- Email: [email protected]
Is monetary policy important for forecasting real growth and inflation?
Journal of Policy Modeling 27 (2005) 177–187
Is monetary policy important for forecasting real growth and inflation? Edward N. Gambera,∗ , David R. Hakesb b
a Department of Economics and Business, Lafayette College, Easton, PA 18042, USA Department of Economics, University of Northern Iowa, Cedar Falls, IA 50614-0129, USA
Received 1 March 2004; received in revised form 12 September 2004; accepted 15 December 2004 Available online 19 January 2005
Abstract We hypothesize that if monetary policy is important in explaining movements in output and inflation then it should follow that more accurate forecasts of monetary policy, on average, will tend to produce more accurate forecasts of growth and inflation. Using data from the Survey of Professional Forecasters we find that improved monetary policy forecast accuracy corresponds to lower variance of forecast errors for growth and inflation but very little reduction in the overall average size of forecast errors for growth and inflation. © 2005 Society for Policy Modeling. Published by Elsevier Inc. All rights reserved. Keywords: Survey of Professional Forecasters; Monetary policy; Forecast accuracy
1. Introduction The literature on measuring the importance of monetary policy is immense. Friedman and Schwartz (1963) mark the beginning of a long series of papers and books looking at the importance of monetary policy in explaining movements in real output and inflation. The techniques used to address these questions are quite varied. The narrative approach pioneered by Friedman and Schwartz has been expanded upon by other authors (see Romer and Romer, 1989, 1994). The strand of literature initiated by Anderson and Jordan (1968), ∗
Corresponding author. Tel.: +1 610 330 5310; fax: +1 610 330 5715. E-mail address: [email protected] (E.N. Gamber).
0161-8938/$ – see front matter © 2005 Society for Policy Modeling. Published by Elsevier Inc. All rights reserved. doi:10.1016/j.jpolmod.2004.12.008
178
E.N. Gamber, D.R. Hakes / Journal of Policy Modeling 27 (2005) 177–187
estimates the impact of monetary policy using reduced-form regressions. Work by Sims (1972) continued this line of research using vector autoregressive techniques. Structural approaches such as Bernanke (1986), Sims and Leeper (1994) and Sims, Leeper, and Zha (1996) are also aimed at assessing the importance of monetary policy in explaining movements in real output and inflation. The general conclusion of this literature is that monetary policy matters for explaining movements in real output (in the short run) and inflation (in the long run). The exact mechanism through which monetary policy affects the economy is still being debated. But the overall effect of monetary policy on the economy is one of the few areas of agreement among macroeconomists. This paper looks at the importance of monetary policy from the perspective of forecasting. If monetary policy is important in explaining the economy we would expect it to be important in forecasting the economy as well. In other words, the importance of monetary policy should be revealed in the forecasts of real growth and inflation. We investigate this question by testing whether more accurate forecasts of monetary policy result in more accurate forecasts of real growth and inflation. If monetary policy is important, we expect to find that lower forecast errors for monetary policy are associated with lower forecast errors for growth and inflation. The questions addressed in this paper are of particular importance to policymakers. Policymakers are consumers and producers of economic forecasts. From the perspective of a consumer of forecasts, policymakers would find it useful to know whether the overall accuracy of forecasts that they purchase depend on the accuracy of the embedded monetary policy forecasts. From the perspective of a producer of forecasts, policymakers may wish to understand how best to improve their modeling. Does it make sense to devote more resources to improve the forecast of monetary policy or is the accuracy of monetary policy forecasts irrelevant for the overall accuracy of an economic forecast? We test this hypothesis using the sample of interest rate, real growth and inflation forecasts contained in the Survey of Professional Forecaster (SFP) data set from the Federal Reserve Bank of Philadelphia.1 In each quarter, a sample of 20–50 professional forecasters report their forecasts to the Federal Reserve Bank of Philadelphia. Variables included in the forecast are real GDP (real GNP before 1992), inflation, corporate profits, unemployment rate and interest rates as well as several other variables. The variables of interest for this paper are real GDP and the 3-month T-bill rate, which we use as a proxy for the stance of monetary policy. We find that the mean and median forecasts of monetary policy are biased and inefficient. We also find that improving monetary policy forecast errors would not reduce the mean absolute forecast error for growth or inflation at three and four quarter ahead forecast horizons. We do, however, find that getting monetary policy right (zero forecast error) would substantially reduce the root means square error of the inflation and output forecast errors. Thus, it appears that the payoff for accurate monetary policy forecasts is a reduction in the variance but not the level of the forecast error for growth and inflation. 1 See http://www.phil.frb.org/econ/spf/index.html. The Survey of Professional Forecasters (SFP) was previously called the ASSA-NBER Survey of Forecasters from 1968 to 1990. The Federal Reserve Bank of Philadelphia took over the survey in 1990. See Croushore (1993) for a complete description of the SFP.
E.N. Gamber, D.R. Hakes / Journal of Policy Modeling 27 (2005) 177–187
179
This paper is organized as follows. In the following section we review the literature on using survey forecasts to measure expectations. Following that, we present a detailed analysis of the data on interest rate forecasts and we justify why it is appropriate to use forecasts of the 3-month T-bill rate as a proxy for monetary policy forecasts. Next we investigate whether forecasts of the 3-month T-bill are rational and unbiased (we find that they are neither). Finally, we look at whether monetary policy forecast errors are correlated with real growth and inflation forecast errors.
2. Previous research using the Survey of Professional Forecaster’s database Several previous authors have tested for rationality and bias in the SPF forecasts but the majority of those studies analyze the rationality and bias in inflation and real growth forecasts. There have been no studies of the rationality and bias in interest rate forecasts and no studies linking the forecast of short-term interest rates to the forecast of monetary policy. For the most part, previous authors cannot reject the hypothesis that the SFP (or its precursor the ASA-NBER2 ) forecast data are unbiased. In a series of papers Zarnowitz (1979, 1985), Zarnowitz and Braun (1993) and Zarnowitz and Lambros (1987) found that both the individual as well as the median SFP forecasts are biased and irrational. Su and Su (1975) conducted a comprehensive evaluation of the mean and median ASA/NBER forecasts. Their results were mixed. They found that the mean and median forecasts of some of the variables were biased for some forecast horizons. But they did not evaluate the biasedness of the T-bill rate because it was not included in the survey until 1981. A recent study by Ball and Croushore (2001) examined whether survey forecasts of output growth are consistent with actual output growth following a change in monetary policy. Ball and Croushore find that survey respondents consistently (and substantially) underpredict the effects of a tightening of monetary policy on output. They conclude that this violates rational expectations.
