Professional Documents
Culture Documents
Trade Credit 1
Trade Credit 1
Abstract
We empirically investigate the proposition that firms charge premia on cash prices in transactions
involving trade credit. Using a comprehensive panel data set on product-level transaction prices and
firm characteristics, we relate trade credit issuance to price setting. In a recession characterized by
tightened credit conditions, we find that prices increase significantly more on products sold by firms
issuing more trade credit, in response to higher opportunity costs of liquidity and counterparty risks.
Our results thus demonstrate the importance of trade credit for price setting and show that trade credit
issuance induces a channel through which financial conditions affect prices. (JEL: E31, E32, D22,
G30, L11)
1. Introduction
Does trade credit issuance affect firms’ price-setting behavior? Research on the links
between firm characteristics and price setting has demonstrated that due to capital
market imperfections, firms’ leverage and liquidity positions make for important
1. In Sweden, 97% of business-to-business transactions involve trade credit (Pärlhem 2016). Jacobson
and von Schedvin (2015) show that the average amount of accounts receivable and payable, scaled by
assets, are 16% and 11% for Swedish firms. Similar reliance on trade credit financing prevails across
countries. For instance, Rajan and Zingales (1995) show that the corresponding numbers for a sample of
US firms are 18% and 15%, whereas Berger and Udell (1998) show that trade credit provides 31% of debt
financing to US SMEs, which is nearly as much as commercial banks.
2. Udell (2015) discusses what he labels “the trade credit pricing puzzle”, and notes that “the ‘all-in’
price—that which matters—must incorporate both the price of the product as well as the financial terms
of trade credit. It is highly unlikely that there are any available data that would allow us to calculate this
all-in price”.
Amberg, Jacobson, and von Schedvin Trade Credit and Product Pricing 3
rate during the recession. Importantly, the richness of our data allows us to carefully
3. “2/10 net 30” contracts stipulate that the buyer must pay within 30 days, whereas offering a discount
of 2% for payments made within ten days. This implies that buyers are charged an implicit annualized
interest rate of 44.6% when they do not avail themselves of the discount offer.
4. Only a small fraction, around 3%, of Swedish firms use factoring services (Pärlhem 2016).
4 Journal of the European Economic Association
enforce late trade credit payments. The results of the cross-sectional heterogeneity
5. For US firms, Ng, Smith, and Smith (1999) find that 25.5% of the firms in a sample drawn from
Compustat mainly offer two-part contracts, whereas Giannetti, Burkart, and Ellingsen (2011), using the
1998 National Survey of Small Business Finances, document that firms on average are offered early
payments discounts from 21.3% of their suppliers. Moreover, Giannetti et al. find that only 7.8% of firms
operate in industries in which discounts are common. Using more recent data from 2005, Klapper, Laeven,
and Rajan (2012) find that 13% of the contracts extended to a sample of large US and European buyers
included early payment discounts. In Sweden, the dominance of net terms contracts over two-part contracts
appears to be even larger. For example, Ellingsen, Jacobson, and von Schedvin (2016) find no evidence
of the use of two-part contracts in a sample comprising over 50 million trade credit contracts issued by
a sample of Swedish suppliers. Moreover, there is no mention of early payments discounts in either of
two recent inquiries on trade credit payment terms commissioned by the Swedish government (Bengtsson
2007; Pärlhem 2016).
6. Note that the argument that product prices may include implicit interest rates also has implications
for two-part contracts—the credit extended during the initial period of a two-part contract is commonly
viewed as a zero-cost loan, but this credit could also be priced through an implicit interest rate.
Amberg, Jacobson, and von Schedvin Trade Credit and Product Pricing 5
cash transactions when buyers are located in countries with weak enforcement of
The 2008–2009 recession in Sweden is well-suited for testing the hypothesis that
product prices include a trade credit premium, since it featured a sharp downturn in
the real economy as well as severe distress in the banking sector, both of which were
caused by external shocks hitting the Swedish economy in the wake of the global
financial crisis.
The banking sector distress had two main causes. The first was the collapse of
international financial markets following the outbreak of the global financial crisis.
