Essays on the properties of financial analysts' forecasts

Restricted until 6 April 2009. This work examines forecast errors in financial analysts' earnings forecasts. First, the relative accuracy of financial analysts' and adaptive time-series forecasts is considered. The central question is whether financial analysts efficiently utilize available information and produce forecasts that are more accurate than predictions of statistical models. The study employs a novel forecasting approach -- artificial neural networks and identifies cognitive anomalies that influence the analysts' forecasting behavior. Financial analysts exhibit systematic optimism for a specific subset of companies. The magnitude of the analysts' optimistic forecast bias increases with the difficulty of the forecasting task, which is represented by statistical characteristics of a firm's earnings as well as the overall economic activity. Both the mean and median forecast errors are largest for companies with the most volatile earnings that move against or independently of the market earnings. The study also presents a model of the analysts' forecasting behavior and provides evidence that the analysts' optimistic forecast error somewhat slowly decreases throughout the forecast horizon. Financial analysts on average are found to overreact to positive earnings releases and underreact to negative. In addition, they tend to ignore the expected economic activity when making earnings forecasts and, furthermore, fail to adjust their forecasts appropriately in periods of economic downturns. It leads to the inverse relationship between the optimistic forecast bias and the overall economic activity. The evidence presented contributes to the understanding of the formation and value of analysts' predictions.