Discovery of a 13-Sharpe OOS Factor: Drift Regimes Unlock Hidden Cross-Sectional Predictability
Mainak Singha · 2025
Read Discovery of a 13-Sharpe OOS Factor: Drift Regimes Unlock Hidden Cross-Sectional Predictability by Mainak Singha (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2511-12490.txt.
Introduction
Mainak Singha documents an equity trading factor that he reports achieves very high out-of-sample performance by switching on only in certain market regimes. The strategy combines value and short-term reversal signals, but activates them only for stocks in a drift regime, defined by a trailing window in which the stock rose on most days.
Singha reports walk-forward tests over twenty years of S and P 500 data with frozen parameters, plus randomization trials and checks that exposure to standard risk factors is small. Those are the author's reported results, presented as evidence that regime conditioning uncovers cross-sectional predictability hidden in unconditional tests. Readers should weigh the extraordinary claimed returns with that framing in mind.
Catalog record
| CREATOR | Mainak Singha |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 28 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2511-12490
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Discovery of a 13-Sharpe OOS Factor: Drift Regimes Unlock Hidden Cross-Sectional Predictability Mainak Singha Abstract We document an high cross-sectional equity factor achieving out-of-sample Sharpe ratios exceeding 13 through regime-conditional signal activation. By combining value and short-term reversal signals exclusively during stock-specific drift regimes—periods when individual stocks exhibit over 60% positive days in trailing 63-day windows—we generate annualized returns of 158.6% with 12.0% volatility and maximum drawdown of − 11.9 % -11.9\% . Through rigorous walk-forward validation spanning 20 years of S&P 500 data (2004–2024), we demonstrate performance approximately 13 times superior to market benchmarks on risk-adjusted basis, achieved entirely out-of-sample with frozen parameters. The strategy passes extensive robustness tests including 1,000 randomization trials yielding p-values below 0.001, maintains Sharpe ratios above 7 across 30% parameter variations, and exhibits virtually zero exposure to traditional risk factors with total R 2 R^{2} below 3%. We provide mechanistic evidence that drift regimes fundamentally alter market microstructure, creating unique conditions where behavioral biases amplify, liquidity dynamics shift, and cross-sectional price discovery mechanisms produce exploitable inefficiencies. Conservative capacity analysis suggests deployable capital of $100–500 million before material performance degradation. Keywords: Cross-sectional momentum, Regime switching, Market microstructure, Anomalies, Factor investing JEL Classification: G12, G14, G17, C58 1 Introduction The efficient market hypothesis has dominated financial theory ( Fama and French 1993 ; Jegadeesh and Titman 1993 ; Carhart 1997 ; Asness et al. 2013 ; Harvey et al. 2016 ) for half a century, yet practitioners and researchers continue to document persistent anomalies that challenge perfectly efficient price discovery. These anomalies represent billions of dollars in systematic trading profits and fundamentally shape our understanding of market dynamics. The tension between theoretical efficiency and empirical evidence of predictability remains one of finance’s most important puzzles, particularly as markets have become increasingly sophisticated with democratized access to complex statistical techniques, compressed inefficiency timescales through high-frequency trading, and proliferation of quantitative strategies arguably arbitraging away many historical sources of excess returns ( Daniel et al. 1998 ; Barberis et al. 1998 ; Hong and Stein 1999 ) . In this environment, discovering new, robust sources of alpha requires fundamentally new ways of thinking about market structure and price formation rather than simply better data or faster computers. Market regimes represent one of the most promising yet underexplored dimensions for alpha generation. While prices may be efficient unconditionally, substantial evidence suggests market dynamics vary considerably across different states, with volatility clustering, changing correlation structures, and time-varying risk premia all pointing to markets operating under multiple distinct regimes ( Ang and Bekaert 2002 ) . The key insight driving our research is that a signal or factor appearing weak or nonexistent when averaged across all market conditions may become extraordinarily powerful when applied selectively in the appropriate regime. Our contribution combines insights from regime-switching literature with simple, interpretable cross-sectional signals, demonstrating that straightforward combinations of value and reversal signals, when conditioned on stock-specific drift regimes, produce out-of-sample risk-adjusted returns far exceeding any previously documented factor. Rather than allowing factor loadings or risk premia to vary continuously with conditioning variables, we implement a binary regime gate that completely activates or deactivates the factor based on clearly defined criteria, recognizing that certain market conditions may fundamentally alter the information content of cross-sectional signals rather than merely modulating their strength. We define regimes at the individual stock level rather than the market level, showing that stock-specific regimes based on recent return patterns provide more granular and actionable information for cross-sectional strategies, allowing different stocks to be in different regimes simultaneously and dramatically expanding the opportunity set compared to strategies requiring specific market-wide conditions. 