Financial and Macroeconomic Indicators of Recession Risk
Michael T. Kiley
Recessions impose sizable hardship, with large increases in the unemployment rate and related dislocations. In addition, recessions can lead to large shifts in financial markets. As a result, economists and financial market professionals have considered prediction models to assess the probability of a recession.
A range of approaches have been considered. One strand of analysis uses financial market variables—often the slope of the yield curve—to assess recession risk. A second strand augments the first approach by adding leading indicators (which summarize confidence and economic activity indicators) to recession prediction models. A third approach considers the state of the macroeconomy—inflation and the business cycle, as measured by the unemployment rate—because historical analyses have emphasized how recessions were preceded by high inflation, an overheated economy, and a shift in monetary accommodation (e.g., Romer and Romer, 1989).
This note highlights the implications of these approaches to assessing recession risk. . . .
The analysis herein highlights how financial, leading indicator, and other macroeconomic variables provide different signals regarding the risk of a recession. For example, the term spread does not suggest much risk of a recession over the near future, while the low level of unemployment and high level of inflation suggest a higher risk of a recession over the next year or two. Notably, a prediction model that includes the unemployment rate and inflation rate, in addition to the term spread, substantially lowers the role of the term spread in predicting recessions at one- or two-year horizons, suggesting that some of the predictive power of the term spread stems from the correlation of the term spread with the business cycle.
The differences across the approaches highlight the need for additional research. For example, the approach that combines inflation, unemployment, the credit spread, and the term spread nests the approach that only looks at the credit spread and the term spread. This nesting would imply that the larger model is superior if the sample of data is sufficiently large to provide reliable results. As highlighted in the figures, the sample period only includes seven episodes of large increases in unemployment, and econometric analysis with such limited historical experience is a challenge. Notably, all results herein are based on estimation over the full sample period and do not reflect "real-time" estimates or predictions.