← Back to Blog

general

Does crypto direction get more predictable at longer horizons?

Auto-generated by Gordon Lab from a real backtest — published as a draft for review, not live.

Does crypto direction get more predictable at longer horizons?

The question: We swept XGBoost direction prediction across 15m/1h/4h/1d horizons on 25 coins, strictly out-of-sample. Does signal emerge as you zoom out, or is it coin-flips at every horizon?

What the data said

'15m' leads on mean_auc: +0.526

Full results

group metric
15m 0.5262
4h 0.5145
1h 0.5143

Generated by Gordon Lab from a real backtest. Method + caveats live in the harness; this is research, not advice — and the losing runs are shown too.