Burg Linear Prediction and the Overfitting Risk of Reoptimization

Summary

This document describes an Expert Advisor that predicts a future price from a sequence of past prices using linear prediction. Burg’s method estimates model coefficients by reducing mean-square prediction error over a training window. The strategy’s configurable inputs include the history length, model order, a minimum predicted profit threshold, risk limits, trade counts, and profit, loss, and trailing-stop controls. It can optionally transform prices into log returns or percentage rate of change, with the two transformations mutually exclusive.

The document offers no performance study or out-of-sample evidence. Instead, it gives a direct caution about the method: like many optimized Expert Advisors, it performs well on training data but can lose money without repeated reoptimization. This warning points to overfitting and unstable live performance as central limitations. The description does not specify how often to reoptimize, how to validate predictions, or whether the listed risk controls mitigate those limitations.

Key ideas

• Linear prediction estimates future prices as a weighted function of prior observations.
• Burg’s method selects model coefficients by reducing mean-square error on a training segment.
• The Expert Advisor offers return transformations, but its momentum and rate-of-change options cannot both be enabled.
• Its configurable controls include risk, position count, predicted profit, and exit thresholds.
• The document warns that training performance may not persist without ongoing reoptimization.

Tags

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