We examine whether management earnings forecast errors exhibit serial correlation and how analysts understand the serial correlation property of management forecast errors (MFEs). MFEs should not exhibit serial correlation if managers efficiently process information in prior forecast errors and truthfully convey their earnings expectations through management forecasts. However, for long‐horizon management forecasts of annual earnings, we find significantly positive serial correlation in MFEs, and sample self‐selection does not seem to drive this phenomenon. Further analyses suggest that managers’ unintentional information processing bias contributes to this positive serial correlation. Analysts anticipate the intertemporal persistence of MFEs but underestimate the persistence level when reacting to management forecasts. Our findings have implications for market participants who rely on management forecasts to form earnings expectations, and also shed light on the efficiency of managerial decision making.