Excess Loan Premium

Corporate Loan Spreads and Economic Activity∗
Anthony Saunders Alessandro Spina

Stern School of Business, New York University Copenhagen Business School
Daniel Streitz
Sascha Steffen Copenhagen Business School
Frankfurt School of Finance & Management Danish Finance Institute
November 14, 2020

Abstract
We study the predictive power of loan versus bond spreads for business cycle fluctu-
ations. Using a novel credit spread measure derived from the secondary loan market,
we show that loan market-based credit spreads have additional predictive power for
macroeconomic outcomes (such as employment and industrial production) compared
to bond spreads as well as other credit spreads and equity returns, both in the U.S.
and Europe. Differences in the composition of firms borrowing in loan or bond mar-
kets are important in understanding the differential predictive power of both credit
spreads. Industry specific loan spreads predict different industry cycles and can be
used to construct alternative weighting schemes which further improve the predictive
power of loan spreads.
JEL classification: E23, E44, G20
Keywords: Secondary loan market, Bonds, Credit supply, Real economic activity
∗
Corresponding author: Sascha Steffen, Frankfurt School of Finance & Management, Adickesallee 32-34.
60322 Frankfurt, Germany. E-mail: s.steffen@fs.de. We thank Klaus Adam, Ed Altman, Yakov Amihud, Gio-
vanni Dell’Ariccia, Tobias Berg, Nina Boyarchenko, Jennifer Carpenter, Itay Goldstein, Arpit Gupta, Kose
John, Toomas Laaritz, Yuearan Ma, Emanuel Moench, Holger Mueller, Cecilia Palatore, Carolin Pflueger,
Martin Oehmke, Moritz Schularick, Marti Subrahmanyam, Elu von Thadden, Olivier Wang, Michael Weber
and seminar participants at Bundesbank, Cass Business School, Frankfurt School, IMF, NYU, Science Po,
WU Vienna and participants at the Bonn Macro Workshop, the Danish Finance Institute Annual Confer-
ence, the Mannheim Banking Workshop, the 2019 Santiago Finance Workshop, the Villanova Webinar in
Financial Intermediation, and the Regulating Financial Markets conference for many helpful suggestions.
Spina and Streitz gratefully acknowledge support from the Center for Financial Frictions (FRIC), grant
no. DNRF102. Streitz gratefully acknowledges support from the Danish Finance Institute (DFI). Steffen
gratefully acknowledges support from the German Science Foundation (DFG), grant number STE 1836/4-1.
Electronic copy available at: https://ssrn.com/abstract=3717358

1 Introduction
Credit spreads are widely used to forecast the business cycle (see, among others, Bernanke,
1990; Friedman and Kuttner, 1992, 1993a; Gertler and Lown, 1999; Gilchrist and Zakrajšek,
2012; López-Salido et al., 2017). This is typically motivated by financial accelerator type
mechanisms or credit channel theories (e.g. Bernanke and Gertler, 1989; Kiyotaki and Moore,
1997). These theories focus on the role of financial frictions – reflected in credit spreads – in
affecting investment and output decisions of firms. Hence, credit spreads can act as signals,
i.e., leading indicators of real economic activity.
The theoretical literature focuses on bank loan supply contractions in propagating and
amplifying shocks to the economy. However, the empirical literature generally relies on
credit spreads derived from the corporate bond market likely because corporate loan prices
have, until recently, not been available. Hence, an implicit assumption in most studies is
that frictions are reflected across corporate debt markets. For example, López-Salido et al.
(2017) argue that “we have in mind that the pricing of credit risk in the bond market is [...]
linked to the pricing of credit risk in the banking system. Although the former is easier for
us to measure empirically, we suspect that the latter may be as or more important in terms
of economic impact” (p. 1398).
We introduce a novel loan market-based credit spread to predict economic outcomes.

Over the last 30 years, a liquid secondary market for syndicated corporate loans has de-
veloped (the annual trading volume reached $742 billion in 2019), enabling us to construct
a novel bottom-up credit spread measure based on granular data from secondary market
pricing information for about 9,100 individual loans to U.S. non-financial firms over the
November 1999 to March 2020 period. Importantly, this market mainly comprises of firms
that do not have access to public debt or equity markets and for which financial frictions are
an important driver of the cost of external finance (Bernanke and Gertler, 1989; Holmström
and Tirole, 1997).
Our main finding is that loan spreads contain information about the future business cycle
1
above and beyond other credit spread indicators. To motivate this result, Figure 1 Panel
A, shows the development of our loan market credit spread measure as well as a corporate
bond market credit spread measure over the 2018 to 2019 period. The figure highlights that
during the 2019 “late-cycle phase”, the loan spread was gradually increasing at the same
time as growth in industrial production had begun to cool off. The bond spread, in contrast,
was not yet signalling any deteriotation in the health of the macroeconomy.
We rigorously scrutinize this effect by means of predictive regressions over the entire
20-year sample period. Figure 1 Panel B summarizes our main results. The details, data
description, and formal statistical analysis are all left to the main text. The figure documents
the predictive power of our monthly loan spread measure for three-month ahead industrial
production in the U.S., benchmarked against a bond credit spread measure. In Panel B1,
the standardized coefficient implies that a one standard deviation (SD) increase in the loan
spread is associated with a 0.410 SD decrease in industrial production over the subsquent
three months, whereas the economic magnitude of the bond spread coefficient is half of this
effect. This result prevails if loan and bond spreads are jointly included in the model. Panel
B2 highlights that the loan spread measure yields a sizable improvement in the in-sample fit,
with a R2 increase of about +15 percentage points (p.p.) relative to a baseline prediction
model (with no credit spreads). Later sections give the same result using different economic
aggregates and time horizons.
We provide a series of robustness tests. First, we compare our loan market measure to a
large range of alternative credit spread measures that have been used in the literature, such
as commercial paper – bill spreads, Baa-Aaa credit spreads, or high-yield corporate bond
spreads. Benchmarking our loan spread measure against a high-yield bond spread suggests
that the superior predictive power of the loan market cannot solely be attributed to the fact
that loans traded in the secondary market are of higher credit risk than bonds (the majority
of loan market borrowers who are rated have a BB or B rating). Second, we run a horse
race against predictors from the equity market, which are potentially more informationally
sensitive instruments. Third, we control for supply-demand conditions in secondary markets
2
Panel A. Industrial Production and Credit Spreads over 2018/19 Panel B1. Spread Coeffecient Panel B2. Incremental R2
0.2 0.2
10
0.0 0.0
Incremental R−Squared
4
Ind Prod YoY%
Spread %
Coeffecient
5 −0.2 −0.2
−0.4 −0.4
0 0
2018−01 2018−07 2019−01 2019−07 2020−01 Loan Bond Loan Bond
Figure 1: Motivating evidence and main results

Panel A plots the loan spread (red), bond spread (black) and YoY% in U.S. industrial
production from January 2018 to December 2019 (blue bars). See Section C for details
on the construction of the credit spread measures and underlying data sources. Panel B
compares the coefficients and incremental R2 of using loan spreads versus bond spreads to
forecast three-month ahead changes in industrial production over the December 1999 to
March 2020 period. Incremental R2 is the change in adjusted R2 that results from adding
the respective credit spread measure to a baseline prediction model that includes the term
spread, the federal funds rate, and lagged industrial production. See Table 1 for underlying
regressions, and Section 3 for a full discussion of the model.
using measures of loan market liquidity. Fourth, we take into account that loan and bond
contracts are structured differently. That is, differences in non-price loan terms (maturity,
collateral or covenants) and callability of loans vis-a-vis bonds could affect our results. We
regress loans spreads on a set of loan level characteristics to extract a loan spread orthogonal
to these characteristics. Fifth, we drop the financial crisis period (2007:Q4 – 2009:Q2) and
the predictive effect of the loan spread drops by approximately half, but remains significant.
The bond market credit spread, in contrast, becomes economically small and insignificant.
This is consistent with the interpretation that credit spreads in particular are good predictors
of “tail events” (Adrian et al., 2019). In all tests our main result remains unchanged.
A further potential objection to our result is that the sample period covers the 1999 to
2020 period and is thus relatively short to be making strong claims regarding the predictive
power of loan spreads for the business cycle. We collect loan and bond spreads as well as
3
economic outcome data for Germany, France, and Spain (some of Europe’s largest economies
for which we have sufficient secondary loan market data coverage) and run the same set of
tests as we did for the U.S. We again find that loan spreads provide additional information
to forecast manufacturing and consumption good production as well as the unemployment
rate compared to other credit spread measures. Overall, outside the U.S. and in arguably
more bank-dependent countries (which exhibit differential cycles over the last 20 years), we
document the same patterns.
After having documented that loan market credit spreads appear to have predictive power
above and beyond other credit spread measures, we next examine potential explanations
for why this might be the case. One important feature of the loan market is that they
are populated with firms that may have limited access to alternative funding sources and
exhibit a higher sensitivity to bank loan supply contractions. For example, more than 80%
of borrowers in the bond market have a credit rating of BBB or higher, while the majority of
loan market borrowers who are rated have a BB or B rating, while others are private firms
with no public rating. Of our entire sample, only 57% are loans to publicly traded firms.
Thus, there is a limited overlap between bond and loan borrowers. Consequently, a repricing
of risk by banks in the loan market might have implications for the overall economy that are
not perfectly reflected by investors in bonds.
Several pieces of evidence support the conjecture that the more robust predictive power
of loan over other credit spreads reflects the differential type of firms in the loan vis-a-vis
bond and equity markets. First, we show that within the loan market, it is the spread of
smaller, younger, and private firms which drives a substantial portion of the loan spread’s
predictive power. Small, young, and private firms are more likely to be financially constrained
(Hadlock and Pierce, 2010), face more severe informational frictions that may add to the
costs of external finance (Gertler and Gilchrist, 1994), and are more likely to borrow using
collateral (Lian and Ma, 2020), i.e., are more dependent on bank financing. That is, these
borrowers are presumably more affected when credit market conditions tighten because of
4
a lack of alternative funding sources, which eventually feeds into the real economy.1 In
particular among the group of small, young, and private firms the overlap between the loan
and bond market is limited. For instance, in our loan sample only 16% of smaller and younger
borrowers also have a bond outstanding, compared to 39% for larger and older borrowers.
Second, we provide evidence consistent with the conjecture that loan market spreads
better reflect credit supply conditions in the primary market than bond market spreads.
To that end, we use information from the Federal Reserve’s quarterly Senior Loan Officer
Opinion Survey on Bank Lending Practices (SLOOS) on changes in credit conditions for
commercial and industrial (C&I) loans at U.S. banks and alternatively a measure of bank
undrawn commitments. We use equivalent information from the European Central Bank’s
(ECB) Bank lending survey (BLS) for Germany, France, and Spain (i.e., the European
countries for which we are able to construct our loan market credit spread). Our evidence
suggests that loan spreads are more strongly correlated with changes in credit standards
compared to bond spreads, both for the U.S. as well as for European countries. This supports
the view that loan spreads better reflect credit supply conditions in the primary loan market
compared to other credit spread measures.
Third, we construct loan and bond spreads on an industry rather than an economy-wide
level classifying U.S. firms into industries using the Bureau of Economic Analaysis (BEA)
sector definitions. We show that industry-specific loan spreads have significant forecasting
power for industry-level production and employment, controlling for any economy-wide fac-
tors through time fixed effects. More importantly, we look at the predictive power of loan
spreads across industries and find a large degree of hetrogeneity in the ability of industry
level spreads to predict industry-level macro variables. Consistent with our finding at the
economy-wide level that smaller, younger, and private firms account for most of the pre-
dictive power of the loan market credit spread, we also find at the sectorial level that loan
1
Cloyne et al. (2020), for instance, provide evidence that younger firms’ investment behaviour responds
more strongly to changes in market interest rates compared to older firms. Begenau and Salomao (2019)
also explore the financing patterns of small and large firms over the business cycle and find smaller firms
financing policy is procyclical due to financial constraints. Pflueger et al. (2020) show that a measure
for the price of volatile firms, which exhibit a behavior similar to private firms, is related to future
macroeconomic activity.
5
spreads have greater predictive power in industries with firms that are more dependent on ex-
ternal finance (Rajan and Zingales, 1998). Overall, compositional differences between firms
who borrow in the loan versus bond market are important in understanding the differential
predictive power of loan and bond spreads.
In a final set of tests, we investigate whether we can further improve the predictive power
of loan spreads by altering the way we aggregate individual loan spreads. Until now, the
literature (including our paper) has used a simple average to construct aggregate spreads.
We show that an aggregate loan spread that puts more weight on industries in which the
loan spread has a higher predictive power increases the in-sample fit by an additional +3 p.p.
relative to the baseline (i.e., unweighted) loan spread measure. That is, industries in which
loan spreads have a higher predictive power also contribute more to the aggregate forecasting
power of loan spreads. A similar improvement can be obtained by assigning more weight to
industries that comprise of firms that are more sensitive to external financing frictions.
To summarize, we provide robust evidence that our loan-market based credit spread has
more predictive power for macroeconomic outcomes vis-a-vis the bond spread (and other
credit spread measures as well as equity returns). Importantly, there are reasons to believe
that the predictive power of the bond spread might even decline in the future. At the start
of the COVID-19 pandemic in March 2020, the Federal Reserve (Fed) responded to a freeze
in the corporate bond market – as reflected in a substantial increase in the bond spread –
and started buying corporate, investment-grade rated bonds (Gilchrist et al., 2020). That
is, the Fed started to directly target corporate bond markets. Bond spreads have declined,
while loan spreads remain highly elevated signaling substantial downside risk to the overall
economy. It is thus unclear whether the bond spread will continue to predict economic
activity in the future.
Related Literature: There exists a long history of research that examines the power of
financial market prices to predict macroeconomic outcomes and financial crises. Previous
research has focused on stock and bond markets (Harvey, 1989), commercial paper spreads
(Bernanke, 1990; Friedman and Kuttner, 1993b), the slope of the yield curve (Estrella and
6
Hardouvelis, 1991), high yield bonds (Gertler and Lown, 1999), corporate bond credit spreads
(Gilchrist and Zakrajšek, 2012; Krishnamurthy and Muir, 2020; López-Salido et al., 2017;
Philippon, 2009; Mueller, 2009), composite financial cycle indices (Borio et al., 2020), and
mutual fund flows (Ben-Rephael et al., 2020). We introduce a novel credit spread measure
derived from the syndicated loan market and explore if loan market based credit spreads can
offer additional information for understanding business cycles fluctuations. While we focus
on credit spreads, there is also a related broad empirical literature on the implications of
credit quantities, i.e., aggregate amount of credit within the banking system, for credit cycles
using cross-country level (Schularick and Tyler, 2012; Jordà et al., 2013), bank level data
(Baron and Xiong, 2017), and data for large (Ivashina and Scharfstein, 2010; Chodorow-
Reich, 2014), and small firms (Greenstone et al., 2020; Giroud and Müller, 2018).
Our paper also relates to the literature on the role of financial frictions in the generation,
propagation and amplification of shocks through the economy as discussed in Bernanke and
Gertler (1989), Gertler and Gilchrist (1994) and Holmström and Tirole (1997). A related
literature explores the role of sentiment as a potential driver of credit spreads, see Bordalo
et al. (2018), López-Salido et al. (2017). In our results we do not distinguish between which
of these mechanisms drives variation in the loan spread and instead focus on it’s use as a
tool for prediction.
Finally, we contribute to the strand of literature which highlights the importance of

smaller, younger, and private firms in the economy. Asker et al. (2015) show that private
firms invest more than public firms holding firm size, industry, and investment oppurtunities
constant. Davis et al. (2006) highlight that private firms show a greater volatility and
dispersion of growth rates than public firms. Pflueger et al. (2020) introduce a measure
of risk perceptions based on the price of volatile firms, which behave similar to private
firms, and provide evidence that the measure helps understand macroeconomic flucuations.
Cloyne et al. (2020) provide evidence that younger firms’ investment behavior responds more
strongly to changes in market interest rates compared to older firms. Begenau and Salomao
(2019) also explore the financing patterns of small and large firms over the business cycle
7
and find smaller firms financing policy is procyclical due to financial constraints. We add to
this literature and show that loan spreads may be a more accurate measure of the cost of
credit for more constrained, private firms, which may be especially prescient in forecasting
macro outcomes.2
Our paper proceeds as follows. In Section C, we provide details on the construction of

