In this paper we show that the errors in the Federal Reserve's weekly preliminary money supply estimates may be treated as an autoregressive, conditionally heteroskedastic (ARCH) process. We present theoretical and empirical evidence concerning the implications of systematic Fed errors for announcement effect studies. The results show that findings in previous studies of structural change in the response of interest rates to unanticipated changes in the money supply and of significant negative effects of anticipated money changes on interest rates are not robust when corrections are made to incorporate efficient market responses to error-ridden announcements.