3. The 3-month T-bill rate as a proxy for the federal funds rate and monetary policy The Survey of Professional Forecasters database does not include forecasts of the federal funds interest rate. Fortunately, it does include forecasts of the 3-month T-bill rate, which as we show below, is very highly correlated with the federal funds rate. Other researchers have used the 3-month T-bill as a proxy for the federal funds interest rate (see, for example, Fair, 1984; Fuhrer and Moore, 1995). As a consequence, we only briefly review the relationship between the T-bill rate and the federal funds rate. Fig. 1 shows the quarterly averages of the 3-month T-bill and federal funds rate for our sample period (1981:4 through 2002:1). The top line is the federal funds rate. The bottom line is the 3-month T-bill rate. It is clear from the figure that movements in the 3-month
2
ASA-NBER stands for the American Statistical Association-National Bureau of Economic Research.
180
E.N. Gamber, D.R. Hakes / Journal of Policy Modeling 27 (2005) 177–187
Fig. 1. The federal funds rate and the 3-month T-bill rate. Table 1 The relationship between the 3-month T-bill rate and the federal funds rate Sample
Number of obs
Intercept a
Slope b
R2
Regression results 1983:4–2002:1 1983:4–1991:4 1992:1–2002:1
82 41 41
0.34 (.08)** 0.27 (.20)* 0.27 (.12)*
0.86 (.01)** 0.87 (.02)** 0.89 (.02)**
0.99 0.98 0.97
Sample
Number of obs
Average spread
Standard deviation of spread
Summary statistics for the spread (Tbill − federal funds) 1983:4–2002:1 82 0.56 1983:4–1991:4 41 0.83 1992:1–2002:1 41 0.29
0.44 0.43 0.23
Standard errors are in parentheses beside the parameter estimates. ∗ Significant at the 0.05 level. ∗∗ Significant at the 0.01 level.
T-bill closely mimic movements in the federal funds rate. This same information can be summarized in the following regression: Tbillt = a + b × fft
(1)
where Tbillt is the 3-month T-bill rate in quarter t and fft the average federal funds rate in quarter t.3 The regression results and the summary statistics for the spread between the 3-month T-bill and the federal funds rate are presented in Table 1. The two series are very closely related. Moreover, the relationship between the T-bill and the federal funds interest rate has become closer over the past decade. 3 Some researchers use the average interest rate during the last month of the quarter instead of the 3-month average. In this study we are constrained to using the 3-month average because the interest rate forecasts contained in the SFP are the 3-month average for the quarter.
E.N. Gamber, D.R. Hakes / Journal of Policy Modeling 27 (2005) 177–187
181
Fig. 2. Forecasted and actual real 3-month T-bill rate.
Thus, we suggest that the 3-month T-bill rate is a reasonable proxy for the federal funds rate. Moreover, work by Bernanke and Blinder (1992) and Bernanke and Mihov (1998) has shown that the federal funds rate is a reasonable measure of monetary policy (especially post-1981). Thus, we conclude that the 3-month T-bill rate is a reasonable measure of the stance of monetary policy. 4. Are forecasts of monetary policy rational and unbiased? Above, we showed that the nominal 3-month T-bill rate closely tracks the nominal federal funds rate. But in empirical studies of monetary policy and in practitioners’ forecasting models, it is the real, not the nominal interest rate that serves as a measure of monetary policy.4 Thus, for the remainder of the paper, we focus on the real 3-month T-bill rate as the measure of monetary policy and the forecast of the real 3-month T-bill rate as the measure of the forecast of monetary policy. The real 3-month T-bill rate is constructed by subtracting the annualized growth rate of the implicit price deflator (GNP deflator before 1992 and GDP deflator afterwards) from the nominal 3-month T-bill rate. The forecast of the real 3-month T-bill rate is constructed by subtracting the forecast of the annualized growth rate of the implicit price deflator from the forecast of the 3-month T-bill rate. The actual and forecasted (1–4-quarter ahead) real T-Bill rates are shown in Fig. 2. The actual real rate is the solid line in the figure. We conducted a battery of tests to determine whether the forecasted 3-month T-bill rate is a rational and unbiased estimate of the actual 3-month T-bill rate. If the forecast of monetary policy is rational, then, the probability distribution of the forecast should be the same as the actual probability distribution (this is a necessary but not sufficient condition as we explain below). As a first step we assessed whether both series have a unit root.5 We rejected the null hypothesis of a unit root at the 5% level for the actual and the forecasted real T-bill series. Thus, it appears that both the actual real 3-month T-bill as well as the median forecasted real 3-month T-bill are stationary time series. In addition we tested for bias and rationality and found that median forecasts of monetary policy are both biased and irrational. 4 For the tests conducted in the previous section it does not matter whether we use the real or nominal 3-month T-Bill and federal funds rate because we would subtract the inflation rate from both sides of Eq. (1). 5 The complete test results discussed in this section are available from the authors upon request.
182
E.N. Gamber, D.R. Hakes / Journal of Policy Modeling 27 (2005) 177–187
5. Do monetary policy forecast errors explain GDP forecast errors? In this section we test the main hypothesis of the paper by looking at whether forecast errors for GDP growth and inflation are correlated with the forecast errors for monetary policy. More specifically, are smaller monetary policy forecast errors associated with smaller errors in forecasting real growth and inflation? To illustrate this hypothesis, we present the following simple three-equation macro model: IS :
yt = −αrt−1 + εt
Phillips curve : Taylor rule :
p
πt = δ(yt − yt ) + µt ep
e e rt = γ(yt+4 − yt+4 ) + θπt+4 + ηt
where y is the log of real output, r the real interest rate and yp the potential output (which we consider exogenous to the model). The superscript ‘e’ denotes the expectation as of time t. The parameters α, δ, γ, and θ are the positive constants. The innovations ε, µ, η are zero mean white noise. The one-step ahead forecast error for real output is: ep
e e yt+1 − E[yt+1 |Ωt ] = −α{g(yt+4 − yt+4 ) + θπt+4 + ηt } ep
e e − yt+4 ) + θπt+4 + ηt } = εt+1 +εt+1 − α{γ(yt+4
Note that the monetary policy forecast error (ηt ) does not appear in this equation because there is a one period lag before monetary policy effects real output. Thus, the actual monetary policy that is to impact y at t + 1 is known at time t—there is no monetary policy forecast error. But if we look at the forecast error for real growth in t + 2 (again, based on time t information) we have: ep
e e yt+2 − E[yt+2 |Ωt ] = −α{γ(yt+5 − yt+5 ) + θπt+5 + ηt+1 } ep
e e − yt+5 ) + θπt+5 } +εt+2 − α{γ(yt+5
= −αηt+1 + εt+2 Here we see that the monetary policy forecast error does enter the equation for the real output forecast error. The IS equation above assumes that monetary policy affects real output with a one-period lag. We can generalize this result. If the lag between monetary policy and real growth is two periods then the forecast error for real growth in t + 3 is: yt+3 − E[yt+3 |Ωt ] = −αηt+1 + εt+3 And if the lag between monetary policy and real growth is three periods then the forecast error for real growth in t + 4 is: yt+4 − E[yt+4 |Ωt ] = −αηt+1 + εt+4
E.N. Gamber, D.R. Hakes / Journal of Policy Modeling 27 (2005) 177–187
183
Table 2 The relationship between monetary policy forecasts and real growth and inflation forecasts Forecast horizon
One-quarter Two-quarter Three-quarter Four-quarter ∗∗ †
Coefficient on unexpected monetary policy Real growth
R2
Inflation
R2
0.27 (0.61) 0.33 (1.70)† −0.25 (−1.23) −0.23 (−1.20)
0.19 0.17 0.16 0.19
−0.66 (−14.5)** −0.10 (−1.12) 0.03 (0.32) −0.17 (−1.64)†
0.74 0.003 0.13 0.18
Significant at the 0.01 level. Significant at the 0.10 level.
