The Macro-GBM chip is the most sophisticated model in the Fear & Greed switcher — and the one to be most careful with. It trains gradient-boosted quantile regressors (predicting q05/q50/q95 with pinball loss) on Fear & Greed history, macro features and the index's sub-indicators, then wraps the interval in a split-conformal (CQR) adjustment so the 90% band’s stated coverage is honest.
Sophistication is not the same as accuracy, and this page won't pretend otherwise. On the committed walk-forward backtest, Macro-GBM has not yet cleared the bar the other models are held to — on coverage, MAPE and Brier score — which is exactly why it is not the widget’s default. It is kept selectable precisely so you can watch it try to prove itself: the "Model performance" table under the Fear & Greed forecast shows Macro-GBM's coverage, MAPE, RMSE, Brier and directional hit-rate beside every other model. If and when it starts beating the OU baseline and the current default, that table is where you'll see it happen.
Because hiding it would be the dishonest choice. Showing a newer, not-yet-proven model alongside the simpler ones — with its shortfall stated plainly and a live table to check — is how you let a model earn trust on evidence rather than on a marketing label. Judge it by the Brier score and MAPE in the table, not by the word 'sophisticated'.
No. On the committed walk-forward backtest it has not yet cleared the bar the other models are held to on coverage, MAPE and Brier — which is exactly why it is not the widget’s default. Sophistication is not the same as accuracy.
Because hiding a newer, unproven model would be the dishonest choice. It is kept selectable — with its shortfall stated plainly — so you can watch it try to prove itself in the live 'Model performance' table on real data.
A distribution-free adjustment that widens or narrows the 90% band on a held-out slice so its stated coverage is honest, rather than assuming a symmetric residual bootstrap. It’s about making the uncertainty band trustworthy, not about the point forecast.
Browse all S&P 500 tickers to see this metric applied to individual companies.
Educational research only — not investment advice.