Backtest Overview





Kerry Back

Preliminaries

  • Get data
  • Filter based on, e.g., size if desired
  • Define industries and industry dummies if desired
  • Transform features and ret in each cross-section
  • Define pipeline

Train and Predict Loop

For each date in a set of training dates,

  • Define training data = past
  • Train (possibly using cross-validation to choose hyperparameters)
  • Use the trained model to make predictions for each month until the next training date

portfolio returns

  • Use predictions to define portfolios at the beginning of each month. Example: best 100 and worst 100, equally weighted
  • Use actual (not transformed) stock returns to compute portfolio returns

Evaluate returns

  • Sharpe ratio, accumulation, drawdowns
  • Compared to beta-adjusted market benchmark: alpha, information ratio
  • Compare to market and other factors (e.g., Fama-French): alpha, information ratio, attribution analysis