Learn

Quantitative research library

Summaries of the papers and rules that the Stochastly assistant cites, each with its source. Read them here, or ask the assistant inside the app.

Definitions of the terms used in these notes: Glossary of quantitative research terms

Backtest overfitting

Sourced summaries of the research on backtest overfitting: deflated Sharpe, PBO, CPCV, data snooping, multiple testing and backtest-to-live decay.

Backtest versus live: a large cohort · Can forecasters forecast? (Cowles, 1933) · Combinatorial purged cross-validation · Data snooping and p-hacking · Deflated Sharpe ratio: the 2014 paper · Deflation vs Selection Bias · Does publication destroy predictability? · Evaluating trading strategies: the threshold · 20 notes

Research method

Short rules for research that can fail: registered hypotheses, positive controls, null construction, detection floors and reporting trials.

Report a backtest with its trials · Conditional samples count in episodes · A forecast is scored, or it is not a forecast · Register the hypothesis before the test · A judge outside its domain abstains · Small-sample effects are exaggerated · A specification chosen after the results · A sub-period difference is a detector · 3 notes

Statistics

Sourced summaries of the statistics behind strategy evaluation: Sharpe ratio errors, bootstrap, forecast comparison, shrinkage and first passage.

AIC and BIC: penalising parameters · Regime changes in financial markets · Why prices are a martingale · Comparing two forecasts: DM, CW, GW · Correlation matrices with few observations · Doob's optional stopping theorem · Efron's bootstrap · Extreme Value Theory for Tail Risk · 12 notes

Machine learning in finance

Sourced notes on machine learning for trading: leakage in cross-validation, calibration, concept drift, low signal-to-noise and model evaluation.

A backtesting protocol for machine learning · Backtest overfitting in ML strategy search · Can Machines Learn Finance? The Sober Answer · Concept drift and online learning · Conformal prediction intervals · Deep learning on the order book · Directional vs sizing models · Empirical asset pricing via machine learning · 9 notes

Risk and portfolio

Sourced summaries on risk and portfolio construction: Kelly sizing, drawdown constraints, diversification, risk parity, shrinkage and impact.

A prop firm challenge is a barrier option · Backtesting a value-at-risk model · Betting against beta · Black-Litterman: equilibrium and views · Correlation and diversification · Correlations rise in a crisis · Drawdown Control & the Calmar/MAR Ratio · Drawdown Stop-Outs and the Triple Penance Rule · 18 notes

Strategies

Sourced summaries of the research on momentum, trend following, carry, value and technical rules, with what survived out of sample and costs.

Time-Series Momentum (TSM) · Cross-Sectional Momentum (Jegadeesh-Titman) · Mean Reversion & Ornstein-Uhlenbeck (half-life) · Pairs Trading / Cointegration · Statistical Arbitrage (Avellaneda-Lee) · Carry Trade / Carry Factor · Value & Quality Factors · Short-Term Reversal · 3 notes

Market practice

Sourced summaries on how markets and traders behave: day-trader results, intraday and overnight returns, stop-loss cascades, crypto and prop firms.

Announcement days carry the premium · Arbitrage across crypto exchanges · Intraday and overnight returns cross-section · Crypto carry and perpetual funding · Why technical levels sometimes work · Day trading for a living: the evidence · Do day traders learn about their ability? · Retail CFD loss rates (ESMA) · 7 notes

Market microstructure

Sourced summaries of market microstructure research: market impact laws, spread estimation from bars, order flow memory and execution costs.

Algorithmic trading textbooks compared · Momentum after trading costs · Beyond the square-root impact law · Crypto wash trading and fake volume · Dynamic trading with costs · Estimating the spread from OHLC bars · How markets digest supply and demand · Implementation shortfall (Perold) · 1 note

Data practices

The data errors that bias a backtest: survivorship, delisting, timestamps and time zones, continuous futures rolls and bid-ask bounce, with sources.

A Wide Panel Empties Itself Silently · Bar timestamps, time zones and DST · Bid-ask bounce and false mean reversion · Cleaning Tick Data, Step by Step · Crypto Data Does Not Agree Across Sources · Is Bitcoin really untethered? · Point-in-time data: the real-time data papers · Rolls and continuous futures contracts · 8 notes

Econophysics

Using Stochastly

How the Stochastly engine handles bar conventions, custom node code, numerical stability and property tests, written for users who write their own nodes.

Bar conventions in the engine · Custom node code: what is allowed · Floating point and numerical stability · NumPy and pandas in custom nodes · Property tests and seeds · 1 note