The expressions for the forecast error for inflation are similar. If inflation is affected by monetary policy actions that occurred at or before time t then the monetary policy forecast error will not affect the forecast error for inflation. Of course, in our empirical application, we do not know the exact lag time between a change in the real T-bill rate and a change in real output and inflation. We therefore consider possible lags in the effect of monetary policy by regressing the value of forecast errors for GDP growth and inflation at horizons 1–4 on the value of the one-step ahead forecast error for monetary policy. Thus, it is likely that forecasters consider the lagged effect of monetary policy actions by modeling the effect of monetary policy actions next quarter on real growth and inflation in subsequent quarters. Our regression takes the following form: e e xt+i − xt+i = a + b(rt+1 − rt+1 ) + νt ,
i = 1, . . . , 4
(2)
where x represents GDP growth and inflation. All regressions were corrected for first order serial correlation. The results are presented in Table 2. Monetary policy forecast errors are positively correlated with real growth for forecast horizons 1 and 2 and negatively correlated with real growth forecast errors at forecast horizons 3 and 4. The interpretation of the signs is as follows. For the one and two quarter ahead forecasts, if monetary policy is tighter than expected (the short-term real interest rate is higher than expected) then real growth will be higher than expected as well. For the three and four quarter horizon, if monetary policy is tighter than expected then real growth will be slower than expected. Although only one coefficient (at the two-quarter horizon) is marginally significant, the signs on the coefficients make sense given the lags associated with monetary policy. At the shorter horizons we are most likely observing some amount of reverse causation. Higher than expected output growth leads to higher than expected short-term interest rates through a Fed reaction function such as the Taylor rule. Even at the two-quarter horizon, because real growth exhibits a strong positive serial correlation, higher than expected output next quarter is likely to generate higher than expected output in the following quarter as well. Thus, the positive coefficient on the unexpected monetary policy variable most likely indicates a continued reaction to higher than expected output by monetary policymakers. The fact that the signs change at three and four quarters, however, indicates that forecasters are anticipating an effect of monetary policy on the economy at these longer horizons. An unexpectedly high short-term real interest rate next quarter will lead to unexpectedly low real growth three and four quarters in the future.
184
E.N. Gamber, D.R. Hakes / Journal of Policy Modeling 27 (2005) 177–187
The results for inflation are presented in the right hand side columns of Table 2. The coefficient on the one-quarter ahead forecast is negative and highly significant. It is also much larger than the rest of the coefficients in the table. The reason for the sign, size and significance of this coefficient is that the Fed policymakers do not react one-for-one to unexpected changes in inflation with the same quarter. As shown by Clarida, Gali, and Gertler (2000) the Fed smoothes its reaction to inflation and output growth innovations over several quarters. Because nominal interest rates do not react to one-for-one with innovations in inflation, innovations in inflation cause the real interest rate to fall. Thus, unexpectedly high inflation leads to unexpectedly low real interest rates at the one-quarter horizon. The negative coefficient at the four-quarter horizon is consistent with lower-than expected real interest rates causing higher-than-expected inflation although the coefficient is only marginally significant. The coefficients shown in Table 2 indicate the direction of correlation and whether the correlation is significant. But they do not answer the question of whether smaller monetary policy forecast errors are associated with smaller growth and inflation forecast errors. The adjusted R-squares give us some idea of the importance of monetary policy forecast errors in explaining growth and inflation forecast errors. The adjusted R-squares range from 16 to 19% for the real growth forecasts and 3–74% for the inflation forecasts. Because the variables in the regressions are forecast errors, the adjusted R-squares tell us the percent of the forecast variance of growth (or inflation) that is explained by the variance of the monetary policy forecast error. Another way to assess the importance of monetary policy in explaining growth and inflation forecasts is to look at the reduction in the “size” of the growth and inflation forecast errors that would result from forecasting monetary policy perfectly, with zero error. To measure this effect we set the monetary policy forecast error to zero and use the estimated intercept and residuals from Eq. (2) to simulate the resulting growth and inflation forecast errors. We then compared the size of those forecast errors with the size of the actual
Table 3 Reduction in real growth and inflation forecast error when monetary policy forecast error is zero Mean absolute error
Root mean square error
Actual
Simulated
Percentage difference
Actual
Simulated
Percentage difference
Real growth One-quarter Two-quarter Three-quarter Four-quarter
1.87 1.99 1.90 1.90
1.18 1.93 1.95 1.94
−3.74 −3.04 2.64 2.11
1.47 1.58 1.58 1.57
1.39 1.48 1.56 1.59
−4.8 −6.1 −1.03 1.29
Inflation One-quarter Two-quarter Three-quarter Four-quarter
0.92 1.04 1.15 1.23
0.48 1.01 1.16 1.19
−48.6 −2.9 0.43 −3.05
0.59 0.73 0.78 0.94
0.39 0.72 0.79 0.92
−34.57 −0.3 0.88 −2.08
Percentage difference = 100 × (simulated − actual)/actual.
E.N. Gamber, D.R. Hakes / Journal of Policy Modeling 27 (2005) 177–187
185
forecast errors where size is defined by the absolute forecast error as well as the root mean square of the forecast error. The results are presented in Table 3. Aside from the quirky result for inflation at the one-quarter horizon, the reductions in the size of the forecast error (in absolute or variance terms) are quite small. These results suggest that (aside from the one-quarter ahead inflation forecast) getting monetary policy right would not improve the median forecast of real output by much.
6. Panel results One potential pitfall with using the median forecast in this test is that it covers up a lot of individual information on the relationship between monetary policy forecast errors and both real growth and inflation forecast errors. We therefore also estimate a (slightly modified) version of Eq. (2) using the individual forecaster’s data from the SFP. The individual forecasts in the SFP do not constitute a balanced panel data set. The number of forecasters reporting interest rate, GDP (GNP before 1992) and deflator forecasts each quarter varies from a low of 8 to a high of 48. Moreover, the actual forecasters who participate varies quarter to quarter. A total of 179 different forecasters reported interest rate and GDP forecasts during the 81 quarters in our sample. The minimum number of times a forecaster participated was 1. The maximum number of times was 57, the mean number of times was 22 and the median number of times was 26. Using the pooled data on individual forecasts of real growth, inflation and the real shortterm interest rate we estimated the following regression: e e xjt+i − xt+i = b(rjt − rjt ) + cj + dt + νt ,
i = 1, . . . , 4
(3)
where x is either inflation or real growth. The matrix j is a 179 × 179 matrix of forecasterspecific dummy variables and the matrix t is an 81 × 81 matrix of time period dummy variables. These two sets of dummy variables are included to hold constant time period and forecaster-specific effects. The results are presented in Table 4. The one and two quarter ahead growth forecast errors are positively related to the one quarter ahead monetary policy forecast error. This sign is consistent with a reverse causality effect—monetary policy is reacting to the economy. The three and four quarter ahead results have the correct sign but are insignificant. The inflation results for the one-quarter horizon again probably picks up the effect of inflation Table 4 The relationship between monetary policy forecasts and real growth and inflation forecasts Forecast horizon
One-quarter Two-quarter Three-quarter Four-quarter ∗ ∗∗
Coefficient on unexpected monetary policy Real growth
R2
Inflation
R2
0.38 (9.21)** 0.37 (8.52)** −0.04 (−0.87) 0.03 (0.04)
0.10 0.13 0.04 0.08
−0.71 (−70.1)** 0.03(1.49) −0.07 (−3.03)** −0.05 (−2.47)*
0.79 0.10 0.14 0.13
Significant at the 0.05 level. Significant at the 0.01 level.
