On the accuracy of Federal Reserve forecasts of the budget deficit-output ratio

Abstract

Monetary authorities, while unable to resolve fiscal imbalances, have to deal with their consequences in formulating monetary policy. This article asks whether the Federal Open Market Committee (FOMC) is provided with accurate forecasts of the federal budget deficit-output ratio. We show that the forecasts made in the period 1982 to 2002 are unbiased with useful predictive information above that contained in time-series forecasts.

Notes

¹See, for instance, then-Chairman Alan Greenspan's testimony to the House Budget Committee on 25 February and 8 September 2004 and to the Joint Economic Committee on 3 November 2005, available on the website of Federal Reserve System.

²Given that the budget deficit should be judged relative to the size of the economy, we evaluate the forecasts of the federal budget deficit-output ratio (instead of the federal budget deficit alone). The Federal Reserve forecasts of nominal federal budget deficits and nominal output are taken from the Greenbook documents available on the Federal Reserve Bank of Philadelphia website. Emphasis changed from Gross National Product (GNP) to Gross Domestic Product (GDP) in 1992. Thus, output is measured by nominal GNP before 1992 and by nominal GDP for the period after.

³Our conclusions remain unchanged when we use the forecasts made in the last month of the quarter.

⁴Re-specifying the interest rates in real terms (nominal interest rate minus inflation rate) does not improve forecast accuracy. The same is true when we include the inflation rate as a separate variable.

⁵Real time data on the federal budget deficits are obtained from various issues of Business Conditions Digest, Survey of Current Business, and the Greenbook documents. Real time data on nominal output and the unemployment rate are available on the website of Federal Reserve Bank of Philadelphia.

⁶With the actual and forecast series present in Equations 1(1) and 2(2), the OLS covariance matrix estimator becomes inconsistent due to the serial correlation inherent in the error term (ε_t+f). As pointed out by Romer and Romer (2000), the order of the serial correlation increases as the forecast horizon becomes longer. In our case, ε_t+f follows an ( f +1)-order moving average process under the null hypothesis of rationality. Since forecast errors are generally heteroscedastic, we utilize the Newey–West procedure to estimate the covariance matrix of Equations 1(1) and 2(2), correcting for both heteroscedasticity and serial correlation over ( f +1) quarters.