Although Swedish banks had little direct exposure to asset-backed securities issued
in the United States, they are highly dependent on external wholesale funding and
are therefore sensitive to funding conditions in international financial markets, which
deteriorated rapidly after the onset of the crisis. The second cause was the severe
economic crisis in the Baltic countries, which two of Sweden’s four major banks were
highly exposed to as a result of having expanded in the Baltic market in the years prior
to the crisis. These two shocks led to elevated distress in the financial system, although
observers’ judgments differ somewhat as to the severity of the distress. According to the
IMF’s banking crisis database, for example, Sweden suffered a “borderline” systemic
banking crisis beginning in 2008 (Laeven and Valencia 2012), whereas Romer and
Romer (2017), using a financial distress measure ranging from 0 to 15, classifies the
level of distress in Sweden during 2008–2009 as 5 on average, with a peak value of 7.
The banking sector distress quickly led to a deterioration in the credit conditions
facing corporate borrowers: beginning in 2008 and continuing throughout 2009,
growth in bank lending to firms fell steadily (Finansinspektionen 2012), and many
firms reported on a worsening access to external finance (Konjunkturinstitutet 2009;
Sveriges Riksbank 2009). Meanwhile, the real economy fell into a sharp recession,
with a decline in real GDP of around 6% in 2009, partly due to the domestic banking
sector distress and partly due to the breakdown in international trade that occurred
during this period (see, e.g., Levchenko et al. 2010), which affected the export-oriented
Swedish economy badly. The recession did thus not only impair firms’ cash flows and
their access to external finance—and thereby their liquidity positions—but also led to
increased counterparty risks, not least manifested in a near-doubling of the aggregate
bankruptcy rate. Moreover, it is conceivable that corporate credit risk premia increased
even more dramatically than the bankruptcy rate. Berndt et al. (2018), for example,
document that risk premia for public firms in the United States—measured as the
difference between a firm’s credit default swap rate and its expected default loss rate—
peaked in the financial crisis of 2008–2009, when the cost for credit insurance per unit
of expected loss increased by a factor 10 on average.
The empirical analysis is based on data from four sources, which we merge
unambiguously by means of the unique identifier (organisationsnummer) attached
Amberg, Jacobson, and von Schedvin Trade Credit and Product Pricing 7
to each Swedish firm. First, we obtain data on prices and quantities from “Industrins
7. These data have previously been used by Carlsson and Skans (2012). To give an idea of the granularity
of the product classification, we can, for example, compare the codes 84212100 and 84212200, which refer
to “machinery and apparatus for filtering or purifying water” and “machinery and apparatus for filtering
or purifying beverages (excl. water)”, respectively.
8 Journal of the European Economic Association
Oi07 is thus a proxy for firm i’s average trade credit maturity, measured in years, across
Table 1 reports descriptive statistics for all variables used in the empirical analysis.
The mean (median) firm-product inflation rate, reported in panel A, is 2.9% (0.7%).
The average value of Oi07 , reported in panel B, is 0.097, which corresponds to a trade
credit contract maturity of 35 days. Average cash holdings amount to 8.3% of total
assets, whereas the average size of the unused part of firms’ credit lines is 4.3% of total
assets. Panel C shows the values of the time-varying firm characteristics. The average
firm has a book value of assets of 319 million SEK and sales of 390 million SEK
(roughly 49 and 60 million USD, respectively, at the exchange rate prevailing at the
end of 2007). The sample thus consists primarily of medium-sized and large firms. In
Figure 1, we show that our sample is representative of the broader economy in terms of
price changes. More specifically, the figure shows that the average annual firm-product
8. A firm’s accounts receivable is the sum of all claims held by the firm on its customers for goods and
services that have been delivered but not paid. Hence, it measures the total amount of trade credit extended
by a firm to all of its customers at a given point in time. By dividing accounts receivable by sales, one
obtains a proxy for the average maturity of the trade credit issued by a given firm, measured in years. More
precisely, Oi07 measures average time to payment, which may differ from contracted payment time due
to either late or premature payments. Note that value-added tax (VAT) is included in accounts receivable
as reported on firms’ balance sheets, but not in the sales figures on the profit-and-loss statements. Hence,
the amount of receivables observed for a given firm-year observation needs to be divided by one plus the
applicable VAT rate for Oi07 to capture the true average maturity (the standard VAT rate in Sweden is 25%,
but for a few of the industries in our sample, a lower VAT rate of 12% applies).
