Walk-Forward Backtesting
An out-of-sample model validation method that trains on a rolling historical window, predicts the next unseen period, then slides forward — preventing look-ahead bias.
Walk-forward backtesting is the institutional standard for validating a predictive trading model. Instead of training on all available history and testing on a random hold-out, you train on a window (say 2018-2022), predict the next out-of-sample period (2023 Q1), then slide the window forward and repeat. Each prediction is made only with data that would have been available at that point in time.
This matters because financial time series are non-stationary: relationships between features and returns drift as market regimes change. A model that looks brilliant on a single static train/test split is often overfit to one regime. Walk-forward testing exposes how a model behaves as conditions evolve, and produces a realistic equity curve you could actually have traded.
endeavr.ai walk-forward backtests every model in its four-architecture ensemble nightly before any weights are promoted to production. Results — directional accuracy, Sharpe, and drawdown versus buy-and-hold — are published per ticker so you can see how the model held up across full cycles, not just a cherry-picked window.