Performance-Driven Causal Signal Engineering for Financial Markets under Non-Stationarity

Battle Count: 71

Directly addresses the core problem of constructing causal, real-time trading signals in non-stationary markets. The walk-forward protocol, hysteresis decision rule, and risk-reshaping properties are highly relevant to practical quantitative trading. The framework provides an interpretable alternative to black-box ML approaches. However, the zero-cost assumption and high turnover limit immediate practical deployment. The emphasis on strict causality and out-of-sample execution aligns with best practices in systematic trading.

Moderate complexity. The signal construction pipeline (normalization, aggregation, derivative, gating) is well-defined and implementable with standard numerical operations. The walk-forward protocol adds orchestration complexity. No ML model training is required, but the grid search over multiple parameter dimensions and the rolling window management require careful implementation. The hysteresis logic and one-step delayed execution are straightforward. Main complexity lies in ensuring strict causality throughout and managing the walk-forward epoch structure.

Code promised to be made available upon request and released publicly after acceptance. Uses standard Python libraries (NumPy, pandas). All parameter grids, indicator definitions, and pipeline steps are explicitly specified. Data sourced from Binance public Spot API. However, code is not yet publicly available, and some implementation details (e.g., exact backtesting logic) are not fully specified.