پویایی تورم در ایالات متحده: آیا می توان تغییرات سیاست های پولی را بیان کرد ؟
|کد مقاله||سال انتشار||مقاله انگلیسی||ترجمه فارسی||تعداد کلمات|
|27486||2012||14 صفحه PDF||سفارش دهید||11037 کلمه|
Publisher : Elsevier - Science Direct (الزویر - ساینس دایرکت)
Journal : Journal of Econometrics, Volume 167, Issue 1, March 2012, Pages 47–60
We investigate the relationship between monetary policy and inflation dynamics in the US using a medium scale structural model. The specification is estimated with Bayesian techniques and fits the data reasonably well. Policy shocks account for a part of the decline in inflation volatility; they have been less effective in triggering inflation responses over time and qualitatively account for the rise and fall in the level of inflation. A number of structural parameter variations contribute to these patterns
The US economy has gone through a number of important structural changes over the last forty years. For example, the level of inflation and of nominal interest rates shows an inverted U-shaped pattern, rising at the end of the 1970s and falling at the beginning of the 1980s; while the persistence and the volatility of inflation have dramatically declined since the mid-1980s; see e.g. Stock and Watson (2002). These patterns are well documented in the literature. What is still to be determined is the cause of these changes. The prevailing view suggests that the run-up of inflation occurred because monetary authorities believed that there was an exploitable trade-off between inflation and output. Since output was low following the oil shocks of the 1970s, the temptation to inflate was strong. However, the option of keeping inflation temporarily high was unfeasible: in the medium run, inflation reached a higher level with output settling at its potential. Since the 1980s, central banks learned that the output–inflation trade-off was not exploitable and concentrated on the objective of fighting inflation. A low inflation regime ensued, and the larger predictability of monetary policy made the macroeconomic environment less volatile (see e.g. Sargent (1999), Clarida et al. (2000), and Lubik and Schorfheide (2004)). There are two alternative views as regards this prevailing wisdom: one focuses on “real” causes (see e.g. McConnell and Perez Quiroz (2000), Campbell and Herkovitz (2006)) and the other hinges on “good luck” (see e.g. Bernanke and Mihov (1998), Leeper and Zha (2003), Sims and Zha (2006)) to explain the changes in the level and in the autocovariance function of inflation. One reason for this heterogeneity of explanations is that the empirical strategy used to study the issue matters. In general, VAR based evidence tends to support the good luck hypothesis; calibration exercises point to real reasons for the changes; and structural econometric analyses favor the idea that monetary policy is responsible for the observed variations (see, e.g., Ireland (2001), Lubik and Schorfheide (2004) and Boivin and Giannoni (2006)). However, while structural VAR exercises allow for time varying coefficients and variances, the evidence produced by more structural calibration or econometric analyses is mostly restricted to arbitrarily chosen subsamples. Because inflation and the nominal rate displayed an inverted U-shaped pattern, subsample evidence may depend on the selected break point. Fernandez Villaverde and Rubio Ramirez (2008) and Justiniano and Primiceri (2008) have estimated evolving structural models but their conclusions are only suggestive, because computational complexities force them to consider variations only in a subset of the parameters. Given that one expects important covariations in the evolution of structural parameters, allowing only a subset of the parameters to change may bias inference. Hence, it is of interest to know whether less computationally intensive and yet intuitively appealing structural methods can tell us more about the nature of the changes experienced by US inflation. This paper provides a step in that direction by estimating a structural model over rolling samples of fixed length with Bayesian techniques. Bayesian methods, which have become popular tools for bringing DSGE models to the data thanks to the work of Smets and Wouters (2003) and Del Negro and Schorfheide (2004) among others, have inferential and computational advantages over traditional limited and full information classical techniques when dealing with models which are known to be a misspecified description of the data. In these situations, unrestricted classical estimates are often unreasonable or on the boundary of the parameter space and tricks must be employed to produce economically sensible estimates. Furthermore, asymptotic standard errors attached to classical estimates–which are constructed assuming that the model is “true”–are meaningless. Rolling samples allow us to use relatively standard techniques to study the nature of the time variations present in interesting parameters while maintaining some form of rationality in the economy and keeping computational costs manageable. For example, in contrast to that of Fernandez Villaverde and Rubio Ramirez (2008), our setup allows the use of Kalman filtering techniques in building the likelihood function and permits time covariations in all the parameters. The specification that we consider deviates somewhat from what is standard in the literature by allowing money to play a role. The stock of money has been neglected in all recent monetary policy discussions (see e.g. Woodford (2003)) and Ireland (2004) provided some empirical evidence supporting this approach. In our setup real balances can potentially affect the Euler equation and the growth rate of real balances is allowed to enter the monetary policy rule. Since we will use loosely specified but proper priors in the estimation, the data will decide whether these features are important in characterizing the experience. Overall, the statistical fit of the model looks satisfactory, in particular, in