the loan spread measure. We present the baseline results in Section 3 and provide robustness
tests as well as evidence from Europe. We explore the mechanisms in Section 4. Section 5
shows results from industry-level tests. We introduce a new weighting scheme in Section 6.
Section 7 concludes.
2 Constructing the loan credit spread measure
Over the last two decades, the U.S. secondary market for corporate loans has developed
into an active and liquid dealer-driven market, where loans are traded much like other debt
securities that trade in over-the-counter (OTC) markets. This allows observing daily price
quotes for private claims, i.e., claims that are not public securities under U.S. securities
law and hence can be traded by institutions such as banks legally in possession of material
non-public information (Taylor and Sansone, 2006).
A nascent secondary market emerged in the 1980’s but it was not until the founding of
the Loan Syndication and Trading Association (LSTA) in 1995, which standardized loan
market contracts and procedures, that the market began to flourish (Thomas and Wang,
2004). In 2019 the annual secondary market trading volume reached $742 billion USD (see
Figure 2).
The majority of loans traded in the secondary market are syndicated corporate loans, i.e.,
2
Our paper also relates to the literature studying secondary loan markets. Altman et al. (2010) and Gande
and Saunders (2012) argue that banks still have an advantage as information provider even when previous
loans of borrowers have been traded on the secondary loan market. Drucker and Puri (2009) document
better access to credit and lower cost of debt for firms with traded loans. Consistent with the forecasting
power of loan spreads documented in this paper, Addoum and Murfin (2020) show that traders can use
infromation from secondary loan market prices to generate positive alpha.
8
loans that are issued to a borrower jointly by multiple financial institutions under one lending
contract. The syndicated loan market is one of the most important sources of private debt
for corporations. For example, about 69% of non-financial firms in the Compustat North
America database are active syndicated loan issuers during the 1999 to 2020 period and the
annual primary market issuance volume in the U.S. exceeded that of public debt and equity
as early as 2005 (Sufi, 2007). Both public and (larger) private firms rely on syndicated loans.
Of our entire sample, described in detail below, about 40% are loans to private firms.
Data: Our analysis utilises a novel dataset comprising of daily secondary market quotes
for corporate loans that trade in the OTC market, which we obtain from the LSTA. Our
sample spans the December 1999 to March 2020 period and contains 13,221 loans issued by
U.S. non-financial firms. Loan sales are usually structured as “assignments”,3 and investors
trade through dealer desks at the large underwriting banks. The LSTA receives bid and ask
quotes, every day, from over 35 dealers that represent the loan trading desks of virtually
all major commercial and investment banks.4 These dealers and their quoted loan prices
represent over 80% of the secondary market trading in syndicated loans. Furthermore it has
been established that loan price quotes provide an accurate representation of secondary loan
market prices for large corporate loans (Berndt and Gupta, 2009).5
We exclude credit lines and special loan types (around 1,703 loans), i.e., restrict our
sample to term loans.6 Term loans are fully funded at origination and are typically repaid
mostly at maturity, i.e., have a cashflow structure similar to bonds. Further, we require
3
In an assignment, the buyer becomes a direct signatory to the loan. Assignments make trading easier
as the loan ownership is assigned, or “transferred”, from seller to buyer. In contrast, in a participation
agreement the lender retains official ownership of the loan.
4
Investors usually trade through the dealer desks at large loan issuing banks. There is little public infor-
mation about dealers who provide quotes that are collected by LSTA. However, we know the identities
of the dealer banks for all loans in 2009. In Table A.5 of the Online Appendix we show that the top 25
dealers account for more than 90% of all quotes. We rank dealers according to their market share in the
secondary loan market and as loan underwriter in the primary loan market and find a correlation of 0.87.
5
We choose to focus our analysis on secondary market credit spreads instead of a primary market spread,
e.g. AISD. Primary market loan spreads may simply reflect the endogenous changes to the composition
of issuers over time. For example, during an economic downturn only the highest rated borrowers may be
able to access the loan market.
6
The vast majority of loans that are traded in the secondary market are term loans, as (non-bank) insti-
tutional investors typically dislike the uncertain cash flow structure of credit lines (Gatev and Strahan,
2009, 2006).
9
that loans can be linked to the DealScan database and restrict the sample to loans with a
remaining maturity of at least one year, resulting in a final sample of around 9,095 term
loans.
As we use monthly measures of economic activity in our forecasting regressions, we

