Eco 519 Take home project#1: Tests for Forecast unbiasedness
预测无偏性测试代写 I have attached one-year ahead consensus forecasts of real GDP made every quarter from 1971:02 till 2014:02. Actual real GDP growth data currently
I have attached one-year ahead consensus forecasts of real GDP made every quarter from 1971:02 till 2014:02. Actual real GDP growth data currently revised and those available in real time (first monthly advance announcements) are also attached. The purpose of this project is to test if these forecasts are “unbiased and efficient” by running a regression of the type: Yt = a + b. Xt + ut, where Yt is the actual GDP growth and Xt is the forecast. ut is the error term.
In your short report include the following: 预测无偏性测试代写
1) Draw the forecast against the two actuals in two separate diagrams and make comments on how they look. How different are the unrevised vintage data compared with the revised data?
2) Compute the variances of two actual Yt series and of forecasts Xt, and explain why variance of forecasts is smaller than that of actuals.
3) Run the regression Yt = a + b. Xt + ut and test jointly the unbiasedness hypothesis that (a=0, b=1) using currently available revised data. Explain why (a=0, b=1 jointly) implies unbiased forecasts.
4) Redo #3 above using real time unrevised data. Comment on any difference you see, and why. Should one depend more on revised or unrevised data in testing the unbiasedness hypothesis?
5) Check if the estimated residuals from the two regressions are serially uncorrelated using the DW test. Will the presence of autocorrelation in residuals suggest that forecasts are not efficient in the sense that the forecasters do not fully incorporate all relevant information in predicting GDP growth? Explain why.
Write a short report and submit a pdf file by email before Monday December 15. You will get a second take-home project next with the same deadline. These two projects will count 30% towards the final.
