آیا بعد از یک سیاست پولی محکم تورم افزایش می یابد ؟ پاسخ ها بر اساس مدل برآوردی DSGE
|کد مقاله||سال انتشار||مقاله انگلیسی||ترجمه فارسی||تعداد کلمات|
|26167||2007||32 صفحه PDF||سفارش دهید||14546 کلمه|
Publisher : Elsevier - Science Direct (الزویر - ساینس دایرکت)
Journal : Journal of Economic Dynamics and Control, Volume 31, Issue 3, March 2007, Pages 906–937
This paper estimates the importance of the cost channel of monetary policy in a New Keynesian model of the business cycle. A model with nominal and real rigidities is extended by assuming that a fraction of firms need to borrow money to pay their wage bill. Hence, a monetary policy tightening increases effective unit labor costs of production, and might imply an increase in inflation. The paper examines the conditions under which the model can generate a positive response of inflation to a monetary contraction, and estimates the model's parameters using Bayesian methods. The paper shows that the estimated demand-side effects of monetary policy dominate the estimated supply side effect, even if restrictions are imposed that make occurrence of a positive inflation response to a monetary contraction more likely.
What is the effect of monetary policy on prices? The conventional view suggests that monetary policy tightenings are associated with declines in output and inflation. However, the results coming from using vector autoregressive (VAR) models are far from conclusive: one of the most controversial findings in the empirical literature on monetary policy shocks is the so-called ‘price puzzle,’ whereby a tightening of monetary policy is associated with an increase, rather than a decrease, of the price level. Two main explanations have been offered for this phenomenon: one implies that the unexpected part of monetary policy shocks is not well measured, while the other suggests that there are ‘cost channel’ effects of monetary policy. The first explanation suggests that VAR models cannot properly measure the forward-looking component of monetary policy, and hence, do not properly measure monetary policy shocks. Suppose that the central bank expects higher inflation in the future, due to productivity shocks, oil price shocks, exchange rate developments, and the like. When the central bank increases interest rates, those shocks may have already been built into the economy, so a simultaneous increase in interest rates and prices is observed. Therefore, the price puzzle arises due to a misidentification of the unexpected component of monetary policy shocks. Sims (1992) suggested that once commodity prices are included in a VAR model, the price puzzle disappears. His explanation was that the information set available to policy makers may include variables useful in forecasting inflation that the econometrician has not considered. A more recent paper by Romer and Romer (2004) constructs series of monetary policy shocks after controlling for the endogenous response of the Federal Reserve to its own forecasts of output growth, inflation and unemployment. Among other results, they find that the price puzzle becomes irrelevant. The second explanation suggests that there is no methodological problem with a price puzzle type of behavior. On the contrary, it is indeed the cost channel of monetary policy that causes prices (or inflation) and nominal interest rates to move in the same direction after a monetary policy shock. When the central bank increases interest rates, some production (financing) costs increase, which will tend to cause an increase in the inflation rate. This ‘supply side’ effect of monetary policy may coexist with and, in fact, dominate the traditional ‘demand-side’ effect. Barth and Ramey (2001) reach this conclusion using industry level and aggregate data for the United States, and show that their finding is robust even when commodity prices are introduced in their VAR. More recent work by Christiano et al. (2005) reaches the same conclusion, using aggregate data. This paper attempts to disentangle these two conflicting explanations by estimating a dynamic stochastic general equilibrium (DSGE) model using a Bayesian approach. The use of DSGE models based on staggered price and wage setting (i.e. New Keynesian models) has become increasingly popular for the analysis of monetary policy, due to their analytical tractability. However, in the baseline model, there is no room for a cost channel of monetary policy: increases in interest rates always cause inflation to decline.2 In this paper, a New Keynesian model is extended by introducing a ‘working capital’ or cost channel assumption: a fraction of firms need to borrow funds to pay for their wage bill before selling their product. As a result, the nominal interest rate is a determinant of real marginal costs, and hence, of inflation. By constructing and estimating a model that allows for an increase of inflation after a monetary policy tightening, I examine to what extent this is a feature of the aggregate data, and its relevance in monetary policy making. The results of the paper can be summarized as follows. First, the estimates of model parameters point at a low elasticity of inflation to changes in the nominal interest rate, with a posterior mean of 0.15. As a result, the posterior probability of observing an increase of inflation after a monetary policy tightening is zero. Second, when the model is estimated assuming that all firms are subject to the ‘working capital’ assumption, it is still not possible to obtain a positive response of inflation to a monetary policy contraction. In the model, inflation depends on the real marginal cost of production, which includes the real wage, the rental rate of capital, and the nominal interest rate. In order to generate an increase of inflation after a monetary contraction, it is necessary that the immediate positive effect of the nominal interest rate on the real marginal cost is not offset by declines in the real wage and the rental rate of capital. Introducing staggered wage setting with indexation makes the response of the real wage smoother, while allowing for high variability in the capital utilization rate makes the response of the rental rate of capital less volatile. But all these necessary conditions are not picked up when the model is estimated. Finally, when all necessary conditions, in addition to the cost channel, are imposed in the estimation procedure, model fit worsens significantly. In particular, the model has trouble explaining the behavior of nominal variables. Other estimated parameters of the model change such that the demand-side effect of monetary policy always dominates the supply side, and inflation and interest rates move in opposite directions after a monetary policy shock. As a result, policy makers should not be concerned about short-run increases in inflation after policy tightenings. This paper is closely related to the recent literature of estimation of New Keynesian models with a cost channel of monetary policy. Ravenna and Walsh (2005) estimate a new Phillips Curve which explicitly incorporates a cost channel of monetary policy, and find a large elasticity of inflation to the nominal interest rate. However, their results vary depending on the choice of the weighting matrix in their Generalized Method of Moments estimation procedure, and the choice of instruments. Single