Multiple Regression for Building and Evaluating Indicator-Based Strategies

Summary

This article explains how multiple regression can combine technical indicator values to estimate subsequent price movement and turn the fitted equation into trading rules. It presents an EURUSD hourly example using indicator inputs, directional decisions based on the regression output, and a threshold to filter entries. The author describes a training sample and a later test period, reporting that the simple model remained profitable in the test period, while an expanded indicator and period search performed worse out of sample.

The discussion emphasizes that regression measures numerical association rather than causation, and that correlated inputs, too many candidate parameters, nonlinear relationships, and outliers can undermine results. It gives a rule of thumb that observations should substantially outnumber predictors and warns that optimization across many indicator settings increases overfitting risk. The examples are historical and limited; reported profitability does not establish robustness under other markets, periods, transaction costs, or execution conditions.

Key ideas

• Multiple regression can map indicator values to an estimate of future price movement.
• A regression equation can be translated into directional rules and entry thresholds.
• Correlated predictors and large parameter searches can make fitted relationships unstable.
• Outliers and nonlinear effects may distort regression estimates.
• The article's examples show mixed test-period outcomes and do not prove broad robustness.

Tags

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