186
E.N. Gamber, D.R. Hakes / Journal of Policy Modeling 27 (2005) 177–187
Table 5 Reduction in real growth and inflation forecast error when monetary policy forecast error is zero Mean absolute error
Root mean square error
Actual
Simulated
Percentage difference
Actual
Simulated
Percentage difference
Real growth One-quarter Two-quarter Three-quarter Four-quarter
2.14 2.22 2.19 2.15
2.07 2.16 2.20 2.14
−3.53 −2.74 0.26 −0.031
2.84 3.01 3.14 2.83
1.89 2.14 2.28 1.91
−33.6 −30.5 −27.4 −32.4
Inflation One-quarter Two-quarter Three-quarter Four-quarter
1.08 1.21 1.29 1.40
0.55 1.23 1.27 1.38
1.45 1.48 1.53 1.50
0.53 1.18 1.24 1.16
−63.4 −20.4 −18.7 −22.8
−49.06 1.10 −1.45 −1.31
Percentage difference = 100 × (simulated − actual)/actual.
on the short-term real interest rate. The coefficients at the three and four quarter horizon are consistent with our story—lower than expected real interest rates next quarter lead to higher than expected inflation three and four quarters in the future. Once again we simulated the economy assuming that forecasters made no forecast error in forecasting monetary policy. The root mean square errors and the mean absolute errors with and without this assumption are presented in Table 5. For the most part, getting monetary policy right does not have a large impact on the mean absolute error. The exception of course is the one-quarter horizon for inflation but because the inflation rate moves the real interest rate almost one-for-one within the quarter, the large reduction in forecast error is simply saying that if we forecasted inflation right we would forecast inflation right! The percentage reductions in the root mean square errors are much larger. These results suggest that getting monetary policy right significantly reduces the variation in the real growth and inflation forecast errors.
7. Conclusion Overall monetary policy forecast errors do not seem to matter very much for forecasting growth and inflation. Smaller monetary policy forecast errors do not imply significantly smaller growth and inflation forecast errors. Smaller monetary policy forecast errors do, however, imply smaller variance of growth and inflation forecast errors. Thus, we conclude that the evidence is somewhat mixed on whether getting monetary policy right matters for forecasting real growth and inflation. It does not seem to matter much in terms of the absolute size of the forecast errors. But it does appear to matter in terms of the reduction in the variance of growth and inflation forecast errors. These results have interesting implications for policymakers. Policymakers clearly must face a tradeoff—getting monetary policy right reduces the variance of forecast errors at the
E.N. Gamber, D.R. Hakes / Journal of Policy Modeling 27 (2005) 177–187
187
expense of overall higher average errors. The choice of whether to focus on monetary policy accuracy in either the consumption or production of forecasts may therefore depend on the degree of risk aversion on the part of the policymaker.
Acknowledgement We thank Fred Joutz for helpful comments and suggestions. All remaining errors are our own.
References Anderson, L. C., & Jordan, J. L. (1968). Monetary and fiscal actions: A test of their relative importance in economic stabilization. Review. Federal Reserve Bank of St. Louis. Ball, L., & Croushore, D. (2001). Expectations and the effects of monetary policy (Working Paper No. 0112). Federal Reserve Bank of Philadelphia. Bernanke, B.S. (1986). Alternative explanations of the money-income correlation. In Proceedings of the CarnegieRochester Conference Series on Public Policy (Vol. 25, pp. 49–99). Bernanke, A., & Blinder. (1992). The federal funds rate and the channels of monetary transmission. American Economic Review, 82(4), 901–921. Bernanke, I., & Mihov. (1998). Measuring monetary policy quarterly. Journal of Economics, 113(3), 869–902. Clarida, Gali, R. J., & Gertler, M. (2000). Monetary policy rules and macroeconomic stability: Evidence and some theory. Quarterly Journal of Economics, 115(1), 147–180. Croushore, D. (1993). Introducing: The Survey of Professional Forecasters, Business Review (pp. 3–13). Federal Reserve Bank of Philadelphia. Fair, R. (1984). Specification, estimation, and analysis of macroeconometric models. Harvard University Press. Friedman, M., & Schwartz, A. J. (1963). A monetary history of the United States, 1867–1960. Princeton, NJ: Princeton University Press. Fuhrer, J. C., & Moore, G. R. (1995). Monetary policy trade-offs and the correlation between nominal interest rates and real output. American Economic Review, 85(1), 219–239. Romer, C., & Romer, D. (1989). Does monetary policy matter? A new test in the spirit of Friedman and Schwartz. In NBER macroeconomics annual (pp. 121–170) MIT Press. Romer, C., & Romer, D. (1994). Monetary policy matters. Journal of Monetary Economics, 34(1), 75–88. Sims, C. A. (1972). Money, income and causality. American Economic Review, 62(4), 540–552. Sims, C. A., & Leeper, E. M. (1994). Toward a modern macroeconomic model usable for policy analysis. In NBER macroeconomics annual (pp. 81–118) MIT Press. Sims, C. A., Leeper, E. M., & Zha, T. (1996). What does monetary policy do? Brookings Papers on Economic Activity, 2, 1–63. Su, V., & Su, J. (1975). An evaluation of ASA/NBER business outlook survey forecasts. Explorations in Economic Research, 2, 588–618. Zarnowitz, V. (1979). An analysis of annual and multiperiod quarterly forecasts of aggregate income, output and the price level. Journal of Business, 52, 1–33. Zarnowitz, V. (1985). Rational expectations and macroeconomic forecasts. Journal of Business and Economic Statistics, 3, 293–311. Zarnowitz, V., & Braun, P. (1993). Twenty-two years of the NBER-ASA quarterly economic outlook surveys: Aspects and comparisons of forecasting performance. In J. H. Stock & M. Watson (Eds.), Business cycles, indicators, and forecasting (pp. 11–94). Chicago: University of Chicago Press. Zarnowitz, V., & Lambros, L. A. (1987). Consensus and uncertainty in economic prediction. Journal of Political Economy, 95, 591–621.