comparison with other structural specifications. We estimate the preferred specification a number of times over rolling samples, analyze the time evolution of interesting inflation statistics, measure the contribution of monetary policy to the observed changes and study the evolution of the structural parameters. Our model captures the fall in inflation volatility over time and attributes part of the changes to monetary policy shocks. We detect level but not shape differences in the transmission of policy shocks which tend to make inflation less reactive to policy disturbances as time goes by. Finally, variations in the level of inflation are qualitatively related to policy shocks: had those been absent, the rise of the 1970s and the fall of 1980s would have been much more modest. A number of structural changes drive these results. We find support for the conjecture that the Fed had a much stronger dislike for inflation but also notice that in the latest samples the coefficient resembles the one obtained at the beginning of the sample. Moreover, the estimate of the long run coefficient on monetary aggregates has been steadily declining over time. We detect, in agreement with the good luck hypothesis, variations in the posterior mean estimate of the variance of the policy shocks. Nevertheless, as in Sims and Zha (2006), the variations that we discover are typically reversed over time. Finally we also find, in consistency with non-monetary explanations of the facts, that important private sector parameters such as the slope of the Phillips curve and the variability of real demand shocks have significantly changed in the later samples. In sum, we find, in consistency with the conclusions of Gambetti et al. (2008), that a combination of causes appears to be responsible for the changes in the level and the autocovariance function of US inflation over the last forty years: changes in the variance of the shocks, in the parameters regulating private sector behavior and in the policy rule all more or less contributed to explain why inflation rose and fell, and why inflation volatility subsided. The rest of the paper is organized as follows. Section 2 describes the model, the estimation technique and the diagnostics used to evaluate the quality of the model’s approximation to the data. Section 3 presents estimation results for the full sample. Section 4 reports the time profile of inflation statistics over the rolling samples. Section 5 interprets these time profiles in terms of rolling structural parameter estimates. Section 6 concludes.
نتیجه گیری انگلیسی
This paper examines the contribution of policy shocks to the dynamics of inflation using a medium scale structural model estimated with US post-WWII data and Bayesian techniques over rolling samples. The model belongs to the class of New Keynesian structures that have been extensively used in the literature but explicitly allows money to play a role. Bayesian techniques are preferable to standard likelihood methods or to indirect inference (impulse response matching) exercises, because the model that we consider is a false description of the DGP of the data and misspecification may be important. We show that our approach delivers interesting estimates of the structural parameters when priors are broadly non-informative and the policy reaction function appropriately chosen. We also demonstrate that the model fits the data reasonably well and that alternative structural specifications produce lower marginal likelihoods and fail to capture important aspects of the data. Our model captures the fall in inflation volatility and attributes parts of the changes to monetary policy shocks. We detect level but not shape differences in the transmission of policy shocks which tend to make inflation and output somewhat less reactive to policy disturbances as time goes by. Finally, variations in the level of inflation are qualitatively related to policy shocks: had those been absent, the rise of the 1970s and the fall of the 1980s would have been much more modest. A number of structural changes drive these results. We find support for the conjecture that the Fed had a much stronger short run “dislike” for inflation in later samples but also notice that the long run coefficient on nominal variables has been steadily decreasing. We detect, in agreement with the good luck hypothesis, variations in the posterior mean estimate of the variance of the policy shocks. Perhaps surprisingly, we find that these variations do not change the way in which monetary shocks are transmitted to the economy. In sum, several causes in combination are responsible for the changes in the level and the autocovariance function of US inflation over the last forty years: changes in the variance of the shocks, in the parameters regulating the private sector and in the policy rule all contributed to a greater or lesser extent to explaining why inflation rose and fell and why inflation variability subsided. These conclusions are rather different than those present in the literature, with the exception of Gambetti et al. (2008). There are a number of ways in which our analysis can be extended. For example, the estimation approach that we employ treats expectations as latent variables. However, measures of output and inflation expectations do exist in the literature. While these proxies are probably contaminated with measurement error, it would be interesting to see whether they provide additional or contrasting information about the issues at stake. Similarly, it is important to consider additional statistics to the evaluation process: while the model seems by and large well specified, it may not capture the time profile of the dynamics of a particular variable well. Finally, the use of alternative rolling estimation techniques, such as those employed by Kapetanios and Yates (2008), can help us to understand whether the conclusions are also robust along this dimension. We leave all these extensions to future research.