rely on monthly mid quotes. That is, for each loan-month we take the average mid quote
across all trading days in the month. This results in about 302,223 loan-month obervations.
On average, our sample comprises around 1,219 outstanding loans per month (min ∼330;
maximum ∼2,293).7
We complement the LSTA pricing data with information about the structure of the
underlying loans from the DealScan database. The databases are merged using the LIN, if
feasable, or else a combination of the borrower name, dates, and available loan characteristics.
DealScan contains information on maturity and scheduled interest payments as of origination,
which are key inputs used to determine our credit spread measure, as described below. Table
A.1 of the Online Appendix contains a full list of the variables used and their sources.
Methodology: To examine the predictive power of loan credit spreads, we use a bottom-
up methodology similar to Gilchrist and Zakrajšek (2012). In contrast to bonds, loans do
not carry a fixed coupon but are floating rate debt instruments based on an interest rate,
typically the three-month LIBOR, plus a fixed spread. Therefore, to construct the sequence
of future cash flows for each term loan, we use the three-month LIBOR forward curve and
the spread obtained from DealScan to estimate projected cash flows. In particular, we add
the three-month forward LIBOR rate for the respective period to the term loan’s fixed all-
in-spread-drawn (AISD). The AISD comprises of the spread over the benchmark rate and
the facility fee, and has been shown to be an adequate measure for the pricing of term loans
7
Figure A.1 in the Online Appendix provides information on liquidity in the secondary loan market over
time. The median bid-ask spread in the 1999 to 2020 period was 81 basis points (bps). For comparison,
Feldhütter and Poulsen (2018) report an average bid-ask spread for the U.S. bond market of 34 bps over
the 2002 to 2015 period. This suggests that while the secondary loan market has become an increasingly
liquid market, it is still somewhat less liquid than the bond market.
10
(Berg et al., 2016, 2017). We assume that cash flows are paid quarterly.8 Let Pit [k] be the
price of loan k issued by firm i in period t promising a series of cash flows C(S). Using this
information we calculate the implied yield to maturity, yit [k], for each loan and each period.
To avoid a “duration mismatch” in the calculation of the spread, for each term loan we
construct a synthetic risk-free security that has exactly the same cash payment profile as
the loan. Let Pitf [k] be the “risk-free equivalent price” of loan k. Pitf [k] is defined as the
sum of the projected cash flows discounted using the continuously compounded zero-coupon
Treasury yields from Gürkaynak et al. (2007). From this price we extract a synthetic risk-
free equivalent yield to maturity, yitf [k]. The loan spread Sit [k] is defined as the difference
between the loan’s implied yield to maturity and its risk-free equivalent yield to maturity. To
ensure the results are not driven by outliers, all loan-month observations with loan spreads
below five bps and above 3,500 bps as well as observations with a remaining maturity below
12 months are excluded.
We take a monthly arithmetic average of all secondary market loan spreads, to create
a loan spread index StLoan . Whilst a variety of alternative weighting mechanisms could be
adopted, we stick to the method used by Gilchrist and Zakrajšek (2012) to minimize any
chance of data mining and to ensure comparability to the existing literature. We discuss
alternative weighting schemes in later sections. Specifically, the loan spread is defined as:
1 XX
S Loan
t = Sit [k], (1)
Nt i k
Figure 4 plots our estimated loan spread as well as the bond credit spread measure by
Gilchrist and Zakrajšek (2012) over time.9 The loan spread and bond spread follow a similar
pattern over time, with sharp movements around the 2001 recession, the 2008/2009 financial
8
We use the same interest period for all loans, as information on the loan-specific interest period is often
missing in the DealScan database. However, in a sub-sample of term loans to U.S. non-financial firms for
which the interest period is reported in DealScan, interest is paid on a quarterly basis for over 70% of
loans.
9
The bond spread measure is provided by Favara et al. (2016), which is an updated version (i.e., avail-
able also for more recent periods) of the bond spread measure by Gilchrist and Zakrajšek (2012). See
https://www.federalreserve.gov/econresdata/notes/feds-notes/2016/files/ebp csv.csv for details.
11
crisis, and at the beginning of the COVID-19 pandemic. The correlation between the loan
spread and bond spread is high but not perfect (0.76 over the entire sample period and
0.65 when excluding the 2008-09 crisis period). In our empirical tests we use spread changes,
which substantially reduces this correlation to 0.45 or 0.23 excluding the 2008-09 crisis period
(details are provided in the following section). Further, the loan spread is significantly more
volatility with a standard deviation (SD) of 2.4% (versus 1.0% for the bond spread) and
an order of magnitude higher than the bond spread. This is consistent with the syndicated
loan market containing a wider universe of borrowers and especially including more lower
credit quality borrowers such as private firms who cannot access public bond markets.10 A
full summary of descriptive statistics for all variables is available in Table A2 of the Online
Appendix.
3 Loan spreads and economic activity
3.1 Baseline results
In this section we start out by examining whether loan spreads contain information that is
useful for predicting aggregate economic variables. We build on López-Salido et al. (2017)
and run forecasting regressions of the following form:
∆y t+h = α + β∆y t−1 + γ∆St + λT S + φRF F + t+h , (2)
where h is the forecast horizon and ∆y is the log growth rate for a measure of economic
activity from t − 1 to t + h. S is a credit spread index, and ∆St is the change in spread from
10
However, Schwert (2020) provides evidence that primary market loan spreads are higher than bond spreads
also in a sample of loans matched with bond spreads from the same firm on the same date and accounting
for other differences between loan and bond contracts.
12
t − 1 to t.11 In most tests we use the loan spread measure as defined in the previous section.
In some tests we benchmark the loan spread results against predictive regressions utilizing
other credit spread measures, such as a bond spread. We further include the term spread,
which is defined as the slope of the Treasury yield curve (i.e., the difference between the
ten-year constant-maturity Treasury yield and the three-month constant-maturity Treasury
yield), and the real effective federal funds rate.12 The regressions are estimated by ordinary
least squares (OLS), with one lag, i.e., economic activity from t − 2 to t − 1.13 Standard
errors are heteroskedasticity and autocorrelation corrected Newey-West standard errors with
a four period lag structure. The sample period covers the November 1999 to March 2020
period.
Table 1 shows the results using a forecast horizon of three months (h=3). To gauge the
contribution of the loan spread to the in-sample fit of the model, at the bottom of each
panel we report the incremental increase in adjusted R2 relative to a baseline model that
uses only the term spread, the real federal funds rate, and the lagged dependent variable.
Panel A of Table 1 reports the results using industrial production as the dependent variable.
Column (1) shows that a model including the loan spread can explain 31.3% of the variation
in changes in industrial production. This represents a sizable R2 increase of 15 percentage
points (p.p.) relative to the baseline model. The loan spread coefficient indicates that a
one SD increase in loan spread is associated with a decrease in three-month ahead industrial
production by 0.410 SD. In economic terms this equates to a 45bps increase in loan spreads
11
We follow López-Salido et al. (2017) and use changes rather than levels in our predictive regressions.
This can also be motivated by the framework provided by Krishnamurthy and Muir (2020) for diagnosing
financial crises. The forecasting power of spread changes can arise for two reasons. First, becasue the
asset side of bank balance sheets are sensitive to credit spreads, changes in spreads will be correlated with
bank losses. Second, increases in credit spreads reflect an increase in the cost of credit which impacts
investment decisions. Finally, first differencing accounts for non-stationary of the credit spread time series.
12
Aggregate macroeconomic data, i.e., monthly (non-farm private) payroll employment [NPPTTL], unem-
ployment rate [UNRATE], and (manufacturing) industrial production [IPMAN], are obtained from the
Federal Reserve’s FRED website. The term spread data comes from the ten-year Treasury constant ma-
turity minus three-month Treasury constant maturity data series [T10Y3MM] available via FRED. The
real effective federal funds rate is estimated using data from the Fed’s H.15 release [FEDFUNDS] and
realised inflation as measured by the core consumer price index less food and energy [CPILFESL].
13
In all specifications we hold the lag structure fixed to facilitate the comparison of R2 across models.
An AR(1) process, i.e., a one period lag structure, captures most of the peristence. However, including
additional lags up to 6 periods, or allowing for an optimal lag length selection based on the AIC leads to
very similiar results.
13
being associated with a 0.74% decrease in industrial producion over the next 3 months,
compared to an unconditional mean of 0.23% growth in industrial production.14
In column (2) we benchmark this result against a commonly used credit spread indicator
from the corporate bond market, the GZ spread (Gilchrist and Zakrajšek, 2012). The eco-
nomic magnitude of the bond spread coefficient is half that of the loan spread coefficient in
Column (1). In particular, a one SD increase in bond spread is associated with a decrease
in industrial production by 0.198 SD. Also the improvement in-sample fit due to the bond
spread is modest with a R2 increase of 3.5 p.p. from the baseline.
When we combine both spreads in one model in column (3), we find that the loan spread
coefficient and incremental R2 are almost unchanged compared to a model with the loan
spread only. In other words, while both bond and loan spreads have predictive power for
industrial production, the loan spread has additional forecasting power. A variance inflation
factor of below 1.5 of both loan and bond spreads suggests that the correlation between
both spreads is not affecting our results. The results remain consistent when we look at the
unemployment rate in Panel B and payroll employment in Panel C of Table 1.
To examine dynamics, we extend our prediction model to consider longer forecast horizons
in a local projections framework (Jordà, 2005). Figure 5 plots the coefficient and 95%
confidence intervals on the loan spread and bond spread at various forecasting horizons (one
to 12 months ahead) using each of our dependent variables. The loan and bond credit spread
models are estimated in separate regressions.
Focusing on industrial production (top-left panel), the predictive power of the loan spread
peaks around h=3, i.e., the loan spread today is most correlated with the growth in industrial
production three months from now, and then dissipates slowly. The bond spread, in contrast,
shows the largest predictive power at a horizon of about eight months and then the effect
levels out. However, even at longer forecasting horizons the loan spread shows a stonger
predictive power compared to the bond spread.
14
See Table A.2/A.3 of the Online Appendix for a full list of the unconditional moments for each variable.
14
3.2 Other credit spread and equity market measures
One potential explanation for why the loan spread possesses additional predictive power
relative to other credit spread measures is that the secondary loan market is populated by
a set of riskier borrowers than the bond market. Figure 3 highlights that more than 80% of
borrowers in the bond market have a credit rating of BBB or higher, while the majority of
loan market borrowers who are rated have a BB or B rating, while others are private firms
with no public rating. Firms with higher credit risk may be more exposed to financial frictions
and face a higher external finance premium. Hence, their credit spread may be particularly
suitable for forecasting economic developments (Gertler and Lown, 1999; Mueller, 2009). A
cleaner evaluation of the loan spread’s additional predictive power might thus be to compare
it to a corporate bond spread conditional on a set of riskier borrowers.
To that end, Table 2, Panel A, columns (2) and (3) use the Baa-Aaa credit spread and
a high-yield credit spread.15 The Baa-Aaa credit spread measures the spread between Aaa
rated corporate bonds and Baa rated corporate bonds and has been used among others
by Gertler and Lown (1999). The high yield corporate bond spread measures the spread
between high yield corporate bonds and AAA rated bonds. The results indicate that the
high-yield corporate bond spread has a somewhat larger predictive power compared to the
baseline bond spread. The coefficient on the Baa-Aaa spread (high-yield spread) is -0.277
(-0.248) and the incremental R2 is +8 p.p. (+6 p.p.). For comparison, the coefficient on
the baseline corporate bond spread is -0.198 and the incremental R2 is +3.5 p.p. [Table
1 column (2)]. For both spreads, however, the economic magnitude and contribution to
in-sample fit remains significantly below that of the loan spread [reported again in Table 2,
Panel A, column (1), as benchmark]. Table A.6 of the Online Appendix repeats Table 1 but
keeps the loan spread in each coloumn and the loan spread retains it significant predictive
power.
We provide additional tests in Table A.7 of the Online Appendix. In particular, we use
15
Baa-Aaa credit spread [BAA AAA] is obtained from Federal Reserve’s FRED website. The spread is
constructed by Moody’s and is based on bonds with 20 years to maturity and above. The high yield index
[BAMLH0A0HYM2EY] also comes from FRED is based on the ICE Bofa US high yield effective index.
15
bond level pricing data from TRACE to create bottom-up corporate bond credit spreads
(Gilchrist and Zakrajšek, 2012) for different rating categories: (i) A or higher, (ii) BBB, and
(iii) below BBB. We find a monotonic increase in the size of the coefficient as we condition
on a risker set of borrowers in the bond market. However, a bottom-up measure comprised
of non-investment grade bonds does not match the predictive power of the loan spread. A
one SD higher non-investment grade bond spread is associated with only a 0.222 SD decrease
in industrial production three months ahead and an incremental R2 of 4.9%. Overall, the
results indicate that while credit risk can explain some of the improvement in predictive
power between loan and bond credit spreads, it cannot account for the entire difference.
Column (4) in Table 2, Panel A, uses the commercial paper - bill spread to forecast three-
month ahead industrial production (see, among others, Friedman and Kuttner, 1993b, 1998;
Estrella and Mishkin, 1998). Commercial paper are unsecured, short-term debt instruments
issued by high credit quality corporations. During our sample period it shows no predictive
power and adds little to the model’s R2 .
Finally, informationally sensitive securities like equity may contain signals about the
development of the economy as well (see e.g. Greenwood et al., 2020; López-Salido et al.,
2017). In Table 2, Panel A, column (5) we use the monthly return of the S&P 500 index
to forecast industrial production. A one SD higher equity market return is associated with
a 0.216 SD increase in industrial production three months ahead and the incremental R2 of
including the equity market return in the prediction model is +4.1 p.p. Again, this is well
below the effect we document for the loan market credit spread [Table 2, Panel A, column
(1)].
3.3 Robustness
This section further discusses the robustness of our main result. First, we control for supply-
demand conditions in the secondary market by using a measure of loan market liquidity
(plotted in Figure A.1 of the Online Appendix). In particular, in Table 2, Panel B, column
16
(2) we include the contemperaneous median bid-ask spread as an additional control. Our
main result remains unchanged. Second, loan and bond contracts might be different with
respect to non-price contract terms. We regress loan spreads on various characteristics and
take the residual spread (see Online Appendix for details). Column (3) in Table 2, Panel B,
uses this “residual loan spread” and finds very little difference in predictive power relative
to the baseline loan spread. Third, results may be exclusively driven by the large changes in
spreads during the 2008-09 financial crisis. Table 2, Panel B, columns (4) and (5) show that
the predictive power of the bond spread becomes economically small and insignificant when
excluding the financial crisis. The predictive effect of the loan spread drops by approximately
half, but remains significant. That is, aggregate loan and, particularly, bond spreads perform
weaker outside of financial crisis periods consistent with the interpretation that credit spreads
perform better as predictors of “tail events” (Adrian et al., 2019).
We conduct a series of additional tests, which are relegated to – and discussed in more
detail in – the Online Appendix, including: i) The baseline forecasting regression assumes
that increases and decreases of spreads have the same relationship with future economic
activity. We test for potential asymmetric impacts (Stein, 2014) and find that both increases
and decreases in loan spreads are correlated with future economic developments, with the
effect of spread increases being somewhat stronger compared to decreases. ii) We test the
predictive ability of loan and bond spreads on pseudo out-of-sample data and continue to
find a higher predictive power for loan versus bond spreads. iii) We follow the methodology
proposed by Gilchrist and Zakrajšek (2012) and decompose the loan spread into a default
risk component and a residual component (“excess loan premium”). We find that both
components have predictive power.
3.4 Evidence from European countries
A time series of secondary market loan prices has only been available for about 20 years,
which is shorter than that of bonds and other debt instruments that often have a longer
data history. While we cannot extend the time-series backwards, we can perform similar
17
predicatbility tests using European data, i.e., exploit the fact that different countries have
different business cycles. We focus on three of Europe’s largest economies, Germany, France,
and Spain, as we have sufficient loan market data available to perform meaningful tests. We
construct the European loan spread following the methodology described in Section C.16 For
the aggregate bond spread, we use the spread provided by Mojon and Gilchrist (2016).
Figure 6 shows aggregate bond and loan spreads for Germany, France, and Spain. Similar
to the U.S., loan spreads are higher in levels compared to bond spreads consistent with
different types of firms issuing debt instruments in both markets. Aggregate bond spreads
also decrease following the 2008-2009 financial crisis, but remain elevated during the sovereign
debt crisis (but at a lower level compared to 2008-2009). Interestingly, absolute bond spreads
in and out of crises are substantially lower compared to U.S. bond spreads. This is consistent
with the interpretation that only the highest quality European firms can access public debt
markets and that European markets are more bank debt dominated. Aggregate loan spreads
increase during the financial crisis up to about 15%, a level comparable to the U.S. spread.
Similar to the U.S., loan spreads remain higher after both crises periods compared to the
period before 2008.
We run similar aggregate forecasting models as in the U.S. setting using a term spread
(defined as the 10-year benchmark Euro area government bond minus the three-month EU-
RIBOR rate) and the real EONIA (defined as EONIA minus HCIP inflation over the previous
12 months) and report the results in Table 3. Our dependent macro variables are the monthly
unemployment rate, manufacturing goods production, and consumption goods production.
We start with the three-month ahead forecasts of macro outcomes in Germany in Panel
A of Table 3. In our baseline model (not shown), we include only the term spread, real
EONIA, and lag of the dependent variable to match the set up of previous tables. In column
(1) we look at the predictive power of the aggregate bond spread and loan spread for the
16
We adjust the methodology by using equivalent EU variables. The term spread, i.e., the difference
between 10-year Euro area government bond (i.e., a GDP weighted average all Euro area government
bonds, Source: OECD’s MEI) and three-month EURIBOR (Source: ECB), and the real EONIA, i.e.,
the overnight rate (Source: ECB) minus realised inflation EURIBOR forward curves to calculate loan
cashflows and a different risk free rate.
18
manufacturing production index. Again the loan spread remains positive and significant
even controlling for the bond spread. The explanatory power increases by +11.1 p.p. above
the baseline model. Decomposing this R2 increase into a part that is due to the inclusion of
the loan spread and a part that is due to the inclusion of the bond spread, we find the 86%
of the increase in in-sample fit comes from the loan spread.
We find consistent results for an index of unemployment rate [column (2)] and an index
of construction activity [column (3)] and the results extend to France and Spain. Overall,
our evidence from the European market is consistent with the U.S. evidence. Loan spreads
have more predictive power for macroeconomic outcomes compared to bond spreads.
4 Exploring the mechanism
In the previous section, we showed that loan spreads have predictive power beyond other
commonly used measures. This result holds for a host of different macroeconomic outcome
variables, additional controls, different time horizons, and for different countries. In this
section we explore potential mechanisms that might explain this result. A testable hypothesis
is that the loan market is populated with firms that have limited access to alternative funding
sources and are more exposed to bank loan supply contractions. In addition, this set of firms
may be particulary sensitive to financial frictions, which makes them an important channel
for amplifying credit supply shocks.
We attempt to explore this hypothesis by (i) analyzing which set of firms account for
most of the predictive power of loan spreads and (ii) examining the link between loan spreads
and credit supply conditions in the economy.
4.1 Effect by firm size, age, and listing
One important feature of the loan market is that it is populated with firms that may have
limited access to alternative funding sources and exhibit a higher sensitivity to bank loan
19
supply contractions. For example, Figure 3 highlights that more than 80% of borrowers in
the bond market have a credit rating of BBB or higher, while the majority of loan market
borrowers who are rated have a BB or B rating, while others are private firms with no public
rating. Of our entire sample, only 57% are loans to publicly traded firms. Thus, there is a
limited overlap between bond and loan borrowers. Consequently, a repricing of risk by banks
in the loan market might have implications for the overall economy that are not perfectly
reflected by investors in bonds.
In this section we examine which types of borrowers account for most of the predicitve
power of loan spreads. As suggested above, the loan market may comprise of firm’s which are
more reliant on external financing but do not have access to public debt or equity markets.
This is specifically the case for small, young, and private firms, which are more likely to be
financially constrained (Hadlock and Pierce, 2010), face more severe informational frictions
that may add to the costs of external finance (Gertler and Gilchrist, 1994), and are more likely
to borrow using collateral (Lian and Ma, 2020), i.e., are more dependent on bank financing.
That is, these borrowers are presumably most affected when credit market conditions tighten
because of a lack of alternative funding sources, which eventually feeds into the real economy.
Table 4 performs the same aggregate forecasting regression as Table 1 for the U.S. econ-
omy, but includes loan spreads that are conditional on the size of the borrower, as measured
by total assets (Panel A), or the age of the borrower, as measured by length of time the firm
has financial information available in the Compustat North America dabase (Panel B). Panel
A, column (1) [column (2)] shows the aggregate loan spread for small (large) borrowers, i.e.,
total assets <= (>) median. The results are significantly stronger for small firms compared
to large firms (coefficent of -0.380 versus -0.260 and incremental R2 of +13.7 p.p. versus
+6.7 p.p.). Panel B provides consistent, albeit weaker, results splitting the firm sample by
firm age.
In addition to the size and age splits, Table 4 reports results using a loan spread measure
constructed from private firms, defined as firms that cannot be linked to Compustat.17 The
17
Hence, a size or age split cannot be performed for these firms. We report the private firm result throughout
Panels A-C, i.e., in all three panels, to facilitate a comparison with the other spread measures.
20
results indicate that the predictive power of a loan spread constructed based on private
firms is stronger even compared to small and young firms with a coefficent of -0.420 and an
incremental R2 of +15.7 p.p.
In Table 4 Panel C for those loan market borrowers for which we do have age and size
information available, we double sort firms by age and size buckets. Again the effect is
stronger for small and young versus old and large firms. The coefficent for the former group
is -0.390 and the incremental R2 +14.7 p.p. compared to a coefficient of -0.210 and an
incremental R2 of +3.7 p.p. for the latter group. Interestingly, the predictive power of large
and old firms is close to that of the bond spread measure [coefficent of -0.210 versus -0.19
and incremental R2 of +3.7 p.p. versus +3.5 p.p., cf. Table 1, colum (2)]. The predictive
power of small and young firms is close to that of private firms [coefficent of -0.390 versus
-0.420 and incremental R2 of +14.7 p.p. versus +15.7 p.p., cf. Table 1, colum (2)].
Taken together, the results suggest that the predictive power of the loan spread is stronger
for younger, smaller, and private borrowers who are more exposed to financial frictions.
Importantly, in particular among the group of small, young, and private firms the overlap
between the loan and bond market is limited. We match borrowers in the loan market with
borrowers in the bond market each year using company names in Mergent. Only 16% of
smaller and younger borrowers in our loan sample also have a bond outstanding, compared
to 39% for larger and older borrowers. This suggests that compositional differences may
explain (part of) the difference in predictive power of loan versus bond markets.
4.2 Effect by ratings
An alternative way to examine the role of financial frictions is to look at loan spreads
conditional on rating groups. Credit ratings are an alternative proxy that may capture
the riskiness of borrowers and their exposure to financial frictions. Loan level ratings are
sourced from Dealscan and Leveraged Commentary and Data (LCD). Table 5 performs the
same aggregate forecasting regression as Table 1, but sorts loans into four groups, BBB,
21
BB, B and below and a group for which no rating can be found. Column (1) highlights
that a loan spread derived from the highest rated loans, BBB, has no predictive power for
three-month ahead macroeconomic outcomes. This is consistent with the safest borrowers
being least exposed to bank loan supply contractions.
Column (2) and (3) show that as we condition on a riskier set of loans, the loan spread
increases in its predictive power. Column (4), which includes loans for which no loan rating
could be identified, shows a very similar pattern to loans rated B or below. Comparing to
the baseline results in Table 1, it appears most of the predictive power of the loan spread is
coming from loans rated B or below and loans with no available rating. These borrowers,
most likely private firms, are the type of firms for which we would expect financial frictions
to matter the most.
4.3 Credit supply
Higher loan spreads could reflect a general tightening of lending standards or a reduction
in loan supply by banks because of deteriorating bank balance sheets, which leads to an
increase in loan spreads and a subsequent reduction in economic activity. In this section we
examine whether loan spreads are associated with a tightening of financial conditions.
In Table 6 we regress a measure of aggregate bank lending standards on our loan spread
and benchmark the effect against the bond spread. The dependent variable in Panel A.1
is the aggregated measure for changing bank lending standards obtained from the Senior
Loan Officer Opinion Survey on Bank Lending Practices administered from the Board of
Governors of the Federal Reserve System (SLOOS). Specfically it is defined as the percentage
who respond “lending tightened”, less the percentage who responded that “lending eased”,
i.e., a net percentage. A higher SLOOS measure signals a tightening of lending standards.
The survey is conducted quarterly and reflects the credit conditions in the previous quarter.
In column (1) [column (2)] we regress the SLOOS indicator on the change in loan (bond)
spread also over the previous quarter – i.e., the contemporaneous relationship between credit
22
conditions and spreads is examined. The loan spread has a higher correlation with the
SLOOS indicator compared to the bond spread and a substantially higher R2 . A one SD
increase in loan spread over the quarter is associated with a 0.43% increase in the net
percentage indicating tighter lending conditions. Including both spreads in the same model
shows consistent results [column (3)]. In line with our prior results, the loan spread retains its
economic and statistical significance while the bond spread becomes small and insignificant.
The dependent variable in Panel A.2 is banks’ unused commitments (as % of total assets)
as a measure of bank credit supply. Banks might curtail their exposure at the beginning of
an economic downturn primarily by reducing the amount of undrawn commitments (Bassett
et al., 2014). In column (1) [column (2)] we regress the change in the undrawn commitments
over the previous quarter on the change in loan (bond) spread over the same quarter – i.e.,
a measure of the contemporaneous relationship. We include both spreads in column (3).
An increase in both loan and bond spreads decreases banks’ unused commitments in the
quarter ahead, but the R2 is higher in the loan spread regression and the coefficient of the
loan spread is higher both individually and collectively when the bond spread is included.
In Panel B of Table 6, we extend these results to Europe using the European Central
Bank’s (ECB)’ equivalent Bank Lending Survey (BLS). We see a similar pattern across all
three European countries, where the loan spread is more highly correlated with bank lending
standards than the bond spread.
Overall, these tests consistently show that loan spreads derived from secondary market
prices reflect supply effects in the primary loan market. These results are consistent with
the interpretation that the pricing of credit risk in the loan market is closely linked to the
supply of credit in the banking system.
5 Industry-level forecasting
In this section we show there is additional information to be captured by going beyond