9. We do not observe labor costs and intermediate input costs at the product level, so we must resort to the
following approximation when calculating UICi;p;t . For each plant and year, we portion out plant-level
labor costs and intermediate input costs across the products produced at the plant in proportion to the plant-
level sales share of each product. We then compute the plant-product level values of UICi;p;t . Finally, we
aggregate UIC i;p;t to the firm-product level using total sales for each product-plant as weights.
Amberg, Jacobson, and von Schedvin Trade Credit and Product Pricing 9
Notes: This table reports descriptive statistics for all variables used in the empirical analysis, as well as for
some additional firm characteristics. Each firm appears only once in panel B and only once per year in panel C,
irrespective of how many products it sells and thus how many firm-product observations it has in a given year.
Definitions of the variables are provided in the text.
inflation rates in our sample track the changes in the aggregate producer price index
for the goods-producing sector of the economy quite closely over the sample period.
In Figure 2, finally, we document an important feature of trade credit maturities
that will help inform our conceptual framework and empirical approach, namely,
that maturities are persistent over time. More specifically, the figure plots average
trade credit maturities, as measured by Oi;t , for firms above and below the sample
medians of Oi;t and Oi07 , respectively. Average trade credit maturities are very stable
over the period 2004–2011 for both above- and below-median firms when using the
yearly classification, whereas above-median firms, as classified based on Oi07 , reduced
maturities slightly during the crisis, with Oi;t falling from 0.129 to 0.112.
In standard formulations of the firm’s price-setting problem, the optimal price for
product p sold by firm i, Pi, p , is equal to the product of the firm’s marginal cost for
10 Journal of the European Economic Association
.05
F IGURE 1. Price changes in sample and in the aggregate economy. The solid line shows the average
annual firm-product inflation rate in our sample and the dashed line the annual change in the
aggregate producer price index for the entire goods-producing economy (SNI/NACE sections A– E).
We calculate the latter as the log change in the annual average of the monthly values of the producer
price index. Source: Statistics Sweden and authors’ calculations.
.08 .16
.12
τ
.04
0
F IGURE 2. Average trade credit maturities over time. The figure shows average trade credit
maturities, as measured by Oi;t , for firms above and below the sample medians of Oi;t and Oi07 ,
respectively, over the period 2004–2011. The solid and short-dashed lines show average maturities
when the classification of firms is done on a year-by-year basis (Oi;t ), and the long-dashed and
dashed-dotted lines when the classification is done based on the trade credit maturity distribution in
2007 (Oi07 ).
Amberg, Jacobson, and von Schedvin Trade Credit and Product Pricing 11
producing p, MCi, p , and a mark-up, i, p , that depends on the firm’s price-setting power
in which the firm-product inflation rate in year t is determined by the change in the
To test the hypothesis outlined in the previous section, we apply the following empirical
specification:
i;p;t D ˇ Crisist Oi07 C ˛i;p C ˛t C Xi;p;t C ı Zi;t1 C "i;p;t ; (5)
where i, p,t is the firm-product inflation rate; Crisist is a dummy variable equal to one
in the years 2008 and 2009 and zero otherwise; Oi07 is the average trade credit maturity
for firm i in 2007; ˛ i, p and ˛ t are firm-product and year fixed effects, respectively; and
Xi;p;t and Zi;t1 are the vectors of control variables defined in Section 2.2. We fix the
maturity variable to each firm’s last precrisis value to mitigate endogeneity concerns,
but we show in what follows that our baseline result is virtually unchanged if we instead
use a time-varying measure of maturities, which reflects the considerable persistence
in Oi;t documented previously. The firm-product fixed effects control for potential
10. Antitrust legislation may to some extent limit firms from engaging in price discrimination. Similarly
to the US setting, EU Competition Law (Article 82(c) of the EC Treaty) dictates what Swedish firms
can and cannot do in terms of setting differential prices across buyers, and under what circumstances.
