those stocks tend to outperform in the short term, but the effect partly reverses, pointing to a role for temporary demand pressure. Key takeaway: Attention from global investors can potentially be traded, but mainly as a short-term signal.

A Levered ETF Anomaly Explained (Bianchi and Goldberg)

Leveraged ETFs underperform in choppy markets; that part is well known. What’s less appreciated is how large the gap can be, and why. Even when the S&P 500 was roughly flat in 2022 to 2023, 2× and 3× ETFs lost about −11% and −28%. It’s not just volatility drag from compounding; it’s amplified by how realized leverage co-moves with returns, systematically eroding performance. Key takeaway: Leveraged ETFs are path-dependent volatility trades, not long-term leverage on market returns.

Hedge Funds

Hedge Fund Awards: Do Investors and Managers Care, and Should They? (Choi, Kang, and Park)

Hedge fund awards look like signals of manager skill, but they mainly reallocate capital rather than predict performance. Winners see sizable inflows, around 0.6% of AUM per month for about a year, yet their returns don’t improve afterward. The effect seems driven by visibility and external validation, not new information. Key takeaway: Awards may boost capital, but they don’t predict alpha, and may even distort it.

Machine Learning and Large Language Models

Getting the Target Right in Return Prediction (Cakici and Zaremba)

Investors try to improve return forecasts by upgrading models, but the bigger lever is the target. Moving from raw to standardized or rank-based returns nearly triples predictive accuracy and lifts returns from 1.0% to 1.6% per month. Raw returns are noisy and market-driven; transformed return targets force models to learn relative performance. Key takeaway: Better targets in return prediction matter more than better models.

From Hypotheses to Factors: Constrained LLM Agents in Cryptocurrency Markets (Huang, Fan, Hu, and Ye)

Most investors chase better models. This paper shows the real edge may come from tighter constraints on the research process. By forcing an LLM to propose testable hypotheses under fixed rules (no data snooping), it discovers a crypto factor portfolio that holds up out-of-sample, 45% annual return, and a Sharpe of 1.55 after costs. Key takeaway: A disciplined discovery process is essential for reliable alpha.

Large Language Models for Asset Pricing: Learning from Earnings Calls (Zhang and Zhou)

This paper uses LLMs to extract features from earnings call transcripts, producing 1.7% per month with a Sharpe of 2.26 out of sample, while also predicting earnings surprises and investment decisions. Key takeaway: Earnings call language contains forward-looking information about fundamentals that markets don’t fully incorporate, but that LLM-based signals can systematically extract.

Do earnings call transcripts predict post-announcement returns? (Molinaro)

Earnings calls contain short-lived alpha. This paper shows that NLP features from transcripts predict post-announcement idiosyncratic returns with 51 to 52% hit rates and statistically significant long–short spreads, concentrated in the first few days (especially 1–3 days) after the call, with the signal fading to noise by one month. Key takeaway: Markets underreact to complex language, but the edge is small and gets priced quickly.

Macro

Could We Predict Macroeconomic Variables With Financial Conditions? (Welch)

Credit-based models used to predict macro conditions break down out of sample, often underperforming simple AR benchmarks and only adding value in crisis periods. Even improved versions require stock returns and winsorization to work (OOS R² of 41%). Key takeaway: Macro predictability is unstable; signals that look strong in-sample often fail out-of-sample.

Portfolio Construction

The economic value of forecasting and strategy gains in volatility timing (Xu, Aschakulporn, and Zhang)

Volatility timing gains come mainly from better risk estimation. Using a stochastic volatility model improves Sharpe ratios, while volatility-managed portfolios don’t create new alpha; they largely mirror optimized mean–variance allocations, even if they can outperform simpler approaches. Key takeaway: Better risk models drive alpha.

Prediction Markets

Arbitrage Analysis in Polymarket NBA Markets (Cheng, Yang, and Zou)

Some think arbitrage in prediction markets is easy. Not in Polymarket’s NBA markets. Across 75M+ order book snapshots, pricing gaps are rare and disappear within seconds. Even when cross-market trades (moneyline vs spread) offer 1% edges, position sizes are tiny, and liquidity is the real constraint. Key takeaway: The challenge isn’t spotting mispricing; it’s executing it, and capacity kills scalability.

Blogs

The Iran War Doesn’t Have to Be a Rerun of ‘That ’70s Show’ (John H. Cochrane)

Which Macro Indicators Actually Predict Market Drawdowns? (Quantseeker)

Deep Learning for Volatility Surface Repair (Jonathan Kinlay)

Curve trades with macroeconomic signals (Macrosynergy)

Geopolitical Shocks: What Moves First and Why It Matters (CFA Institute)

For The Love of The Game (Robot Wealth)

How to Build a Futures Database in Python using Norgate Data (Concretum Group)

You Can Trade (Almost) Like Mulvaney (Concretum Group)

Podcasts

Ex-Tudor Quant PM: “There Hasn’t Been a New Idea in Trading for 15 Years” (Odds on Open)

What 43 Years in the Markets Teaches You · Stephen Kalayjian (Chat with Traders)

The Last Moat | Chris Mayer and Ian Cassel on the Stock Picking Edge AI Can’t Replicate (Excess Returns)

Social Media & Industry Research

Rethinking Trend Following: Optimal Regime-Dependent Allocation (Alpha Architect)

Last Week’s Most Popular Links

A Comprehensive Review of Statistical Methods in Quantitative Finance: From Classical Inference to Machine Learning Frontiers (Rahaman)

Call the Zookeeper: A Unified Framework for Commodity Risk Premiums (Fan, Li, Qiao, and Zhang)

A Comparative Analysis of Overnight vs. Daytime Static and Momentum Strategies Across Sector ETFs (Salotra, Katikireddy, Anumolu, and Pinsky)

Paid subscriptions may not be available in all jurisdictions and may change without notice.