aggregate spreads and looking at the cross-sectional heterogeneity in spreads. Bottom-up
23
credit spread measures create the ability to aggregate spreads not only at the economy-wide
but also at less aggregated levels, such as the industry level. There are several reasons for
studying the predictive power of credit spreads at disaggregated levels. First, it allows for
more nuanced tests as to the predictive power of credit spreads and economic aggregates.
Second, in cross sectional tests it is easier to shut down potential confounding factors that
may affect real outcomes but are correlated with credit spreads. Hence, one can isolate the
credit spread specific forecasting power more cleanly, as will become clear below. Third, by
exploiting variation in the cross section, we are able to study in which industries loan credit
spreads have greater predictive power. This can further our understanding as to why loan
spreads are informative.
5.1 Baseline results
Industry credit spread: To construct a loan spread measure at the industry level, we
classify U.S. firms into industries using the Bureau of Economic Analysis (BEA) sector
definitions, excluding financial and government owned firms. Industry level loan spreads,
Loan
Sbt , are constructed following Section C, but instead of aggregating across all firms in the
economy we now aggregate loan spreads using an arithmetic average across all firms in a
BEA sector b. We exclude industry-months with less than 5 loans. Overall, we construct
spreads for 11 distinct BEA sectors.18
Figure 7 plots industry loan spreads over time. Loan spreads are not perfectly correlated
across industries. For example, while the “Construction” and “Transportation” sectors expe-
rienced a significant spread increase during the 2008-09 crisis, this increase is less pronounced
for more stable sectors such as “Education and health care” and “Utilities”. Further, some
industries experienced industry-specific crisis periods. The “Mining” sector (which includes
volatile oil and gas companies), for instance, experienced a wave of defaults in 2015 fueled
by collapsing oil and metal prices, which is reflected in a spread increase that even surpassed
18
The “Agriculture, forestry, fishing, and hunting” and “Other services, except government” sectors are
excluded due to an insufficent number of observations.
24
the 2008-09 level. Figure 7 also highlights the heterogenous impact of COVID-19 across
industries, with exposed industries such as “Mining” and “Retail Trade” experiencing larger
spikes in spreads as the crisis unfolded.
Industry forecasting results: To assess the relationship between industry specific spreads
and industry specific macroeconomic variables, we use quarterly total employment and to-
tal establishment figures from the Bureau of Labour Statistic’s (BLS) Quarterly Census of
Employment and Wages (QCEW). In addition we use quarterly industry gross output from
the BEA’s industry accounts. This data is only available from Q1 2005 to 2019 Q4.19 The
baseline results are reported in Table 7.
The first column in Panel A starts with a model that includes the industry loan spread
in a pooled regression. Note that in contrast to the aggregate forecasting regressions, we
now include the loan spread level and not the change in the spread. This is because by later
including industry fixed effects we effectively run a demeaned regression, i.e., we capture
spread deviations from the industry mean. The dependent variable is the one quarter ahead
change in industry employment. Controlling for the aggregate loan spread, a one SD increase
in industry loan spreads is associated with a decrease in employment by 0.13 SD. The
incremental R2 is +8.6 p.p. In column (2), we include time fixed effects, which absorbs any
common time trends that affect all industries. In particular, this captures variables such
as aggregate credit spreads but also the stance of monetary policy, aggregate business cycle
fluctuations (such as the overall effect of the 2008-09 crisis), or overall regulatory changes.
Interestingly, industry specific loan spreads remain highly statistically and economically
significant. In fact, the coefficient remains hardly changed relative to column (1), indicating
that omitted aggregate variables do not bias the coefficient to a significant extent. This shows
that there is significant information contained in loan spreads that are not captured by other
aggregate economic factors. In column (3), we further include industry fixed effects to absorb
any time-invariant unobserved cross-industry differences. Again the statistical significance
19
The underlying macroeconomic data obtained from the BEA and BLS is not seasonally adjusted. To
ensure that any monthly seasonal variation does not interfere with our analysis, we use a seasonal trend
decomposition to remove any predictable monthly seasonal variation from the raw data. What remains
in the de-seasonalised macroeconomic data is any underlying time trend and residual component.
25
and economic magnitude of industry loan spreads remains virtually unchanged.20 We find
consistent results in Panels B and C using alternative industry specific economic outcomes
as dependent variables.
5.2 Across industry heterogeneity
Table 7 reveals that industry level loan spreads have predictive power for industry-specific
outcomes, above and beyond aggregate level information. The predictive power, however,
may vary across industries. As discussed in Section 4, the loan market comprises of firms
that may have limited access to alternative funding sources and exhibit a higher sensitivity to
bank loan supply contractions. Hence, loan spreads may have predictive power in particular
in industries that comprise of firms that are more dependent on external finance.
The top panel of Figure 8 summarises the results of seperate OLS regressions for each
industry, where industry-level employment growth is regressed on the industry-level loan
spread. The blue bars indicate the regression coefficient and highlight the heterogeneity
in correlations. Loan spreads in “Manufacturing”, “Wholesale trade”, “Construction”, and
“Mining” seem to be relativly more associated with future growth in employment than other
industries.
The bottom panel of Figure 8 summarises the external finance dependence (hereafter
EFD) of each industry. We define a sector’s dependence on external finance following Rajan
and Zingales (1998). External dependence is defined as the ratio of total capital expenditures
minus current cash flow to total capital expenditure.21 The correlation between the top and
bottom panel is 0.50. Table 8 mirrors the specificaion in column (3) of Table 7 but interacts
loan spreads with indicator variables for the sector’s dependence on external finance. Note
that the base EFD effect is absorbed by the industry fixed effects. Column (1) interacts
20
In untabulated robustness tests, we also include industry-level bond spread measures, constructed using
bond price data from TRACE, in the model. Controlling for the industry-specific bond spread has little
impact on magnitude or significance of the industry loan spread coefficient.
21
Specifically, the measure is calculated using firm-level data from Compustat on capital expenditures
(CAPX) and free cash flow (OANCF). The industry-level EFD is based on the median firm within each
industry over the 2000-2019 period.
26
loan spreads with a dummy variable equal to one for the five industries most reliant on
external funding. The result indicates that industries more reliant on external funding indeed
show a significantly stronger relationship between total employed and industry loan spread.
Alternatively, Column (2) interacts loan spreads with the continuous EFD measure for a
given industry, with similiar results. Finally, Column (3) includes dummy variables for the
top three industries with the largest EFD, middle four, and bottom four and we see industries
with less reliance on external finance have a weaker relationship with loan spreads. We find
consistent results using industry establishments and gross output as dependent variables
(untabulated).
Overall, these results are consistent with our finding at the economy-wide level that
presumably more external (bank) finance dependent firms, such as smaller, younger, and
private firms, account for most of the predictive power of the loan market credit spread.
That is, compositional difference across loan and bond markets contribute to the differential
predictive power.
6 Weighting schemes using industry-level information
Finally, we explore if the heterogeneity in forecasting power across industries can be ex-
ploited to improve forecasting results at the aggregate level. In constructing the aggregate
loan spread a simple arithmetic average of all loan spreads available each month is used, fol-
lowing Gilchrist and Zakrajšek (2012). This method puts an equal weight on all loan-month
observations. For instance, at an industry level this implies that a higher weight is assigned
to industries with a greater number of loans outstanding. However, industries may differ
in their aggregate importance and loan spreads may have a differential information content
across industries, as implied by the tests in the previous section. This may or may not be
reflected in the number of loans outstanding across industries. In this section we explore
alternative weighting schemes to construct an aggregate loan spread.
One can envisage a number of alternative weighting schemes that instead put a higher
27
weight on the spreads from some industries relative to others, for example based on that
industry’s predictive power. Table 9 reports aggregate level regressions using the three-
month ahead industrial production as the dependent variable. That is, the table mirrors
Table 1, Panel A. Column (1) reports the baseline aggregate loan spread, constructed as a
simple arithmetic average across all individual loan-month observations, for comparison. In
columns (2) to (5) we use aggregate loan spreads employing alternative weighting schemes.
Column (2) uses a loan spread constructed by weighting each industry loan spread by
that industry’s contribution to GDP in the respective year. Interestingly, a GDP weighted
loan spread performs similarly to the arithmetic average in column (1). This implies that
assigning a higher weight to industries that account for a larger share of aggregate economic
outcome does not improve the prediction.
In column (3), we take the loan spread coefficients from the top panel of Figure 8 as
weights (rescaled to sum to one). That is, this aggregate spread puts more weight on indus-
tries in which the loan spread has a higher predictive power. This weighting scheme results
in a sizable improvement in adjusted R2 of +3 p.p. That is, industries in which the loan
spread has a higher predictive power also contribute more to the aggregate forecasting power
of the loan spread.
In column (4), the loan spread is constructed using the industry’s EFD as weights, as
defined in the previous section. That is, in the construction of the spread more weight is put
on industries that comprise of firms that are more sensitive to external financing frictions.
This approach yields similar results as in column (3). This is a reflection of the evidence
reported in Table 7 that the predictive power of the loan spread is larger in industries that
are more externally finance dependent. That is, both the weighting scheme employed in
column (3) and column (4) put a larger weight on high EFD industries. Again, these results
are consistent with the conjecture that (part of) the predictive power of the loan spread can
be explained by loan markets comprising of firms that are particularly sensitive to financial
frictions.
Finally, in column (5) we weight industries using optimal weights chosen by an elastic
28
net regression.22 That is, we use a statistical appraoch to improve the in-sample fit of the
model. As expected, this results in an improvement in R2 relative to the baseline loan spread
[cf. column (1)]. More interestingly, the data mining approach does not improve upon the
economically motivated weighting schemes in column (3) and (4). Specifically, the elastic
net approach also places a larger weight on high EFD sectors, consistent with the conjecture
that the predictive power of the loan market credit spread is driven by a better coverage
of firms that are more sensitive to external financing frictions. These results are robust to
using other macro outcomes (payroll employment and unemployment, results untabulated).
Overall, this section highlights the usefulness of bottom-up credit spread measures in un-
covering cross-sectional heterogeneity. Further, deviating from simple arithmetic averaging
when construcing aggregate measures from microdata can help improve aggregate forecasting
results.
7 Conclusion
We introduce a novel measure of credit spreads based on the prices of traded syndicated
corporate loans. We document extensive evidence that this new measure outperforms cor-
porate bond spreads as well as other existing credit spread measures and equity returns used
in the previous literature. Importantly, the predictive power of our loan spread measure
is consistent with existing theories on the role of financial frictions in amplifying business
cycles. Compositional differences between firms borrowing in the loan vis-a-vis the bond
market explain part of the predictive power of loan over bond spreads. We show how the
forecasting power can be even enhanced further through different aggregation methods.
This is the first paper to use secondary loan market prices to construct a bottom-up
measure of loan spreads to forecast business cycles. We only scratch the surface with respect
to many issues that we raise in our paper. For example, we provide a very simple way
to aggregate the loan spread measure. We clearly need more research on how to improve
22
Results are similar if LASSO or Ridge regressions are used to choose optimal weights.
29
the forecasting power of our loan spread (and of all bottom-up measures, such as the bond
spread). Moreover, the forecasting power of the loan spread might be interesting for other
applications and on different aggregation levels, e.g., the industry- or even the firm-level.
Even though our time-series covers the last 20 years (due to data availability) we are able
to provide very consistent evidence as to its predictive power across different specifications
as well as cross-country evidence. We believe that the additional predictive power of the
loan over the bond spread will likely grow in the years ahead. The development of both
spreads has already substantially diverged during the 2018 to 2019 period. Moreover, mon-
etary policy interventions that have been introduced in March 2020 at an early stage of the
COVID-19 pandemic have directly targeted corporate bonds with bond spreads declining
below pre-COVID levels at times when the economy was far from recovering (while loan
spreads remain elevated). In other words, the information content of bond spreads might
be severely impaired if targeted by monetary policy. We look forward to future research in
these promising areas.
30
References
Addoum, J. M. and Murfin, J. R. (2020). Equity price discovery with informed private
debt. Review of Financial Studies, 33 (8), 3766–3803.
Adrian, T., Boyarchenko, N. and Giannone, D. (2019). Vulnerable growth. American
Economic Review, 109 (4), 1263–1289.
Altman, E., Gande, A. and Saunders, A. (2010). Bank debt versus bond debt: Evidence
from secondary loan prices,. Journal of Money, Credit and Banking, 42, 755–767.
Asker, Farre-Mensa and Ljungqvist (2015). Corporate investment and stock market
listing: A puzzle? Review of Financial Studies, 28 (2), 342–390.
Baron, M. and Xiong, W. (2017). Credit expansion and neglegted crash risk. Quarterly
Journal of Economics, 132 (2), 713–764.
Bassett, W. F., Chosak, M. B., Driscoll, J. C. and Zakrajšek, E. (2014). Changes
in bank lending standards and the macroeconomy. Journal of Monetary Economics, 62,
23–40.
Begenau, J. and Salomao, J. (2019). Firm financing over the busines cycle. Review of
Financial Studies, 32 (4), 1235–1274.
Ben-Rephael, A., Choi, J. and Goldstein, I. (2020). Mutual fund flows and fluctuations
in credit and business cycles. Journal Financial Economics.
Berg, T., Saunders, A. and Steffen, S. (2016). The Total Costs of Corporate Borrowing
in the Loan Market: Don’t Ignore the Fees. Journal of Finance, 71 (3), 1357–1392.
—, —, — and Streitz, D. (2017). Mind the gap: The difference between u.s. and european
loan rates. Review of Financial Studies, 30 (3), 948–987.
Bernanke, B. S. (1990). On the predictive power of interest rates and interes rate spreads.
New England Economic Review, pp. 51–68.
— and Gertler, M. (1989). Agency costs, net worth, and business fluctuations. American
Economic Review, 79 (1), 14–31.
Berndt, A. and Gupta, A. (2009). Moral hazard and adverse selection in the originate-
to-distribute model of bank credit. Journal of Monetary Economics, 56 (5), 725–743.
Bharath, S. T. and Shumway, T. (2008). Forecasting default with the merton distance
to default model. Review of Financial Studies, 21 (3), 1339–1369.
Bordalo, P., Gennaioli, N. and Shleifer, A. (2018). Diagnostic expectations and
credit cycles. Journal of Finance, (1), 199–227.
Borio, C., Drehmann, M. and Xia, D. (2020). Predicting recessions: financial cycle
versus term spread. BIS Working Papers, (818), 1–27.
Chodorow-Reich, G. (2014). The employment effects of credit market disruptions: Firm-
level evidence from the 2008–9 financial crisis. Quarterly Journal of Economics, 129 (1),
1–59.
31
Cloyne, J., Ferreira, C., Froemel, M. and Surico, P. (2020). Monetary policy,
corporate finance and investment. Working Paper, University of California Davis.
Davis, Haltiwanger, Jarmin and Miranda (2006). Volatility and dispersion in business
growth rates: Publicly traded versus privately held firms. NBER Macroeconomics Annual,
21, 107–179.
Dick-Nielsen, J. and Rossi, M. (2019). The cost of immediacy for corporate bonds.
Review of Financial Studies, 32 (1), 1–41.
Drucker, S. and Puri, M. (2009). On loan sales, loan contracting, and lending relation-
ships. Review of Financial Studies, 22 (7), 2835–2872.
Estrella, A. and Hardouvelis, G. (1991). The term structure as a predictor of real

economic activity. Journal of Finance, 46 (2), 555 – 576.
— and Mishkin, F. S. (1998). Predicting u.s. recessions: Financial variables as leading

indicators. Review of Economics and Statistics, 80 (1), 45–61.
Favara, G., Gilchrist, S., Lewis, K. F. and Zakrajšek, E. (2016). Updating the re-
cession risk and the excess bond premium. FEDS Notes. Washington: Board of Governors
of the Federal Reserve System.
Feldhütter, P. and Poulsen, T. K. (2018). What determines bid-ask spreads in over-

