Definition
Cross-Validation (Trading Models)
Cross-validation is a model-evaluation method that partitions data into multiple training and testing folds to assess how well a predictive model generalises, adapted carefully for the time-ordered nature of markets.
Standard k-fold cross-validation shuffles data, which leaks future information in a time series and is therefore unsafe for trading. Indian quants instead use time-aware variants such as expanding-window or purged, embargoed cross-validation that respect chronology and prevent look-ahead.
Used correctly, cross-validation gives a distribution of performance estimates rather than a single number, helping distinguish robust signals from lucky ones. It complements walk-forward analysis and out-of-sample testing as a defence against overfitting the noise in historical prices.
Related terms
- BacktestingBacktesting is the process of simulating a trading strategy on historical data to estimate how it would have performed, including returns, drawdowns and risk, before committing real capital.
- Walk-Forward AnalysisWalk-forward analysis is a backtesting technique that repeatedly optimises a strategy on one window of historical data and tests it on the immediately following out-of-sample window, rolling forward through time.
- Out-of-Sample TestingOut-of-sample testing evaluates a strategy on data that was deliberately withheld during model development, providing an unbiased check of whether the discovered edge generalises beyond the fitting period.
- OverfittingOverfitting, or curve-fitting, occurs when a strategy is tuned so closely to historical data that it captures random noise rather than a genuine pattern, and consequently fails on new data.
Plain-English explainer from The Dispatch Investors Encyclopedia. General information, not financial advice.