Whereas it is true that firms cannot fully customize prices—although Article 82(c) has quite strict and
precise requirements for making a case of price discrimination—free pricing is not a prerequisite for our
analysis, neither conceptually, nor empirically. Indeed, the average interest rate and maturity parameters
in equation (2) are consistent with firms setting standardized prices equal for all their buyers on the basis
of the opportunity costs of liquidity they face and the average counterparty risks in the pool of buyers they
service.
11. This treatment of trade creditors in bankruptcy proceedings is similar to international practice (see
Cuñat and Garcia-Appendini 2012 for an overview).
Amberg, Jacobson, and von Schedvin Trade Credit and Product Pricing 13
time-invariant differences in price setting between firms with low and high trade credit
12. In exploiting an aggregate shock combined with cross-sectional variation in firms’ exposure to the
shock, our empirical set-up is similar to those used in several papers in the previous literature on the financial
crisis of 2008–2009, such as Duchin, Ozbas, and Sensoy (2010), Garcia-Appendini and Montoriol-Garriga
(2013), and Gilchrist et al. (2017).
14 Journal of the European Economic Association
Cash/Assets07
i
0.076 0.112 0.091 0.141 0.110 0.235 0.938 0.991
Bank debt/Assets07 i
0.146 0.165 0.111 0.159 0.217 0.038 0.976 0.971
Tangible assets/Assets07
i
0.235 0.174 0.270 0.195 0.190 0.116 0.912 0.974
Cash flow/Assets07 i
0.139 0.127 0.145 0.149 0.043 0.160 0.914 0.974
07
Inventories/Salesi 0.150 0.109 0.134 0.111 0.141 0.023 0.906 0.974
ln.Assets07
i
/ 10.947 1.281 11.166 1.585 0.153 0.213 0.877 0.983
Export share07j.i/
0.305 0.229 0.293 0.234 0.052 0.019 0.950 0.966
Unused LC/Assets07 i
0.047 0.071 0.039 0.068 0.115 0.045 0.977 0.974
CustPDc(i) 0.707 0.134 0.710 0.133 0.021 0.012 0.964 0.890
Salesc(i) 0.083 0.100 0.056 0.105 0.264 0.048 0.856 0.955
SumInji 0.121 0.740 0.108 0.748 0.018 0.011 1.000 1.000
NoInji 0.119 0.705 0.100 0.725 0.027 0.029 1.000 1.000
No. firms 1,752 1,656
Notes: This table reports descriptive statistics for firms with average trade credit maturities, Oi07 , above and below
the sample median (panels A and B), as well as four measures of covariate balance proposed by Imbens and
Rubin (2015) (panel C). The set of covariates comprises all firm-level control variables, measured in 2007, as
well as all sample-split variables used in the empirical analysis. h ` denotes a normalized difference and is
q
calculated as .XN h XN ` /= .Sh2 C S`2 /=2, where XN is the mean, S is the standard deviation, and subindices h and
` denote firms with trade credit maturities above and below the sample median, respectively. h=` is the log of
the ratio of the standard deviations for above-median and below-median firms. Finally, h:95 measures the share
of below-median firms for which the value of a given variable lies in the 95% central range of the distribution of
the same variable for above-median firms (and vice versa for `:95 ).
with average trade credit maturities above and below the sample median, respectively,
on all firm-level covariates used in the empirical analysis, measured in 2007. The
comparison is conducted on the basis of four measures of covariate balance proposed
by Imbens and Rubin (2015): normalized differences in means, logs of the ratios
of standard deviations, and two measures of coverage frequency. The results of this
comparison, presented in Table 2, show that firms issuing trade credit with long and
short maturities are similar across the entire set of covariates under consideration.13
This is consistent with the findings in the previous literature, which has struggled to
provide unambiguous evidence regarding the determinants of variation in trade credit
maturities across sellers (see Ellingsen, Jacobson, and von Schedvin 2016 for empirical
evidence and a review of the literature).
We assess potential threats to the identifying assumption in the following ways.
First, we estimate the baseline specification augmented with, in turn, product-year
fixed effects and industry-year fixed effects, which means that we effectively compare
firms with high and low trade credit issuance selling the same product, or operating
13. To see this, note that the largest normalized difference is 0.26 in absolute terms, whereas the lowest
coverage frequency is 0.86. As a comparison, Imbens and Rubin (2015), in an evaluation of the experimental
LaLonde (1986) data set, observe a maximum normalized difference of 0.30 and a minimum coverage
frequency of 0.91, which they judge to be excellent covariate balance.
