Reading the Evidence Without Overreading It
The evidence question is whether observed bond markets show the same shape-based pattern as the theory: do high-volatility conditions merely raise spreads, or do they make the spread–leverage curve steeper? This chapter focuses on how the Federal Reserve study constructs an empirical credit surface and what its data can support. The payoff for the reader is a calibrated judgment: the evidence is consistent with uncertainty concentrating pressure at higher leverage, but the measured curve is an approximation to theoretical LTV, not an assumption-free view of it.
What was measured and compared
The study uses corporate bond markets because bonds are an important source of nonfinancial corporate credit and because the data provide both spreads and characteristics of the bond issue and issuer. The initial sample covers 1998 through 2023 and comes from the constituents of widely followed investment-grade and high-yield bond indices. The final matched sample contains 11,774 bonds issued by 3,100 firms.
The sample is narrowed to make the spread–leverage relationship more comparable. It includes domestic nonfinancial firms, bonds with seven to ten years of contractual remaining maturity, and ratings from CCC− upward. Observations with option-adjusted spreads, or OAS, below 5 basis points or above 3,500 basis points are removed. OAS is useful here because it adjusts the spread measure for embedded bond options and makes bonds with different option features more comparable.
Ratings are divided into five groups: AAA to A−, BBB+ to BBB−, BB+ to BB−, B+ to B−, and CCC+ to CCC−. This separation is not a minor presentation choice. A bond’s rating already summarizes information about creditworthiness, so pooling all ratings could make the leverage relationship look different simply because the sample mixes fundamentally different risk groups. The study instead asks how spreads vary with leverage within each rating band.
The leverage measure is an empirical approximation rather than the theoretical LTV used in the model. It combines the book value of debt with balance-sheet information and the market value of equity. The resulting empirical leverage, or EL ratio, is intended to approximate the market value of debt divided by the market value of total assets, but it uses book debt because market values for other liabilities are difficult to obtain consistently.
The uncertainty comparison is organized around the CBOE Volatility Index, or VIX. The study defines an uncertainty-shock month as one in which average daily VIX is above 29.4, the 90th percentile of its distribution over 1997–2023. The remaining months form the baseline regime. This is an operational definition of high market volatility, not a complete definition of economic uncertainty. It gives the researchers a reproducible way to compare spread–leverage relationships in relatively calm and unusually volatile conditions.
The empirical surface is estimated separately for each rating group and each regime. Rather than assume that the relationship must be linear, the authors use a local-linear kernel estimator.