Macro Variables and Term Structure Models

National Bureau Of Economic Research Inc, Cambridge MA

Investigators

Abstract

Two distinct areas of economics use linear tractable frameworks for modeling tem structure. The first area, financial asset pricing, uses no-arbitrage latent factor models for bond pricing, derivatives pricing and for risk management. These affine models forecast poorly out of sample, do not incorporate non-linearities and the latent factors are given no economic interpretation. The second area of empirical macroeconomics uses Vector Auto Regression (VARs) to identify and compute the effects of monetary policy on inflation, economic growth and the behavior of the term structure. The VARs say little about how yields not included in the VAR may move, do not rule out arbitrage in the dynamics of the yield curve, and do not efficiently use forward-looking financial information. The goal of this proposal is to develop a framework that uses the best of the asset pricing and macroeconomic fields. The research has the following implications for macro and asset pricing term structure modeling. First, the framework identifies the effects of monetary policy and shocks from macro variables with priced time-varying factor risk. Second, incorporating inflation and real activity directly gives economic content to the asset pricing factor models. Third, imposing no-arbitrage provides greater efficiency, and preliminary results from the first project show that no-arbitrage models forecast better than unrestricted VARs, and incorporating macro variables further improves forecasting ability. Finally, incorporating regime switches has policy implications, as the impact of various shocks on the term structure may be very dissimilar across the regimes. More specifically, the new framework is no-arbitrage, incorporates latent and observable macro variables and introduces non-linearities. With time-varying risk premiums, the full specification is consistent with deviations from the Expectations Hypothesis, the Fisher Hypothesis, and in an international extension, Uncovered Interest Rate Parity and Purchasing Power Parity. The project builds to the full model by incorporating additional features in each individual project. The first two projects, joint with Monika Piazzesi, build a Gaussian no-arbitrage model with observable macro variables and latent variables. The model is related to the Taylor (1993) rules of monetary policy, where the surprise shocks to the short rate are identified by no-arbitrage cross-equation restrictions. The first project assumes that macro and latent variables are orthogonal, to focus on the ability of inflation and real activity to forecast the dynamics of the yield curve, before resorting to latent variables. The second project introduces feedback between the macro and latent factors. It identifies monetary policy shocks implied by no-arbitrage assumptions and investigates how shocks affect inflation, real activity, and yield curve dynamics. The third project, joint with Geert Bekaert and Bob Hodrick, introduces non-linearities in interest rates by incorporating regime switches. The model allows means, covariances and risk premia to switch across the regimes. Impulse responses for yields from shocks to macro factors, and impulse responses for macro variables from monetary policy shocks, differ across the regimes. In an international extension, expected currency depreciations and forward rates are affine functions of state variables, conditional on the regime.