equation (or limited information) estimation techniques are most robust and can help reduce potential misspecification problems by leaving some relationships unspecified. However, they cannot capture the linkages between several variables in a larger scale model, they are less efficient, and can suffer from identification problems. Christiano, Eichenbaum, & Evans (2005; CEE henceforth) conduct parameter estimation by minimizing the distance between estimated (VAR-based) and model-based impulse responses of several variables to a monetary policy shock. Since their VAR evidence displays an increase in inflation after a tightening of monetary policy, their parameter estimates cause their model to reflect that property.3 The work of this paper is complementary to CEE's approach. CEE's focus is to match the VAR-based impulse responses to a monetary policy shock, and they assume a specific value for a subset of parameters of their model that turn out to be key in generating an increase of inflation after a monetary policy tightening in the model. CEE introduce full indexation to lagged inflation in both price and wage setting, a large elasticity of capital utilization with respect to the rental rate of capital, and an elasticity of the real marginal cost of production to the nominal interest rate of one. This paper takes a different approach by estimating the parameters of the model using a likelihood-based method: the estimation procedure tries to fit all the second moments of the data in a model that incorporates more shocks than just monetary policy shocks. The parameter estimates suggest that it is not possible to observe a positive inflation response after a monetary contraction. Assuming parameter values that would generate such effect worsens model fit significantly: the autocorrelations and standard deviations of inflation and interest rates, and the correlation between inflation and interest rates are much higher in the model than in the data. There are several advantages of using a Bayesian estimation approach in a DSGE framework. First, it is a flexible method that allows the researcher to introduce prior information about the model's parameters. Classical methods make it difficult to accommodate even the most noncontroversial prior information. Second, since it is a likelihood-based method, it takes advantage of the general equilibrium approach: all the theoretical restrictions implied by the model for the likelihood function and the full dimension of the data are taken into account for estimation.4 The model's accuracy of fit is addressed using the ratio of marginal likelihoods or Bayes factor. The marginal likelihood averages all possible likelihoods across the parameter space, using the prior as a weight, and is a concept of fundamental importance in Bayesian model comparison, because of its role in determining the posterior model probability.5 A main shortcoming with respect to limited information methods is that the fit of the model is based on a large number of moments, which requires more structure on the model's relationships. In addition, the marginal likelihood depends on the priors chosen by the researcher. The rest of the paper is organized as follows: in Section 2, I present a model with nominal and real rigidities, which incorporates the cost channel; Section 3 presents the linearized version of the model, while Section 4 shows the response of inflation and output to a monetary policy shock under CEE's parameterization. Section 5 explains the econometric methodology for parameter estimation and model comparison using Bayesian methods. In Section 6, the main results are discussed, while concluding remarks are left for Section 7.
نتیجه گیری انگلیسی
The results of the present paper are based on a likelihood-based estimated structural model, and support the view that inflation and interest rates move in opposite directions after a monetary policy shock. Therefore, the results contradict the view that the supply side effect of monetary policy dominates the more traditional demand-side effect. As in any other research paper, the results are conditional on the choice of a model, econometric strategy, and a particular data set. However, the paper has shown that by imposing several structural relationships aimed at generating an increase of inflation to a monetary policy shock, the overall model fit to the data worsens. In particular, when some parameters of the model are fixed to values that would allow for an increase of inflation after a monetary policy shock, the model cannot fit the behavior of nominal variables: it implies too high inflation and nominal interest rate persistence and volatility, and a correlation of one between inflation and interest rates, which is at odds with the data. In addition, the more restricted models deliver a too high volatility of all variables. The results of the paper contradict those of Ravenna and Walsh (2005) and CEE, who suggest that the presence of the cost channel is more important, and that it is possible, in the context of this model, to observe an increase of inflation to a monetary policy shock. In those papers, the econometric methodology was based in matching a smaller set of moments. But when the set of moments to be explained is expanded, their results disappear. Why? In other to fit the additional observed moments, the estimated parameters move away from the choices that would generate a positive response of inflation to a contractionary monetary policy shock. Interestingly, Altig et al. (2005) estimate the parameters of a New Keynesian model similar to the one presented here (and in CEE) by minimizing the distance between model-based and VAR-based impulse responses to monetary, investment specific and neutral technology shocks. Among other results, their estimate of ψ falls from 100 when trying to match the response to only monetary policy shocks, to 0.5 when trying to match the response to the three shocks at the time. In the latter case, their estimated model-based impulse responses no longer display an increase of inflation after a contractionary monetary policy shock.30 The methodology presented in the present paper delivers different results about the behavior of inflation after a monetary policy shock than a strand of the VAR literature, in particular the results obtained by Barth and Ramey (2001) for the 1960–1979 period. Therefore, if a price puzzle type of behavior arises in a VAR, it is likely to come from misspecification. The interesting question to study in future research would be if VARs can properly identify the effects of monetary policy shocks using simulated data.31 Finally, it is important to highlight that the estimate of average duration of wage contracts is surprisingly low. It could well be that the Calvo (1983) model for wage setting does not seem to explain wage dynamics as well as it explains inflation dynamics. Competing models of wage setting should be further studied in a DSGE setup. In addition, it would be worthwhile estimating a dynamic general equilibrium model using Bayesian methods and industry-level data, and examine the sectoral properties of inflation dynamics. This would allow to study which sectors are affected by the cost channel, and whether there are washing out effects once the step from industry-level to aggregate data is done.