the-countermarkets? Working Paper, Copenhagen Business School.
Friedman, B. M. and Kuttner, K. N. (1992). Money, income, prices and interest rates.
American Economic Review, 82, 472–492.
— and — (1993a). Economic activity and the short-term credit markets: An analysis of
prices and quantities. Brookings Papers on Economic Activity, 24, 192–283.
— and — (1993b). Why Does the Paper-Bill Spread Predict Real Economic Activity?, Uni-
versity of Chicago Press, pp. 213–254.
— and — (1998). Indicator properties of the paper-bill spread: Lessons from recent experi-
ence. Review of Economics and Statistics, 80 (1), 34–44.
Gande, A. and Saunders, A. (2012). Are banks still special when there is a secondary
market for loans? Journal of Finance, 67 (5), 1649–1684.
Gatev, E. and Strahan, P. E. (2006). Banks’ advantage in hedging liquidity risk: Theory
and evidence from the commercial paper market. Journal of Finance, 61 (2), 867–892.
— and — (2009). Liquidity risk and syndicate structure. Journal of Financial Economics,
93, 490–504.
Gertler, M. and Gilchrist, S. (1994). Monetary policy, business cycles, and the behavior
of small manufacturing firms. Quarterly Journal of Economics, 109 (2), 309–340.
— and Lown, C. S. (1999). The information in the high yield bond spread for the business
cycle: Evidence and some implications. Oxford Review of Economic Policy, 15 (3), 132–
150.
32
Gilchrist, S., Wei, B., Yue, V. and Zakrajsek, E. (2020). The fed takes on corporate
credit risk: An analysis of the efficacy of the smccf. NBER Working Paper.
— and Zakrajšek, E. (2012). Credit spreads and business cycle fluctuations. American
Economic Review, 102 (4), 1692–1720.
Giroud, X. and Müller, H. (2018). Firm leverage and regional business cycles. Working
Paper, New York University.
Greenstone, M., Mas, A. and Nguyen, H.-L. (2020). Do credit market shocks affect
the real economy? Quasi-experimental evidence from the great recession and “normal”
economic times. American Economic Journal: Economic Policy, 12 (1), 200–225.
Greenwood, R., Hanson, S. G., Shleifer, A. and Sørensen, J. A. (2020). Pre-
dictable financial crises. NBER Working Paper No. 27396.
Gürkaynak, R. S., Sack, B. and Wright, J. H. (2007). The u.s. treasury yield curve:
1961 to the present. Journal of Monetary Economics, 54 (8), 2291–2304.
Hadlock, C. J. and Pierce, J. R. (2010). New evidence on measuring financial con-
straints: Moving beyond the KZ index. Review of Financial Studies, 23 (5), 1909–1940.
Harvey, C. (1989). Forecasts of economic growth from the bond and stock markets. Fi-
nancial Analysts Journal, 45 (5), 38 – 45.
Holmström, B. and Tirole, J. (1997). Financial intermediation, loanable funds, and the
real sector. Quarterly Journal of Economics, 112 (3), 663–691.
Ivashina, V. and Scharfstein, D. (2010). Bank lending during the financial crisis of
2008. Journal of Financial Economics, 97, 319–338.
Jordà, Ò. (2005). Estimation and inference of impulse responses by local projections. Amer-
ican Economic Review, 95 (1), 161–182.
Jordà, Ò., Schularick, M. and Tyler, A. M. (2013). When credit bites back. Journal
of Money, Credit and Banking, 45 (s2).
Kiyotaki, N. and Moore, J. (1997). Credit cycles. Journal of Political Economy, 105 (2),
211–248.
Krishnamurthy, A. and Muir, T. (2020). How credit cycles across financial crisis. NBER
Working Paper No. 23850.
Lian, C. and Ma, Y. (2020). Anatomy of corporate borrowing constraints. Quarterly Jour-
nal of Economics, forthcoming.
López-Salido, D., Stein, J. C. and Zakrajšek, E. (2017). Credit-market sentiment
and the business cycle. Quarterly Journal of Economics, 132 (3), 1373–1426.
Mojon, B. and Gilchrist, S. (2016). Credit risk in the euro area. The Economic Journal,
(2), 118–158.
Mueller, P. (2009). Credit spreads and real activity. Working Paper, London School of
Economics.
33
Pflueger, C., Siriwardane, E. and Sunderam, A. (2020). Financial market risk per-
ceptions and the macroeconomy. Quarterly Journal of Economics, 135 (3), 1443 – 1491.
Philippon, T. (2009). The bond market’s q. Quarterly Journal of Economics, 124 (3),
1011–1056.
Rajan, R. G. and Zingales, L. (1998). Financial dependence and growth. American

Economic Review, 88 (3), 559–586.
Schularick, M. and Tyler, A. M. (2012). Credit booms gone bust: Monetary policy,
leverage cycles, and financial crises, 1870-2008. American Economic Review, 102 (2),
1029–1061.
Schwert, M. (2020). Does borrowing from banks cost more than borrowing from the
market? Journal of Finance, 75 (2), 905–947.
Stein, J. C. (2014). Incorporating financial stability considerations into a monetary policy

framework. Remarks at the International Research Forum on Monetary Policy.
Sufi, A. (2007). Information asymmetry and financing arrangements: Evidence from syn-
dicated loans. Journal of Finance, 62 (2), 629–668.
Taylor, A. and Sansone, A. (2006). The Handbook of Loan Syndications and Trading.
McGraw-Hill and the Loan Syndications and Trading Association (LSTA).
Thomas, H. and Wang, Z. (2004). The integration of bank syndicated loan and junk bond
markets. Journal of Banking & Finance, 28 (2), 299–329.
34
600
Loan Amount (USD Bn)
400
200
2000 2005 2010 2015 2020
Figure 2: Secondary loan market trading volume

This figure plots the development of total loan volume traded in the secondary U.S. syndi-
cated loan market over the 1999 to 2019 period. Source: LSTA.
35
Dist Combined.pdf
0.4
Fraction of loans 0.3
0.2
0.1
0.0
AAA AA A BBB BB B CCC CC C D SD NR NA
0.4
0.3
Fraction of bonds
0.2
0.1
0.0
AAA AA A BBB BB B CCC CC C D SD NR NA
Rating category
Figure 3: Rating Distribution

This figure plots the security level rating distribution across the loan market (top panel) and
the bond market (bottom panel). Loan level ratings come from Standard & Poor’s Leveraged
Commentary & Data (S&P LCD) and Dealscan. Bond level ratings come from TRACE.
36
15
10
Spread %
2000 2005 2010 2015 2020

spread.pdf
Figure 4: Corporate credit spreads

This figure plots monthly credit spread measures over time. Depicted are: (i) the loan spread
(red line), defined as the average credit spread of syndicated loans issued by non-financial
firms that are traded in the secondary market, and (ii) the bond spread (black line), defined
following Gilchrist and Zakrajšek (2012) as the average credit spread on senior unsecured
bonds issued by non-financial firms. Bars indicate NBER recessions. The sample period is
1999:11 to 2020:03.
37
Projections - All - Loand And Bond Spread.pdf
Industrial production Unemployment
Coeffecient on delta spread at t+h

0.50
0.0
0.25
−0.2
0.00
−0.4
−0.25
1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 12
Forecast horizon (months) Forecast horizon (months)
Payroll employment
0.0
−0.1
−0.2
1 2 3 4 5 6 7 8 9 10 11 12
Forecast horizon (months)
Figure 5: Local Projections

This figure plots the impulse response function using a Jordà (2005) local projections frame-
work. In the top left panel the dependent variable is the h-month ahead growth in industrial
production. In the top right panel the dependent variable is the h-month ahead change in
unemployment rate. In the bottom left panel the dependent variable is the h-month ahead
growth in payroll employment. The x-axis indicates the forecast horizon (in months). The
coefficient, at each forecast horizon, for the loan spread is in red. The black line is the
coeffecient on the bond spread. Shaded areas indicate 95% confidence intervals. The sample
period is 1999:11 to 2020:03.
38
Germany France
15 15
10 10
Spread %
Spread %
5 5
0 0
2000 2005 2010 2015 2020 2000 2005 2010 2015 2020
Spain
15
10
Spread %
0
2000 2005 2010 2015 2020
Figure 6: European country loan and bond spreads

This figure plots monthly loan (red lines) and bond (black lines) spread measures over time
for Germany (top left), France (top right) and Spain (bottom left). Observations based on
less than five loans are excluded. Bars indicate OECD downturns. The sample period for the
loan spread is 2001:01 to 2020:03 for Germany, 2004:04 to 2020:03 for France, and 2004:05
to 2020:03 for Spain. The sample period for the bond spread is 1999:11 to 2020:03.
39
Arts and entertainment Construction Education and healthcare Information
20
15
10
Manufacturing Mining Professional services Retail trade
20
% Spread
15
10
2000
2005
2010
2015
2020
Transportation Utilities Wholesale trade
20
15
10
0
2000
2005
2010
2015
2020
2000
2005
2010
2015
2020
2000
2005
2010
2015
2020
Figure 7: Industry loan spreads

This figure plots monthly loan spread measures over time for 11 non-financial sectors. Firms
are classified into sectors following the BEA sector definition. The sample period is 1999:11
to 2020:03 (except for “Construction” and “Mining” due to limited data availablity in the
early sample period). Bars indicate NBER recessions.
40
Manufacturing
Wholesale trade
Construction
Mining
Retail trade
Arts and entertainment
Professional services
Information
Transportation
Utilities
Education and healthcare
−0.5 −0.4 −0.3 −0.2 −0.1 0.0
Coefficient on loan spread
Manufacturing
Mining
Professional services
Transportation
Construction
Utilities
Arts and entertainment
Information
Retail trade
Education and healthcare
Wholesale trade
−0.5 0.0 0.5 1.0
External finance dependence
Figure 8: Industry heterogeneity

The top panel plots regression coefficients using seperate OLS regressions of industry level
employment growth on the industry level loan spread. The bottom panel plots the Rajan
and Zingales (1998) measure of external finance dependence for each industry over the 1999-
2020 sample period. External dependence is defined as the ratio of total capital expenditures
minus current cash flow to total capital expenditure.
41
Table 1: Baseline forecasting results
This table relates credit spread measures to future economic outcomes for the U.S. economy. The unit of observation in Panels
A, B, and C is the monthly level t. The sample period is 1999:11 to 2020:03. The dependent variable in Panel A is the three-
month ahead percentage change in industrial production, i.e., growth from t − 1 to t + 3. The dependent variable in Panel B is
the three-month ahead change in unemployment rate. The dependent variable in Panel C is the three-month ahead percentage
change in non-farm payroll employment. Each specification includes (not reported) a one period lag of the dependent variable,
i.e., growth from t − 2 to t − 1, the term spread, i.e., the difference between 10-year and three-month U.S. treasury and the
real FFR, i.e., the effective federal funds rate minus realized inflation. Incremental R2 refers to the difference between the
adjusted R2 in the respective column and the adjusted R2 of a baseline forecasting model, i.e., a model that only includes a one
period lag of the dependent variable, the term spread, and the real FFR (but no credit spread). Reported OLS coefficients are
standardized. t-statistics, based on heteroskedasticity and autocorrelation corrected Newey-West standard errors with a four
period lag structure, are reported in parentheses.
Forecast horizon: h = 3 months
(1) (2) (3)

Panel A. Industrial Production
∆StLoan -0.410 -0.396
(-5.727) (-3.831)
∆StBond -0.198 -0.030
(-2.257) (-0.267)
Adjusted R2 0.313 0.198 0.311

Incremental R2 +0.150 +0.035 +0.148
Observations 241 241 241
Panel B. Unemployment Rate

∆StLoan 0.355 0.392
(2.808) (2.943)
∆StBond 0.099 -0.081
(0.623) (-0.812)
Adjusted R2 0.272 0.156 0.274

Incremental R2 +0.122 +0.006 +0.124
Panel C. Payroll Employment

∆StLoan -0.207 -0.207
(-6.332) (-4.080)
∆StBond -0.096 0.004
(-2.273) (0.009)
Adjusted R2 0.839 0.806 0.838

Incremental R2 +0.041 +0.008 +0.040
42
Table 2: Other Measures and Robustness
This table relates different credit spread measures and additional controls to future economic outcomes for the U.S. economy.
The unit of observation is the monthly level t. The sample period is 1999:11 to 2020:03. The dependent variable is the three-
month ahead percentage change in industrial production, i.e., the growth from t − 1 to t + 3. Panel A focuses on alternative
credit spreads. Column (1) uses the loan spread as a baseline comparision. Column (2) uses the Baa-Aaa corporate bond
spread. Column (3) uses the corporate high yield minus AAA spread. Column (4) uses the Commercial paper - three-month
Treasury bill spread. Column (5) uses the S&P500 monthly return from t − 1 to t. Panel B focuses on additional robustness
checks. Column (1) uses the loan spread as a baseline comparision. Column (2) controls for the median bid-ask spread in the
loan market at time t. Column (3) uses a residual loan spread from a regression of the loan spread on loan contract terms such
as size, age, amount, AISD, and indicators for secured, senior, and financial covenants. Column (4-5) contain the loan and
bond spread removing the 2008-09 crisis period. Each specification includes (not reported) a one period lag of the dependent
variable, i.e., growth from t − 2 to t − 1, the term spread, i.e., the difference between 10-year and three-month U.S. treasury and
the real FFR, i.e., the effective federal funds rate minus realized inflation. Incremental R2 refers to the difference between the
Baseline Other credit spreads Equity market
(1) (2) (3) (4) (5)

Panel A. Other measures
∆StLoan -0.410
(-5.727)
∆ Baa-Aaa spread -0.277
(-3.918)
∆ HY-Aaa spread -0.248
(4.013)
∆ CP-bill spread 0.080
(0.898)
S&P500 return 0.216
(2.921)
Adjusted R2 0.313 0.237 0.222 0.166 0.204
Incremental R2 +0.150 +0.077 +0.062 +0.006 +0.041
Observations 241 241 241 241 241
Baseline Loan mkt liquidity Contract terms Ex 2008-09 crisis
(1) (2) (3) (4) (5)

Panel B. Robustness
∆StLoan -0.410 -0.360 -0.207
(-5.727) (-5.337) (-3.047)
∆StBond -0.058
(-0.720)
Loan bid-ask spread -0.328
(-3.141)
Residual ∆StLoan -0.405
(-5.646)
Adjusted R2 0.313 0.386 0.318 0.150 0.115
Incremental R2 +0.150 +0.226 +0.120 +0.034 +0.001
Observations 241 241 241 241 241
43
Table 3: Evidence from European countries
This table relates credit spread measures to future economic outcomes across European countries. The unit of observation
is the monthly level t. The sample period is 2001:01 to 2020:03 for Germany, 2004:04 to 2020:03 for France, and 2004:05
to 2020:03 for Spain. The dependent variable in column (1) is the three-month ahead percentage change in manufacturing
production index, i.e., growth from t − 1 to t + 3. The dependent variable in column (2) is the three-month ahead change in
the unemployment rate. The dependent variable in column (3) is the three-month ahead percentage change in the construction
index. Each specification includes (not reported) a one period lag of the dependent variable, i.e., growth from t − 2 to t − 1, the
term spread, i.e., the difference between 10-year Euro government bond (a GDP weighted average of all Euro area government
bonds) and three-month EURIBOR, and the real EONIA, i.e., the overnight rate minus realized inflation. Incremental R2
refers to the difference between the adjusted R2 in the respective column and the adjusted R2 of a baseline forecasting model,
i.e., a model that only includes a one period lag of the dependent variable, the term spread, and real EONIA (but no credit
spread). Contribution from ∆StLoan measures the proportion of the increase in adjusted R2 in the respective column that
results from the inclusion ∆StLoan as opposed to ∆StBond . Reported OLS coefficients are standardized. t-statistics, based on
heteroskedasticity and autocorrelation corrected Newey-West standard errors with a four period lag structure, are reported in
parentheses.
Manufacturing Index U/E Construction Index

(1) (2) (3)
Panel A. Germany
∆StLoan -0.360 0.160 -0.093
(-2.300) (2.600) (-1.300)
∆StBond -0.048 -0.0019 0.029
(-0.690) (-0.350) (0.600)
Adjusted R2 0.260 0.410 0.090

Incremental R2 +0.111 +0.029 -0.031
% Contribution from ∆StLoan 0.86 0.95 0.94
Panel B. France
∆StLoan -0.34 0.290 -0.068
(-2.100) (2.200) (-1.200)
∆StBond -0.009 -0.002 0.010
(-0.100) (-0.019) (0.140)
Adjusted R2 0.190 0.210 0.082

Incremental R2 +0.071 +0.035 -0.044
Panel C. Spain
∆StLoan -0.200 0.130 -0.057
(-1.900) (2.800) (-0.910)
∆StBond -0.130 0.052 -0.160
(-1.000) (0.830) (-1.800)
Adjusted R2 0.190 0.710 0.140