Which Macro Indicators Actually Predict Market Drawdowns?

QuantSeeker — Thu, 30 Apr 2026 11:53:52 GMT

Investors obsess over a range of indicators to gauge market risk. VIX spikes, so risk is elevated. The yield curve inverts, so a recession is coming. M2 contracts, so equities are vulnerable.

Few ask a simpler question: Which of these actually predicts drawdowns, and over what horizons?

The answer is less obvious than it seems.

This week I look at research that tackles exactly this question: Which macro-financial variables predict drawdowns, and how that depends on the forecast horizon. I test part of the framework using my own data.

Weekly Research Recap

QuantSeeker — Tue, 28 Apr 2026 21:26:05 GMT

This week’s Tuesday Roundup features the most valuable investing insights I came across over the last seven days, spanning new academic research, market analysis, high-quality blog posts, and thoughtful online discussions. Links are included throughout for further reading.

Commodities

Call the Zookeeper: A Unified Framework for Commodity Risk Premiums (Fan, Li, Qiao, and Zhang)

Past research has documented many types of risk premiums in commodity markets. This paper tests 81 predictors spanning commodity, macro, and sentiment data, then builds a dynamic long-short factor using 37 commodity-specific characteristics in a forecast model to predict next-month returns and rank commodities. The best-performing long/short portfolio delivered an annualized return of 5.8% with a Sharpe ratio of 0.64. Key takeaway: Combining signals tends to outperform single-factor commodity investing.

Crypto

Cross-Sectional Dispersion and the State Dependence of Cryptocurrency Momentum (Makgolo and Zhang)

Many investors believe that momentum crashes when volatility spikes, but in crypto, the better warning sign may be return dispersion. This paper finds that the dispersion of returns across coins predicts momentum breakdowns better than Bitcoin volatility. In extreme dispersion regimes, momentum weakened sharply. A dispersion-scaled strategy increased Sharpe from 0.63 to 0.80 and cut max drawdown from -42.5% to -17.1%. Key takeaway: For crypto momentum, watch coin dispersion, not just Bitcoin volatility.

Size-Momentum Puzzle in Cryptocurrencies (Xu and Wu)

Momentum in crypto markets depends on size. The authors show that small coins mean-revert sharply, with recent losers rebounding strongly, while large coins display momentum. Using 2018 to 2025 data, the smallest crypto quintile showed a 4.9% weekly reversal spread, while the largest delivered +1.0% weekly momentum, rising further over longer holding periods. Key takeaway: One momentum model does not fit all of crypto; small coins tend to reverse, while large coins trend.

Equities

Short-Term Reversal Persists Globally-If Properly Measured (Stosik and Zaremba)

Standard short-term reversals have weakened considerably over the years. This paper argues that the bigger issue is measurement. Recent losers often rebound once returns are judged relative to industry peers rather than raw price moves. Across 64 countries, the standard reversal signal earned just +0.05% per month, while the industry-adjusted version returned +0.53% monthly with a Sharpe of 0.74. Key takeaway: Short-term mean reversion still exists, but within industries.

Integrating Geopolitical Risk Into Low Volatility Factor Construction (Kallali, Binh, Roseneberg, Tilly, and Sekine)

Low-volatility strategies typically rely only on price data, but this paper finds geopolitical signals can improve them using firm-level news sentiment and regime indicators. The baseline model returned 10.4% annually with a Sharpe of 0.70, while the best geopolitical version delivered 12.8% with similar volatility and a 0.84 Sharpe; a more aggressive version reached 14.5% with a 0.91 Sharpe. Key takeaway: Defensive investing may improve when conditioned on geopolitical regimes.

A Comparative Analysis of Overnight vs. Daytime Static and Momentum Strategies Across Sector ETFs (Salotra, Katikireddy, Anumolu, and Pinsky)

Overnight vs. intraday returns for sector ETFs differ substantially. Across 10 U.S. sector ETFs (1999–2025), overnight returns were stronger than daytime returns, while long overnight plus intraday reversal often performed best. The top strategy turned $100 into over $31,000 in XLK before costs, though much of the edge faded after transaction costs. Key takeaway: Sector ETFs show distinct overnight and intraday return patterns, but monetizing them in practice is challenging.

Save The Date: Analyst/Investor Days as a Trading Signal (Cabrera, Kolokolova, and Zhang)

Analyst/Investor Days may be a stronger signal than most investors realize. Across 1,000+ U.S. firms, announcing an Investor Day led to abnormal returns and a run-up into the event. A long-short strategy buying announcing firms and shorting matched peers earned a Sharpe ratio of 1.06. A machine-learning timing version improved this to a Sharpe of 1.22. Key takeaway: Corporate communication events offer scalable event-driven alpha.

FX

Cross-sectional currency momentum and order flow (Sakemoto and Suda)

Cross-sectional FX momentum has struggled recently, but this paper shows that combining momentum with short-term FX swap order flows can restore returns. Currencies with strong recent performance and positive swap buying pressure earned 3.8% annually (Sharpe 0.57) versus -1.0% for standard momentum. In high-volatility periods, returns rose to 8–10%. Key takeaway: FX trends are strongest when confirmed by hedging flows.

Machine Learning and Large Language Models

ChatGPT as a Time Capsule: The Limits of Price Discovery (Lehner and Lopez-Lira)

Markets may be slower to price public narratives than many assume. This study uses older ChatGPT model snapshots to rank 7,000 U.S. stocks based only on what was knowable at the time. Firms with stronger outlooks earned 1.22% higher next-month returns per standard deviation, while a top-minus-bottom spread reached 2.55% monthly in-sample. The scores also predicted stronger sales growth and analyst upgrades. Key takeaway: LLMs may help process public information that investors aggregate only gradually.

