Gold Dust: Robustness Testing with Consensus Across Optimized Models

Summary

Gold Dust proposes a robustness check for optimized trading systems that addresses the instability of financial-market statistics. Instead of optimizing one parameter set on one historical interval and forward-testing it, the method optimizes separate perceptron parameter sets on different intervals. In a later test, it trades only when those models agree on direction; disagreement suppresses the trade. The example system uses a single-layer neural network, with genetic-algorithm optimization of its weights and selected trading parameters.

The walkthrough divides EURUSD hourly history into three-month intervals: two are used for optimization and an earlier one for testing without optimization. It outlines staged optimization, a consensus test, and a subsequent comparison of the individual models. The author argues that this approach can help identify overfitting, but offers no reported performance results or guarantee of robustness. The method remains sensitive to interval choice and market non-stationarity, and the document recommends checking the process on another untouched historical interval.

Key ideas

• The method optimizes separate parameter sets on multiple historical intervals.
• It takes a trade only when the optimized models agree on direction.
• The example uses two perceptrons and a genetic algorithm to optimize their weights.
• A historical interval excluded from optimization is used for the consensus test.
• The author cautions that the approach cannot guarantee future performance and proposes testing another untouched interval.

Tags

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