Incremental R2 +0.058 +0.102 +0.051
44
Table 4: Impact of financial constraints
This table relates credit spread measures to future economic outcomes for the U.S. economy. The unit of observation is the
monthly level t. The sample period is 1999:11 to 2020:03. The dependent variable is the three-month ahead percentage change
in industrial production, i.e., growth from t − 1 to t + 3. Panel A reports results using loan spreads constructed separately for
firms with total assets below and above the median level. Panel B reports results using loan spreads constructed separately
for firms with age below and above the median level. In Panel C, firms are double sorted by age and size buckets, i.e., loan
spreads are constructed separately for “young and small firms” (below median total assets and below median age) and “old
and large firms” (above median total assets and above median age). Each panel includes a group of “private firms”, defined
as firms that cannot be matched to the Compustat North America database. Each specification includes (not reported) a one
period lag of the dependent variable, i.e., growth from t − 2 to t − 1, the term spread, i.e., the difference between 10-year and
three-month U.S. treasury and the real FFR, i.e., the effective federal funds rate minus realized inflation. Incremental R2 refers
to the difference between the adjusted R2 in the respective column and the adjusted R2 of a baseline forecasting model, i.e., a
model that only includes a one period lag of the dependent variable, the term spread, and the real FFR (but no credit spread).
Reported OLS coefficients are standardized. t-statistics, based on heteroskedasticity and autocorrelation corrected Newey-West
standard errors with a four period lag structure, are reported in parentheses.
(1) (2) (3)

Panel A. By Size
∆StLoan [Small firms] −0.380
(−4.20)
∆StLoan [Large firms] −0.260
(−3.400)
∆StLoan [Private firms] −0.420
(−5.500)
Adjusted R2 0.300 0.230 0.320

Incremental R2 +0.137 +0.067 +0.157
Panel B. By Age
∆StLoan [Young firms] −0.340
(−4.500)
∆StLoan [Old firms] −0.290
(−2.800)
(−5.500)
Adjusted R2 0.270 0.240 0.320

Incremental R2 +0.107 +0.077 +0.157
Panel C. By Size and Age

∆StLoan [Small & young firms] −0.390
(−4.500)
∆StLoan [Large & old firms] −0.210
(−1.800)
(−5.500)
Adjusted R2 0.310 0.200 0.320

Incremental R2 +0.147 +0.037 +0.157
45
Table 5: Impact of loan rating
This table relates credit spread measures conditional on loan ratings to future economic outcomes for the U.S. economy. The
unit of observation in Panels A, B, and C is the monthly level t. The sample period is 1999:11 to 2020:03. Each panel reports
results using loan spreads conditional on a loan level rating of BBB in Column (1), BB in Column (2), B and below in Column
(3), and loans without available rating information in Column (4). The dependent variable in Panel A is the three-month
ahead percentage change in industrial production, i.e., growth from t − 1 to t + 3. The dependent variable in Panel B is the
three-month ahead change in unemployment rate. The dependent variable in Panel C is the three-month ahead percentage
change in non-farm payroll employment. Each specification includes (not reported) a one period lag of the dependent variable,
i.e., growth from t − 2 to t − 1, the term spread, i.e., the difference between 10-year and three-month U.S. treasury and the
real FFR, i.e., the effective federal funds rate minus realized inflation. Incremental R2 refers to the difference between the
(1) (2) (3) (4)

∆StLoan [BBB] -0.105
(-1.557)
∆StLoan [BB] -0.260
(-3.538)
∆StLoan [B and below] -0.425
(-5.425)
∆StLoan [Not Available] -0.415
(-4.040)
Adjusted R2 0.170 0.226 0.322 0.315

Incremental R2 +0.007 +0.063 +0.159 +0.152
Observations 241 241 241 241

∆StLoan [BBB] 0.093
(0.654)
∆StLoan [BB] 0.228
(1.424)
∆StLoan [B and below] 0.341
(2.374)
∆StLoan [Not Available] 0.401
(3.019)
Adjusted R2 0.155 0.199 0.260 0.305

Incremental R2 +0.005 +0.049 +0.110 +0.155
Panel C. Payroll employment

∆StLoan [BBB] -0.089
(-1.793)
∆StLoan [BBB/BB] -0.174
(-5.742)
∆StLoan [B and below] -0.221
(-6.578)
∆StLoan [Not Available] -0.199
(-3.902)
Adjusted R2 0.805 0.828 0.845 0.834

Incremental R2 +0.007 +0.030 +0.047 +0.036
46
Table 6: Credit Conditions
This table relates credit supply conditions to credit spreads in the U.S. and Europe. The unit of observation is the quarterly
level t. The sample period is 1999:11 to 2020:03 for the U.S., 2001:01 to 2020:03 for Germany, 2004:04 to 2020:03 for France, and
2004:05 to 2020:03 for Spain. Panel A focuses on U.S. data. Panel A1 uses as a dependent variable the Federal Reserve’s Senior
Loan Officer Survey, and is defined as the percentage of loan officers who respond that “lending tightened” less the percentage
of loan officers who responded that “lending eased” over the previous quarter. For panel A2 we use the bank level ratio of
total unused commitments/total assets and construct an aggregate ratio as a weighted average across banks each quarter.
Specifically, the dependent variable is then the change in the aggregate ratio from the previous quarter. Panel B focuses on
European data. The dependent variables in Panel B come from the European Central Banks’ Banking Lending Survey and
are defined in a similar way as the Federal Reserve’s Senior Loan Officer Survey, i.e., are measures for country-specific credit
supply conditions based on loan officer survey data. In all specifications we regress the credit conditions over t − 1 to t on the
change in credit spread over the same period i.e. spreads and credit conditions are measured contemporaneously. Coefficients
are standardized. t-statistics, based on heteroskedasticity and autocorrelation corrected Newey-West standard errors with a
four period lag structure, are reported in parentheses.
Panel A. U.S. (1) (2) (3)
Panel A1. Credit supply conditions

∆StLoan 0.430 0.418
(3.810) (5.176)
∆StBond 0.290 0.019
(1.879) (0.118)
Adjusted R2 0.171 0.073 0.164

Panel A2. Banks’ unused commitments

∆StLoan -0.351 -0.287
(-2.435) (-2.166)
∆StBond -0.306 -0.223
(-1.922) (-1.512)
Adjusted R2 0.112 0.082 0.148

Panel B. Credit supply conditions in Europe
Panel B1. Germany

∆StLoan 0.376 0.458
(3.748) (3.214)
∆StBond 0.159 -0.130
(1.182) (-1.031)
Adjusted R2 0.128 0.011 0.126

Panel B2. France

∆StLoan 0.480 0.417
(3.545) (3.533)
∆StBond 0.329 0.140
(1.436) (0.778)
Adjusted R2 0.218 0.094 0.221

Panel B3. Spain

∆StLoan 0.370 0.357
(2.018) (1.951)
∆StBond 0.176 0.031
(1.008) (0.352)
Adjusted R2 0.122 0.015 0.109

47
Table 7: Baseline industry forecasting results
This table relates industry credit spread measures to future industry outcomes. The unit of observation is the industry-quarter
level bt. The sample period is 1999:11 to 2019:12. The dependent variable in Panel A is the one quarter ahead percentage
change in employment for industry b, i.e., the growth from t − 1 to t + 1. The dependent variable in Panel B is the one quarter
ahead percentage change in establishments for industry b, i.e., the growth from t − 1 to t + 1. The dependent variable in Panel C
is the one quarter ahead percentage change in gross output for industry b, i.e., the growth from t − 1 to t + 1. Each specification
includes (not reported) a one period lag of the dependent variable, i.e., the growth from t − 2 to t − 1. The model reported
in column (1) further includes the aggregate loan spread, term spread, i.e., the difference between 10-year and three-month
U.S. treasury and the real FFR, i.e., the effective federal funds rate minus realized inflation. Year × quarter and industry fixed
effects are included when indicated. Incremental R2 refers to the difference between the adjusted R2 in the respective column
and the adjusted R2 of a baseline forecasting model, i.e., a model that only includes a one period lag of the dependent variable,
the term spread, and the real FFR (but no credit spread or fixed effects). Standard errors are clustered by industry. t-statistics
are reported in parentheses. Coefficients are standardized.
(1) (2) (3)
Panel A. Industry total employed

Loan
Sbt −0.130 −0.171 −0.292
(−3.491) (−3.534) (−4.609)
StLoan −0.239
(−3.818)
Year × quarter fixed effects No Yes Yes
Industry fixed effects No No Yes
Adjusted R2 0.452 0.558 0.590
Incremental R2 +0.086 +0.192 +0.224
Panel B. Industry total establishments

Loan
Sbt −0.321 −0.304 −0.413
(−3.373) (−2.713) (−2.834)
StLoan 0.056
(0.746)
Adjusted R2 0.196 0.294 0.395
Incremental R2 +0.063 +0.151 +0.252
Panel C. Industry gross output

Loan
Sbt −0.003 −0.071 −0.099
(−0.039) (−1.075) (−1.542)
StLoan −0.330
(−3.553)
Adjusted R2 0.183 0.379 0.387
Incremental R2 +0.082 +0.233 +0.241
48
Table 8: Industry heterogeneity
This table relates industry credit spread measures to future industry outcomes. The unit of observation is the industry-quarter
level bt. The sample period is 1999:11 to 2019:12. The dependent variable is the one quarter ahead change in employment
for industry b, i.e., the growth from t − 1 to t + 1. Column (1) interacts the industry loan spread with a dummy for the five
industries with the largest external finance dependent (EFD) ratios. Column (2) interacts the industry loan spread with the
continuous EDF value of the industry. Column (3) simultaneosuly interacts the loan spread with a dummy for the industries
with the three largest external finance dependent (EFD) ratios, middle four and bottom four. Each specification includes (not
reported) a one period lag of the dependent variable, i.e., the growth from t − 2 to t − 1, the term spread, i.e., the difference
between 10-year and three-month U.S. treasury, and the real FFR, i.e., the effective federal funds rate minus realized inflation.
EFD is defined following Rajan and Zingales (1998). Standard errors are clustered by industry. t-statistics are reported in
parentheses. Coefficients are standardized.
(1) (2) (3)
Loan x Top 5 EFD

Sbt -0.311
(-4.527)
Loan x Continuous EFD
Sbt -0.319
(-2.698)
Loan x Top 3 EFD
Sbt -0.519
(-5.408)
Loan x Middle 4 EFD
Sbt -0.269
(-2.754)
Loan x Bottom 4 EFD
Sbt -0.139
(-1.606)
Industry fixed effects Yes Yes Yes
Time fixed effects Yes Yes Yes
Adjusted R2 0.271 0.268 0.269
49
Table 9: Alternative Weighting Schemes
This table relates alternative credit spread measures to future economic outcomes for the U.S. economy. The unit of observation
is the monthly level t. The sample period is 1999:11 to 2020:03. The dependent variable is the three-month ahead change in
industrial production, i.e., growth from t − 1 to t + 3. Column (1) reports the baseline aggregate loan spread results for
comparison [cf. Table 1 Column (1)]. In Columns (2) - (5) an aggregate loan spread is constructed as a weighted average
across industry loan spreads using different weighting schemes. Specifically, Column (2) GDP-weights each industry-level loan
spread. Column (3) weights each industry by its correlation between loan spread and industry employment [cf. Figure 8 Top
panel]. Column (4) weights each industry according to its external finance dependence. Column (5) weights each industry by
coefficients from an Elastic net regression. Each specification includes (not reported) a one period lag of the dependent variable,
i.e., the growth from t − 2 to t − 1, the term spread, i.e., the difference between 10-year and three-month U.S. treasury and
the real FFR, i.e., the effective federal funds rate minus realized inflation. Incremental R2 refers to the difference between the
(1) (2) (3) (4) (5)

∆StLoan [Base] -0.410
(-5.727)
∆StLoan [GDP] -0.396
(-5.006)
∆StLoan [Industry] -0.445
(-6.236)
∆StLoan [EFD] -0.443
(-4.805)
∆StLoan [Elastic Net] -0.449
(-5.162)
Adjusted R2 0.313 0.305 0.343 0.337 0.339

Incremental R2 +0.150 +0.142 +0.180 +0.174 +0.176
Observations 241 241 241 241 241
50
Online Appendix: Corporate Loan Spreads and Eco-
nomic Activity
A Variable definitions
This section outlines the key variables used in the main paper and their construction. Table
A.1 describes each variable. Column(1) indicates the country for which the data applies and
the name of the variable used throughout the paper. Column(2) provides a brief description
and the source of the data. Column (3) indicates the data frequency.
Table A.1: Description of variables
Economic Variables Description Frequency

USA - Unemployment rate Unemployment rate (FRED) M
USA - Payroll employment Total private non-farm payroll employment (FRED) M
USA - Industrial production Total industrial production index (FRED) M
USA - Industry output Gross output by industry (Bureau of Economic Analysis) Q
USA - Industry employment Employment level by industry (Bureau of Labour Statistics) Q
USA - Industry establishments Count of establishments by industry (Bureau of Labour Statistics) Q
USA - FSLOSS Fed senior loan officer survey (Federal Reserve) Q
USA - Bank commitments Bank unused commitments (FDIC Call Reports) Q
USA - S&P500 S&P500 monthly return (CRSP) M
USA - Loan bid-ask spread Median bid-ask spread (Authors) M
USA - EFD External finance dependence Rajan and Zingales (1998) (Authors) M
Europe - EONIA EONIA minus HCIP inflation previous 12months (ECB) M
Europe - Unemployment rate Unemployment rate (Eurostat) M
Europe - Manufacturing index Manufacturing production index (Eurostat) M
Europe - Construction index Construction production index (Eurostat) M
Interest rates
USA - Real federal funds rate Avg effective federal funds rate minus core PCE index (FRED) M
USA - Baa-Aaa spread 10year Baa minus 10year Aaa corporate bond spread (FRED) M
USA - Commercial spread Paper/bill spread: 1 month A1/P1 commercial paper minus 3m UST (FRED) M
USA - HY-AAA spread High Yield minus AAA yield (FRED) M
USA - ∆StLoan Monthly aggregate loan spread constructed from loan market (Authors) M
USA - ∆StBond Monthly aggregate bond spread Gilchrist and Zakrajšek (2012) M
Loan
USA - Sbt Monthly industry loan spread (Authors) M
Europe - Term spread 10yr Euro area government bond minus 3m EURIBOR (FRED and Eurostat) M
Europe - ∆StLoan Monthly aggregate loan spread constructed from loan market (Authors) M
Europe - ∆StBond Monthly aggregate bond spread constructed by Mojon and Gilchrist (2016) M
A.1
B Descriptive statistics
Tables A.2 and A.3 summarize basic descriptive statistics for the dependent and indepen-
dent variables used in each table of the main paper in order to interpret the standardized
coeffecients reported.
Table A.2: Summary Statistics

This table shows summary statistics for independent and dependent variables in each table of the main paper. See the table
descriptions in the main paper for more details.
Variable Mean SD Min Median Max
Table 1:
∆yt+3 Industrial production (%) 0.23 1.79 -7.72 0.65 3.28

∆yt+3 Unemployment rate (pp) 0.00 0.42 -0.80 -0.10 1.90
∆yt+3 Payroll employment (%) 0.27 0.68 -2.67 0.51 0.98
∆StLoan (bps) 0.79 44.52 -129.11 -3.78 357.18
∆StBond (bps) -0.23 28.51 -129.49 -2.53 226.86
Table 2:
∆Baa − aaaspread (bps) -0.52 69.08 -324.0 -4.0 455.0

