🤖 ML Models
ML - Walk-Forward Validation
The only honest way to backtest a predictive model on time-series data. Train on a window of past data, predict on the next unseen period, slide the window forward, repeat. Simulates how the model would have performed if deployed in real time.
2
Minutes
6
Concepts
+15+30
Read+Quiz
1
Why It Matters
Standard train/test splits (or k-fold cross-validation) leak future information in time-series data. A model trained on Jan-Dec and tested on random samples from that range has "seen" data that follows its test points — this is lookahead bias, the #1 sin in quantitative finance.