AUTOREGRESSIVE CONDITIONAL HETEROSCEDASTICITY IN DAILY WIND-SPEED MEASUREMENTS

It is argued that the predictability of meteorological variables is no t constant but shows regular variations. This is shown for the daily m ean wind speeds and its meridional and zonal components at Shearwater, Canada, for the period 1963-1988. To capture this feature, a Generali sed Auto Regressive Conditional Heteroscedastic model is proposed. In this model, the conditional variance of an observation depends linearl y on the conditional variances of the previous observations and on the previous prediction errors. Here, conditional heteroscedasticity mode ls are used which let the variance depend on previous prediction error s, in conjunction with an autoregressive model for the mean, using the Gamma distribution for the wind speed and the Normal distribution for its components. It is shown that these heteroscedastic models outperf orm their homoscedastic versions, and that heteroscedastic features ar e more clear in the wind speed component records.