Guides

Trading model guides that stay close to the problem.

This hub is built for the searches people actually make: walk-forward validation, train-test leakage, signal evaluation, threshold calibration, slippage, baselines, and OHLCV feature design.

Start here if you want the theory before you touch the knobs.

Open the Training Simulator See how the guides map to a live model run.

Reading order

Core path first, then the adjacent questions.

The order matters. Start with chronology and leakage, then move into the label, the baseline, the cost model, and the features that feed the whole thing.

Walk-forward validation Why chronological retraining beats a shuffled split when regimes move and scores drift. Best first read for leakage, retraining, and threshold drift. Triple-barrier targets How the label changes when a trade has a stop, a target, and a timeout. Use this before you tune the model on raw returns. Boosted trees vs linear baselines What a tree buys you, and why the linear check still matters. Read this before you decide the model family matters more than the signal. Transaction costs and slippage Where paper edges go to die, and how to measure the damage. Use this before you trust gross PnL. Feature engineering for OHLCV models Which rolling features matter, and where threshold calibration fits. Best read when the model is learning on bar data.

Adjacent questions

Use these when you know the problem already.

These cluster pages answer the searches that sit next to the core guides: split choice, signal quality, slippage, threshold setting, and model family selection.

Walk-forward validation vs train-test split See why a time-series split is not a detail when leakage is the whole problem. Best next read if the backtest is still too optimistic. How to evaluate a trading signal A compact checklist for score quality, phase stability, and live-trading usefulness. Use this when you want the full signal test in one place. What is slippage in model backtesting A plain explanation of where the fill moves away from the price you expected. Read this before you blame the model for a bad fill model. Ridge vs boosted trees for alpha models When ridge is the harder benchmark, and when boosting really does add signal. Use this for baseline comparisons on the same features. How to choose a prediction threshold How to turn a score into trades without letting noise run the book. Best for activation levels, trade frequency, and threshold drift.