Post by ehsanulh125 on Jan 9, 2024 13:51:45 GMT 7
As has been discussed many times on the Economania blog, although one of the most commonly used simplifying assumptions in modern macroeconomics is that of rational expectations, there is considerable empirical evidence to the contrary. I do not intend to argue for either approach in this post – that is for the interested readerhereandherecollected posts for your consideration - however, I present a recent, very simple methodological, and precisely because of this very compelling, study that examines the forecast biases of professional forecasters and American households. In their study entitled "A Comprehensive Empirical Evaluation of Biases in Expectation Formation", Kenneth Eva and Fabian Winkler look for an answer to the question of whether a more accurate forecast could have been given in real time (i.e., for example, in quarterly surveys using only the data available up to the given quarter) for a wide range of macro variables with a simple bias correction.
Eva and Winkler draw parallels with Buy Bulk SMS Service Welch and Goyal's (2008) paper on finance, in which the authors argued that it is not enough to use the entire sample to say that one variable was able to significantly predict the return premium over and above the risk-free return - that for an investor to find the story interesting, it must be shown that a model would have been able to do this in real time . Eva and Winkler use the following general regression equation to examine the forecasts given in the t -th quarter: where y t+h is the realization of the variable (e.g. GDP growth) h -quarters later (in the study, the forecast horizon is basically three quarters), y e t+h | t is the prediction of the predictors for the t+h quarter (taking the average of the individual forecasts as the consensus), which was given in the t -th quarter, x t is a K -dimensional vector in which the predictors known at the t -th time can be included, b is a vector of coefficients belonging to, while u t+h is an error term. If the consensus forecast is rational, i.e.
the forecasts did not differ systematically and predictably from the outcome, i.e. the forecast error on the left side of the equation cannot be predicted, then the coefficient b must be zero . However, if the forecasts are not rational, but follow some pattern of behavioral economics, then it is reasonable to assume that we can use the estimated coefficients to correct the forecasts, which is given by the following equation: It is important to estimate the coefficients only on the past sample, lasting at most t period, for example using a rolling window or recursive window procedure.
Eva and Winkler draw parallels with Buy Bulk SMS Service Welch and Goyal's (2008) paper on finance, in which the authors argued that it is not enough to use the entire sample to say that one variable was able to significantly predict the return premium over and above the risk-free return - that for an investor to find the story interesting, it must be shown that a model would have been able to do this in real time . Eva and Winkler use the following general regression equation to examine the forecasts given in the t -th quarter: where y t+h is the realization of the variable (e.g. GDP growth) h -quarters later (in the study, the forecast horizon is basically three quarters), y e t+h | t is the prediction of the predictors for the t+h quarter (taking the average of the individual forecasts as the consensus), which was given in the t -th quarter, x t is a K -dimensional vector in which the predictors known at the t -th time can be included, b is a vector of coefficients belonging to, while u t+h is an error term. If the consensus forecast is rational, i.e.
the forecasts did not differ systematically and predictably from the outcome, i.e. the forecast error on the left side of the equation cannot be predicted, then the coefficient b must be zero . However, if the forecasts are not rational, but follow some pattern of behavioral economics, then it is reasonable to assume that we can use the estimated coefficients to correct the forecasts, which is given by the following equation: It is important to estimate the coefficients only on the past sample, lasting at most t period, for example using a rolling window or recursive window procedure.