Bivariate and multivariate factor evaluation
To evaluate the numerous factors, we rely on the various performance measures introduced in this book, including the following:
- Bivariate measures of the signal content of a factor with respect to the one-day forward returns
- Multivariate measures of feature importance for a gradient boosting model trained to predict the one-day forward returns using all factors
- Financial performance of portfolios invested according to factor quantiles using Alphalens
We will first discuss the bivariate metrics and then turn to the multivariate metrics; we will conclude by comparing the results. See the notebook factor_evaluation for the relevant code examples and additional exploratory analysis, such as the correlation among the factors, which we'll omit here.
Information coefficient and mutual information
We will use the following bivariate metrics, which we introduced in Chapter 4, Financial Feature Engineering...