Amberg, Jacobson, and von Schedvin Trade Credit and Product Pricing 15
.05
F IGURE 3. Average firm-product inflation rates over time. The figure shows average firm-product
inflation rates in each year of the sample period for firms above (solid line) and below (dashed line)
the median of the trade credit maturity distribution in year t 1, Oi;t1 .
in the same industry, in a given year. These specifications thus address threats from
the set of potentially confounding factors suggested by theories that posit that trade
credit maturities are determined by product- or industry-specific factors (e.g., Lee and
Stowe 1993; Long, Malitz, and Ravid 1993; Kim and Shin 2012; Kalemli-Özcan et al.
2014). Second, we estimate the baseline specification augmented with interactions
between the crisis variable and specific potential confounders that can vary within
product and industry clusters, and which are therefore not absorbed by product-year
or industry-year fixed effects. Third, we address an alternative explanation for why
trade credit issuance and prices may be positively related during periods of tight credit.
Suppose that demand increases for goods sold by firms issuing trade credit with longer
maturities as a consequence of trade credit becoming more valuable for buyers when
other sources of financing dry up. Increased demand could then push up the prices
set by these firms relative to the prices of firms that provide short maturities. To
distinguish between the demand- and supply-side explanations, we estimate a version
of the baseline specification with changes in the quantities of goods sold replacing
changes in prices as outcome variable; if the explanation based on demand is correct,
we should observe an increase in both prices and quantities for goods sold by firms
that issue long trade credit maturities, whereas we would observe higher prices, but
lower or unchanged quantities if the effects are supply-driven.
4. Main Results
(dashed line) the sample median in year t 1. Inflation rates for the two groups of firms
Table 3 reports results for various estimations of the model specified in equation (5).
We quantify the magnitude of the coefficients for the main explanatory variable, the
interaction term Crisist Oi07 , in two ways. First, we report the coefficients themselves,
which, as noted previously, capture the difference in r between crisis and noncrisis
years. Second, we calculate what the coefficients imply in terms of the difference in
annual firm-product inflation rates between firms that issue long and short trade credit
maturities by multiplying the coefficient with the difference in Oi07 between firms
located at the 75th and 25th percentiles, respectively, of the Oi07 -distribution.14
The baseline result is reported in column (1). The coefficient on the interaction
term Crisist Oi07 is positive and statistically significant, which suggests that firm-
product inflation rates are increasing in trade credit maturities in the crisis period.
The magnitude of the coefficient shows that r on average was 20.9 percentage
points higher during the crisis than in noncrisis years.15 This estimate implies that
the difference in annual firm-product inflation rates between firms that issue long and
short trade credit maturities was 1.3 percentage points higher in crisis years than in
noncrisis years.
Next, we re-estimate the baseline specification using weights that adjust for
differences in the shares of each firm’s total sales accounted for by each of its
products. More specifically, we estimate a weighted regression where the weight
for each observation, ! i, p, t , is calculated as firm i’s sales of product p divided by
firm i’s total sales. Hence, we effectively estimate the baseline regression at the firm
level instead of at the firm-product level. The results are reported in column (2). The
coefficient on the interaction term Crisist Oi07 implies a difference in r of 15.4
14. The difference between the 75th and the 25th percentiles of the Oi07 -distribution is 0.060, which
corresponds to a difference in trade credit maturity of around 22 days. Throughout the rest of this section,
we refer to firms located at the 75th and 25th percentiles of the trade credit maturity distribution as firms
that issue long and short trade credit maturities, respectively.
15. We obtain virtually the same point estimate, 0.211 (3.4), when we estimate the baseline specification
with lagged, time-varying trade credit maturities, Oi;t1 , instead of the time-invariant explanatory variable,
Oi07 .
TABLE 3. Baseline results.
