This dissertation contains three self-contained chapters dealing with specific aspects of financial analysts' earnings forecasts. After recent accounting scandals, much attention has turned to the incentives present in the career of professional financial analysts. The literature points to several reasons why financial analysts behave overoptimistically when providing their predictions. In particular, analysts may wish to maintain good relations with firm management, to please the underwriters and brokerage houses at which they are employed, and to broaden career choice. While the literature has focused more on analysts' strategic behavior in these situations, less attention has been paid to the implications these factors have on financial analysts' loss functions. The loss function dictates the criteria that analysts use in order to build their forecasts. Using a simple compensation scheme in which the sign of prediction errors affect their incomes differently, in the first chapter we examine the implications this has on their loss function. We show that depending on the contract offered, analysts have a strict preference for under-prediction or over-prediction and the size of this asymmetric behavior depends on the parameter that governs the financial analyst's preferences over wealth. This is turn affects the bias in their forecasts. Recent developments in the forecasting literature allow for the estimation of asymmetry parameters after observing data on forecasts. Moreover, they allow for a more general test of rationality once asymmetries are present. We make use of forecast data from financial analysts, provided by I/B/E/S, and present evidence of asymmetries and weak evidence against rationality. In the second chapter we study the evolution over time in the revisions to financial analysts' earnings estimates for the 30 Dow Jones firms over a 20 year period. If analysts' forecasts used information efficiently, earnings revisions should not be predictable. However, we find strong evidence that earnings revisions can in fact be predicted by means of the sign of the last revision or by using publicly available information such as short interest rates and past revisions. We propose a three-state model that accounts for the very different magnitude and persistence of positive, negative and `no change' revisions and find that this model forecasts earnings revisions significantly better than an autoregressive model. We also find that our forecasts of earnings revisions predict the actual earnings figure beyond the information contained in analysts' earnings estimates. Finally, the empirical literature on financial analysts' forecast revisions of corporate earnings has focused on past stock returns as the key determinant. The effects of macroeconomic information on forecast revisions is widely discussed, yet rarely tested in the literature. In the third chapter, we use dynamic factor analysis for large data sets to summarize a large cross-section of macroeconomic variables. The estimated factors are used as predictors of the average analyst's forecast revisions for different sectors of the economy. Our analysis suggests that factors extracted from macroeconomic variables do, indeed, improve on the current model with only past stock returns. In trying to explain what drives financial analysts' forecast revisions, the factors representing the macroeconomic environment must be considered to avoid a potential omitted variable problem. Moreover, the explanatory power and direction of such factors strongly depend on the industry in question.