Fitness Measures for Genetic Programming of Equity Factors
Summary
The document lists fitness measures for a genetic programming module that evolves stock factors. It describes ICIR as the mean information coefficient divided by its standard deviation, with the information coefficient measuring the relationship between factor values and subsequent stock returns. Mutual information is also listed, but the post gives no calculation details for it.
The remaining measures evaluate factor-sorted portfolios. Each day, stocks are divided into ten groups by factor value. Long-short measures use the return difference between the highest and lowest groups; long-only measures use the highest-factor group. For each approach, the listed objectives include cumulative summed returns, Sharpe ratio, and annualized volatility, with volatility scaled by the square root of 250. These are definitions rather than a comparison of fitness functions: the document gives no sample results, implementation details, or guidance on choosing an objective, and it does not discuss costs, turnover, or bias controls.
Key ideas
- ICIR evaluates the average information coefficient relative to its variability.
- Mutual information is named as a fitness measure, but its computation is not explained.
- The listed portfolio measures sort stocks into ten groups by factor value each day.
- Long-short measures compare the highest and lowest factor groups, while long-only measures use the highest group.
- Return, Sharpe ratio, and annualized volatility are listed as objectives for both portfolio approaches.
Tags
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.