ETARE: Evolving Trading Strategies with Neural Networks and Reinforcement Learning
Summary
The article describes ETARE, a proposed trading system that combines an LSTM model, reinforcement learning, and an evolutionary population of strategies. Strategies carry neural network weights, which can be recombined or mutated; periodic selection removes individuals that fail specified performance thresholds and replenishes the population. The article also outlines prioritized experience memory, batch training, confidence scoring that blends model predictions with historical patterns, and adjustments linked to market volatility.
The evidence is primarily architectural description and code examples, alongside the author’s account of development and live testing. It does not provide independently verifiable performance statistics, a detailed test design, or enough information to assess robustness. The article mentions DCA and separate position closing in its conclusion but gives little detail about how those methods work. Its claims of effectiveness and profitability should therefore be treated as the author’s assertions, not as demonstrated results.
Key ideas
- ETARE represents candidate trading strategies as neural network weights that can be inherited, recombined, and mutated.
- Periodic selection removes strategies that fail specified profit-factor or win-rate thresholds and replaces them with new individuals.
- The described LSTM and reinforcement learning process use recorded state, action, reward, and next-state experience to update decisions.
- A confidence score combines model output and historical pattern performance, then adjusts for current volatility.
- The article offers implementation sketches and personal testing claims but no detailed, independently verifiable performance evidence.
Tags
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.