Testing HMM Regime Filters and GRU Confirmation for Trend Entries

Summary

The article presents an Expert Advisor design that combines a Hidden Markov Model with regime-specific Gated Recurrent Units. The HMM estimates whether conditions are ranging, trending, or unusually volatile from trend-strength and relative-volatility inputs. A conventional rule proposes a directional continuation trade, and the GRU may veto it if recent price sequence does not support the same direction with sufficient output strength. The intended benefit is more selective entries through a state-first, sequence-second process.

The main contribution is a testing framework: compare the raw rule, the HMM-gated version, and the HMM-GRU version, while logging candidate trades that each layer admits or rejects. The reported forward comparisons are inconclusive: both variants were profitable, but the non-GRU option scored better on several return and risk measures, while the GRU version had a higher win rate and occasional much larger losses. The setups differed in timeframe, signal settings, and trade management, limiting attribution. The article therefore calls for matched comparisons and rolling evaluation; it does not establish that recurrent confirmation adds predictive value.

Key ideas

• The HMM estimates market regime while the trading rule proposes direction and the GRU evaluates recent directional sequence.
• A fair evaluation should compare the raw rule, HMM-only filter, and HMM-GRU filter under matched settings.
• Trades rejected by the GRU should be tracked to determine whether the vetoes remove weaker candidates.
• The reported forward results do not isolate the GRU effect because the compared configurations differed.
• Higher win rate alone does not demonstrate improved expectancy when losses may be larger.

Tags

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