∆HY − aaaspread (bps) 0.03 12.19 -63.0 -1.0 94.0
∆CP − billspread (bps) -0.09 12.92 -55.0 0.0 65.0
∆S&P 500return (%) 0.44 4.19 -16.94 0.94 10.77
Loan bid-ask spread (%) 1.01 0.80 0.43 0.81 5.53
Residual ∆StLoan (bps) 0.43 44.33 -132.63 -4.19 356.24
Table 3:
∆yt+3 Unemployment rate Germany (pp) -0.09 0.29 -0.70 -0.10 0.50
∆yt+3 Manufacturing index Germany (index points) 0.36 3.02 -17.4 0.90 6.80
∆yt+3 Construction index Germany (index points) 0.05 4.67 -13.70 0.20 19.40
∆StLoan Germany (bps) 1.10 38.64 -152.7 -0.17 320.2
∆StBond Germany (bps) 0.28 17.50 -40.0 -1.00 163.40
∆yt+3 Unemployment rate France (pp) -0.02 0.28 -0.70 0.00 1.00
∆yt+3 Manufacturing index France (index points) -0.27 3.11 -20.30 0.25 5.10
∆yt+3 Construction index France (index points) -0.44 4.39 -42.10 -0.40 9.80
∆StLoan France (bps) 1.56 38.23 -110.80 -1.07 256.10
∆StBond France (bps) 0.89 21.59 -62.6 -1.59 151.60
∆yt+3 Unemployment Rate Spain (pp) 0.06 0.97 -1.10 -0.20 3.90
∆yt+3 Manufacturing Index Spain (index points) -0.53 3.49 -18.60 0.20 4.30
∆yt+3 Construction Index Spain (index points) -1.50 8.52 -64.50 -0.80 39.60
∆StLoan Spain (bps) 3.08 66.43 -489.60 -0.51 370.00
∆StBond Spain (bps) 0.63 28.24 -72.3 -2.38 163.80
A.2
Table A.3: Summary Statistics
This table shows summary statistics for independent and dependent variables in each table of the main paper. See the table
descriptions in the main paper for more details.
Table 4:
∆StLoan [Small firms] (bps) 0.93 46 -142 -1.1 310

∆StLoan [Large firms] (bps) -1.50 50 190 -2.3 394
∆StLoan [Young firms] (bps) -0.23 48 -159 -0.66 378
∆StLoan [Old firms] (bps) 0.22 47 -180 -1.24 331
∆StLoan [Small & young firms] (bps) 0.52 48 -162 -1.67 305
∆StLoan [Large & old firms] (bps) -0.08 54 -369 -0.82 319
∆StLoan [Private firms] (bps) 1.2 48 -117 -4.5 352
Table 5:
∆StLoan [BBB] (bps) -0.48 33 -131 -0.34 224

∆StLoan [BB] (bps) -0.37 41 -133 -2.78 370
∆StLoan [B and below] (bps) 1.45 52 -158 -0.97 395
∆StLoan [Not available] (bps) 0.77 42 -101 -2.51 386
Table 6:
FSLOSS (%) 5.63 24.15 -24.10 -3.80 83.60

Bank unused commitments
∆StLoan (bps) 4.80 114.48 -268.00 -9.46 782.83
∆StBond (bps) 2.33 55.84 -220.24 -5.17 218.85
BLS Germany (%) 3.50 14.29 -18.75 0.00 56.25
∆StLoan Germany (bps) 8.04 99.30 -242.63 -8.31 579.69
∆StBond Germany (bps) 1.40 31.98 -92.68 -3.81 134.96
BLS France (%) 6.05 20.36 -18.49 0.00 95.16
∆StLoan France (bps) 4.82 108.55 -343.80 -0.91 540.90
∆StBond France (bps) -1.26 40.83 -119.91 -5.28 123.06
BLS Spain (%) 8.86 22.75 -30.00 0.00 80.00
∆StLoan Spain (bps) 13.95 115.74 -344.69 5.04 470.71
∆StBond Spain (bps) -2.56 59.62 -184.10 -2.58 293.52
Table 7:
∆yt+3 Industry employment (%) 0.23 2.24 -13.28 0.55 7.20

∆yt+3 Industry establishments (%) 0.50 1.36 -13.19 0.51 5.41
∆yt+3 Industry gross output (%) 1.93 5.98 -37.70 2.30 32.55
Loan
Sbt (bps) 688.68 308.65 261.00 599.67 2328.82
StLoan (bps) 583.75 192.18 352.23 553.60 1466.37
Table 8:
∆yt+3 Industry employment (%) 0.23 2.24 -13.28 0.55 7.20

Loan (bps)
Sbt 688.68 308.65 261.00 599.67 2328.82
Table 9:
Loan [GDP] (bps)
∆Sbt 1.19 42.30 -125.98 -2.02 323.89
Loan [Industry] (bps)
∆Sbt 1.94 37.19 -119.96 -1.31 260.16
Loan [EFD] (bps)
∆Sbt 2.79 42.75 -122.38 0.07 270.06
Loan [Elastic Net] (bps)
∆Sbt 2.52 43.09 -120.86 0.38 291.65
A.3
Table A.4 provides a comparison of the bond and loan market. Loan market data is
a combination of LSTA data for secondary market quotes complemented with information
about the underlying loans from the DealScan database. Bond market data is a combination
of monthly TRACE data for prices combined with Mergent FISD for information about the
underlying bond.
The loan and bond market differ in many dimensions. Bonds exhibit a longer maturity
at issuance (11.6 years) and term to maturity (8.2 years) than loans (6.0 and 4.6 years,
respectively). On average, bond issues tend to be of larger size ($537million v $454million)
than loans. The vast majority of loans are structured as senior and secured, whereas only
9% of bonds are secured.
Table A.4: Bond and loan market comparison

This table shows summary statistics for the bond and loan market. Loan market data comes from LSTA and Dealscan. Bond
market data comes from TRACE.
Panel A. Bond Market Characteristics
No. Bonds per month 4297.2 906.2 2980.0 4010.0 5817.0

No. Bonds per firm/month 5.4 10.3 1.0 2.0 138.0
Offering Amount ($mill) 536.9 544.3 0.0 400.0 15000.0
Maturity at issue (years) 11.6 8.4 0.5 9.8 50.0
Term to Maturity (years) 8.2 7.7 0.0 5.6 30.0
Duration (years) 6.0 4.1 0.0 5.0 19.2
Coupon (%) 6.2 2.5 0.4 6.2 18.0
Secured (%) 9.0 - - - -
Bond Spread (bps) 405.1 437.5 13.6 260.2 3495.6
Panel B. Loan Market Characteristics
No. Loans per month 1183.2 530.3 443.0 977.0 2125.0

No. Loans per firm/month 2.7 2.5 1.0 2.0 31.0
Facility Amount ($mill) 454.4 666.6 1.0 250.0 24000.0
Maturity at issue (years) 6.0 1.4 0.1 6.0 27.0
Term to Maturity (years) 4.6 1.7 0.0 4.6 26.6
Facility Amount ($mill) 454.4 666.6 1.0 250.0 24000.0
All In Spread Drawn (bps) 401.5 194.4 12.5 350.0 1600.0
Secured (%) 92 - - - -
Senior (%) 99 - - - -
Loan Spread (bps) 577.8 429.0 12.7 448.6 3477.0
A.4
C Additional institutitional background
C.1 Loan market liquidity
Figure A.9 plots the median bid-ask spread (scaled by the mid-quote) over the 1999-2020
period as well as the interquartile range in grey. The median bid-ask spread in the pre
2008-2009 crisis period was 68 basis points (bps). For comparison, Feldhütter and Poulsen
(2018) report an average bid-ask spread for the U.S. bond market of 34 bps over the 2002 to
2015 period. This suggests that the secondary loan market is still somewhat less liquid than
the bond market.
Section 3.3 finds the inclusions of the median bid-ask spread as an additional control,
has very little impact on the main result.
0.09
Bid−Ask (Median)
0.06
0.03
0.00
2000 2005 2010 2015 2020
Figure A.9: U.S. secondary loan market liquidity

This figure plots the median bid-ask spread (scaled by the mid quote) over the 1999:11 to
2020:03 period as well as the interquartile range. The sample is restricted to term loans
issued by U.S. firms. Source: LSTA.
As might be expected, bid-ask spreads surged during the 2008-2009 period, reaching a
median of 500bps in 2009. Following the crisis period the bid-ask spreads fell and stabilized
again at a level higher than prior to the crisis. A similar phenomenon can be observed in the
bond market. However, Dick-Nielsen and Rossi (2019) provide evidence that liquidity provi-
A.5
sion has become more expensive after the financial crisis and link this fact to a more stringent
regulatory environment, which disincentivized dealers from taking on large inventories in the
post-Volcker rule world.23
0.25
0.20
0.15
Fraction of loans
0.10
0.05
0.00
AAA AA+ AA AA− A+ A A− BBB+ BBB BBB− BB+ BB BB− B+ B B− CCC+ CCC CCC− CC D SD
Rating category
Figure A.10: Rating distribution of traded and non-traded loans

This figure plots the security level rating distribution across the loan market for loans that
are traded (grey bars) and loans that are not traded (white bars). Loan level ratings come
from Dealscan.
Figure A.10 shows the distribution of rating categories across the loan market for loans
which are traded (grey bars) and not traded (white bars) over the 1999 to 2020 period.
While the rating distribution is similar (particularly for the very low-risk and very high-risk
loans), the distribution of traded (non traded) loans has a larger mass in the BB to B (BBB)
rating categories, indicating that a significant portion of traded loans is below investment
grade.
C.2 Loan market dealers
Table A.5 shows the lead arranger (underwriter) market share in the primary market for the
top 10 dealers in the secondary market for 2009. The five largest dealer banks are Credit
23
Although the Volcker-rule restrictions on secondary market positions by banks was part of the Dodd-Frank
Act of 2011, it has ony really become operational at the regulatory level post 2015.
A.6
Suisse, Bank of America, Barclays, Citigroup, and JP Morgan with a combined market share
of 35%. These banks are also the largest underwriters in the primary loan market.
Table A.5: Top 10 dealers in the secondary loan market in 2009

This table shows the lead arranger (underwriter) market share in the primary syndicated loan market as well as the dealer
market share in the secondary market for syndicated loans for the top 10 dealers in 2009.
Name Dealer Market Share Underwriter Market Share

Credit Suisse 9.0% 6.6%
Bank of America 8.3% 12.3%
Barclays 7.7% 7.7%
Citigroup 6.0% 7.3%
JP Morgan 5.8% 12.4%
Morgan Stanley 5.7% 6.4%
Deutsche Bank 5.2% 5.1%
BNP Paribas 3.9% 2.9%
Wells Fargo 3.5% 3.2%
Royal Bank of Canada 3.3% 1.8%
A.7
D Aggregate forcasting results
D.1 Other measures and robustness
Table A.6 replicates Table 2, Panel A of the main paper but includes the loan spread in each
column.
Table A.6: Other Measures and Robustness

This table relates different credit spread measures and additional controls to future economic outcomes for the U.S. economy.
The unit of observation is the monthly level t. The sample period is 1999:12 to 2020:03. The dependent variable is the three-
month ahead percentage change in industrial production, i.e., the growth from t − 1 to t + 3. Panel A focuses on alternative
credit spreads. Column (1) uses the loan spread as a baseline comparision. Column (2) uses the Baa-Aaa corporate bond
spread. Column (3) uses the corporate high yield minus AAA spread. Column (4) uses the Commercial paper - three-month
Treasury bill spread. Column (5) uses the S&P500 monthly return from t − 1 to t. Panel B focuses on additional robustness
checks. Column (1) uses the loan spread as a baseline comparision. Column (2) controls for the median bid-ask spread in the
loan market at time t. Column (3) uses a residual loan spread from a regression of the loan spread on loan contract terms such
as size, age, amount, AISD, and indicators for secured, senior, and financial covenants. Column (4-5) contain the loan and
bond spread removing the 2009-09 crisis period. Each specification includes (not reported) a one period lag of the dependent
variable, i.e., growth from t − 2 to t − 1, the term spread, i.e., the difference between 10-year and three-month U.S. treasury
and the real FFR, i.e., the effective federal funds rate minus realized inflation. Incremental R2 refers to the difference between
the adjusted R2 in the respective column and the adjusted R2 of a baseline forecasting model, i.e., a model that only includes
a one period lag of the dependent variable, the term spread, and the real FFR (but no credit spread or equity market return).
Reported OLS coefficients are standardized. t-statistics, based on heteroskedasticity and autocorrelation corrected Newey-West
standard errors with a four period lag structure, are reported in parentheses.
Baseline Other credit spreads Equity market
(1) (2) (3) (4) (5)

Panel A. Other measures
∆StLoan -0.410 -0.380 -0.364 -0.407 -0.382
(-5.727) (-3.757) (-4.968) (-6.040) (-5.459)
∆ Baa-Aaa spread -0.044
(-0.567)
∆ HY-Aaa spread -0.117
(1.901)
∆ CP-bill spread 0.069
(0.939)
S&P500 return 0.148
(2.858)
Adjusted R2 0.313 0.312 0.323 0.315 0.331
Incremental R2 +0.150 +0.149 +0.160 +0.152 +0.168
Observations 241 241 241 241 241
D.2 Conditional on bond rating groups
As discussed in Section 3.2 of the main paper, the loan market is populated by a set of
riskier borrowers than the bond market. Hence, the superior predictive power of the loan
spread could reflect credit risk. A cleaner evaluation of the loan spread’s additional predictive
power would be to compare the loan spread to a bond spread conditional on a set of riskier
A.8
borrowers.
Table A.7: Aggregate forecasting results by bond ratings

This table relates credit spread measures to (future) economic outcomes. The unit of observation in Panels A, B, and C is the
monthly level t. The sample period is 1999:12 to 2020:03. The dependent variable in Panel A is the 3 month ahead change
in industrial production payroll employment (h=3) i.e growth from t − 1 to t + 3. The dependent variable in Panel B is the
3 month ahead change in unemployment rate. The dependent variable in Panel C is the 3 month ahead change in non-farm
payroll employment. Col (1) uses a loan spread calculated from all firms and time periods. Col(2)-(4) uses alternative versions
of the bond spread for different ratings classes. The ratings are bond level ratings available throught TRACE. Each specification
includes (not reported) one period lag of the dependent variable i.e growth from t − 2 to t − 1, the term spread i.e difference
between 10year and 3month US treasury and the real FFR i.e effective federal funds rate minus realised inflation. Reported
OLS coeffecients are standardised. t-statistics, based on heteroskedasticity and autocorrelation corrected Newey-West standard
errors with a 4 period lag structure, are reported in parentheses.
Forecast horizon: 3 months
(1) (2) (3) (4)

∆StLoan -0.410
(-5.727)
0
∆StBond−A s
-0.144
(-1.417)
∆StBond−BBB -0.212
(-3.127)
∆StBond−HY -0.222
(-2.322)
Adjusted R2 0.313 0.184 0.207 0.212