Is monetary policy important for forecasting real growth and inflation? Edward N. Gambera,∗ , David R. Hakesb b
a Department of Economics and Business, Lafayette College, Easton, PA 18042, USA Department of Economics, University of Northern Iowa, Cedar Falls, IA 50614-0129, USA
Received 1 March 2004; received in revised form 12 September 2004; accepted 15 December 2004 Available online 19 January 2005
Abstract We hypothesize that if monetary policy is important in explaining movements in output and inflation then it should follow that more accurate forecasts of monetary policy, on average, will tend to produce more accurate forecasts of growth and inflation. Using data from the Survey of Professional Forecasters we find that improved monetary policy forecast accuracy corresponds to lower variance of forecast errors for growth and inflation but very little reduction in the overall average size of forecast errors for growth and inflation. © 2005 Society for Policy Modeling. Published by Elsevier Inc. All rights reserved. Keywords: Survey of Professional Forecasters; Monetary policy; Forecast accuracy
1. Introduction The literature on measuring the importance of monetary policy is immense. Friedman and Schwartz (1963) mark the beginning of a long series of papers and books looking at the importance of monetary policy in explaining movements in real output and inflation. The techniques used to address these questions are quite varied. The narrative approach pioneered by Friedman and Schwartz has been expanded upon by other authors (see Romer and Romer, 1989, 1994). The strand of literature initiated by Anderson and Jordan (1968), ∗
Corresponding author. Tel.: +1 610 330 5310; fax: +1 610 330 5715. E-mail address: [email protected] (E.N. Gamber).
0161-8938/$ – see front matter © 2005 Society for Policy Modeling. Published by Elsevier Inc. All rights reserved. doi:10.1016/j.jpolmod.2004.12.008
178
E.N. Gamber, D.R. Hakes / Journal of Policy Modeling 27 (2005) 177–187
estimates the impact of monetary policy using reduced-form regressions. Work by Sims (1972) continued this line of research using vector autoregressive techniques. Structural approaches such as Bernanke (1986), Sims and Leeper (1994) and Sims, Leeper, and Zha (1996) are also aimed at assessing the importance of monetary policy in explaining movements in real output and inflation. The general conclusion of this literature is that monetary policy matters for explaining movements in real output (in the short run) and inflation (in the long run). The exact mechanism through which monetary policy affects the economy is still being debated. But the overall effect of monetary policy on the economy is one of the few areas of agreement among macroeconomists. This paper looks at the importance of monetary policy from the perspective of forecasting. If monetary policy is important in explaining the economy we would expect it to be important in forecasting the economy as well. In other words, the importance of monetary policy should be revealed in the forecasts of real growth and inflation. We investigate this question by testing whether more accurate forecasts of monetary policy result in more accurate forecasts of real growth and inflation. If monetary policy is important, we expect to find that lower forecast errors for monetary policy are associated with lower forecast errors for growth and inflation. The questions addressed in this paper are of particular importance to policymakers. Policymakers are consumers and producers of economic forecasts. From the perspective of a consumer of forecasts, policymakers would find it useful to know whether the overall accuracy of forecasts that they purchase depend on the accuracy of the embedded monetary policy forecasts. From the perspective of a producer of forecasts, policymakers may wish to understand how best to improve their modeling. Does it make sense to devote more resources to improve the forecast of monetary policy or is the accuracy of monetary policy forecasts irrelevant for the overall accuracy of an economic forecast? We test this hypothesis using the sample of interest rate, real growth and inflation forecasts contained in the Survey of Professional Forecaster (SFP) data set from the Federal Reserve Bank of Philadelphia.1 In each quarter, a sample of 20–50 professional forecasters report their forecasts to the Federal Reserve Bank of Philadelphia. Variables included in the forecast are real GDP (real GNP before 1992), inflation, corporate profits, unemployment rate and interest rates as well as several other variables. The variables of interest for this paper are real GDP and the 3-month T-bill rate, which we use as a proxy for the stance of monetary policy. We find that the mean and median forecasts of monetary policy are biased and inefficient. We also find that improving monetary policy forecast errors would not reduce the mean absolute forecast error for growth or inflation at three and four quarter ahead forecast horizons. We do, however, find that getting monetary policy right (zero forecast error) would substantially reduce the root means square error of the inflation and output forecast errors. Thus, it appears that the payoff for accurate monetary policy forecasts is a reduction in the variance but not the level of the forecast error for growth and inflation. 1 See http://www.phil.frb.org/econ/spf/index.html. The Survey of Professional Forecasters (SFP) was previously called the ASSA-NBER Survey of Forecasters from 1968 to 1990. The Federal Reserve Bank of Philadelphia took over the survey in 1990. See Croushore (1993) for a complete description of the SFP.
E.N. Gamber, D.R. Hakes / Journal of Policy Modeling 27 (2005) 177–187
179
This paper is organized as follows. In the following section we review the literature on using survey forecasts to measure expectations. Following that, we present a detailed analysis of the data on interest rate forecasts and we justify why it is appropriate to use forecasts of the 3-month T-bill rate as a proxy for monetary policy forecasts. Next we investigate whether forecasts of the 3-month T-bill are rational and unbiased (we find that they are neither). Finally, we look at whether monetary policy forecast errors are correlated with real growth and inflation forecast errors.
2. Previous research using the Survey of Professional Forecaster’s database Several previous authors have tested for rationality and bias in the SPF forecasts but the majority of those studies analyze the rationality and bias in inflation and real growth forecasts. There have been no studies of the rationality and bias in interest rate forecasts and no studies linking the forecast of short-term interest rates to the forecast of monetary policy. For the most part, previous authors cannot reject the hypothesis that the SFP (or its precursor the ASA-NBER2 ) forecast data are unbiased. In a series of papers Zarnowitz (1979, 1985), Zarnowitz and Braun (1993) and Zarnowitz and Lambros (1987) found that both the individual as well as the median SFP forecasts are biased and irrational. Su and Su (1975) conducted a comprehensive evaluation of the mean and median ASA/NBER forecasts. Their results were mixed. They found that the mean and median forecasts of some of the variables were biased for some forecast horizons. But they did not evaluate the biasedness of the T-bill rate because it was not included in the survey until 1981. A recent study by Ball and Croushore (2001) examined whether survey forecasts of output growth are consistent with actual output growth following a change in monetary policy. Ball and Croushore find that survey respondents consistently (and substantially) underpredict the effects of a tightening of monetary policy on output. They conclude that this violates rational expectations.
3. The 3-month T-bill rate as a proxy for the federal funds rate and monetary policy The Survey of Professional Forecasters database does not include forecasts of the federal funds interest rate. Fortunately, it does include forecasts of the 3-month T-bill rate, which as we show below, is very highly correlated with the federal funds rate. Other researchers have used the 3-month T-bill as a proxy for the federal funds interest rate (see, for example, Fair, 1984; Fuhrer and Moore, 1995). As a consequence, we only briefly review the relationship between the T-bill rate and the federal funds rate. Fig. 1 shows the quarterly averages of the 3-month T-bill and federal funds rate for our sample period (1981:4 through 2002:1). The top line is the federal funds rate. The bottom line is the 3-month T-bill rate. It is clear from the figure that movements in the 3-month
2
ASA-NBER stands for the American Statistical Association-National Bureau of Economic Research.
