Forward-Oriented Causal Observables for Non-Stationary Financial Markets

Battle Count: 68

The paper is highly relevant to quantitative trading as it addresses the critical practical constraint of strict causality in signal construction, which is often violated in academic backtests. The emphasis on online computability, regime dependence, and transparent decision rules aligns with practitioner needs. However, the lack of transaction cost modeling, single-asset focus, and absence of ML baselines limit direct applicability. The insight that causal derivative-based operators can inject forward-oriented structure without look-ahead bias is practically valuable for signal design.

The methodology is relatively straightforward to implement: standard technical indicators (RSI, MFI, MACD, BB%), a simple 1D Kalman filter, linear aggregation, finite-difference derivative with moving-average smoothing, and a threshold-based hysteresis rule. All parameters are fixed and explicitly stated. The main implementation challenges are ensuring strict causality in all operations (especially the causal median), correctly handling trading hours filtering, and implementing the state-dependent mixing coefficients. No complex optimization or training is required.

The methodology is described with explicit formulas, fixed parameters (Kalman q=0.01, r=0.1; threshold θ=0.06; indicator lookback windows at conventional values; derivative smoothing span=4; scaling constants α specified in Figure 1). However, no code repository is provided, no explicit data source URL is given, and the EURUSDT 1-minute data source is not specified. The trading hours filter (removing weekends, specific Sunday/Friday cutoffs) is described but implementation details may vary. No parameter optimization was performed, which aids reproducibility.