To determine the appropriate level of risk capital financial institutions are required to empirically estimate and predict specific risk measures. Although regulation commonly prescribes the forecasting horizon and the frequency with which risk assessments have to be reported, the scheme with which the underlying data are sampled typically remains unspecified. It is shown that, given assessment frequency and forecasting horizon, the choice of the sampling scheme can greatly affect the outcome of risk assessment procedures. Specifically, sequences of variance estimates are prone to exhibit spurious seasonality when the assessment frequency is higher than the sampling frequency of non-overlapping asset return data. The autocorrelation function of such sequences is derived for a general class of weak white noise processes and for a general class of variance estimators. The problem of spurious seasonality can be overcome by using overlapping return data for estimation of risk measures.