180
E.N. Gamber, D.R. Hakes / Journal of Policy Modeling 27 (2005) 177–187
Fig. 1. The federal funds rate and the 3-month T-bill rate. Table 1 The relationship between the 3-month T-bill rate and the federal funds rate Sample
Number of obs
Intercept a
Slope b
R2
Regression results 1983:4–2002:1 1983:4–1991:4 1992:1–2002:1
82 41 41
0.34 (.08)** 0.27 (.20)* 0.27 (.12)*
0.86 (.01)** 0.87 (.02)** 0.89 (.02)**
0.99 0.98 0.97
Sample
Number of obs
Average spread
Standard deviation of spread
Summary statistics for the spread (Tbill − federal funds) 1983:4–2002:1 82 0.56 1983:4–1991:4 41 0.83 1992:1–2002:1 41 0.29
0.44 0.43 0.23
Standard errors are in parentheses beside the parameter estimates. ∗ Significant at the 0.05 level. ∗∗ Significant at the 0.01 level.
T-bill closely mimic movements in the federal funds rate. This same information can be summarized in the following regression: Tbillt = a + b × fft
(1)
where Tbillt is the 3-month T-bill rate in quarter t and fft the average federal funds rate in quarter t.3 The regression results and the summary statistics for the spread between the 3-month T-bill and the federal funds rate are presented in Table 1. The two series are very closely related. Moreover, the relationship between the T-bill and the federal funds interest rate has become closer over the past decade. 3 Some researchers use the average interest rate during the last month of the quarter instead of the 3-month average. In this study we are constrained to using the 3-month average because the interest rate forecasts contained in the SFP are the 3-month average for the quarter.
E.N. Gamber, D.R. Hakes / Journal of Policy Modeling 27 (2005) 177–187
181
Fig. 2. Forecasted and actual real 3-month T-bill rate.
Thus, we suggest that the 3-month T-bill rate is a reasonable proxy for the federal funds rate. Moreover, work by Bernanke and Blinder (1992) and Bernanke and Mihov (1998) has shown that the federal funds rate is a reasonable measure of monetary policy (especially post-1981). Thus, we conclude that the 3-month T-bill rate is a reasonable measure of the stance of monetary policy. 4. Are forecasts of monetary policy rational and unbiased? Above, we showed that the nominal 3-month T-bill rate closely tracks the nominal federal funds rate. But in empirical studies of monetary policy and in practitioners’ forecasting models, it is the real, not the nominal interest rate that serves as a measure of monetary policy.4 Thus, for the remainder of the paper, we focus on the real 3-month T-bill rate as the measure of monetary policy and the forecast of the real 3-month T-bill rate as the measure of the forecast of monetary policy. The real 3-month T-bill rate is constructed by subtracting the annualized growth rate of the implicit price deflator (GNP deflator before 1992 and GDP deflator afterwards) from the nominal 3-month T-bill rate. The forecast of the real 3-month T-bill rate is constructed by subtracting the forecast of the annualized growth rate of the implicit price deflator from the forecast of the 3-month T-bill rate. The actual and forecasted (1–4-quarter ahead) real T-Bill rates are shown in Fig. 2. The actual real rate is the solid line in the figure. We conducted a battery of tests to determine whether the forecasted 3-month T-bill rate is a rational and unbiased estimate of the actual 3-month T-bill rate. If the forecast of monetary policy is rational, then, the probability distribution of the forecast should be the same as the actual probability distribution (this is a necessary but not sufficient condition as we explain below). As a first step we assessed whether both series have a unit root.5 We rejected the null hypothesis of a unit root at the 5% level for the actual and the forecasted real T-bill series. Thus, it appears that both the actual real 3-month T-bill as well as the median forecasted real 3-month T-bill are stationary time series. In addition we tested for bias and rationality and found that median forecasts of monetary policy are both biased and irrational. 4 For the tests conducted in the previous section it does not matter whether we use the real or nominal 3-month T-Bill and federal funds rate because we would subtract the inflation rate from both sides of Eq. (1). 5 The complete test results discussed in this section are available from the authors upon request.
182
E.N. Gamber, D.R. Hakes / Journal of Policy Modeling 27 (2005) 177–187
5. Do monetary policy forecast errors explain GDP forecast errors? In this section we test the main hypothesis of the paper by looking at whether forecast errors for GDP growth and inflation are correlated with the forecast errors for monetary policy. More specifically, are smaller monetary policy forecast errors associated with smaller errors in forecasting real growth and inflation? To illustrate this hypothesis, we present the following simple three-equation macro model: IS :
yt = −αrt−1 + εt
Phillips curve : Taylor rule :
p
πt = δ(yt − yt ) + µt ep
e e rt = γ(yt+4 − yt+4 ) + θπt+4 + ηt
where y is the log of real output, r the real interest rate and yp the potential output (which we consider exogenous to the model). The superscript ‘e’ denotes the expectation as of time t. The parameters α, δ, γ, and θ are the positive constants. The innovations ε, µ, η are zero mean white noise. The one-step ahead forecast error for real output is: ep
e e yt+1 − E[yt+1 |Ωt ] = −α{g(yt+4 − yt+4 ) + θπt+4 + ηt } ep
e e − yt+4 ) + θπt+4 + ηt } = εt+1 +εt+1 − α{γ(yt+4
Note that the monetary policy forecast error (ηt ) does not appear in this equation because there is a one period lag before monetary policy effects real output. Thus, the actual monetary policy that is to impact y at t + 1 is known at time t—there is no monetary policy forecast error. But if we look at the forecast error for real growth in t + 2 (again, based on time t information) we have: ep
e e yt+2 − E[yt+2 |Ωt ] = −α{γ(yt+5 − yt+5 ) + θπt+5 + ηt+1 } ep
e e − yt+5 ) + θπt+5 } +εt+2 − α{γ(yt+5
= −αηt+1 + εt+2 Here we see that the monetary policy forecast error does enter the equation for the real output forecast error. The IS equation above assumes that monetary policy affects real output with a one-period lag. We can generalize this result. If the lag between monetary policy and real growth is two periods then the forecast error for real growth in t + 3 is: yt+3 − E[yt+3 |Ωt ] = −αηt+1 + εt+3 And if the lag between monetary policy and real growth is three periods then the forecast error for real growth in t + 4 is: yt+4 − E[yt+4 |Ωt ] = −αηt+1 + εt+4
E.N. Gamber, D.R. Hakes / Journal of Policy Modeling 27 (2005) 177–187
183
Table 2 The relationship between monetary policy forecasts and real growth and inflation forecasts Forecast horizon
One-quarter Two-quarter Three-quarter Four-quarter ∗∗ †
Coefficient on unexpected monetary policy Real growth
R2
Inflation
R2
0.27 (0.61) 0.33 (1.70)† −0.25 (−1.23) −0.23 (−1.20)
0.19 0.17 0.16 0.19
−0.66 (−14.5)** −0.10 (−1.12) 0.03 (0.32) −0.17 (−1.64)†
0.74 0.003 0.13 0.18
Significant at the 0.01 level. Significant at the 0.10 level.