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
C
i, p, t i, p, t i, p, t i;p;t i, p, t i, p, t i, p, t i, p, t i, p, t Qi, p, t
Crisist Oi07 0.209 0.154 0.415 0.232 0.125 0.182 0.214 0.230 0.008
(3.4) (3.2) (2.4) (3.9) (2.5) (3.4) (3.6) (3.8) ( 0.1)
Crisist Oi07 ; High 0.018
(2.8)
j
Crisist Export share07 0.046
(3.4)
i
Crisist Inventories/Sales07 0.083
(2.9)
High versus low Oi07 0.013 0.009 — 0.025 0.014 0.008 0.011 0.013 0.014 0.000
Firmproduct FE Yes Yes Yes Yes Yes No Yes Yes Yes Yes
Productyear and firm FE No No No No No Yes No No No No
Industryyear FE No No No No No No Yes No No No
Firm- and product-level controls Yes Yes Yes Yes No Yes Yes Yes Yes Yes
Weights No Yes No No No No No No No No
R2 0.429 0.420 0.429 0.319 0.214 0.590 0.458 0.430 0.429 0.306
No. firms 3,408 3,408 3,408 3,408 3,408 3,174 3,401 3,408 3,408 3,408
No. observations 45,953 45,953 45,953 45,953 45,953 35,971 45,921 45,953 45,953 45,953
Notes: This table reports results for estimations of various specifications based on equation (5). The dependent variable is the firm-product inflation rate, i, p, t , in all specifications
except those in column (4), where it is a dummy equal to one for price increases and zero otherwise, and in column (7), in which it is the change in the quantity of goods sold,
Amberg, Jacobson, and von Schedvin Trade Credit and Product Pricing
Qi, p, t . The regression in column (2) is estimated using WLS, where the weight for each observation, ! i, p, t , is calculated as firm i’s sales of product p divided by firm i’s total
sales in year t. The product fixed effects are based on 8/9-digit CN codes and the industry fixed effects on three-digit SNI/NACE codes. All regressions include year fixed effects
except that in column (7), in which they are redundant. High versus low Oi07 is calculated as the estimated difference in the dependent variable between firms located at the 75th
and 25th percentiles of the Oi07 -distribution. The estimation period is 2004–2011 in all columns. t-statistics calculated using robust standard errors clustered at the firm-level are
reported in parentheses. Significant at 5%; Significant at 1%.
17
percentage points and a difference in the annual firm-product inflation rate between
resulting coefficient, reported in column (6), is positive and statistically significant, but
reduce these through price cuts. The different signs of the estimated effects of trade
5. Mechanisms
In this section we propose to scrutinize the underlying mechanisms for the baseline
finding that firms issuing longer trade credit maturities tended to raise product prices
more during the crisis. Thus, we conjecture—in accordance with the hypothesis
outlined in the conceptual framework—that the association between trade credit
issuance and price changes is stronger for firms subject to larger increases in liquidity
costs and counterparty risk. To this end, we conduct cross-sectional heterogeneity
analyses, in which we estimate the baseline specification on subsamples of firms
defined by proxies for changes in liquidity costs and counterparty risk during the
crisis. The proxies we use capture firm characteristics obtaining in the crisis and we
therefore restrict the sample period to 2006–2009 throughout this section.
We use two approaches to identify firms that experienced particularly large increases in
liquidity costs during the crisis. First, we follow a common practice in the literature and
identify firms based on precrisis characteristics that are likely to have been important
determinants of their liquidity costs during the crisis. Second, we use the Baltic crisis
and its detrimental effect on two Swedish banks as a quasi-experiment along the lines
of Chodorow-Reich (2014).
three deciles, the former of which corresponds to firms experiencing large increases
16. We only observe lending from the four major Swedish banks and will therefore underestimate
Unused LC/Assets07
i;t1
for firms obtaining credit from minor banks. This will, if anything, lead us to
underestimate the difference between the two subsamples.
17. This empirical approach has subsequently been employed in several other studies, for example, Huber
(2018), Bentolila, Jansen, and Jiménez (2018), and Berton et al. (2018).
18. For more comprehensive overviews of the Baltic crisis, see the European Commission (2010) and
Hansson and Randveer (2013). A more thorough description of the Swedish financial system is provided
by Sveriges Riksbank (2015). Bryant, Henderson, and Becker (2012), finally, provide a detailed analysis
of the Swedish experience during the global financial crisis.
