RMSE (root-mean-squared error) measures how far a point forecast lands from what actually happened. You square each error, average the squares over the backtest windows, and take the square root — so large misses are penalized far more than small ones. Lower is better.
RMSE is expressed in whatever the target is measured in — Fear & Greed index points (a 0–100 scale) for the sentiment forecast, or dollars for a price forecast. An RMSE of 18 index points means the typical miss is roughly 18 points on that scale. Because the number carries units, it is only meaningful compared across models on the same target and horizon; lower is better.
On Quantustik’s committed Fear & Greed forecast leaderboard (the checked-in walk-forward baseline the calibration page reflects), the leading model posts an RMSE of about 18.3 index points across the backtest. That leaderboard is a live table with one RMSE column per model, so the ranking is verifiable, not taken on trust. Treat it as a model-comparison diagnostic, not a promise about any single future forecast — and note the sample is small (a few dozen non-overlapping windows), so narrow gaps are noise.
RMSE and MAPE answer related but different questions. MAPE is a percentage error, unit-free and comparable across targets; RMSE is in the target’s own units and leans harder on the big misses. A model can look better on one and worse on the other, which is why the leaderboard shows several scores side by side.
The same units as the forecast target — Fear & Greed index points for the sentiment forecast, dollars for a price forecast. That means an RMSE is only comparable across models forecasting the same target over the same horizon.
Squaring makes large misses count for much more than small ones, so RMSE rewards a model that avoids big blow-ups. A model that is usually close but occasionally far off scores worse on RMSE than one that is consistently a little off.
MAPE is a unit-free percentage error, comparable across targets; RMSE is in the target’s own units and penalizes large errors more. A model can rank better on one and worse on the other, which is why the leaderboard shows both.
Browse all S&P 500 tickers to see this metric applied to individual companies.
Educational research only — not investment advice.