Credit Spreads and Business Cycle Fluctuations

The paper asks what information corporate bond credit spreads carry about future economic activity, and specifically what component of those spreads carries that information: variation in expected defaults, or something else. It does so by constructing a new, more precise credit spread index (the “GZ spread”) from micro-level secondary-market bond prices and then decomposing it into a default-risk component and a residual “excess bond premium” (EBP), asking which component actually drives the spread’s forecasting power for real activity and the size of financial shocks in a VAR (Introduction, pp. 1692-1694).

For each bond, the authors compute a credit spread as the yield on the bond minus the yield of a hypothetical Treasury security with exactly the same cash flows (discounting the bond’s own cash flows at zero-coupon Treasury yields from Gürkaynak, Sack, and Wright 2007), rather than comparing to a Treasury security of the same stated maturity. This avoids the “duration mismatch” that biases conventional indexes when a bond’s effective duration differs from its nominal maturity (pp. 1694-1695). The GZ spread is then the simple average of these bond-level spreads across all bond-firm observations each month (Eq. 1, p. 1696), computed from 346,126 bond-month observations covering 5,982 securities and 1,112 US nonfinancial firms, January 1973-September 2010 (Table 1, p. 1695). The resulting series is considerably more volatile (SD ~= 100 bp) than the Baa-Aaa spread (SD ~= 50 bp) or the paper-bill spread (SD ~= 67 bp), and only modestly correlated with either (0.38 and -0.17 respectively) (pp. 1696-1697).

The GZ spread is “statistically a highly significant predictor of all three measures of economic activity at both the short- and longer-term horizons,” unlike the Baa-Aaa spread (marginal, “economically negligible” at 12 months for unemployment) or the paper-bill spread (predicts all three but with “only a small increase in adjusted R-squared”) (Table 2, pp. 1698-1699). At the 12-month horizon the GZ spread raises adjusted R-squared by 12-15 percentage points over the baseline, and a 100 bp increase predicts industrial production growth about 3 pp lower (annualized) over the next three months. For real GDP, a 100 bp increase in the GZ spread predicts growth more than 1.25 pp lower over the subsequent four quarters (coefficient -0.482, t = 5.74), again outperforming Baa-Aaa and paper-bill, the latter significant only at the one-quarter horizon (Table 3, pp. 1699-1700).

They regress each bond’s log credit spread on the issuing firm’s distance to default (DD) plus bond characteristics (duration, par amount, coupon, age, callability, and for callable bonds also Treasury curve level/slope/curvature and realized yield volatility), with industry and credit-rating fixed effects (Eq. 3, pp. 1700-1704). Distance to default is estimated from a Merton (1974) contingent-claims model solved iteratively via the Bharath-Shumway (2008) procedure across the full Compustat-CRSP universe (14,458 firms); it is strongly procyclical, reaching record lows in autumn 2008 (Eqs. 4-8, Figure 2, pp. 1701-1703). The predicted GZ spread is the fitted value from this regression averaged across bonds; the excess bond premium (EBP) is defined as the actual GZ spread minus this predicted spread – the average pricing error left over after controlling for measured default risk (p. 1701). The EBP was low in 2004-2006 (“lax credit standards, excessive credit growth”), rose from summer 2007, and hit a record high of 275 bp in October 2008 before declining (Figure 4, pp. 1707-1708).

The excess bond premium, not the predicted (default-risk) component, accounts for most of the GZ spread’s forecasting power, and this becomes even starker after 1985. Over the full sample both components have significant independent forecasting power for monthly and quarterly activity measures, but the EBP’s coefficients are consistently larger in magnitude (e.g., 12-month payroll employment coefficient of -0.369 [t=14.5] for EBP vs. -0.355 [t=9.63] for predicted spread; Table 6, p. 1708-1710). In the post-1985 subsample, the predicted GZ spread has no forecasting power for real GDP at all (coefficient -0.023, t = 0.20), while the EBP remains highly significant (t = 6.80), with a 100 bp EBP increase predicting about a 2 pp GDP decline over four quarters (Table 7, Panel B, p. 1711). The authors conclude: “Since the mid-1980s, most of the predictive content of the GZ credit spread for economic activity can be attributed to variation in the excess bond premium rather than to variation in default risk” (p. 1711).

