یادگیری تطبیقی​​، قواعد ابزار مبتنی بر پیش بینی و سیاست پولی

This paper argues that recently popular forecast-based instrument rules for monetary policy may fail to stabilize economic fluctuations. In a New Keynesian model of output gap and inflation determination in which private agents face multi-period decision problems, but have non-rational expectations and learn over time, if the monetary authority adopts a forecast-based instrument rule and responds to observed private forecasts then this class of policies frequently induce divergent learning dynamics. A central bank that correctly understands private behavior can mitigate such instability by responding to the determinants of private forecasts. This suggests gathering information on the determinants of expectations to be useful.

The recent monetary policy rules literature argues that private-sector forecasts are an important part of central bank decision procedures for the determination of the nominal interest rate. Hall and Mankiw (1994) propose nominal GDP forecast targeting. Batini and Haldane (1999) argue that a simple interest-rate rule that posits the nominal interest rate as depending on private-sector inflation forecasts provides a robust formulation of policy. Such rules are also found in a number of large-scale macroeconomic models used in policy evaluation (see, for instance, the Reserve Bank of New Zealand or Bank of England forecasting models). Clarida et al., 1998 and Clarida et al., 2000 also provide evidence that reaction functions of a number of central banks find an important role for expectations in the current stance of policy. Giannoni and Woodford (2002a) demonstrate optimal targeting rules invariably imply an instrument setting that depends on expectations. More recently, Levin et al. (2003) provide evidence that appropriately designed forecast-based instrument rules are robust to model uncertainty. In evaluating the merit of such policy proposals, this literature typically assumes that both the central bank and private agents possess the same model of the economy and have rational expectations. It follows that all economic actors hold identical expectations regarding the evolution of the economic variables of interest and therefore that there is no important distinction between internal central bank forecasts and external private forecasts. To the extent that good monetary policy depends on expectations, it is sufficient for the policy problem to be cast in terms of internal central bank forecasts. In practice, however, internal central bank forecasts and external private forecasts rarely coincide. As a result, both of these sources of forecasts provide potentially important information for the monetary policy decision process. This is evidenced by the considerable resources that central banks spend on forecasting the near-term evolution of the economy. In addition, external forecasts of various private agents are monitored using an array of surveys on the ground that policy can be improve by having more information about the state of the economy. But if these two sources of information about the near-term evolution of the economy diverge, what then is the appropriate dependency of the central bank's instrument setting on such forecasts. Is it appropriate for monetary policy to depend on private forecasts and if so in exactly what way? Or are there reasons for a central bank to concern itself solely with internal forecasts? This paper therefore seeks to examine whether policy rules that posit the interest rate to depend on private expectations are desirable as a means to stabilize economic fluctuations when agents and the central bank have differing expectations about the evolution of the macroeconomy. We will be interested to learn what is the appropriate dependency of optimal monetary policy decision procedures on private forecasts. In particular, we shall explore whether desirable policy can be described by an instrument rule that naively responds to observed private expectations, or whether more sophisticated uses of the information embodied in these forecasts is required in internal central bank forecasting procedures for economic stability. In so doing, it can be adjudged whether central banks need only devote resources to the measurement of forecasts themselves or whether greater resources need to be devoted to understanding the underlying determinants of such forecasts. Furthermore, if knowledge of the determinants of private forecasts is desirable, the analysis will shed light on their appropriate use in monetary policy design. This paper proceeds as follows. Section 2 outlines the analysis of Preston (2005a), which develops a model in which agents face multi-period decision problems as in the microfoundations used in recent analysis of the implications of monetary policy rules under rational expectations—see Bernanke and Woodford (1997), Clarida et al. (1999) and Woodford (1999). In contrast to these papers, private agents are not assumed to possess a complete economic model in making their spending and pricing decisions and therefore cannot infer the true probability laws that govern the evolution of the economy and must instead attempt to learn them over time. They do this by use of a simple econometric model, which is updated each period as additional data become available. Section 3 discusses private agents beliefs and the learning algorithm in some detail. Section 4 outlines the optimal policy problem. Two forecast-based instrument rules are proposed that are consistent with implementing the resulting optimal equilibrium. Several approaches to constructing the required forecasts to implement these rules are then considered. Section 5 analyzes a decision procedure in which the monetary authority responds to observed private forecasts. It shows that this approach to policy is likely to facilitate divergent learning dynamics, and is therefore undesirable as a means to stabilizing economic fluctuations in the presence of private agent learning. Importantly, this contrasts with the analysis of Bullard and Mitra (2002), which finds the Taylor principle to be necessary and sufficient for stability under learning dynamics. However, it is also shown that if private agent's are endowed with knowledge of the monetary policy rule such instability is mitigated—the so-called Taylor principle is necessary and sufficient for stability under learning dynamics. This finding is similar in spirit to Orphanides and Williams (2005) which shows that transparency about the central bank's long-run inflation objective can engender a more favorable inflation-output trade off when agents must learn about the economy's inflation dynamics. Section 6 proposes a second decision procedure that assumes the monetary authority to have correct knowledge of private agents’ learning behavior and decision rules. This allows the monetary authority to construct optimal forecasts of the evolution of the economy conditional on agents’ behavior. In this case, the instability problems associated with the first decision procedure are largely avoided. Indeed, as found by Bullard and Mitra (2002), the forecast-based instrument rules lead to stability under learning dynamics if and only if the so-called Taylor principle is satisfied. Importantly, these results highlight the importance of gathering information about the determinants of private sector expectations. Section 7 offers some remarks on alternative approaches to implementing optimal policy under learning—a topic that is more thoroughly treated in Preston (2004). The final section concludes.

This paper applies the framework of Preston (2005a) to understand the appropriate use of private forecasts in the design of monetary policy. The analysis demonstrates that recently popular forecast-based instrument rules may give rise to divergent learning dynamics and therefore be undesirable as a means to stabilize economic fluctuations. In particular, the Taylor principle is not a sufficient condition for private agents to be able to learn the associated rational expectations equilibrium. This result contrasts with recent work by Bullard and Mitra (2002) which finds in an economy where only one-period-ahead expectations matter, that the Taylor principle is in fact necessary and sufficient conditions for stability under learning dynamics. Evidence on the importance of a transparent monetary policy is also adduced by demonstrating that such instability can be mitigated if private agents are informed about the form of the central bank's instrument rule. In this case, the Taylor principle is once more necessary and sufficient for stability under learning. However, if the central bank correctly understands the learning mechanism of private agents, it can construct optimal forecasts conditional on private agents’ behavior. In this case, forecast-based instrument rules have the central bank respond to the determinants of private forecasts, rather than the actual forecasts themselves, and this mitigates observed instability problems of the former decision procedure. Indeed, the Taylor principle is again necessary and sufficient for E-Stability. This underscores the importance of gathering information on the nature and form of private forecasting methods. Moreover, it emphasizes that the concern of Bernanke and Woodford (1997), that policy rules which naively depend on observed private forecasts might be susceptible to problems of indeterminacy, has greater ambit than rational expectations models.