Suppose there is no recession in the United States until February 2023. What do the duration and credit spreads, the financial conditions index, the debt service ratios predict?
First, probit regression estimates, 1985-2023M02:
Remarks: Bold type indicates significance at 10% msl.
Term spread is 10-year-3-month Treasury spread, Short rate is 3-month Treasury yield, Credit spread is Excess Premium (EBP) from Gilchrist & Zakrajsek (AER, 2012) , the NFCI is the Chicago Fed’s national financial conditions index, the DSR is the debt service ratio.
Note that increasing the standard term spread and short rate with EBP increases the proportion of variance explained, and EBP comes in with an expected and significant sign. Alternatively, financial stress as measured by the NFCI comes in with the expected sign, and explains recessions even better. Finally, as argued Borio, Drehmann and Fan (J.Macro, 2020)the debt service ratio measurably adds to the explanatory power of the duration gap.
Interestingly, only the DSR survives as having statistical significance when the three additional financial indicators are added. EBP and NFCI are quite strongly correlated, while DSR is not particularly correlated with EBP and NFCI, so this result is not entirely surprising.
While the DSR adds considerably to the sample fit, it is interesting to consider what this specification implies for the prediction of recessions.
Figure 1: Probability of recession estimated from duration gap plus short rate (bold blue), duration and credit gap plus short rate (tan), duration gap plus short rate and NFCI (green), duration spread plus short rate and DSR (red), and from term and credit spreads plus short rate, NFCI and DSR (chartreuse). The NBER has defined peak-to-trough recession dates as shaded. Source: Author’s calculations.
Note that the DSR data ends at 2022M12, so forecasts based on the DSR end at 2023M12.
Interestingly, any specification incorporating DSR imputes a significantly lower probability of recession than the corresponding specifications excluding DSR. For example, at the end of 2023, a short term plus rate spread model indicates a probability of 44%, while a model integrating all measures, including the DSR, displays a probability of 22%.