SVM for Quarterly Multi-Factor Stock Selection

Summary

The document introduces support vector machines, outlining hard-margin classification for linearly separable data, soft margins for data with outliers, and kernel methods for nonlinear boundaries. It then applies an SVM with a radial basis function kernel to quarterly stock selection using financial statement and valuation factors. Features are standardized, and cross-validation is proposed to tune the penalty and kernel parameters.

The example labels a stock positive when its return over the following quarter exceeds 5%, and negative otherwise. It describes buying predicted positive stocks when not held and selling held stocks predicted negative. The universe is the CSI 300, with 21 accounting and valuation features drawn from a 2014 reporting date. A daily backtest is specified for March 2016 to March 2017 with stated initial capital, but the supplied text contains no performance figures or visible equity curve. It also gives little detail on portfolio weights, transaction costs, point-in-time data handling, or validation across other periods, so the example alone does not establish robustness.

Key ideas

• SVMs can use hard margins, soft margins, or kernels depending on whether the training data are linearly separable.
• The example uses a radial basis function kernel to classify stocks from standardized financial factors.
• Quarterly training labels distinguish stocks whose subsequent return exceeds 5% from those that do not.
• Cross-validation is used to choose the penalty and kernel parameters.
• The described backtest lacks reported performance measures and details needed to assess robustness.

Tags

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