Essays on macroeconomic forecasting

Since the early 1970s, the dominant approach to expectation formation has been to assume rational expectations. But there is a growing concern among the economists about the validity of strong assumptions of rational expectations. In this work we evaluate rational expectations hypothesis using survey forecasts. In the first essay, we study whether deviations from rational expectations can be justified if forecasters use asymmetric loss functions. We question the reasonability of the magnitude and the behavior of the asymmetric loss function parameters over horizons. We propose a Nordhaus type of forecast efficiency test that can be used under the most popular forms of the asymmetric loss functions. Analyzing the monthly real GDP growth forecasts of 18 developed countries, we find strong evidence against informational efficiency and do not detect any evidence for the use of asymmetric loss function.; Finding strong evidence against rational expectations, we settle to investigate the sources of the deviations from rational expectations by constructing the stylized facts of expectation formation mechanisms. In the second essay, we analyze the differences between the long-run and short-run expectations. By developing two measures for the content of new information in forecasts, and using the real GDP growth forecasts of 18 countries, we find that the content of the news in the fixed-event forecasts varies significantly over horizons and resembles a bell-curve. The peak of the bell-curve, which denotes the timing of the most important information, is located when the horizon is approximately 12 months.; In the third essay, we characterize the nature of inefficiency in the real GDP growth forecasts of the G7 countries by developing a new tool, "intertemporal variance decompositions". We find that the degree of inefficiency tends to be smaller in forecasts of the USA than in forecasts of some European countries. We also find that one significant source of inefficiency is in utilization of foreign news.; In the last essay, we develop a framework to study forecasters' reaction to news under very general conditions. We explicitly model the reaction of forecasters by assuming that they may correct their earlier mistakes and have noisy information. Using 3-month Treasury Bill rate forecasts from a panel of professional forecasters, we find that the forecasters significantly underreact to the news when forecast horizon is more than two quarters. The degree of underreaction tends to increase as the horizon increases.