were taken from a smoothed yield curve estimated on a large sample of off-the-run
Treasury coupon securities using the technique proposed by Svensson (1994).11 The
resulting spread between corporate and Treasury securities, however, is distorted by
the differential tax treatment of corporate and government debt—corporate bond
coupons are subject to taxes at the state level whereas Treasury coupons are not.
Because investors compare returns across instruments on an after-tax basis, yields on
corporate bonds will be systematically higher than yields on government securities
to compensate for the payment of state taxes. Indeed, Elton et al. (2001) estimate
that, on average, these tax factors can account for as much as 20 percent of corporate
credit spreads.
We used the method proposed by Cooper and Davydenko (2002) to estimate the
distortionary effect of the state-level taxation on corporate bond spreads. According
to Elton et al. (2001), the relevant tax rate for the tax-adjusted spread between
corporate and government securities is given by τ = ts(1 - tg), where ts and tg are
the state and the federal tax rates, respectively. As suggested, we set τ equal to
4.875% and compute for each corporate security the portion of the spread due to
taxes according to
∆ yτ = — ln
tM
1 — τ
1 — τeχp(—rtMtM)_ ’
where tM is the corporate security’s maturity and rtM is the corresponding Treasury
coupon yield (see Cooper and Davydenko (2002) for further details). To calculate an
overall firm-specific credit spread, we averaged the tax-adjusted spreads on the firm’s
outstanding bonds, using the product of market values of bonds and their effective
durations as weights.12 We matched the firm-specific daily spreads to quarterly bal-
ance sheet information by averaging the daily spreads over the first month of the
quarter.13
11On-the-run securities were excluded from the sample because yields on those securities are
strongly influenced by liquidity premiums, which can affect the shape of the estimated yield curve
and, moreover, can shift the curve around the auction cycle.
12 The use of the dollar duration of bonds as a weight in computing the yield on a portfolio of
bonds represents a first-order Taylor series approximation to the portfolio yield; see Choi and Park
(2002) for details. Our results were virtually identical when portfolio spreads were averaged using
market values of bonds as weights only.
13That is, credit spreads matched to any first quarter of balance sheet data are averages of the daily
spreads in January, spreads during the second quarter are averages of the daily spreads during April,
and so on. We also converted daily spreads to a quarterly frequency by averaging over the entire
quarter. All of the results reported in this paper were robust to this alternative timing assumption.
13