The expressions for the forecast error for inflation are similar. If inflation is affected by monetary policy actions that occurred at or before time t then the monetary policy forecast error will not affect the forecast error for inflation. Of course, in our empirical application, we do not know the exact lag time between a change in the real T-bill rate and a change in real output and inflation. We therefore consider possible lags in the effect of monetary policy by regressing the value of forecast errors for GDP growth and inflation at horizons 1–4 on the value of the one-step ahead forecast error for monetary policy. Thus, it is likely that forecasters consider the lagged effect of monetary policy actions by modeling the effect of monetary policy actions next quarter on real growth and inflation in subsequent quarters. Our regression takes the following form: e e xt+i − xt+i = a + b(rt+1 − rt+1 ) + νt ,
i = 1, . . . , 4
(2)
where x represents GDP growth and inflation. All regressions were corrected for first order serial correlation. The results are presented in Table 2. Monetary policy forecast errors are positively correlated with real growth for forecast horizons 1 and 2 and negatively correlated with real growth forecast errors at forecast horizons 3 and 4. The interpretation of the signs is as follows. For the one and two quarter ahead forecasts, if monetary policy is tighter than expected (the short-term real interest rate is higher than expected) then real growth will be higher than expected as well. For the three and four quarter horizon, if monetary policy is tighter than expected then real growth will be slower than expected. Although only one coefficient (at the two-quarter horizon) is marginally significant, the signs on the coefficients make sense given the lags associated with monetary policy. At the shorter horizons we are most likely observing some amount of reverse causation. Higher than expected output growth leads to higher than expected short-term interest rates through a Fed reaction function such as the Taylor rule. Even at the two-quarter horizon, because real growth exhibits a strong positive serial correlation, higher than expected output next quarter is likely to generate higher than expected output in the following quarter as well. Thus, the positive coefficient on the unexpected monetary policy variable most likely indicates a continued reaction to higher than expected output by monetary policymakers. The fact that the signs change at three and four quarters, however, indicates that forecasters are anticipating an effect of monetary policy on the economy at these longer horizons. An unexpectedly high short-term real interest rate next quarter will lead to unexpectedly low real growth three and four quarters in the future.
184
E.N. Gamber, D.R. Hakes / Journal of Policy Modeling 27 (2005) 177–187
The results for inflation are presented in the right hand side columns of Table 2. The coefficient on the one-quarter ahead forecast is negative and highly significant. It is also much larger than the rest of the coefficients in the table. The reason for the sign, size and significance of this coefficient is that the Fed policymakers do not react one-for-one to unexpected changes in inflation with the same quarter. As shown by Clarida, Gali, and Gertler (2000) the Fed smoothes its reaction to inflation and output growth innovations over several quarters. Because nominal interest rates do not react to one-for-one with innovations in inflation, innovations in inflation cause the real interest rate to fall. Thus, unexpectedly high inflation leads to unexpectedly low real interest rates at the one-quarter horizon. The negative coefficient at the four-quarter horizon is consistent with lower-than expected real interest rates causing higher-than-expected inflation although the coefficient is only marginally significant. The coefficients shown in Table 2 indicate the direction of correlation and whether the correlation is significant. But they do not answer the question of whether smaller monetary policy forecast errors are associated with smaller growth and inflation forecast errors. The adjusted R-squares give us some idea of the importance of monetary policy forecast errors in explaining growth and inflation forecast errors. The adjusted R-squares range from 16 to 19% for the real growth forecasts and 3–74% for the inflation forecasts. Because the variables in the regressions are forecast errors, the adjusted R-squares tell us the percent of the forecast variance of growth (or inflation) that is explained by the variance of the monetary policy forecast error. Another way to assess the importance of monetary policy in explaining growth and inflation forecasts is to look at the reduction in the “size” of the growth and inflation forecast errors that would result from forecasting monetary policy perfectly, with zero error. To measure this effect we set the monetary policy forecast error to zero and use the estimated intercept and residuals from Eq. (2) to simulate the resulting growth and inflation forecast errors. We then compared the size of those forecast errors with the size of the actual
Table 3 Reduction in real growth and inflation forecast error when monetary policy forecast error is zero Mean absolute error
Root mean square error
Actual
Simulated
Percentage difference
Actual
Simulated
Percentage difference
Real growth One-quarter Two-quarter Three-quarter Four-quarter
1.87 1.99 1.90 1.90
1.18 1.93 1.95 1.94
−3.74 −3.04 2.64 2.11
1.47 1.58 1.58 1.57
1.39 1.48 1.56 1.59
−4.8 −6.1 −1.03 1.29
Inflation One-quarter Two-quarter Three-quarter Four-quarter
0.92 1.04 1.15 1.23
0.48 1.01 1.16 1.19
−48.6 −2.9 0.43 −3.05
0.59 0.73 0.78 0.94
0.39 0.72 0.79 0.92
−34.57 −0.3 0.88 −2.08
Percentage difference = 100 × (simulated − actual)/actual.
E.N. Gamber, D.R. Hakes / Journal of Policy Modeling 27 (2005) 177–187
185
forecast errors where size is defined by the absolute forecast error as well as the root mean square of the forecast error. The results are presented in Table 3. Aside from the quirky result for inflation at the one-quarter horizon, the reductions in the size of the forecast error (in absolute or variance terms) are quite small. These results suggest that (aside from the one-quarter ahead inflation forecast) getting monetary policy right would not improve the median forecast of real output by much.
6. Panel results One potential pitfall with using the median forecast in this test is that it covers up a lot of individual information on the relationship between monetary policy forecast errors and both real growth and inflation forecast errors. We therefore also estimate a (slightly modified) version of Eq. (2) using the individual forecaster’s data from the SFP. The individual forecasts in the SFP do not constitute a balanced panel data set. The number of forecasters reporting interest rate, GDP (GNP before 1992) and deflator forecasts each quarter varies from a low of 8 to a high of 48. Moreover, the actual forecasters who participate varies quarter to quarter. A total of 179 different forecasters reported interest rate and GDP forecasts during the 81 quarters in our sample. The minimum number of times a forecaster participated was 1. The maximum number of times was 57, the mean number of times was 22 and the median number of times was 26. Using the pooled data on individual forecasts of real growth, inflation and the real shortterm interest rate we estimated the following regression: e e xjt+i − xt+i = b(rjt − rjt ) + cj + dt + νt ,
i = 1, . . . , 4
(3)
where x is either inflation or real growth. The matrix j is a 179 × 179 matrix of forecasterspecific dummy variables and the matrix t is an 81 × 81 matrix of time period dummy variables. These two sets of dummy variables are included to hold constant time period and forecaster-specific effects. The results are presented in Table 4. The one and two quarter ahead growth forecast errors are positively related to the one quarter ahead monetary policy forecast error. This sign is consistent with a reverse causality effect—monetary policy is reacting to the economy. The three and four quarter ahead results have the correct sign but are insignificant. The inflation results for the one-quarter horizon again probably picks up the effect of inflation Table 4 The relationship between monetary policy forecasts and real growth and inflation forecasts Forecast horizon
One-quarter Two-quarter Three-quarter Four-quarter ∗ ∗∗
Coefficient on unexpected monetary policy Real growth
R2
Inflation
R2
0.38 (9.21)** 0.37 (8.52)** −0.04 (−0.87) 0.03 (0.04)
0.10 0.13 0.04 0.08
−0.71 (−70.1)** 0.03(1.49) −0.07 (−3.03)** −0.05 (−2.47)*
0.79 0.10 0.14 0.13
Significant at the 0.05 level. Significant at the 0.01 level.
