شوک سیاست پولی: تست شرایط شناسایی تحت نوسانات مشروط متغیر با زمان

We propose an empirical procedure, which exploits the conditional heteroscedasticity of fundamental disturbances, to test the targeting and orthogonality restrictions imposed in the recent VAR literature to identify monetary policy shocks. Based on U.S. monthly data for the post-1982 period, we reject the non-borrowed-reserve and interest-rate targeting procedures. In contrast, we present evidence supporting targeting procedures implying more than one policy variable. We also always reject the orthogonality conditions between policy shocks and macroeconomic variables. We show that using invalid restrictions often produces misleading policy measures and dynamic responses. These results have important implications for the measurement of policy shocks and their temporal effects as well as for the estimation of the monetary authority's reaction function.

There has been in recent years a considerable interest in the identification of monetary policy shocks and measurement of their effects on the economy.1 An important strand of literature uses vector autoregressions (VAR) to generate various data-based measures of policy shocks. These shocks are typically identified by imposing targeting and orthogonality restrictions. The targeting restrictions define the monetary policy indicator, while the orthogonality conditions imply that the policy shocks have no current effects on macroeconomic variables such as output and price indices. Unfortunately, it is impossible to formally verify the validity of these identifying restrictions by performing joint statistical tests. Rather, the selection of the restrictions relies on prior beliefs about the Federal Reserve operating procedures and about the signs, shapes, and persistence of certain dynamic responses to policy shocks. Thus, this approach entails a certain amount of subjectivity. This paper proposes a procedure which permits for the first time formal testing of the identifying conditions assumed in the VAR-based literature. For this purpose, we use a flexible structural VAR (SVAR) that displays three important features. First, unlike previous studies, it relaxes the assumption that the fundamental disturbances are conditionally homoscedastic. Importantly, accounting for time-varying conditional volatilities leads to the overidentification of the SVAR (e.g. Sentana, 1992, King et al., 1994, Sentana and Fiorentini, 2001 and Normandin, 2004). Hence, the restrictions typically imposed in earlier work to identify monetary policy shocks become individually and jointly testable. Second, our SVAR incorporates a standard model of the market for bank reserves (e.g. Brunner, 1994, Gordon and Leeper, 1994 and Bernanke and Mihov, 1998). This model nests the most popular monetary policy indicators. This allows us to test the indicators related to interest-rate targeting (e.g. Bernanke and Blinder, 1992 and Sims, 1992), non-borrowed-reserve targeting (e.g. Eichenbaum, 1992 and Christiano and Eichenbaum, 1992), borrowed-reserve targeting (e.g. Cosimano and Sheehan, 1994), adjusted non-borrowed-reserve targeting (e.g. Strongin, 1995), and mixed interest-rate and reserve targeting (e.g. Bernanke and Mihov, 1998). Third, our SVAR admits current interactions between the monetary policy variables and macroeconomic aggregates such as output and prices. This allows us to test the orthogonality conditions. To do so, we verify whether the policy variables directly affect current output and prices. Moreover, we check whether the policy variables indirectly affect contemporaneous output and prices through their current impacts on other non-policy variables. We estimate our SVAR using U.S. monthly data for the post-1982 period. The estimates reveal that all, but one, structural innovations display time-varying conditional variances. In particular, the policy shocks exhibit pronounced volatilities for the 1984:05–1985:02 and 1988:04–1991:03 periods. Interestingly, the first episode coincides almost exactly with the Continental Illinois incident, where the Fed has sterilized the effects of its extensive lending to this commercial bank. The second episode is consistent with the 1988 contractionary monetary policy reported by Romer and Romer (1994), and accords with common observations about changes in monetary policy through the 1990–1991 recession (e.g. Strongin, 1995). These major volatility shifts allow the identification of the policy shocks, without having to resort to the traditional restrictions. We test the identifying restrictions behind the various targeting procedures. The restrictions associated with interest-rate or non-borrowed-reserve targeting are strongly rejected, whereas those implying the other targeting procedures are not. These results sharply discriminate between interest-rate and borrowed-reserve targetings, which many observers believe to be very close in practice and empirically hard to distinguish. Our findings also help to isolate the causes for rejecting some policy indicators. For example, interest-rate targeting is rejected because the assumption that the Fed fully offsets shocks to the borrowing demand is inconsistent with evidence, while non-borrowed-reserve targeting is refuted since the requirement that the Fed does not respond to shocks to total reserves is not supported by the data. We also find that the orthogonality conditions are strongly rejected. Specifically, the direct effects of policy shocks are significant for interest-rate and mixed interest-rate and reserve targetings, where for both procedures the interest rate represents a policy variable. The indirect effects are statistically important for the other procedures, where the interest rate is a non-policy variable. Consequently, our results suggest that the policy shocks have most of their current effects on output and prices through the adjustment of interest rates. Next, we document the implications of these test results for policy. To do this, we first compare key policy measures obtained from various sets of restrictions with the valid counterparts derived from our flexible SVAR. The measures decompose the monetary authority's reaction function into policy shocks and feedback effects, and distinguish between the Fed's exogenous changes in policy and systematic responses to fluctuations in output and prices. Interestingly, the true targeting restrictions produce policy shocks and feedback effects that track remarkably well the valid policy measures. In contrast, the false interest-rate targeting restrictions yield policy shocks and feedback effects that often display the wrong signs, while the invalid non-borrowed-reserve targeting restrictions lead to reasonable policy shocks but misleading feedback effects. In addition, the false orthogonality conditions always distort the measures of policy shocks and feedback effects. Overall, these findings reveal that the specification of the Fed's reaction function must involve valid policy indicators. These indicators are combinations of the different reserve variables, rather than a single variable such as the interest rate or non-borrowed reserves. Also, the estimation of the Fed's feedback rule must rely on methods that relax the orthogonality conditions. Such methods include the instrumental-variable approach, but not the ordinary-least-square technique. We complete the analysis of the implications for policy by confronting the temporal effects of policy shocks derived from different sets of restrictions with the valid dynamic responses computed from our flexible SVAR. The true targeting restrictions produce dynamic responses that are very close to their valid counterparts. However, the invalid restrictions associated with the interest-rate indicator substantially overpredict the response of output, underpredict the response of prices, and overestimate the liquidity effect. The invalid restrictions behind the non-borrowed-reserve targeting greatly underestimate the response of output and overstate the response of prices. Finally, the false orthogonality restrictions always overstate the magnitude and persistence of the responses of output. The paper is organized as follows. Section 2 presents our flexible SVAR specification. Section 3 discusses identification issues. Section 4 reports the estimates of the SVAR parameters. Section 5 tests the standard targeting and orthogonality restrictions. Sections 6 and 7 analyze the consequences of the various sets of restrictions for policy measures and there dynamic effects, respectively. Section 8 concludes

