Sliced Inverse Regression for Equity Return Forecasting and Portfolio Ranking

Summary

The document explains sliced inverse regression (SIR) as a supervised dimension-reduction method for financial prediction. Unlike principal component analysis, which summarizes variation in predictors without using the target, SIR uses the forecast variable when deriving lower-dimensional predictors. The resulting target-specific directions are then used in a regression to estimate future stock returns.

The described application ranks stocks by predicted returns for portfolio allocation. Its example uses a rolling history to form principal components, applies SIR to relate lagged predictors to returns, takes the highest-ranked names as long positions and the lowest-ranked names as shorts, and rebalances periodically with equal weights. The text presents this as a possible input to active equity, index enhancement, and hedging decisions. It refers to a strategy illustration but supplies no numerical performance results, transaction costs, or robustness analysis, so the example does not establish profitability. The usefulness of the rankings would need to be assessed out of sample and under realistic trading constraints.

Key ideas

• SIR uses the response variable when reducing predictor dimensions, unlike ordinary PCA.
• The reduced predictors are specific to the target being forecast.
• A regression on the SIR-derived predictors produces stock return estimates for cross-sectional ranking.
• The example forms equal-weight long and short portfolios from the highest and lowest predicted-return ranks and rebalances periodically.
• The document gives no numerical performance evidence or analysis of trading costs and robustness.

Tags

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.