Financial Market Perceptions of Recession Risk

Over the Great Moderation period in the United States, we find that corporate credit spreads embed crucial information about the one-year-ahead probability of recession, as evidenced by both inand out-of-sample fit. Furthermore, the incidence of “false positive” predictions of recession is dramatically reduced by utilizing a bivariate model that includes a measure of credit spreads along with the slope of the yield curve; indeed, these bivariate models provide much better forecasting performance than any combination of univariate models. We also find that optimal (Bayesian) model combination strongly dominates simple averaging of model forecasts in predicting recessions. * Division of Monetary Affairs, Board of Governors of the Federal Reserve System. Corresponding author: [email protected]; (202)452-2867. We thank Jonathan Wright and other Federal Reserve Board staff for helpful discussions and Kristen Payne and Ben Johannsen for superb research assistance. The views expressed here are those of the authors and do not represent official positions of the Federal Reserve. This draft: 14 August 2007 Introduction Central banks watch financial markets closely, at least in part because financial data may embed valuable signals about the state of the economy and risks to the macroeconomic outlook. For example, conventional wisdom holds that negative term spreads—that is, a downward-sloping yield curve—generally reflect an elevated risk of recession over the subsequent year. Furthermore, a long tradition has emphasized the extent to which a widening of credit spreads—that is, the difference in yields between lowand high-grade securities—can provide advance warning of a deterioration in macroeconomic conditions. However, recent research has cast significant doubt on the extent to which statistical models can provide robust signals in forecasting economic growth or assesssing the near-term risk of recession. For example, several studies have documented a weakening link in the relationship between term spreads and economic activity over the past couple of decades. More generally, the analysis of Stock and Watson (2003)—henceforth referred to as SW03— has highlighted the unsatisfactory performance of statistical models in forecasting GDP growth, although their study focused on univariate models and did not include any measures of corporate credit spreads. In this paper, we perform a systematic comparison of statistical models of U.S. recession risk over the Great Moderation period. In particular, we consider monthly data for the period 1988 to 2007 for 54 different financial-market variables, including all of the series considered by SW03 as well as a variety of corporate credit spreads and several other distinct measures of market liquidity. Furthermore, we examine the entire space of bivariate statistical models to determine whether any of these models can provide a substantial improvement in the Bayesian information criterion or out-of-sample forecasting performance of the best univariate models. Finally, we investigate the benefits of combining individual model forecasts via either simple unweighted averaging or Bayesian model averaging (which places relatively greater weight on models with higher posterior probability). Our analysis indicates that corporate credit spreads embed crucial information about the one-year-ahead probability of recession, as evidenced by both in-sample fit and out-of-sample 1 See Estrella and Hardouvelis (1991), Estrella and Mishkin (1998), and Estrella, Rodrigues, and Schich (2003). 2 See Bernanke (1983), Friedman and Kuttner (1992), Gertler and Lown (2000), and Bordo and Haubrich (2004). 3 See Dotsey (1998), Chauvet and Potter (2002, 2005), and Giacomini and Rossi (2006). More recently, Wright (2006) investigated the extent to the predictive performance of a univariate yield-curve model can be improved by including a measure of the current level of interest rates.