Abstract

We document the relationship between the CBOE Volatility Index (VIX) and subsequent realized volatility of the S&P 500 over the period 1990–2025 using 9,066 daily observations. Across the full sample, the VIX has averaged 19.5 (percent annualized), with realized volatility over the subsequent 21 trading days averaging 15.4 percent. The implied-minus-realized gap—the volatility risk premium—has averaged +4.02 percentage points, with 85.1% of trading days showing a positive premium. A linear regression of forward 21-day realized volatility on contemporaneous VIX yields a slope of 0.887 (Newey-West HAC t = 18.9 at 21-lag truncation), an intercept of -1.82, and R² of 0.516, implying that the VIX systematically over-predicts realized volatility by an amount consistent with a positive volatility risk premium. We document the variation in the premium across twelve historically motivated market regimes, identifying three notable episodes in which the premium turned substantially negative: the September–October 2008 phase of the global financial crisis (VRP = -0.42 average), the COVID shock of March–April 2020 (VRP = -11.10 average, the most negative in the sample), and brief windows during the 2018 “Volmageddon” episode. Three findings are specifically novel relative to the pre-2020 literature. First, the COVID-shock negative VRP of -11.10 exceeds the GFC negative VRP by a factor of 26, a magnitude that the pre-2020 calibration of rare-disaster models would not have predicted. Second, the post-COVID recovery period (mid-2020 through 2021) exhibited a sustained positive VRP of +7.82, nearly double the unconditional mean, suggesting an upward shift in the price of volatility insurance following the pandemic shock. Third, the persistence of the premium (+3.57) through the post-2022 monetary-tightening cycle—the most aggressive since 1981—constitutes an out-of-sample confirmation that the premium is structural rather than regime-dependent. We supplement the descriptive analysis with a formal return-predictability regression controlling for the term spread, credit spread, and dividend yield, confirming that contemporaneous VIX retains incremental predictive power for subsequent 21-day equity returns (HAC t = 3.42) after accounting for standard macro-financial predictors. We discuss the implications for option-writing strategies, for the interpretation of VIX as a fear gauge, and for the theoretical literature on the variance risk premium.