Incremental R2 +0.150 +0.021 +0.044 +0.049
Panel B. Unemployment
∆StLoan 0.355
(2.808)
0
∆StBond−A s
0.068
(0.439)
∆StBond−BBB 0.157
(1.006)
∆StBond−HY 0.180
(1.244)
Adjusted R2 0.272 0.154 0.174 0.182
Incremental R2 +0.122 +0.004 +0.024 +0.032
∆StLoan -0.207
(-6.332)
0
∆StBond−A s
-0.051
(-1.151)
∆StBond−BBB -0.098
(-2.304)
∆StBond−HY -0.097
(-2.235)
Adjusted R2 0.839 0.832 0.839 0.845
Incremental R2 +0.041 +0.034 +0.041 +0.041
Table A.7 performs the same aggregate forecasting regression as Table 1 in the main
paper, but instead includes a bond spread conditional on bond level rating. We use bond
level pricing data from TRACE to create bottom-up corporate bond credit spreads following
the methodology of Gilchrist and Zakrajšek (2012). This gives us the ability to create bond
A.9
spreads aggregated for different rating categories. Column (1) shows the baseline loan spread
results for comparison, Column (2) shows results using a bond spread comprised of A-rated
bonds, Column(3) uses BBB-rated bonds, and Column (4) uses below investment grade-rated
bonds.
Firstly, we see a montonic increase in the bond coefficient size as we condition on a

risker set of borrowers in the bond market. Secondly, we see that even a bond spread
conditional on being non-investment grade does not match the predictive power of the loan
spread. A one standard deviation increase in loan spreads is associated with a decrease in
industrial production by about 0.410 standard deviations, whereas a one standard deviation
increase in bond spread conditional on non-investment grade bonds, is associated with a
decrease in industrial production by only 0.222 standard deviations. A similar conclusion
can be drawn from comparing the incremental R2 . This highlights that the riskiness of the
underlying borrowers is not solely responsible for the wedge between the loan and bond
market predictability.
D.3 Conditional on increases versus decreases
As discussed in Section 3.3 of the main paper, we also consider potential asymmetries in the
impact of the loan and bond spread. The baseline aggregate forecasting regressions presented
in Table 1 of the main paper assume that an increase and a decrease of the loan spread has
the same relationship with future economic activity. In Table A.8 we test for potential
asymmetric impacts. The results suggests that both increases and decreases in loan spreads
are significantly correlated with future economic developments. However, the effect of spread
increases is somewhat stronger than the effect of spread decreases. Consistent with Stein
(2014), for the bond spread we only find an effect for spread increases. This suggest that
bond spreads primarily capture deteriorations in economic conditions, while loan spreads
capture both improving as well as deteriorating conditions.
A.10
Table A.8: Baseline forecasting results conditional on increases versus decreases
This table relates credit spread measures to future economic outcomes for the U.S. economy. The unit of observation is the
monthly level t. The sample period is 1999:12 to 2020:03. The dependent variable is the three-month ahead percentage change
in industrial production (h=3), i.e., growth from t − 1 to t + 3. Column(1) interacts the loan spread with a dummy for increases
and decreases in the loan spread. Column(2) interacts the bond spread with a dummy for increases and decreases in the bond
spread. Each specification includes (not reported) one period lag of the dependent variable, i.e., growth from t − 2 to t − 1,
the term spread, i.e., the difference between 10-year and three-month U.S. treasury and the real FFR, i.e., the effective federal
funds rate minus realised inflation. Incremental R2 refers to the difference between the adjusted R2 in the respective column
and the adjusted R2 of a baseline forecasting model, i.e., a model that only includes a one period lag of the dependent variable,
the term spread, and the real FFR (but no credit spread). Reported OLS coeffecients are standardised. t-statistics, based
on heteroskedasticity and autocorrelation corrected Newey-West standard errors with a 4 period lag structure, are reported in
parentheses.
(1) (2)
∆StLoan x Increase -0.487
(-6.043)
∆StLoan x Decrease -0.225
(-1.756)
∆StBond x Increase -0.294
(-3.748)
∆StBond x Decrease -0.040
(-0.253)
t-stat (∆StLoan x Increase = ∆StLoan x Decrease) (-1.62) (-1.52)

Adjusted R2 0.317 0.207
Incremental R2 +0.154 +0.044
Observations 241 241
D.4 Out of sample evaluation
We further test the predictive ability of loan and bond spreads on pseudo out-of-sample data.
To do this we use a standard expanding rolling window root mean square error (RMSE) test.
We first run the aggregate forecasting regressions on an initial training window (initially 150
observations) and use the resulting coefficients to predict the 151st macroeconomic variable.
We repeat this exercise including the next observation and so forth. We compute the RMSE
of the predictions against the realized macroeconomic outcome.
Table A.9 reports the RMSE across of variety of model specifications. Column (1) exam-
ines a baseline model with no credit spreads (only control variables). Column (2) examines
a model with only the loan spread. Column (3) uses only the bond spread and Column (4)
uses a model with both loan and bond spreads. Across all panels it is the model with the
loan spread that performs best, i.e. that has the lowest RMSE. A Diebold-Mariano test of
the difference in forecasting power (between Column 2 and 3) confirms that the loan spread
is superior at forecasting out of sample for industrial production and payroll employment,
A.11
consistent with in sample results reported in the main paper.
Table A.9: Aggregate forecasting results - Out of sample forecast

This table reports the RMSE of an out of sample regression of credit spread measures to (future) economic outcomes. The
unit of observation in Panels A, B, and C is the monthly level t. The training period is 1999:12 to 2012:05 or 150 observations.
We use an expanding rolling window to calculate RMSE error between the predictions (1 period ahead) and realized macro
outcome. The dependent variable in Panel A is the 3 month ahead change in industrial production (h=3) i.e growth from t − 1
to t + 3. Panel B is the 3 month ahead change in industrial production unemployment (h=3) i.e growth from t − 1 to t + 3.
Panel C is the 3 month ahead change in payroll employment (h=3) i.e growth from t − 1 to t + 3. Col (1) uses a baseline model
with no credit spreads. Column (2) includes the loan spread. Column (3) includes the bond spread. Column (4) includes both
loan and bond spread. Each specification includes (not reported) one period lag of the dependent variable i.e growth from t − 2
to t − 1, the term spread i.e difference between 10year and 3month US treasury and the real FFR i.e effective federal funds
rate minus realised inflation. The Diebold-Mariano (DM) Test null hypothesis is that the two models have the same forecast
accuracy.
(1) (2) (3) (4)

(Baseline) (∆StLoan ) ( ∆StBond ) (Both)
RMSE 0.0132 0.0118 0.0131 0.0118

DM Test p-value (Col(2) = Col(3)) (0.03)
RMSE 0.3271 0.3242 0.3287 0.3248

RMSE 0.0018 0.0019 0.0019 0.0019

D.5 Controlling for loan level contract terms
Loan and bond contracts might be different with respect to e.g. nonprice contract terms (such
as maturity, collateral and covenants and other characteristics such as size, age, amount). To
control for the impact of contract terms on loan spreads we regress loan spreads on various
characteristics, such as loan age, loan size, (log) loan amount, the loan’s initial all-in-drawn
spread, remaining time to maturity, as well as indicators for secured loans, senior loans, and
financial covenants. We run the following regression
ln S it [k] = αb + β1 ln(Age) + β2 ln(Size) + β3 ln(Amt) + β4 ln(AISD) + β5 ln(T imem at)
+ β6 Secured(0/1) + β7 Senior(0/1) + β8 Covenants(0/1) + it [k]. (3)
A.12
Column 5, Panel A of Table A.10 summarizes the result of this regression. We then take
the residual and Column (8) Table 2 of the main paper uses this “residual loan spread” and
finds very little difference in predictive power relative to the baseline loan spread in Table 1.
D.6 Excess loan premium
We follow Gilchrist and Zakrajšek (2012) and decompose the loan spread into two compo-
nents: a component that captures changes in default risk based on the fundamentals of a
firm or differences in contractual terms and a residual component that captures the price of
risk above a default risk premium. We implement the “naive distance to default” approach
described in Bharath and Shumway (2008). We start with an estimate of σE using the pre-
vious years daily returns. From this σD is approximated as σD = 0.05 + 0.25(σE ). We next
calculate a value of σV = σE [E/(D + E)] + σD [D/(D + E)]. µ is taken to be the stock return
over the previous year. The implied distance-to-default over the next one year horizon from
this procedure is then calculated as:

DD = (ln((E + D)/D) + µV − 0.5σV2 / (σV ) . (4)
Decomposition: To isolate the portion of the loan spread driven by variation in the
expected default of the firm or contractua terms, we regress the natural logarithm of the
loan spread of loan k on the average distance-to-default across all firms in the industry in
the respective month (DDbt ). We do not use the firm-specific DD as this measure cannot be
calculated for private firms, which comprise about 40% of our sample.24 We further include
2
a squared term (DDbt ) to capture a possible non-linear effect of DD on loan spreads, as well
as the volatility of DD across firms in the industry (σDDbt ). We run the following regression
2 2 0
ln S it [k] = αb + β1 DDbt + β2 DDbt + β3 σDDbt + γ Zit [k] + it [k]. (5)
24
Further, DD can also only be calculated for a subset of public firms with sufficient coverage in CRSP and
Compustat. Overall, we are able to calculate DD for about one-third of the firms in our sample.
A.13
Table A.10: Decomposing the Loan Spread
This table shows the results of the loan spread decomposition based on Gilchrist and Zakrajšek (2012). Panel A regresses the
loan spread on loan level characteristics. The dependent variable is the loan spread for facility i at time t. DDbt is the average
2
distance-to-default across all firms in the industry in the respective month based on Bharath and Shumway (2008). DDbt is
the distance to default squared. σDDbt is the volatility of DDbt across firms in the same industry.AISD is the all-in-spread-
drawn measured in basis points. Age is measured as the time elapsed since the loan is first reported in Dealscan. Amount is
measured as the par amount of the loan at issuance. Covenants is a dummy variable that equals 1 if the loan contract includes
covenants. Secured is a dummy variable that equals 1 if the loan is secured by collateral. Senior is a dummy variable that
equals 1 if the loan is senior. t-statistics, based on time and loan clustered standard errors, are reported in parentheses. Panel
B relates the predicted spread and residual spread from Panel A to future economic outcomes. The dependent variable in
Panel B is the three-month ahead percentage change in industrial production, i.e., growth from t − 1 to t + 3. Contribution
from ∆StLoan [Predicted] measures the proportion of the increase in adjusted R2 in the respective column that results from the
inclusion ∆StLoan [Predicted] as opposed to ∆StLoan [Residual]. Reported OLS coefficients are standardized. t-statistics, based
on heteroskedasticity and autocorrelation corrected Newey-West standard errors with a 4 period lag structure, are reported in
parentheses.
Panel A. Decomposing loan spreads

(1) (2) (3) (4) (5)
DDbt −0.357 −0.434 −0.435 −0.417
(−35.251) (−51.707) (−52.299) (−51.264)
2
DDbt 0.022 0.028 0.028 0.027
(26.631) (41.476) (41.888) (39.779)
σDDbt 0.023 0.010 0.010 0.010
(6.965) (3.648) (3.582) (4.734)
Ln(AISD) 0.735 0.732 0.642 0.685
(38.270) (34.482) (29.518) (32.143)
Ln(Age) 0.075 0.075 0.067 0.040
(31.564) (31.618) (30.144) (13.797)
Ln(Amount) −0.078 −0.078 −0.061 −0.093
(−12.127) (−11.963) (−9.842) (−13.592)
Secured(0/1) −0.018 0.012 0.086
(−0.760) (0.499) (3.284)
Covenants(0/1) −0.011 0.011 0.035
(−0.826) (0.870) (2.611)
Senior(0/1) 0.018 0.089 0.025
(0.404) (1.006) (0.464)
Loan type fixed effects No No No Yes No
Industry fixed effects No No No Yes No
Rating fixed effects No No No Yes No
Adjusted R2 0.087 0.407 0.407 0.456 0.315
Observations 287,811 287,811 287,811 287,811 287,811
Panel B. Aggregate Forecasting regression

∆StLoan [Predicted] −0.345
(−6.196)
∆StLoan [Residual] −0.265
(−4.181)
(−6.369)
(−4.767)
(−6.373)
(−4.766)
(−6.519)
(−4.649)
Adjusted R2 0.332 0.332 0.332 0.334
Incremental R2 +0.169 +0.169 +0.169 +0.170
% Contribution from ∆StLoan [Predicted] 0.63 0.69 0.69 0.71
A.14
The top panel of Table A.10 shows the results of these regressions. Column (1) begins
by including only the DD regressors. As expected, a higher DD reduces loan spreads and
2
the positive coefficient on DDbt is consistent with a non-linear effect. Column (2) then adds
a vector of loan-level control variables (Z(it) [k]), including the (log) loan amount, the (log)
age of the issue and (log) AISD. Column (3) further includes a dummy variable indicating
whether the loan includes financial covenants, is a secured loan, and is senior. The signs of
the coefficents are as expected: larger loans or those that include covenants have lower loan
spreads. Loans that are secured, have a higher AISD and have been on the market longer
have higher spreads. These loan terms, which are designed to address firms’ default risk, have
considerable explanatory power for spreads increasing the adjusted R2 to about 40%. The
coefficients remain similar across these specifications, but the explanatory power somewhat
improves. Column (4) further includes fixed effects for loan type, borrower industry and
loan rating category, the main results remain unchanged. We then calculate the predicted
loan spread as
 
2 0 σˆ2
Loan
Ŝbt = exp βˆ1 DDbt + βˆ2 DDbt + βˆ3 σDDbt
2
+ γ̂ Zit [k] +  (6)
2
The predicted component of the loan spread Ŝit [k] reflects the fundamental default risk of firm
i. We also aggregate the predicted component across all firms and obtain an aggregate time
series ŜtLoan . The residual loan spread, in the spirit of Gilchrist and Zakrajšek (2012)’s excesss
bond premium, is then defined as the difference (Residual Loan Spreadt = StLoan − ŜtLoan ),
i.e., the part of the loan spread that cannot be explained by default risk or contract terms.
Panel B of Table A.10 then uses the “predicted” and “residual” components of the loan
spread in an aggregate forecasting regression to mimic Panel A of Table 1 in the main paper.
The dependent variable here is the three-month ahead growth in industrial production. We
find both components of the loan spread are statistically significant at the three-month
horizon. Interestingly, it is the predicted component that contributes around two thirds of
the incremental adjusted R2 , suggesting that the predictive power of the loan spread is at
least partly driven by borrower fundamentals.
A.15

Excess Loan Premium

Uploaded by

Copyright:

Available Formats

Excess Loan Premium

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Excess Loan Premium

Uploaded by

Copyright:

Available Formats

Corporate Loan Spreads and Economic Activity∗

Anthony Saunders Alessandro Spina

November 14, 2020

JEL classification: E23, E44, G20

Electronic copy available at: https://ssrn.com/abstract=3717358

We introduce a novel loan market-based credit spread to predict economic outcomes.

2018−01 2018−07 2019−01 2019−07 2020−01 Loan Bond Loan Bond

Figure 1: Motivating evidence and main results

Finally, we contribute to the strand of literature which highlights the importance of

Our paper proceeds as follows. In Section C, we provide details on the construction of

2 Constructing the loan credit spread measure

As we use monthly measures of economic activity in our forecasting regressions, we

3 Loan spreads and economic activity

3.1 Baseline results

∆y t+h = α + β∆y t−1 + γ∆St + λT S + φRF F + t+h , (2)

3.4 Evidence from European countries

4 Exploring the mechanism

4.1 Effect by firm size, age, and listing

4.2 Effect by ratings

4.3 Credit supply

In this section we show there is additional information to be captured by going beyond

5.1 Baseline results

5.2 Across industry heterogeneity

6 Weighting schemes using industry-level information

Estrella, A. and Hardouvelis, G. (1991). The term structure as a predictor of real

— and Mishkin, F. S. (1998). Predicting u.s. recessions: Financial variables as leading

Feldhütter, P. and Poulsen, T. K. (2018). What determines bid-ask spreads in over-

Rajan, R. G. and Zingales, L. (1998). Financial dependence and growth. American

Stein, J. C. (2014). Incorporating financial stability considerations into a monetary policy

2000 2005 2010 2015 2020

Figure 2: Secondary loan market trading volume

Fraction of loans 0.3

Figure 3: Rating Distribution

2000 2005 2010 2015 2020

Figure 4: Corporate credit spreads

Coeffecient on delta spread at t+h

Coeffecient on delta spread at t+h

Figure 5: Local Projections

Figure 6: European country loan and bond spreads

Manufacturing Mining Professional services Retail trade

Figure 7: Industry loan spreads

Figure 8: Industry heterogeneity

Forecast horizon: h = 3 months

(1) (2) (3)

Adjusted R2 0.313 0.198 0.311

Panel B. Unemployment Rate

Adjusted R2 0.272 0.156 0.274

Panel C. Payroll Employment

Adjusted R2 0.839 0.806 0.838

Forecast horizon: h = 3 months

Baseline Other credit spreads Equity market

(1) (2) (3) (4) (5)

Baseline Loan mkt liquidity Contract terms Ex 2008-09 crisis

(1) (2) (3) (4) (5)

Forecast horizon: h = 3 months

Manufacturing Index U/E Construction Index

Adjusted R2 0.260 0.410 0.090

Adjusted R2 0.190 0.210 0.082

Adjusted R2 0.190 0.710 0.140

Forecast horizon: h = 3 months

∆y t+h = α + β∆y t−1 + γ∆St + λT S + φRF F + t+h , (2)