In this paper, we proposed a procedure to test the targeting and orthogonal restrictions traditionally imposed to identify monetary policy shocks. The novel aspect of this approach is that it accounts for the time-varying conditional volatility of fundamental disturbances. In this context, the SVAR becomes over-identified, so that the restrictions can be tested individually and jointly. Our estimates indicate that all, but one, structural innovations display time-varying conditional variances. Interestingly, the pronounced movements in these variances coincide with specific events, such as the Continental Illinois incident and the 1990–1991 recession. Also, the major volatility shifts allow the identification of the policy shocks. The test results reveal that the targeting restrictions associated with the interest-rate or non-borrowed-reserve indicator are strongly rejected, while those behind the other policy indicators are not. Also, the orthogonality conditions are strongly rejected, given that the policy shocks contemporanously affect output and prices mainly through current adjustments of interest rates. These findings have important implications for policy. Specifically, the policy shocks and their dynamic effects on the economy are adequately measured from the valid targeting restrictions. In contrast, misleading policy measures and dynamic responses are obtained from the invalid restrictions associated with interest-rate targeting, non-borrowed-reserve targeting, or orthogonality conditions. Finally, policy indicators combining several reserve variables and estimation techniques relaxing the orthogonality conditions are required to appropriately decompose the monetary authority's reaction function into policy shocks and feedback effects, and to distinguish between the Fed's exogenous changes in policy and systematic responses to fluctuations in output and prices.