اقتصاد کلان و منحنی بازده: رویکرد عامل نهفته پویا

We estimate a model that summarizes the yield curve using latent factors (specifically, level, slope, and curvature) and also includes observable macroeconomic variables (specifically, real activity, inflation, and the monetary policy instrument). Our goal is to provide a characterization of the dynamic interactions between the macroeconomy and the yield curve. We find strong evidence of the effects of macro variables on future movements in the yield curve and evidence for a reverse influence as well. We also relate our results to the expectations hypothesis.

Macroeconomists, financial economists, and market participants all have attempted to build good models of the yield curve, yet the resulting models are very different in form and fit. In part, these differences reflect the particular modeling demands of various researchers and their different motives for modeling the yield curve (e.g., interest rate forecasting or simulation, bond or option pricing, or market surveillance). Still, an unusually large gap is apparent between the yield curve models developed by macroeconomists, which focus on the role of expectations of inflation and future real economic activity in the determination of yields, and the models employed by financial economists, which eschew any explicit role for such determinants. This paper takes a step toward bridging this gap by formulating and estimating a yield curve model that integrates macroeconomic and financial factors. Many other recent papers have also modeled the yield curve, and they can be usefully categorized by the extent and nature of the linkages permitted between financial and macroeconomic variables. Many yield curve models simply ignore macroeconomic linkages. Foremost among these are the popular factor models that dominate the finance literature—especially those that impose a no-arbitrage restriction. For example, Knez et al. (1994), Duffie and Kan (1996), and Dai and Singleton (2000) all consider models in which a handful of unobserved factors explain the entire set of yields. These factors are often given labels such as “level,” “slope,” and “curvature,” but they are not linked explicitly to macroeconomic variables. Our analysis also uses a latent factor model of the yield curve, but we also explicitly incorporate macroeconomic factors. In this regard, our work is more closely related to Ang and Piazzesi (2003), Hördahl et al. (2002), and Wu (2002), who explicitly incorporate macro determinants into multi-factor yield curve models. However, those papers only consider a unidirectional macro linkage, because output and inflation are assumed to be determined independently of the shape of the yield curve, but not vice versa. This same assumption is made in the vector autoregression (VAR) analysis of Evans and Marshall, 1998 and Evans and Marshall, 2001 where neither contemporaneous nor lagged bond yields enter the equations driving the economy. In contrast to this assumption of a one-way macro-to-yields link, the opposite view is taken in another large literature typified by Estrella and Hardouvelis (1991) and Estrella and Mishkin (1998), which assumes a yields-to-macro link and focuses only on the unidirectional predictive power of the yield curve for the economy. The two assumptions of these literatures—one-way yields-to-macro or macro-to-yields links—are testable hypotheses that are special cases of our model and are examined below. Indeed, we are particularly interested in analyzing the potential bidirectional feedback from the yield curve to the economy and back again. Some of the work closest to our own allows a feedback from an implicit inflation target derived from the yield curve to help determine the dynamics of the macroeconomy, such as Kozicki and Tinsley (2001), Dewachter and Lyrio (2002), and Rudebusch and Wu (2003). In our analysis, we allow for a more complete set of interactions in a general dynamic, latent factor framework. Our basic framework for the yield curve is a latent factor model, although not the usual no-arbitrage factor representation typically used in the finance literature. Such no-arbitrage factor models often appear to fit the cross-section of yields at a particular point in time, but they do less well in describing the dynamics of the yield curve over time (e.g., Duffee, 2002 and Brousseau, 2002). Such a dynamic fit is crucial to our goal of relating the evolution of the yield curve over time to movements in macroeconomic variables. To capture yield curve dynamics, we use a three-factor term structure model based on the classic contribution of Nelson and Siegel (1987), interpreted as a model of level, slope, and curvature, as in Diebold and Li (2002). This model has the substantial flexibility required to match the changing shape of the yield curve, yet it is parsimonious and easy to estimate. We do not explicitly enforce the no-arbitrage restriction. However, to the extent that it is approximately satisfied in the data—as is likely for the U.S. Treasury bill and bond obligations that we study—it will also likely be approximately satisfied in our estimates, as our model is quite flexible and gave a very good fit to the data. Of course, there may be a loss of efficiency in not imposing the restriction of no arbitrage if it is valid, but this must be weighed against the possibility of misspecification if transitory arbitrage opportunities are not eliminated immediately. In Section 2, we describe and estimate a basic “yields-only” version of our model—that is, a model of just the yield curve without macroeconomic variables. To estimate this model, we introduce a unified state-space modeling approach that lets us simultaneously fit the yield curve at each point in time and estimate the underlying dynamics of the factors. This one-step approach improves upon the two-step estimation procedure of Diebold and Li (2002) and provides a unified framework in which to examine the yield curve and the macroeconomy. In Section 3, we incorporate macroeconomic variables and estimate a “yields-macro” model. To complement the nonstructural nature of our yield curve representation, we also use a simple nonstructural VAR representation of the macroeconomy. The focus of our examination is the nature of the linkages between the factors driving the yield curve and macroeconomic fundamentals. In Section 4, we relate our framework to the expectations hypothesis, which has been studied intensively in macroeconomics. The expectation hypotheses of the term structure is a special case of our model that receives only limited support. We offer concluding remarks in Section 5.

We have specified and estimated a yield curve model that incorporates both yield factors (level, slope, and curvature) and macroeconomic variables (real activity, inflation, and the stance of monetary policy). The model's convenient state-space representation facilitates estimation, the extraction of latent yield-curve factors, and testing of hypotheses regarding dynamic interactions between the macroeconomy and the yield curve. Interestingly, we find strong evidence of macroeconomic effects on the future yield curve and somewhat weaker evidence of yield curve effects on future macroeconomic developments. Hence, although bidirectional causality is likely present, effects in the tradition of Ang and Piazzesi (2003) seem relatively more important than those in the tradition of Estrella and Hardouvelis (1991), Estrella and Mishkin (1998), and Stock and Watson (2000). Of course, market yields do contain important predictive information about the Fed's policy rate. We also relate our yield curve modeling approach to a traditional macroeconomic approach based on the expectations hypothesis. The results indicate that the expectations hypothesis may hold reasonably well during certain periods, but that it does not hold across the entire sample. From a finance perspective, our analysis is unusual in that we do not impose no-arbitrage restrictions. However, such an a priori restriction may in fact be violated in the data due to illiquidity in thinly traded regions of the yield curve, so imposing it may be undesirable. Also, if the no-arbitrage restriction does indeed hold for the data, then it will at least approximately be captured by our fitted yield curves, because they are flexible approximations to the data. Nevertheless, in future work, we hope to derive the no-arbitrage condition in our framework and explore whether its imposition is helpful for forecasting, as suggested by Ang and Piazzesi (2003).