Using LDA to Map Lagged Price Data to Moving-Average Forecasts

Summary

This article applies the category-theory idea of a natural transformation to time-series forecasting with differently structured datasets. One dataset contains a moving-average value; another contains the constituent lagged prices. The author treats the mapping between these tables as a natural transformation and uses linear discriminant analysis (LDA) to relate the price columns to a future moving-average value. Because the datasets are time-staggered by the moving-average period, the forecast horizon is also that period rather than the next bar.

The discussion explains LDA as a classifier that learns weights for separating labeled groups, then describes using those weights to map between the simpler and more detailed data representations. The article reports limited optimization results from only a few generations and says the runs did not walk forward. It also flags the short historical test window and suggests longer testing or combining the signal with another component. The example demonstrates a modeling construction, not established forecasting or trading performance.

Key ideas

• Natural transformations are presented as mappings between related datasets with different column structures.
• The simple dataset contains moving averages, while the compound dataset contains their constituent prices.
• LDA supplies a learned mapping from lagged prices to a moving-average value at a later horizon.
• The forecast horizon equals the moving-average period, so it is not an immediate next-bar forecast.
• The reported optimization was limited and did not walk forward, leaving predictive robustness unproven.

Tags

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