In a recursively ordered, 2-lag, 8-variable quarterly VAR (consumption, business fixed investment, real GDP, GDP-deflator inflation, EBP, equity excess return, the 10-year Treasury yield, and the federal funds rate, identified so that EBP shocks affect real activity and inflation only with a lag), a one-standard-deviation EBP shock (about 20 bp) produces a decline in output that bottoms roughly 0.5 percentage points below trend five quarters after the shock, alongside a more severe and persistent fall in investment, a significant and persistent fall in consumption, appreciable disinflation, a cumulative stock market decline of about 7.0 percentage points, and an easing of the federal funds rate beginning about one quarter after the shock (Section IV.B, Figure 5, pp. 1711-1714). In variance decompositions, EBP shocks account for more than 10 percent of output variance and more than 25 percent of business-fixed-investment variance at business-cycle frequencies – “proportions that exceed the amount of variation typically explained by monetary policy shocks” (Figure 6, p. 1712).

The authors interpret the EBP as reflecting cyclical variation in the effective risk-bearing capacity of the financial sector – the price investors demand for bearing corporate credit risk exposure, above and beyond what measured default risk justifies – rather than as a pure default-risk signal (p. 1701). This interpretation is explicitly linked to financial-accelerator theory (Bernanke and Gertler 1989; Kiyotaki and Moore 1997; Bernanke, Gertler, and Gilchrist 1999) and intermediary asset-pricing theory (He and Krishnamurthy 2010; Adrian, Moench, and Shin 2010a, 2010b), under which a reduction in intermediaries’ risk-bearing capacity raises the price of credit risk and tightens the supply of credit even absent any change in borrowers’ fundamental default probabilities (Section IV.C, pp. 1713-1714).

The EBP is “strikingly” highly correlated with the Senior Loan Officer Opinion Survey (SLOOS) measure of the net percentage of banks tightening commercial and industrial lending standards, especially after 2000, and is inversely related to financial-sector return on assets – periods of high intermediary profitability coincide with a low EBP (Figure 7, p. 1714). The EBP also tracks the average one-year CDS spread of broker-dealers closely over 2003-2010, including around the September 2008 Lehman Brothers collapse (Figure 8, pp. 1715-1716). In a five-variable monthly VAR over broker-dealer variables (VIX, broker-dealer excess return, one-year and five-year broker-dealer CDS spreads, and the EBP, with a September 2008 dummy), an adverse shock to broker-dealer profitability produces an immediate rise in CDS spreads and “a sustained increase in the excess bond premium,” with the EBP’s response tracking the one-year CDS spread response closely – evidence the authors read as showing that “increases in risk aversion lead to a decline in asset prices and a contraction in the supply of credit, both through the corporate bond market and the broader commercial banking sector” (Figure 9, p. 1717).

The recursive VAR identification – assuming EBP shocks affect macro variables only with a lag – is an application-specific identifying assumption not validated by formal tests, though the authors report similar results ordering the EBP first in a “fast-moving” financial block and at monthly frequency (fn. 12, p. 1711). About two-thirds of the sample bonds are callable, which embeds interest-rate sensitivity that complicates interpreting spreads as pure default-risk indicators; this is addressed, not fully eliminated, by controlling for term-structure level/slope/curvature and realized volatility (pp. 1704-1706). The forecasting regressions “do not establish causality from credit spreads to economic activity; they document predictive content” (p. 1697). The Merton distance-to-default framework assumes geometric Brownian motion for firm value and a simplified capital structure; the Bharath-Shumway iterative procedure mitigates but does not resolve all of its limitations (pp. 1701-1703). Finally, the sample ends in September 2010, so the paper does not analyze EBP behavior during the subsequent zero-lower-bound period.