Improving earnings forecasts using intra-industry information transfer
Faculty Advisor: Jacob Thomas
This study examines whether forecasts for quarterly earnings per share (EPS) from analysts (obtained from IBES) and a time series model (seasonal random walk with drift) can be improved by using intra-industry information transfers (historic co-movement in the unexpected earnings of firms within an industry). After confirming that co-movement patterns are relatively persistent through time, I combine the historic co-movement between firm pairs with the forecast error of the firm that announces earlier to update the forecast for the firm that has not yet announced. I then measure the improvement in forecast accuracy of these updated forecasts, relative to the corresponding forecasts available just prior to the earlier announcer's earnings release, for both time-series and analysts' forecasts. Though I also examine the bias and accuracy of the available and updated forecasts, I focus primarily on precision (the inverse of the root mean square error from the regression of actual EPS on forecast EPS) when measuring performance. My results are as follows: (a) while I find unambiguous benefits to updating time series forecasts with information transfer, there is no obvious improvement when I update analyst forecasts with information transfers; (b) analysts appear to update their forecasts for firms after observing the unexpected earnings for earlier announcers; and (c) aggregating updated forecasts from different pairs containing that later announcer improves performance, especially for updated forecasts derived from analyst forecasts.