2 Data and Methodology 2.1 Strategy Construction We constructed our empirical analysis using daily data for all current constituents of the S&P 500 index spanning twenty years from January 2004 through December 2024, deliberately choosing current constituents despite introducing survivorship bias to ensure results are based on the most liquid and economically significant segment of the US equity market where implementation would be most feasible for institutional investors. Our base factor combines two well-established cross-sectional predictors whose innovation lies not in individual components but in how we condition their application on specific market regimes. The value component employs a price-based measure requiring no accounting information, computing inverse price for each stock then converting to cross-sectional ranks through percentile scores between 0 and 1. The reversal component captures the well-documented tendency for recent losers to outperform recent winners over short horizons, computing trailing 10-day returns and negating them to create contrarian signals ( Lo and MacKinlay 1990 ) , then standardizing cross-sectionally to z-scores ensuring comparability across varying market conditions. The BASE factor combines these components through the convex combination: \text{BASE}_{i,t}=0.7\times\text{value}_{i,t}+0.3\times\text{reversal}_{i,t} (1) with heavier weight on value reflecting its generally more stable signal properties ( Jegadeesh and Titman 1993 ; Asness et al. 2013 ) . The crucial innovation is our regime filter that gates the BASE signal, discovering that BASE, while mediocre on average, becomes extraordinarily powerful when applied selectively during specific market conditions. For each stock i at time t , we calculate the fraction of positive return days over trailing 63 trading days: \text{UpFraction}_{i,t}=\frac{1}{63}\sum_{k=1}^{63}\mathbb{I}[r_{i,t-k}>0] (2) defining stocks as being in drift regime when this fraction exceeds 60%: \text{REGIME}_{i,t}=\mathbb{I}[\text{UpFraction}_{i,t}>0.60] (3) The Unicorn Edge combines the BASE signal with regime filter through simple multiplication: \text{EDGE}_{i,t}=\text{BASE}_{i,t}\times\text{REGIME}_{i,t} (4) ensuring we only trade stocks exhibiting specific conditions where our signal proves most effective. Approximately 35% of stock-days qualify on average, varying from 8% during market crashes to 67% during strong bull markets. 2.2 Portfolio Implementation and Risk Management Converting EDGE scores into portfolio weights requires careful consideration of risk, diversification, and market neutrality. Each day we identify all stocks with valid, non-zero EDGE scores, compute standardized z-scores ensuring zero mean and unit variance, then construct market-neutral long-short portfolios by separating stocks into long and short buckets based on z-scores. We normalize within each side to control gross exposure, with long positions summing to 50% and short positions to 50%, ensuring constant gross exposure of 100% and approximately zero net exposure for market neutrality. Professional investment strategies require sophisticated risk management ensuring survival during adverse periods. We implement volatility and drawdown controls through scaling factor: s^{*}=\min\left(\frac{12\%}{\text{TrainingVol}},\frac{15\%}{|\text{TrainingMaxDD}|}\right) (5) satisfying both 12% annual volatility cap and 15% maximum drawdown constraint, with all portfolio weights multiplied by this factor determined once during training and applied without modification during testing. Beyond static scaling, we implement dynamic kill-switch protection shutting down the strategy entirely if experiencing losses beyond historical norms, monitoring absolute drawdown from peak (30% threshold) and rolling 63-day performance ( -10\% threshold). The kill-switch prioritizes capital preservation over return maximization, with switches that cannot reset within evaluation periods preventing emotional re-entry attempts. We employ rigorous walk-forward validation with true out-of-sample testing where all parameters are fixed based on historical data then applied to future periods without modification. Each evaluation consists of five-year training periods for parameter estimation followed by one-year test periods where we apply frozen strategies. We conduct three walk-forward windows: Window 1 covering post-financial crisis recovery (train 2005–2010, test 2010–2011), Window 2 representing stable markets (train 2010–2015, test 2015–2016), and Window 3 encompassing COVID disruption (train 2015–2020, test 2020–2021). Transaction costs of 0.6 basis points