عدم تقارن چرخه های کسب و کار و سیاست های پولی: تحقیقات بیشتر با استفاده از مدل مرستار

This paper investigates the asymmetric effects of monetary shocks when the impact of monetary policy on real activity works through state-dependent variables. We use a nonlinear model, the multiple regime smooth transition autoregressive model, that allows the effects of shocks to vary across the business cycles when monetary innovations modify both the endogenous and state variables. Our impulse response functions show a history-dependence property. Indeed, hitting the economy at a given time induces persistence and asymmetric responses across histories and shocks. The empirical application concerns the US over the period 1975:1–1998:2.

The past years have witnessed an increasing number of papers dealing with the asymmetry of business cycles. Although the idea is ancient, empirical studies have grown rapidly since the beginning of the 1980s. The recognition that structural changes affect the economies at any period has encouraged the use of multiple regimes models, instead of previous empirical approaches that mostly distinguished between two phases in business cycles: expansion and contraction. The varying slopes of expansion and contraction phases that induce time variations from the mean to the trough or peak of cycles, is an old stylized fact—at least it dates back to Mitchell (1927). However, for a long time, the problem has been the following: how can the theoretical concept be made operational? In an attempt to answer this question, econometricians have suggested the use of nonlinear time series models that enable the study of different dynamics over the business cycles. A plethora of papers on this topic started emerging in the 1980s and in the 1990s (see, among others, Neftçi, 1984, Falk, 1986, Lüükkonen and Teräsvirta, 1991, Teräsvirta and Anderson, 1992, Emery and Koening, 1992, Sichel, 1994, Ramsey and Rothman, 1996, Verbrugge, 1997, Pesaran and Potter, 1997 and van Dijk and Franses, 1999). Among the arguments that motivate the use of nonlinear structures, a simple idea is that the output fluctuations are influenced by variables that distort the business cycle shape. Such variables cause changes in regime in the sense that output variations follow a different time series process over different periods. This may be a cause of asymmetric dynamics. With regard to linear or VAR models, the ‘asymmetry’ of business cycles suggests that contractions last a longer period than expansions, or that shocks have stronger effects on certain variables during one of the two phases. With regard to nonlinear models, the meaning of ‘asymmetry’ is more general in the sense that we simply say that shocks have time-varying effects on the real activity. This variability occurs because the parameters of the equations describing the dynamics of the output change as a result of a regime-shift variable. Such a view modifies our comprehension of how demand and supply shocks contribute to movements in the real GDP over the business cycle. Indeed, when one perturbs the present to produce information on the dynamics of a nonlinear model, the response does not only depend on the sign of the shocks, but it is also a function of the history and of the magnitude of the shocks. This is a new challenge to econometricians. In this paper, we study the effects of monetary policy on the real sector of the US economy, assuming that output fluctuations are governed by regime-shift models, here the multiple regime smooth transition autoregressive (henceforth MRSTAR) models. These models were introduced by van Dijk and Franses (1999) who analyzed how regime-shift variables cause asymmetries in the US business cycle. They generalized the smooth transition autoregressive (STAR) models that were extensively used in the literature.1 Why is it interesting to use an MRSTAR model to evaluate the asymmetric effects of monetary policy on real GDP? If we were using a linear model (for instance a VAR process), we would proceed as follows. We would, firstly estimate a money–output equation, secondly create two series of respectively positive and negative monetary shocks, and thirdly study the properties of impulse response functions (IRFs). In such a framework, the usual results obtained in the literature may be summarized as follows: (1) money does affect output strongly when monetary policy is restrictive and raises inflation when it is expansive; (2) the effects of money on output is greater during the contraction phases of business cycles and their impact on inflation are greater during expansion phases; (3) if prices adjust slowly, then only negative shocks affect the output. In a MRSTAR model, contractionary and expansionary monetary shocks lead asymmetric effects that differ significantly from those just mentioned. Indeed, the IRFs exhibit a time dependence property. The coefficients of the money–output equation are indeed state-dependent and vary according to transition variables that generate changes in the business cycle regimes. The regime-shift variables are economic indicators characterizing both the aggregate supply and the aggregate demand. For instance, the reaction of output to negative monetary shocks may be undetermined because the level of stocks and the production capacity act as state variables that condition the reaction of the GDP to money variations (see for similar arguments Wong (2000)). There are other state variables that may induce time variation of the elasticity of output to money. Firstly, due to the imperfect structure of the credit market, initial shocks by the central bank can be either smoothed or amplified by commercial banks. A variable representing the credit channel may thus be hypothesized as being regime-shifting (Galbraith, 1996). Secondly, the impact of monetary shocks on activity is also conditioned by the credibility of monetary policy. Financial variables such as interest rate differentials reflect the agents’ expectations about future conditions of the business cycle. People may want to increase saving if they foresee a slowdown. In this case an expansive monetary policy might be ineffective. There is evidence in the literature that such behaviors induce asymmetric dynamics in the business cycle (Aftalion, 1997). Other examples of regime-shift variables could be evoked: the indexing rules that characterize the wage-price loop, the pricing rules on the good markets, the growth rate of federal expenditures, the output-gap. Whatever the case, it seems difficult to assume that a money–output equation has parameters that are invariant across alternative values of the regime-shift variables. In this paper, we use an MRSTAR model to see whether the state-dependent approach helps capturing the money nonneutrality on the business cycle. Our study concerns US quarterly data over the period 1975:1–1998:2. The paper is organized as follows. Section 2 presents the MRSTAR model that is used to estimate the money–output equation. The endogenous variable is the variation of the GDP. The exogenous variables are, respectively, the growth rate of M1, a total productivity index variable and the federal budget deficit. The regime-shift—or transition—variables include the output-gap and financial variables that are indicators of the credit channel and interest rate term structure variables. Section 3 presents the econometric methodology and the results obtained for the US economy in Section 4, we give simulation results from generalized IRFs and compare the results obtained for STAR and MRSTAR models. This allows us to show evidence of asymmetry. Section 5 concludes the paper.

In this paper, we have explored a new approach for studying the quantitative effects of monetary policy. The framework of regime-switching models such as the MRSTAR models allows reproducing some stylized facts, notably the asymmetric responses of the GDP. Also, the MRSTAR models help reproducing phenomena such as history-dependence, time variability of the impulse functions and sensitivity to the regime observed when the initial shock is produced. This paper offers several extensions. First, it may be interesting to compare the results obtained here for the US economy with those of other O.E.C.D. countries. Secondly, the financial transmission channels of monetary policy are often considered without evoking the impact of volatility. Volatility can be a source of instability in the response functions. In this view, it may be worth extending the MRSTAR model by including nonlinear components in the error term. Thirdly, it might be interesting to calibrate and simulate MRSTAR models (instead of estimating them from data) and find the transition function parameters for which the models best reproduce the usual stylized facts on monetary policy.