A novel approach to trading strategy parameter optimization, using double out-of-sample data and walk-forward techniques

Battle Count: 136

Directly addresses core quantitative trading challenges: walk-forward optimization methodology, overfitting mitigation, out-of-sample testing integrity, transaction cost sensitivity, and portfolio construction. The novel contribution of parameterizing walk-forward window lengths is highly relevant to strategy development. The double out-of-sample validation approach (global training + unseen period) provides a rigorous framework. Results on intraday cryptocurrency data with realistic cost assumptions are directly applicable to algorithmic trading practice.

Moderate complexity. The EMA crossover strategy itself is simple, but the full framework involves: (1) 81 combinations of walk-forward window sizes, (2) nested optimization of EMA parameters within each WF step, (3) grid smoothing with weighted averages, (4) block bootstrap with 1000 iterations, (5) cost sensitivity across 7 levels, (6) multi-asset cross-validation, and (7) portfolio construction. The GitHub repository provides implementation code. Computationally intensive due to nested loops and bootstrap procedures.

GitHub repository with all calculations is publicly available (https://github.com/tmr-crypto/wf_optim_crypto_analysis). Data sourced from Kaggle G-Forecast challenge. Detailed algorithm pseudocode provided. However, the specific Kaggle dataset version and preprocessing steps may need clarification. Smoothing weights follow Gómez and Ślepaczuk (2021) methodology.