Walk-forward tests
Models are trained on past data and tested on the period that follows, repeated across many folds and securities.
Product
Validation shows how forecasts performed out-of-sample, using walk-forward tests on stored history, so you can judge how much weight to give them.
Models are trained on past data and tested on the period that follows, repeated across many folds and securities.
Direction accuracy, how often prices land inside the 90% range, average error and Brier score are shown per horizon.
New models are only promoted when they pass accuracy gates. Promotion and rollback are audited.
Informational research tool. Not investment advice.