186
E.N. Gamber, D.R. Hakes / Journal of Policy Modeling 27 (2005) 177–187
Table 5 Reduction in real growth and inflation forecast error when monetary policy forecast error is zero Mean absolute error
Root mean square error
Actual
Simulated
Percentage difference
Actual
Simulated
Percentage difference
Real growth One-quarter Two-quarter Three-quarter Four-quarter
2.14 2.22 2.19 2.15
2.07 2.16 2.20 2.14
−3.53 −2.74 0.26 −0.031
2.84 3.01 3.14 2.83
1.89 2.14 2.28 1.91
−33.6 −30.5 −27.4 −32.4
Inflation One-quarter Two-quarter Three-quarter Four-quarter
1.08 1.21 1.29 1.40
0.55 1.23 1.27 1.38
1.45 1.48 1.53 1.50
0.53 1.18 1.24 1.16
−63.4 −20.4 −18.7 −22.8
−49.06 1.10 −1.45 −1.31
Percentage difference = 100 × (simulated − actual)/actual.
on the short-term real interest rate. The coefficients at the three and four quarter horizon are consistent with our story—lower than expected real interest rates next quarter lead to higher than expected inflation three and four quarters in the future. Once again we simulated the economy assuming that forecasters made no forecast error in forecasting monetary policy. The root mean square errors and the mean absolute errors with and without this assumption are presented in Table 5. For the most part, getting monetary policy right does not have a large impact on the mean absolute error. The exception of course is the one-quarter horizon for inflation but because the inflation rate moves the real interest rate almost one-for-one within the quarter, the large reduction in forecast error is simply saying that if we forecasted inflation right we would forecast inflation right! The percentage reductions in the root mean square errors are much larger. These results suggest that getting monetary policy right significantly reduces the variation in the real growth and inflation forecast errors.
7. Conclusion Overall monetary policy forecast errors do not seem to matter very much for forecasting growth and inflation. Smaller monetary policy forecast errors do not imply significantly smaller growth and inflation forecast errors. Smaller monetary policy forecast errors do, however, imply smaller variance of growth and inflation forecast errors. Thus, we conclude that the evidence is somewhat mixed on whether getting monetary policy right matters for forecasting real growth and inflation. It does not seem to matter much in terms of the absolute size of the forecast errors. But it does appear to matter in terms of the reduction in the variance of growth and inflation forecast errors. These results have interesting implications for policymakers. Policymakers clearly must face a tradeoff—getting monetary policy right reduces the variance of forecast errors at the
E.N. Gamber, D.R. Hakes / Journal of Policy Modeling 27 (2005) 177–187
187
expense of overall higher average errors. The choice of whether to focus on monetary policy accuracy in either the consumption or production of forecasts may therefore depend on the degree of risk aversion on the part of the policymaker.
Acknowledgement We thank Fred Joutz for helpful comments and suggestions. All remaining errors are our own.
References Anderson, L. C., & Jordan, J. L. (1968). Monetary and fiscal actions: A test of their relative importance in economic stabilization. Review. Federal Reserve Bank of St. Louis. Ball, L., & Croushore, D. (2001). Expectations and the effects of monetary policy (Working Paper No. 0112). Federal Reserve Bank of Philadelphia. Bernanke, B.S. (1986). Alternative explanations of the money-income correlation. In Proceedings of the CarnegieRochester Conference Series on Public Policy (Vol. 25, pp. 49–99). Bernanke, A., & Blinder. (1992). The federal funds rate and the channels of monetary transmission. American Economic Review, 82(4), 901–921. Bernanke, I., & Mihov. (1998). Measuring monetary policy quarterly. Journal of Economics, 113(3), 869–902. Clarida, Gali, R. J., & Gertler, M. (2000). Monetary policy rules and macroeconomic stability: Evidence and some theory. Quarterly Journal of Economics, 115(1), 147–180. Croushore, D. (1993). Introducing: The Survey of Professional Forecasters, Business Review (pp. 3–13). Federal Reserve Bank of Philadelphia. Fair, R. (1984). Specification, estimation, and analysis of macroeconometric models. Harvard University Press. Friedman, M., & Schwartz, A. J. (1963). A monetary history of the United States, 1867–1960. Princeton, NJ: Princeton University Press. Fuhrer, J. C., & Moore, G. R. (1995). Monetary policy trade-offs and the correlation between nominal interest rates and real output. American Economic Review, 85(1), 219–239. Romer, C., & Romer, D. (1989). Does monetary policy matter? A new test in the spirit of Friedman and Schwartz. In NBER macroeconomics annual (pp. 121–170) MIT Press. Romer, C., & Romer, D. (1994). Monetary policy matters. Journal of Monetary Economics, 34(1), 75–88. Sims, C. A. (1972). Money, income and causality. American Economic Review, 62(4), 540–552. Sims, C. A., & Leeper, E. M. (1994). Toward a modern macroeconomic model usable for policy analysis. In NBER macroeconomics annual (pp. 81–118) MIT Press. Sims, C. A., Leeper, E. M., & Zha, T. (1996). What does monetary policy do? Brookings Papers on Economic Activity, 2, 1–63. Su, V., & Su, J. (1975). An evaluation of ASA/NBER business outlook survey forecasts. Explorations in Economic Research, 2, 588–618. Zarnowitz, V. (1979). An analysis of annual and multiperiod quarterly forecasts of aggregate income, output and the price level. Journal of Business, 52, 1–33. Zarnowitz, V. (1985). Rational expectations and macroeconomic forecasts. Journal of Business and Economic Statistics, 3, 293–311. Zarnowitz, V., & Braun, P. (1993). Twenty-two years of the NBER-ASA quarterly economic outlook surveys: Aspects and comparisons of forecasting performance. In J. H. Stock & M. Watson (Eds.), Business cycles, indicators, and forecasting (pp. 11–94). Chicago: University of Chicago Press. Zarnowitz, V., & Lambros, L. A. (1987). Consensus and uncertainty in economic prediction. Journal of Political Economy, 95, 591–621.