5-fold time-series cross-validation with 24-month rolling training window and 3-month forward validation period. Daily model retraining with no look-ahead bias.
Data Quality
Institutional-grade tick data from live exchange API feeds. Survivorship-bias-free historical database with robust missing-data handling.
Conservative Assumptions
Realistic fees and slippage applied throughout. No data snooping on test sets. Walk-forward testing under executable market conditions.
