نمایش ساختار شرایط سیاست های پولی تحت مدل های جایگزین از انتظارات نماینده

Term structure models and many descriptions of the transmission of monetary policy rest on the empirical relevance of the expectations hypothesis. Small differences in the perceived policy reaction function in VAR models of agent expectations strongly influence the relevance in the transmission mechanism of the expected short rate component of bond yields. Mean-reverting or difference-stationary characterizations of interest rates require large and volatile term premiums to match the observable term structure. However, short rate descriptions that capture shifting perceptions of long-horizon inflation evident in survey data support a more substantial term structure role for short rate expectations.

The term structure of interest rates is often characterized as a crucial transmission channel of monetary policy where accurate market perceptions of policy are required for effective policy execution. This description rests on three propositions: First, a short-term interest rate, such as the overnight federal funds rate, provides an adequate summary of monetary policy actions. Second, long-term bond yields are important determinants of the opportunity cost of investments. Third, under the expectations hypothesis, the anticipated path of the policy short-term rate is the main determinant of the term structure of bond rates. Although the empirical validity of each of these three propositions has been criticized, the accumulation of empirical evidence against the third proposition — the expectations hypothesis — is impressively large.1 If variations in current bond rates are not well-connected to current and expected movements in the policy-controlled short-term rate, then the conventional description of monetary policy transmission is vacuous. In no-arbitrage formulations of the term structure, variation in bond rates due to current and expected movements in short rates is determined by the description of short rate dynamics. Under the standard finance assumption that short rates are mean-reverting, the average of expected future short rates is considerably smoother than historical bond rates. Thus, the expectations hypothesis is empirically rejected by tests which assume constant term premiums, and postwar shifts in the term structure are often attributed to sizable movements in ‘liquidity’ and ‘term’ premiums. However, postwar data are consistent with descriptions of short rate movements other than mean reversion. This paper shows that small differences in the specification of the stochastic process for the short rate strongly influence the relative importance of short rate expectations and residual term premiums in bond rate movements. As an alternative to conventional descriptions of short rate dynamics, a simple class of time-varying descriptions of short rate behavior is examined. Given well-documented shifts in postwar monetary policy, it seems highly probable that the short rate expectations of bond traders are heavily influenced by shifting perceptions of monetary policy. Short rate descriptions that capture shifting perceptions of the long-run objective of monetary policy, formulated as a long-run inflation target, are supportive of a substantial term structure role for short rate expectations.2 In addition to rejecting the implication that high bond rates in the 1980s reflect large term premiums, the preferred description of short rate behavior also does not support interpretations that perceived inflation targets were rapidly reversed in the 1980s. More likely, the long-run inflation target perceived by the market behaved similarly to survey data on long-run expected inflation. Survey data suggest agents were cautious in adjusting their perceptions of long-run inflation. This paper illustrates the term structure implications of alternative representations of monetary policy perceptions of agents. Section 2 develops a discrete-time, no-arbitrage model of the term structure where the stochastic discount factor is related to a vector of macroeconomic variables. Forecasts of this state vector are generated by a VAR with flexible specifications of long-run behavior. Section 3 discusses three VARs with different characterizations of agent perceptions of long-run behavior. Owing to the different characterizations of perceived long-run behavior, the VARs embed perceived policy reaction functions with different implied inflation targets. Section 4 indicates how estimates of expected short-rate components of bond yields and conventional ‘residual’ estimates of term premiums depend critically on assumptions regarding the long-run behavior of policy targets in VAR proxies of agent expectations. Section 5 concludes.