Cross-Stock Predictability via LLM-Augmented Semantic Networks (Huang, Fan, Hu, and Ye)

This paper uncovers alpha from convergence trades between economically linked firms. Using 10-K filings, the authors build stock networks, then use an LLM to classify real firm relationships and remove noisy links. When linked stocks diverge, the strategy bets on mean reversion. A long-short S&P 500 strategy improved Sharpe from 0.74 to 0.82 and cut max drawdown from -10.5% to -7.9%. Key takeaway: LLMs can be used to identify relative-value opportunities across connected firms.

Prediction Markets

Beating the Earnings Game: Why Do Prediction Markets Outperform Professional Analysts? (Rabetti, Shao, and Zhang)

Prediction markets may be coming for Wall Street analysts. Studying 469 Polymarket contracts, this paper finds crowd-implied probabilities beat analyst consensus in forecasting earnings surprises: 78.5% vs 43.7% hit rate. The edge appears linked to continuous updating, diverse participants, and fewer institutional forecast biases. Key takeaway: Decentralized forecasts may become a serious new information signal for investors.

Prediction Market Accuracy: Crowd Wisdom or Informed Minority? (Gomez Cram, Guo, Jensen, and Kung)

Prediction markets are usually sold as collective intelligence in action. This paper suggests the opposite: Prices are made accurate by a small group of informed traders. Using complete Polymarket trading records, the authors estimate that roughly 3% of accounts account for most price discovery, react fastest to news, and earn persistent profits. The crowd supplies liquidity and losses. Key takeaway: What looks like crowd wisdom in prediction markets may simply reflect the decisions of a small informed minority.

Quant Finance

A Comprehensive Review of Statistical Methods in Quantitative Finance: From Classical Inference to Machine Learning Frontiers (Rahaman)

This is a very comprehensive review of quant finance topics, spanning topics such as probability theory, econometrics, time-series modeling, portfolio construction, derivatives pricing, tail-risk methods, Monte Carlo simulation, and machine learning.

Blogs

When Big Gets Small: Trading the Lower Tier of Large Caps and Upper Mid Caps (Quantpedia)

Backtests Lie: Building a Stress-Test Framework for Trading Signals (Vertox)

Podcasts

Sam Hartzmark: The Dividend Mistake Most Investors Still Make (Meb Faber)

How to Trade Futures | Rob Carver (Capital Horizons)

“Concentrated Strategies Will Do Extremely Well” - Sean Emory on Outperforming the Index (Odds on Open)

The $17M+ Prop Firm Roundtable - How The World’s BEST Traders Really Win (Words of Rizdom)

Social Media & Industry Research

The Skip-Month Mystery: What Last Month’s Returns Are Really Telling You (Alpha Architect)

The Inflation Diversification Problem (Man Group)

The MOST Important Thing: Unlocking Investment Wisdom with Antti Ilmanen (AQR)

Last Week’s Most Popular Links

Five-Factor Market-Neutral Strategy Across Korean and U.S. Equity Markets: Structural Alpha Without Regime Filters (Quinn)

Spot-Based Basis and Basis Momentum in Commodity Futures Markets (Luo and Xue)

Survival of the Fittest: A Three-Factor Model in the Currency Market (Liu, Wang, and Zhao)

Paid subscriptions may not be available in all jurisdictions and may change without notice.

Weekly Research Recap

QuantSeeker — Tue, 21 Apr 2026 16:26:27 GMT

This week’s Tuesday Roundup brings together the most useful investing ideas I found recently, from fresh academic studies and market research to thoughtful blog posts and smart discussions online. Links are included throughout so you can explore further.

Commodities

Spot-Based Basis and Basis Momentum in Commodity Futures Markets (Luo and Xue)

A smarter way to trade commodity futures may be to look beyond the futures curve itself. Using 41 Chinese commodities, basis and basis-momentum signals built from actual spot prices and futures prices beat traditional basis/carry measures. The strongest spread implied a return of +1.73% per month with a Sharpe of roughly 1.40. Key takeaway: Spot-market information may contain carry signals that are missed by futures-only models.

Equities

Five-Factor Market-Neutral Strategy Across Korean and U.S. Equity Markets: Structural Alpha Without Regime Filters (Quinn)

A market-neutral strategy spanning Korean and U.S. equities reports unusually strong performance: A Sharpe ratio of 3.64, a maximum drawdown of just -0.3%, and an 12.8% annual alpha after controlling for six standard factors. It combines five low-correlation signals from distinct return sources. Key takeaway: Diversifying across multiple niche inefficiencies usually matters more than chasing a few crowded factors.

Commodity Price Risk and the Cross-Section of Equity Returns (Dong and Palazzi)

U.S. stocks whose returns deviate from levels implied by market + commodity risk exposures tend to reverse next month. Underpriced minus overpriced stocks earned 1.86% monthly, equal-weighted, with a Sharpe of 1.31, the strongest in smaller, less-followed names. Key takeaway: Commodity-linked mispricing may be a niche source of alpha.

Momentum Returns and the Role of Liquidity Improvements (Bro)

Momentum may not be a standalone anomaly at all. This paper argues that much of classic stock momentum reflects improving liquidity: Past winners become easier to trade, while losers become harder. A long-short liquidity-improvement factor returned 0.65% per month (roughly 0.69 annualized Sharpe) and rendered momentum alpha insignificant in factor tests. Key takeaway: Some momentum profits may stem from gradual repricing to changing liquidity conditions.

