It is widely known that conditional covariances of asset returns change over time. Researchers adopt many strategies to accommodate conditional heteroskedasticity. Among the most popular: (a) chopping the data into short blacks of time and assuming homoskedasticity within the blocks, (b) performing one-sided rolling regressions, in which only data from, say, the preceding five year period is used to estimate the conditional covariance of returns at a given date, and (c) two-sided rolling regressions which use, say, five years of leads and five years of lags. GARCH amounts to a one-sided rolling regression with exponentially declining weights. We derive asymptotically optimal window lengths for standard rolling regressions and optimal weights for weighted rolling regressions. An empirical model of the S&P 500 stock index provides and example.

[Many researchers have employed ARCH models to estimate conditional variances and covariances. How successfully can ARCH models carry out this estimation when they are misspecified? This paper employs continuous record asymptotics to approximate the distribution of the measurement error. This allows us to (a) compare the efficiency of various ARCH models, (b) characterize the impact of different kinds of misspecification (e.g., "fat-tailed" errors, misspecified conditional means) on efficiency, and (c) characterize asymptotically optimal ARCH conditional variance estimates. We apply our results to derive optimal ARCH filters for three diffusion models, and to examine in detail the filtering properties of GARCH(1, 1), AR(1) EGARCH, and the model of Taylor (1986) and Schwert (1989).]