The OU chip on the VIX forecast is an Ornstein–Uhlenbeck process (its discrete-time twin is the AR(1) autoregression) — but fitted to log(VIX) rather than VIX itself. It assumes one thing: that volatility, once it spikes, is pulled back toward a long-run equilibrium, and the further it strays the stronger the pull. It is the switcher’s simplest, most honest benchmark.
This is the detail that matters for VIX. Working in log-space hands you two of VIX's most stubborn features for free. First, positivity: VIX can never go negative, and because the exponential of any number is positive, log-space paths can't wander below zero the way a naive model on the raw level might. Second, right-skew: VIX spends most of its time low and calm, then spikes violently and decays — a lopsided, long-right-tailed shape. Modelling the log reproduces that spike-and-decay asymmetry without any extra machinery. That is why every VIX model in the switcher is built in log-space.
The OU chip is the bar the other VIX models must clear. If the AR(2) or Market Conditions model can't beat this one-line log-space baseline on the walk-forward backtest, its extra complexity isn't earning its keep. You can check that directly: the "Model performance" table under the VIX forecast re-runs the backtest on current data and reports coverage, MAPE, RMSE, Brier and directional hit-rate for every model side by side.
Because it hands you two of VIX's features for free: positivity (exponentiated paths can't go negative, and VIX never does) and right-skew (VIX sits low then spikes and decays). Modelling log(VIX) reproduces that spike-and-decay shape without extra machinery.
That it mean-reverts: once VIX spikes, it is pulled back toward a long-run equilibrium, and the further it strays the stronger the pull. It has just three quantities — equilibrium, reversion speed and noise scale, all in log-space.
It's the baseline, not necessarily the best. Open the 'Model performance' table under the VIX forecast to see OU's coverage, MAPE, RMSE, Brier and hit-rate next to the AR(2) and Market Conditions models on current data.
Browse all S&P 500 tickers to see this metric applied to individual companies.
Educational research only — not investment advice.