Combining SSA Forecasts and Bayesian Classification for Short-Term Trading
Summary
The article describes a short-term recommendation system that combines Singular Spectrum Analysis (SSA) forecasts with Bayesian classification. It builds three SSA-based indicators for trend, MACD-like direction and stochastic behavior, then groups their readings into rising, falling or uncertain states. Agreement among indicators is intended to strengthen a near-term directional forecast; an epsilon band marks readings too close to zero for a confident direction.
The author evaluates the approach on historical price series for gold and Brent futures, a ruble-dollar instrument, and EUR/USD, across intraday timeframes. The described comparisons include roughly 1,000-point samples and charts of predicted versus observed values. The article notes that forecasts can lag and that directional errors remain, motivating the combined classification. It presents the system as a recommendatory method, with practical issues still to resolve before automation. SSA may be unreliable in highly volatile, noisy or thinly traded markets, and the reported illustrations do not establish broad profitability or robustness.
Key ideas
- SSA separates price series into components that can be extended to produce forecasts.
- Three SSA-based indicators estimate trend, MACD-like direction and stochastic movement.
- An epsilon zone labels near-zero indicator readings as uncertain rather than directional.
- Combining indicator states through Bayesian classification is intended to reduce critical forecast errors.
- The method is evaluated on selected instruments and intraday histories, but noisy or illiquid markets limit its reliability.
Tags
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.