Factor Investing

Factor Investing in Emerging Markets: From Multi-Factor Industry Portfolios to a Market-Neutral Alpha Strategy (Zhang, Dong, He, and Zheng)

This paper tests factor investing in emerging markets using strict out-of-sample splits. A beta-hedged EM strategy built from six classic signals (size, value, quality, momentum, low vol, yield) delivered a gross Sharpe of 0.83, 0.67 net after costs, 4.16% annual alpha, and just -5.2% max drawdown (2019 to 2024). Key takeaway: In emerging markets, diversified simple factors can still deliver strong risk-adjusted returns.

FX

Survival of the Fittest: A Three-Factor Model in the Currency Market (Liu, Wang, and Zhao)

The crowded “currency factor zoo” may boil down to three core drivers: Dollar (DOL), Carry (CAR), and Output Gap (GAP). Using Bayesian model selection on 1991 to 2024 FX data, this trio ranked highest versus popular alternatives and best explained rival factors. In the authors’ recursive portfolio test, it posted a 1-year out-of-sample annualized Sharpe of 2.99. Key takeaway: In FX, parsimonious macro factor models may beat complexity.

Machine Learning and Large Language Models

Machine Learning and Technical Analysis in International Market (Chin and Lin)

Across 25 equity markets, machine learning outperformed OLS when using 107 technical price/volume signals. Global long-short monthly returns reached 1.15% vs 0.58%, with ML also adding value among large firms and in down markets. Key takeaway: Technical indicators may still contain alpha, and flexible models seem better at extracting it.

Option Return Predictability via Large Language Models (Liu, Zhou, Cheng, Zou, and Wang)

GPT-5 is tested as an autonomous generator of option-return factors in U.S. and Chinese markets using only option characteristics. In a post-Sept 2024 out-of-sample test, many generated factors beat benchmarks and delivered significant positive alpha. Many reported Sharpe ratios were strong, based on backtests of delta-hedged portfolios. Key takeaway: LLMs may become valuable tools for idea generation in niche, high-dimensional markets.

Timing Currency Factors with Machine Learning (Torok)

Machine learning may help time FX factors, but gains are concentrated in specific cases. This paper tests 264 signals across currencies, options, equities, and macro data to time carry and USD factors. Most individual predictors added little, but Developed Market USD timing improved Sharpe by 0.4 to 0.5 in 2012–2024 out-of-sample tests. Key takeaway: In FX, ML may be most useful for selective risk timing rather than broad forecasting.

Latency Alpha: Real-Time LLM-Based Semantic Extraction of EIA Petroleum Data for Forecasting and Trading Crude Oil Futures (Pape)

This paper tests an LLM system that reads the weekly EIA petroleum report in real time, converts report surprises into trading signals, and trades WTI futures. In a 2022 to 2023 backtest (88 releases), it reports a Sharpe of 1.82, a Sortino ratio of 2.89, 31.2% cumulative return, and a 68.5% hit rate. Key takeaway: In event-driven markets, faster semantic interpretation of public data may create temporary alpha opportunities.

Interpretable Systematic Risk around the Clock (He)

The author uses U.S. equity market data plus an open-source LLM to identify what drives major market jumps. Macroeconomic jump risk earned the strongest premium: 3.65% annually with a Sharpe of 0.78 (vs market 0.53). A real-time strategy rotating into the best-priced jump-risk theme each year reached a Sharpe of 0.95 out of sample. Key takeaway: Not all volatility is equal; macro shocks appear to earn the richest compensation.

Portfolio Construction

Measuring Strategy-Decay Risk: Minimum Regime Performance and the Durability of Systematic Investing (Alexander and Fabozzi)

Many quant strategies look strong in backtests but weaken across regimes. A new paper proposes Minimum Regime Performance (MRP): The worst Sharpe observed across historical environments. Among the tested equity factors, only Quality kept a positive MRP, while Size and Investment looked fragile. Key takeaway: Judge the durability of strategies across regimes, not just average Sharpe.

Dynamic Allocation with Macro Factor-Mimicking Portfolios (Molinaro and Chaudron)

Inflation may be more tradable than investors think. This paper ranks stocks by inflation beta, builds high- and low-inflation-beta portfolios, then uses macro forecasts to rotate between them. The best strategy delivered 21.4% annual return with a Sharpe of 1.09. Key takeaway: Forecasting macro regimes may add value when expressed through targeted inflation-sensitive equity exposures.

Blogs

Inflation as a trading signal (Macrosynergy)

Pairs Trading in the Metals Complex: A Reality Check (Quantseeker)

From Risk Premia to Constraints: How Markets Really Clear (CFA Institute)

Annual performance update- year 12 (Rob Carver)

Exploiting Mean-Reversion in Decentralized Prediction Markets: Evidence from Polymarket Binary Contracts (Quantpedia)

Podcasts

“It’s the Dumbest Market in the World” - Quant Trader Scott Phillips on Edge in Crypto (Odds on Open)

Markets Look Calm… But Are They? ft. Rob Carver (Top Traders Unplugged)

$1,900 Small Cap Trader Hit $2.9 Mil in Profits · Eduardo Briceño (Chat with Traders)

Social Media & Industry Research

Quantitative Trading Is Going To Eat All Markets (SystematicLongShort)

How should you size your trend trades? (Concretum Research)

Mastering Claude Code in 30 Minutes (via Movez)

Can you trust your intraday database? (Concretum Research)

Last Week’s Most Popular Links

Optimal Buy-and-Hold Asset Allocation: A Multi-Horizon Drawdown-Constrained Approach (Wang and Wang)

Can AI Do Financial Research? LLM-Guided Hypothesis Discovery in Asset Pricing (Liu, Liu, Liu, and Mei)

Don’t Mix What Should Be Separated: Why Combining Value and Momentum Signals Destroys Alpha (Morales)

Paid subscriptions may not be available in all jurisdictions and may change without notice.

