Forecasting Inflation and Growth: Do Private Forecasts Match Those of Policymakers?

Abstract

FOMC projections are important because they provide information for evaluating current monetary policy intentions and because they indicate what FOMC members think will be the likely consequence of their policies. Results here show that the Blue Chip consensus forecasts are a good proxy for the FOMC views. For example, they match the policymakers' views as closely as do the Board staff forecasts presented at FOMC meetings. Using alternative forms of the Taylor Rule, we show that the Blue Chip consensus and the Fed Policymakers' forecasts have almost identical implications for the monetary policy process.

The Effectiveness of Monetary Policy

“…57 At the 1-year projection horizon conditioned on market interest rate expectations, the reported uncertainty measure is 0.76 percent; at the 2-year horizon it is 1.0 percent and at the 3-year horizon it is 1.10 percent. These values are on the order of 50 percent of the root-mean-squared error of the RBNZ and FOMC projections at comparable horizons, 54 Gavin and Mandal (2001), Table 2. Forecasts are fourth-quarter over fourth-quarter growth rates.…”

“…Absent instantaneous reaction of the economy to policy actions, effect stabilization actions require an assessment of the future state of the economy. Gavin and Mandal (2001) found the accuracy of the forecasts by FOMC participants as recorded in Monetary Policy Reports from 1983 through 1994 for real 53 For the United States a limited amount of vintage data has been reconstructed by the research staff of the Federal Reserve Bank of Philadelphia. Complete archives of the FRED data base have been preserved since the web version of this service was introduced in 1996 at least at monthly intervals; since 1999 at weekly intervals.…”

Bias in Federal Reserve Inflation Forecasts: Is the Federal Reserve Irrational or Just Cautious?

“…23,24 To correct for any autocorrelation in excess of j = h + 1 and for heteroskedasticity, expected from a non-constant variance in Figures 1 and 2, autocorrelation and heteroskedasticity corrected standard errors using Newey and West's (1997) method are employed. 25 The results are presented in Tables 1 (real-time data) and 2 (revised data).…”

“…24 The sample starts in 1968 rather than in 1965, as the Romers sample, because the quarterly data put together by the Philadelphia Fed starts the same date than the SPF data. 25 The bandwidth was chosen so that h lags were included to calculate the variance covariance matrix because under the null of forecast optimality the errors from the regressions should have autocorrelations up to order h different from zero. Newey-West method was chosen to avoid ending with a non positive definite matrix.…”

“…If the sample average of the errors is zero, then the estimate of θ would not be significantly different from zero, as the sample counterpart of orthogonality condition (25) would be satisfied only if θ is zero. If the sample average of the errors is not zero (i.e., if there is a bias), the value of θ is adjusted until the sample counterpart of the orthogonality condition is satisfied.…”