Forecast accuracy
Hit rate = share of realizations that landed inside the bull/bear cone around AFV at forecast time. Thin buckets show "insufficient data" instead of fake percentages.
Accuracy not published yet
We only show hit rates after a backtest run writes enough samples to forecast_accuracy_daily. Until then we will not display placeholder percentages.
Does the Buy/Sell label predict direction?
A different question from the table above — that one checks whether the fair-value forecast (AFV) itself was close to the truth. This checks whether the signal derived from it (STRONG BUY / BUY / HOLD / SELL / OVERPRICED) actually predicts which way the price moves next: mean/median forward price move over the following 30 days, grouped by the signal that was active at forecast time. Thin buckets show "insufficient data" instead of fake percentages — same rule as the table above.
Signal accuracy not published yet
We only show forward-return stats after a backtest run writes enough samples to signal_accuracy_daily. Until then we will not display placeholder percentages.
Run backend/scripts/eval_arbitrage_signal.py to populate this section.
Source: nightly eval_arbitrage_signal.py → signal_accuracy_daily. Informational only — not financial advice.
Source: nightly / on-demand backtest_afv_hit_rate.py → forecast_accuracy_daily. Informational only — not financial advice.