Pairs Trading in the Metals Complex: A Reality Check

QuantSeeker — Mon, 20 Apr 2026 18:18:05 GMT

A recent paper reports attractive Sharpe ratios from a simple mean-reversion strategy in precious metals ETF spreads. But do those results survive real-world parameter uncertainty? I replicate the study, test it under walk-forward parameter selection, and show why a simple ensemble approach may be more robust.

Weekly Research Recap

QuantSeeker — Tue, 14 Apr 2026 19:34:47 GMT

Welcome to this week’s Tuesday roundup, a collection of practical investing insights drawn from new academic papers, market research, blogs, and sharp discussions across the web, with links provided throughout.

Asset Allocation

Optimal Buy-and-Hold Asset Allocation: A Multi-Horizon Drawdown-Constrained Approach (Wang and Wang)

Using ETF data from 1996 to 2026, the paper finds that drawdown-constrained optimization often favors unconventional portfolios: About 40% growth equities, 40% gold, and 20% Bitcoin in its 5-, 10-, and 20-year tests. Reported Sharpe ratios range from 0.75 to 1.20, well above benchmark portfolios. Key takeaway: Focusing on drawdowns can lead to very different portfolios than traditional allocation rules.

Bond Hedge Effectiveness (Lachana)

Using data from 2007 to 2023 in the U.S., the study shows that Treasuries are not a constant equity hedge. Protection improves when investor mood turns fearful, as a pessimistic media tone is linked to more negative stock–bond comovement. A signal-driven strategy modestly beats a static 50/50 stock-bond mix on Sharpe ratio, though performance depends on the inflation backdrop. Key takeaway: Stock–bond hedging varies with sentiment and macro conditions.

Equities

Formal Equity Valuation: Overview and Limits (Ohlson and Rueangsuwan)

The authors argue that valuation formulas have real-world limits, especially sensitivity to growth and discount-rate assumptions, but still add value. DCF, dividend, and residual-income models can mislead when used mechanically, yet remain useful frameworks for thinking about earnings, growth, risk, and payout policy. Best practice is to pair them with forward EPS and realistic multiples. Key takeaway: Use formal valuation models as guides, while recognizing their limitations.

Is Return seasonality Due to Risk or Mispricing? Evidence from Anomaly seasonality (Wang)

Using 125 anomaly portfolios, the paper argues that seasonal return patterns in anomaly portfolios largely originate from stock-level seasonality rather than genuine seasonal factor premia. A stock-based seasonality factor cuts alpha on seasonality strategies from 1.45% to 0.27% per month, while anomaly-based seasonality factors only reduce it to 0.86–1.23%. Key takeaway: Seasonal alpha is strongest at the individual-stock level, not in factor portfolios.

Don’t Mix What Should Be Separated: Why Combining Value and Momentum Signals Destroys Alpha (Morales)

Using a 2000 to 2026 backtest of the top 1,000 U.S. stocks, the paper finds that merging value and momentum into one composite rank dilutes the diversification benefit of their negative correlation. Running them as separate strategies produced a higher Sharpe, lower volatility, and a smaller max drawdown. At matched risk, the separate-strategy approach outperformed by 52 bps annually. Key takeaway: Combine factors at the portfolio level rather than merging signals into one rank.

Machine Learning and Large Language Models

Can AI Do Financial Research? LLM-Guided Hypothesis Discovery in Asset Pricing (Liu, Liu, Liu, and Mei)

Using U.S. equity data from 1963 to 2024, the study finds that an LLM can generate understandable return signals rather than only summarize prior work. It produced 280 accounting-based ideas; 159 passed an initial predictive screen, 38 remained after redundancy checks, and roughly 6 to 9 survived the toughest validation hurdles. Key takeaway: AI can expand idea generation, but rigorous testing determines what truly adds value.

Large Language Models and Stock Investing: Is the Human Factor Required? (Crisostomo and Mykhalyuk)

Using a 10-month live IBEX-35 test, the paper finds that LLM-generated stock rankings can outperform the benchmark, especially when prompts are structured and outputs are reviewed by humans. Naive prompts delivered just 0.35% monthly excess return (insignificant), structured prompts 2.24%, and supervised step-by-step workflows 3.04% with IR 0.68. Reasoning and data errors remained frequent. Key takeaway: LLMs can add value, but perform best under human supervision.

Options

Call option pressure and option return predictability: A U-shaped nonlinearity (Cai)

Using China’s SSE 50 ETF options (2019–2026), the paper finds that a net call-positioning measure predicts next-day option returns in a U-shape: Very low readings align with reversals, very high readings with continuation, while moderate readings add little signal. An ETF timing strategy reports a Sharpe of 0.97, rising to 2.43 with additional filtering. Key takeaway: Extreme call-positioning signals contain more information than normal flow.

Options Volume as Noise: Evidence from Three Decades of Earnings Announcements (Taheri Hosseinkhani)

Studying U.S. earnings events, the paper finds that the option-to-stock volume ratio predicts earnings-window returns, but the relationship changes over time. High O/S stocks underperformed by 39 bps around earnings and 118 bps over the next 60 days in full-sample portfolio sorts, while the O/S coefficient flipped sign in 2020–2024 regressions. The put/call ratio also predicted earnings surprises. Key takeaway: The predictive power of options volume depends on the market regime.

