This site has moved to
The posts below are backup copies from the new site.

December 9, 2007

Glenn Rudebusch and John Williams: Using the Yield Curve to Forecast Recessions

The WSJ Economics blog notes this paper from the San Francisco Fed on the usefulness of the yield curve as an indicator of coming recessions. It may be more useful than we thought:

Forecasting Recessions: The Puzzle of the Enduring Power of the Yield Curve, by Glenn D. Rudebusch and John C. Williams, July 2007: Abstract We show that professional forecasters have essentially no ability to predict future recessions a few quarters ahead. This is particularly puzzling because, for at least the past two decades, researchers have provided much evidence that the yield curve, specifically the spread between long- and short-term interest rates, does contain useful information at that forecast horizon for predicting aggregate economic activity and, especially, for signalling future recessions. We document this puzzle and suggest that forecasters have generally placed too little weight on yield curve information when projecting declines in the aggregate economy.

Here's more from the introduction:

1. Introduction The word "recession" conjures a variety of fears -- for workers who suffer job losses, for investors who endure asset price declines, for entrepreneurs who risk bankruptcy. Recessions are periods of greater dislocation and anxiety, higher unemployment and suicide rates, and lower output and profits. In the United States, recessions have become less frequent and less severe in the past two decades; however, non-recessionary episodes have also become more stable, so in relative terms, as the market sensitivity in the epigraph suggests, recessions appear to many to be as perilous as before. Therefore, any ability to predict recessions remains highly profitable to investors and very useful to policymakers and other economic agents. Accordingly, there remains a keen and widespread interest in predicting recessions, and our paper examines what economic forecasters know about the likely occurrence of a recession and, most importantly, when do they know it.

Our analysis focuses on two divergent strands in the recession prediction literature. First, it is common wisdom that economists are not very good at forecasting recessions. For example, Zarnowitz and Braun (1993) showed that economic forecasters made their largest prediction errors during recessions, and Diebold and Rudebusch (1989, 1991a, b) provide a pessimistic assessment of the ability of the well-known index of leading indicators to actually provide useful signals of future recessions. In this paper, we provide new evidence on this issue by examining the information content of economic forecasts provided by participants in the Survey of Professional Forecasters (SPF). We find that these forecasters have little ability to predict recessions -- especially at a forecast horizon of a few quarters ahead.

A second strand of the recession prediction literature uses financial data, notably, the slope of the yield curve, to predict recessions. Early research that described the predictive power of the yield spread for real activity include Harvey (1989), Stock and Watson (1989), and Estrella and Hardouvelis (1991). In distinct contrast to the widely acknowledged weak performance of professional forecasters, the received wisdom from the yield curve studies is that the spread between long- and short-term interest rates is a fairly good predictor of recessions. This point 1 was made by a variety of authors in the late 1980s and has been reinforced by a large subsequent literature.[1]

This paper documents that the puzzling conflict between these two literatures has not disappeared over the past two decades. In particular, even after the predictive power of the yield curve had been well publicized, it appears that SPF participants did not incorporate all of the available yield curve information into their forecasts of economic activity. Indeed, we find that a simple model for predicting recessions that uses only real-time yield curve information would have produced better forecasts of recessions than the professional forecasters provided.

The paper proceeds as follows. In the next section, we provide a simple definition of recessions in terms of real GDP growth. In Section 3, we describe a variety of alternative real-time probability forecasts for these GDP-based recessions. In Section 4, we assess the accuracy of these forecasts, and we conclude with some speculation about possible resolutions to the puzzle of the enduring relative power of the yield curve for predicting recessions. [ to paper...]

No comments: