Using QDA to Filter Losing Trades in an Intraday Momentum Strategy

Summary

The article explains how Quadratic Discriminant Analysis (QDA) differs from Linear Discriminant Analysis (LDA), then applies QDA to an intraday momentum strategy using three-minute e-mini S&P 500 futures data. LDA assumes classes share a covariance matrix, while QDA estimates a separate covariance matrix for each class. QDA can better fit data when that shared-covariance assumption fails, though its added flexibility can raise variance and its parameter count makes it more demanding, especially with many predictors.

The example builds a momentum strategy around RSI, then uses return direction and lagged volatility measures alongside RSI to classify trades as likely winners or losers. The author reports 92% accuracy in identifying losing trades and says filtering trades changed win and loss rates only marginally while increasing returns by $600 over the tested period, leaving the adjusted strategy positive. These results are specific to the example and its data; the author says the initial strategy was intentionally imperfect and does not present the exercise as a production-ready system. The article’s performance figures alone do not establish robustness or out-of-sample reliability.

Key ideas

• LDA assumes all classes share a covariance matrix, whereas QDA estimates a separate covariance matrix for each class.
• QDA offers more flexibility but can overfit and requires more parameters to estimate.
• The example applies QDA to classify likely losing trades in an RSI-based intraday futures strategy.
• The author reports high accuracy for identifying losing trades and a positive change in the example’s returns.
• The strategy was deliberately imperfect, and the reported results do not demonstrate production readiness.

Tags

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