Prediction Markets

Trend Quality and Predictability in Prediction Markets: Evidence from Minute-Level Kalshi Data (Greene)

Using 4.9 million minute-level Kalshi observations across 29,590 contracts (Jan–Mar 2026), the paper shows that a simple linear trend-regression signal, trend slope divided by residual noise over the final 30-to-12 minutes before expiry, contains information. Highest-quality trends continued 70.8% vs. 51.2% for the weakest group. Key takeaway: Smooth late price trends often signal short-term continuation.

Volatility

The Role of Price-Volatility Cojumps in Volatility Forecasting (Liao)

Using 5-minute S&P 500 and VIX data, the paper finds that simultaneous price-volatility jumps contain information missed by standard jump measures. Downside cojumps predict higher future volatility, while upside cojumps signal calmer markets. Adding these variables to HAR models significantly improves out-of-sample volatility and VaR forecasts. Key takeaway: Joint price-volatility shocks matter more than price jumps alone.

Blogs

Does “Optimal” Portfolio Construction Actually Pay Off? (Quantseeker)

To Trend or Not To Trend? (Wrong question) (Robot Wealth)

Systematic Tactical Allocation in Emerging Markets vs. U.S.: A Momentum-Based Approach (Quantpedia)

Top 10 Most Read Q1 Enterprising Investor Blogs (CFA Institute)

Book Review: Financial Data Science (CFA Institute)

Podcasts

Now Is the Best Time to Become a Junior Analyst - Ex-Citadel and D. E. Shaw PM Brett Caughran (Odds on Open)

The ICT Trader Who Made $936K In 25 Days - Dhesi (Words of Rizdom)

Samir Varma - Classify Risk Don’t Chase Alpha! (The Algorithmic Advantage)

The Hedge Fund that Polls Regular People Where Markets Are Going? (RCM Alternatives)

Social Media & Industry Research

A Positive Stock-Bond Correlation Is a Terrible Reason to Add More Equity Risk to Your Portfolio (AQR)

Factor MAX: A New Signal for Predicting Factor Returns (Alpha Architect)

The Many Facets of Stock Momentum: Distinguishing Factor and Stock Components (Alpha Architect)

Being Front-Run On DEXes (SystematicLongShort)

Last Week’s Most Popular Links

Intraday Stylized Facts and the Shape of Volatility Build-Up in ICE Brent Crude Oil Futures (Haugom, Ewald, Chen, and Smith-Meyer)

The 52-Week High Momentum Strategy: Evidence in Chinese Stock Market (Lan, Truong, and Zhang)

Dividend Flows and the Foreign Exchange Rate (Zheng)

Paid subscriptions may not be available in all jurisdictions and may change without notice.

Does "Optimal" Portfolio Construction Actually Pay Off?

QuantSeeker — Sun, 12 Apr 2026 21:14:07 GMT

Most investors spend their time deciding what to own. Fewer think deeply about how to weigh those holdings. Yet portfolio construction often determines diversification, drawdowns, turnover, and long-run returns just as much as asset selection and signal generation.

Many theoretical portfolio construction methods seem elegant, but few survive real market conditions.

In this post, I test six popular portfolio construction approaches across two very different universes, a multi-asset ETF portfolio and a diversified commodity futures basket, to see which methods actually held up out of sample, after trading costs.

Weekly Research Recap

QuantSeeker — Tue, 07 Apr 2026 18:25:15 GMT

Here’s this week’s Tuesday roundup, a selection of the most practical investing ideas from recent academic studies, industry research, blogs, and insightful discussions across social platforms, with links included throughout.

Commodities

Intraday Stylized Facts and the Shape of Volatility Build-Up in ICE Brent Crude Oil Futures (Haugom, Ewald, Chen, and Smith-Meyer)

Using tick-level Brent futures data (2006–2025), the paper documents a range of intraday regularities: 1-minute returns display short-term reversals (first-lag autocorrelation of about −0.1), while volatility is persistent and follows a pronounced daily cycle. Volatility builds up unevenly over the day, rising more rapidly after market openings and information releases, and does so more strongly and rapidly in near-dated contracts. Key takeaway: Intraday volatility in Brent is predictable, making execution timing, dynamic risk control, and time-aware volatility modeling essential.

Crypto

Cryptocurrencies: Asset Classification, Trading and Portfolio Management (Gottwald, Sun, Chan-Lau, and Mitra)

Cryptocurrencies exhibit commodity-like behavior, including pronounced fat tails and frequent extreme moves. Momentum strategies are strongly regime-dependent, delivering gains mainly during periods of market stress and elevated volatility. Combining crypto with commodities improves portfolio efficiency and Sharpe ratios relative to commodities alone. Key takeaway: Crypto behaves like a speculative, commodity-like risk factor best used as a diversifier rather than a standalone allocation.

Equities

The 52-Week High Momentum Strategy: Evidence in Chinese Stock Market (Lan, Truong, and Zhang)

The 52-week high strategy emerges as a robust momentum signal in Chinese equities while traditional momentum is negative and industry momentum is weak. Its predictive power remains after controls and dominates in cross-sectional tests, but weakens beyond a 12-month holding period. Key takeaway: Price proximity to the 52-week high is a behaviorally driven momentum signal that seems to hold up in Chinese equities.

Geopolitical Risk and Equity Returns: Evidence From Global Markets (Coqueret and Zhou)

Geopolitical risk is priced immediately in equities: A one-unit shock lowers market returns by about 1% on impact, with stronger effects in emerging markets. At the firm level, exposures vary widely, with many stocks reacting immediately while others, particularly in emerging economies, adjust with a lag. Key takeaway: While geopolitical risk is an important risk factor, its effect in equities is primarily contemporaneous, with limited predictive power.