5. Concluding remarks In many macroeconomic models, an important transmission channel of monetary policy is variation in the value of wealth and the cost of borrowed funds due to policy-induced changes in long-term interest rates. Because most central banks directly control interest rates only in short-term banking markets, this implies that monetary policy is conducted more by auction market perceptions of current and anticipated policy actions than by recent activities of the central bank trading desk. Consequently, auction market prices are monitored by policy authorities and observers, not only because quotations are available on a more timely basis than other measurements of economic activity but also to discern market expectations. As noted earlier, this stylized characterization of monetary policy transmission rests on a number of assumptions, of which the most beleaguered is the expectations hypothesis. Not only are various implications of the expectations hypothesis routinely rejected in postwar empirical literature but, as illustrated in this paper, multiperiod predictions of standard time series models of short-term interest rates provide relative poor tracking estimates of historical bond rates. The difficulty, of course, is that researchers cannot reject the hypothesis that bond yields are nonstationary for samples that include the 1970s and 1980s. This is noteworthy because, by the Fisher equation, an important component of bond rates is expected inflation, and nonstationarity is not rejected also for postwar rates of inflation in consumer price indexes. Further, if the stochastic process of the short rate controlled by the policy authority is sufficient under the expectations hypothesis to explain much of the variation in bond yields, then the model of the short rate process needs to capture the response of the short policy rate to the behavior of inflation. The approach adopted in this paper is to represent bond trader expectations by small VARs that accommodate relationships among expectations of inflation and of interest rates. A significant departure in this paper, however, is to represent market expectations by linear expansions around long-run expectations, termed ‘endpoints’. These endpoints may be fixed, may shift, or may be nonstationary. All models in this class can be written as endpoint-reverting. An important empirical finding, shown in various ways in this paper, is that the estimated eigenvalues of endpoint-reverting VARs are relatively unimportant for long-horizon predictions, such as long-maturity bonds, because all endpoints, fixed, shifting, or nonstationary, are reached relatively early in the forecast horizon. By contrast, the specification of long-horizon expectations is of pivotal importance. Standard VARs offer two options: that variables are mean reverting or difference stationary. The mean-reverting VAR is the workhorse of modern macroeconomic analysis, and many empirical features of this specification have become stylized facts in analyses of monetary policy. Unfortunately, the endpoints of VAR variables are fixed in this specification, with the unrealistic implication that market perceptions of the long-run inflation target of policy are independent of actual inflation. This ‘fixed’ endpoints specification provides nearly constant predictions of long-horizon inflation and the expected short rate component of long bond yields. Consequently, the implied term premiums of bonds that are required to match historical bond rates swamp the forward short rate predictions of the mean-reverting VAR.24 Thus, by default under fixed endpoints, the main channel for short rate expectations to influence bond yields is through time-varying term premiums. This channel, however, requires that the level of the short rate is a source of variation in term premiums — such as the Cox–Ingersoll–Ross square-root specification of short rate volatility. Turning to the other conventional specification, the endpoints of the difference-stationary VAR are moving averages of difference-stationary states and, as a result, are also I(1). As illustrated earlier, these ‘moving average’ endpoints move closely with realizations of state variables and provide poor predictions of long-horizon expectations and long-maturity yields under constant term premium assumptions. Not all nonstationary variables are I(1), and the preferred specification in this paper is that endpoints of expectations shift if agents detect shifts in the long-run targets of monetary policy. By the Fisher equation, and assuming after-tax real rates are stationary, the tax-adjusted endpoint of nominal interest rates is linked to the long-run policy target for inflation. However, important results of this paper are that bond rates are driven by market perceptions of policy targets and there appear to be lengthy learning lags between historical shifts in indicators of policy targets, such as the inflation changepoints used here, and shifts as detected by market agents in ‘real time’. An additional issue is whether or not the empirical results presented in this paper are robust to the construction of the shifting endpoints. Other approaches being pursued by the authors include: fixed-weight aggregation over simulated forecast newsletters; Kalman filters of structural time series models; and inference of forecast endpoints from surveys of short-horizon expectations. None of the alternatives appear to undermine several basic conclusions: First, in contrast to the long-horizon forecast implications of fixed endpoint VARs, forecasts of expected short-term interest rates from VARs with shifting endpoints can explain a substantial portion of the variability of long-term bond rates. Second, long-horizon inflation forecasts generated by shifting endpoint VARs provide a much closer match to survey measures of expected long-run inflation than do forecasts provided by the conventional stationary and difference-stationary VAR models. Third, as discussed in Rudebusch (1998), the monetary policy reaction rules in many conventional VAR models are not likely to be stable over lengthy samples, reflecting episodic changes in the stabilization policies of central banks. For the purpose of capturing episodic changes in long-run equilibria, there are several alternatives to a shifting intercepts specification, including multiple regimes models and stochastic coefficient specifications. However, the shifting endpoints model is relatively simple to implement and can provide useful economic insights into plausible and significant sources of nonstationarity in agent expectations. Finally, models that represent how rational agents cope with limited and heterogeneous perceptions of the economy, such as Sargent (1993), Brock and Durlauf (1995), and Kurz (1998), provide explicit interpretations of nonstationarity and endogenous uncertainty. The model of ‘real time’ detections of changepoints discussed in this paper is a rudimentary step toward an endogenous description of agent ignorance.