Using Time-Lagged ICA to Filter Market Noise in Trading Models

Summary

The article explores whether independent components analysis (ICA) can separate structure from noise in market data already smoothed by moving averages. It constructs a dataset from raw and SMA-filtered open, high, low, and close prices, along with lagged observations, then applies ICA to seek independent sources of temporal variation. Because ICA components are not labeled as signal or noise, the number of extracted components is treated as a tuning choice; more components need not improve forecasts.

The proposed workflow uses one model to approximate ICA embeddings and a second to forecast smoothed returns, with the output intended for an MQL5 expert advisor. The article describes model and data preparation, but the supplied excerpt does not give enough quantitative results to establish trading effectiveness. It recognizes that ICA depends on assumptions such as independent sources and stable distributions, and that errors from the first model can propagate into the second. It also leaves the choice of separation method and component count as open design questions.

Key ideas

• ICA is applied to moving-average-filtered price data and its lags to explore latent temporal structure.
• The number of extracted components must be tuned because components are not explicitly identified as noise or signal.
• A two-stage model first approximates ICA embeddings and then forecasts smoothed returns.
• The approach assumes stable, independent sources and may compound errors across its two models.

Tags

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