Fixed Income

The Co-Pricing Factor Zoo (Dickerson, Julliard, and Mueller)

Sparse models fail to explain corporate bond returns, particularly when pricing bonds and equities jointly. Risk premia are driven by a broad, dense set of factors, not a few dominant ones. After accounting for duration risk, bond-specific factors add little beyond equity and macro factors, while combining many signals delivers strong performance (Sharpe of 1.5 to 1.8). Key takeaway: Once duration risk is handled, equity and macro factors price corporate bond returns, but you need a broad set of them.

FX

Dividend Flows and the Foreign Exchange Rate (Zheng)

Dividend payments, though known in advance, generate predictable FX pressure: Currencies with large payouts depreciate by 4.7 bps within 2 days and 6.5 bps within about a week. A short-term strategy earns 30 bps/month in alpha, unexplained by standard FX factors. The effect is driven by benchmark investors’ dividend repatriation and is stronger when intermediary constraints bind. Key takeaway: Even predictable, mechanical flows can move currencies and create small but consistent trading opportunities.

Machine Learning and Large Language Models

Do LLMs Make Markets More Efficient? (Lu, Xu, and Vulicevic)

LLM access speeds up how quickly markets react to news. When LLMs are available, next-day return predictability from news sentiment falls by about 46 to 61%, with no reversal. During outages, predictability, and the profits of a simple news-sentiment strategy, roughly double to triple. Key takeaway: LLMs accelerate price discovery, but outages create brief, exploitable delays.

Multiple Testing

The False Discovery Rate in Finance: Identification Failure and Search-Adjusted Estimation (Lopez de Prado and Fabozzi)

Published factor results are typically the most favorable outcomes from many tested specifications, not single-trial estimates. The paper shows this search-and-selection process breaks standard inference and leads to systematically underestimated false discovery rates (FDRs). Under realistic assumptions, search-adjusted FDR can exceed 70 to 80%, even when nominal significance appears strong. Key takeaway: Most reported factors likely reflect selection bias; robust inference requires out-of-sample validation or explicit modeling of the search process.

Options

Time variation of size premium in the options market (Nguyen)

Delta-hedged options on large firms outperform small-firm options by 1.0 to 1.4% per month, but this spread disappears in recessions and down markets. The effect is strongest in expansions, consistent with demand-based pricing: Options on small, lottery-like firms become overpriced when investor demand is high, while flight-to-quality reduces this mispricing in downturns. Key takeaway: Time-varying demand for lottery-like options, not risk, drives the inverse size effect in options.

Prediction Markets

From Iran to Taylor Swift: Informed Trading in Prediction Markets (Mitts and Ofir)

Prediction markets appear efficient, but a small subset of traders consistently outperforms by trading just before major events. Accounts flagged for unusual timing, size, and concentration win about 70% of the time and capture roughly $143M in profits, far above random outcomes, suggesting informational advantages rather than skill alone. Key takeaway: Prediction markets aggregate information but also reward those with private information.

Blogs

How imputation helps statistical learning for macro trading signals (Macrosynergy)

VIX vs. Policy Uncertainty: What They Signal for Risk (CFA Institute)

One Year Later: Is ChatGPT Finally Worth Using for Quantitative Analysis? (Quantpedia)

Uncertainty (Quantitativo)

GitHub

Financial Datasets MCP Server

Podcasts

Faheem Osman – Commodity QIS: An Under-Appreciated Source of Systematic Returns? (Flirting with Models)

“Positions Can Be LESS Risky at Higher Prices” - Derek Pilecki on Finding Edge in Financials (Odds on Open)

Private Equity’s Low Volatility Isn’t Real (Meb Faber Show)

The ICT Trader Who Made $2 Million After SWITCHING To Orderflow - Yush (Words of Rizdom)

Social Media & Industry Research

A Practical Approach to Weighting Signals (Factset Insights)

Mean Reversion in Play: Carry is BACK?! (Alpha Architect)

Winning the Long Game with RAFI (Research Affiliates)

Last Week’s Most Popular Links

Rethinking Trend Following: Optimal Regime-Dependent Allocation (Zakamulin)

Momentum and Reversal on the Short-Term Horizon: Evidence from Commodity Markets (Ding, Kang, Yu, and Zhao)

Bimodality Everywhere: International Evidence of Deep Momentum (Han and Qin)

Paid subscriptions may not be available in all jurisdictions and may change without notice.

Weekly Research Recap

QuantSeeker — Tue, 31 Mar 2026 22:33:37 GMT

This week’s Tuesday roundup brings together the most actionable investing ideas from recent academic work, industry research, blogs, and high-quality discussions across social media, with links throughout.

Commodities

Momentum and Reversal on the Short-Term Horizon: Evidence from Commodity Markets (Ding, Kang, Yu, and Zhao)

Using CFTC position data, the paper shows that short-horizon returns are driven by two offsetting mechanisms: Speculators’ net trading predicts reversals, while the return component orthogonal to trading flows exhibits continuation. The latter generates about 12 bps per week (about 6.2% annualized) and, when used as a sorting signal, improves intermediate-term momentum returns from 5.2% to 9.9% annually. Key takeaway: Decomposing returns into trading-flow and orthogonal components improves momentum signals.

Equities

Prediction reconditioned: Revisiting relevance (Polakow, Flint, Turro, and van Rooyen)

Relevance-Based Prediction (RBP) focuses on observations similar to current market conditions, but this selectivity often backfires. In predicting S&P 500 returns, the paper finds that OLS achieves higher explanatory power and lower forecast error. Simulations show why: Conditioning on “relevant” data reduces bias but increases variance due to smaller samples. Key takeaway: In noisy financial markets, using more data can outperform using “better” data.

Measuring Equity Factor Uncertainty (Huovinen)

The paper introduces a Factor Uncertainty Index (FUI) based on a broad set of equity factors. FUI predicts future factor returns, with explanatory power rising to 30% at a 12-month horizon and out-of-sample R² around 10 to 13%. Using FUI for risk scaling improves performance, lifting Sharpe from 1.25 to 1.53. Key takeaway: Factor uncertainty predicts returns and provides a simple, effective signal for dynamic risk scaling.

FX

One Leg at a Time: The Hidden Structure of Carry Trade Profitability (Panayotov)

Carry trades look noisy because exchange-rate swings drown out the underlying signal. By stripping out this noise, the paper shows that carry profits arrive in distinct regimes and come from only one side of the trade at a time. Timing which leg to hold using policy-rate dynamics roughly doubles Sharpe from 0.45 to 0.87. Key takeaway: Carry is a regime-dependent strategy and performance improves when you trade it selectively.

Machine Learning and Large Language Models

Bimodality Everywhere: International Evidence of Deep Momentum (Han and Qin)

Momentum returns are inherently unstable: Past winners and losers both face a meaningful risk of sharp reversals. A machine learning “Deep Momentum” approach that models the full return distribution (via predicted probabilities) delivers 41% annual returns with a Sharpe of about 2.5, versus 21% and 1.0 for standard momentum, with consistent global outperformance. Key takeaway: Modeling return distributions, not just averages, improves momentum investing.

Reviving Anomalies (Beckmeyer, Berg, Wiedemann, and Wortmann)

Most anomalies fail after costs, but the paper shows they become profitable when conditioned on ML-based expected returns and transaction costs. By filtering each anomaly to retain only high-conviction stocks, the number of implementable anomalies rises from 10 to 100. A simple 1/N portfolio delivers 12 to 16% annual returns with Sharpe ratios up to 1.0. Key takeaway: Alpha comes from selectively implementing existing factors, keeping only trades with strong expected returns after costs.

Options

The Changing Information Content of Options Markets: Evidence from 1996–2024 (Taheri Hosseinkhani)

Option markets still contain valuable information, but not all signals survive structural change. The implied volatility gap between calls and puts delivers 25 to 30 bps/week alpha with strong significance, while volume-based signals weaken and even reverse after 2020 as retail trading distorts flows, remaining informative mainly in hard-to-short stocks. Key takeaway: In option markets, price-based signals are more robust, while volume-based signals become unreliable when the composition of traders shifts.

Portfolio Allocation

Geopolitical Risk and Asset Pricing Across Market Regimes (Abou Rjaily)

Geopolitical risk is priced unevenly across assets and regimes. In a two-state framework, equities exhibit significant negative exposure, while bonds and currencies respond more to regional and bilateral shocks than to domestic risk. Safe-haven assets such as JPY and CHF benefit during stress, and responses vary widely across asset classes and market conditions. Key takeaway: Geopolitical risk is not a single factor; its impact depends on regime, geography, and asset class.

Sports Betting

The ‘Sleeping Giant’: A Behavioral Anomaly in Prediction Markets (Vaze)

Playoff betting markets systematically overestimate the win probability of recent “dynasty” teams when they are underdogs, by about 10 percentage points. This creates a profitable strategy, with 16% gross returns (63% win rate), while comparable non-dynasty underdogs show no bias. Key takeaway: Even liquid markets can misprice assets when investors overweight recent success.

Trend Following

Rethinking Trend Following: Optimal Regime-Dependent Allocation (Zakamulin)

Trend-following performance depends critically on how exposure is allocated across regimes. While most research focuses on refining signals, the paper shows that optimizing regime-dependent position sizing alone delivers large gains. A Sharpe-optimal rule lifts Sharpe from 0.41–0.57 to 0.56–0.73 in U.S. equities, from 0.05 to 0.30 internationally, and from 0.21 to 0.51 in diversified portfolios. Key takeaway: Optimizing momentum exposures across regimes is a powerful and underexplored source of improvement.

Blogs

What Drives the Commodity Skewness Premium? (Quantseeker)

Breaking the Rules of Intraday Trading (Concretum Group)

Brave New Backtest (Robot Wealth)

War and Interest Rates (John H. Cochrane)

Why Alternatives Still Command High Fees (CFA Institute)

Podcasts

How Billionaire Hedge Fund Managers Are Using Generative AI to Invest (Odds on Open)

David Bush - Build a High-Performance Quant Crypto Portfolio Without Blowing Yourself Up! (The Algorithmic Advantage)

Sports Betting and Prediction Markets—Through a Trading Lens · Stefan Stadie (Chat with Traders)

Social Media & Industry Research

If You Can’t Beat It, Stack It: How Portable Alpha Overlays May Enhance Equity Returns (Quantica Capital)

I Did Not Predict What is Going on in Privates (Cliff Asness)

Unlocking Hidden Patterns: How Daily Returns Predict Future Stock Performance (Alpha Architect)

Last Week’s Most Popular Links

Understanding the Convolutional Neural Network’s Stock Return Predictability (Guan, Li, and Si)

Empirical Asset Pricing via Learning-to-Rank (Lin, Su, and Zhu)

Expected Returns with Trends and Cycles (Hillenbrand and McCarthy)

Paid subscriptions may not be available in all jurisdictions and may change without notice.

What Drives the Commodity Skewness Premium?

QuantSeeker — Mon, 30 Mar 2026 15:30:55 GMT

Over the past week, I have continued to explore skewness strategies in commodities. In a previous post, I showed that a time-series skewness strategy delivers a meaningful Sharpe ratio. A natural next question is whether this performance is broad-based across the commodity universe or concentrated in a smaller set of sectors or contracts. I examine and test this below.