22 Journal of the European Economic Association
Notes: This table reports results for estimations of equation (5) on various subsamples of firms for the period
2006–2009. Column (1) in panel A reports results for estimations on subsamples consisting of firms in the top three
and bottom three deciles, respectively, of the Cash=Assets07i;t 1
distribution, whereas column (2) reports results
for a corresponding sample-split based on the Unused LC/Assets07 i;t 1
distribution. The cutoffs used to construct
the subsamples are defined at the firm level; hence, the number of firms in each subsample is approximately
the same, whereas the number of observations differ somewhat (note, though, that the number of firms differs
somewhat in the split based on credit lines due to the fact that more than 30% of firms have no unused credit
line). Columns (1)–(3) in panel B concern estimations of equation (5) on treated firms and control firms in the
Baltic crisis experiment. The samples underlying the estimations in the three columns comprise all firms with
a precrisis bank relationship (column 1); all single-bank firms (column 2); and single-bank firms except those
with bank D as their main lender (column 3). Reported p-values correspond to one-sided tests, where the null
hypothesis is that the estimates of ˇ are equal in each pair and the alternative hypothesis that the coefficients are
larger for firms in the bottom deciles (panel A) and in the treatment group (Panel B). t-statistics calculated using
robust standard errors clustered at the firm-level are reported in parentheses. Significant at 10%; significant
at 5%; significant at 1%.
The Swedish bank market is dominated by four major banks, which we refer to as
banks A, B, C, and D. These four banks operate countrywide and lend to households
and firms in all sectors of the Swedish economy; at the time of the crisis, they accounted
for about 85% of banking sector assets and 75% of corporate lending. Prior to the global
financial crisis, all of these banks had entered the Baltic market, but the scale of their
Baltic operations differed greatly, with two banks (C and D) having substantial shares
of their loan portfolios allocated to the Baltic market, whereas the other two had only
Amberg, Jacobson, and von Schedvin Trade Credit and Product Pricing 23
minor (bank B) or even negligible (bank A) shares of their lending there. The Baltic
19. The choices of banks C and D about whether or not to participate in the guarantee program were
of course deliberate decisions on the parts of the banks’ leaderships, rather than random outcomes. This
is, however, only problematic for our analysis to the extent that these choices were dictated by the
characteristics of the banks’ Swedish corporate borrowers. Although there is no consensus on what the
determinants of the banks’ choices were, it is unlikely that differences in customer characteristics was one,
as there is little sorting between firms and banks in the corporate segment of the Swedish bank market.
24 Journal of the European Economic Association
150
100
50
0
Mar−07
Jun−07
Sep−07
Dec−07
Mar−08
Jun−08
Sep−08
Dec−08
Mar−09
Jun−09
Sep−09
Dec−09
Mar−10
Jun−10
Sep−10
Dec−10
(b) Stock prices
125
A
100
Stock price (2006Q4=100)
75 B
50
D
C
25
0
Mar−07
Jun−07
Sep−07
Dec−07
Mar−08
Jun−08
Sep−08
Dec−08
Mar−09
Jun−09
Sep−09
Dec−09
Mar−10
Jun−10
Sep−10
Dec−10
110
Loan volume (2007Q4=100)
105
100
95
90
C A/B/D A/B
85
2007Q1
2007Q2
2007Q3
2007Q4
2008Q1
2008Q2
2008Q3
2008Q4
2009Q1
2009Q2
2009Q3
2009Q4
2010Q1
2010Q2
2010Q3
2010Q4
F IGURE 4. Swedish banks and the Baltic crisis. The figure shows five-year senior unsecured CDS
spreads (panel (a)); stock prices (panel (b)); and the total volume of outstanding loans to Swedish
nonfinancial firms (panel (c)) for the four major Swedish banks between 2007 and 2010. The dashed
line in panel (c) shows the joint loan volumes of banks A, B, and D, and the dashed-dotted line the
joint loan volumes of banks A and B. Sources: Thomson Reuters and Statistics Sweden.
Amberg, Jacobson, and von Schedvin Trade Credit and Product Pricing 25
bank A, B, or D to a control group.20 In this setting we can test the hypothesis that
5.1.3. The Baltic Crisis as a Quasi-Experiment: Results. The results for the Baltic
crisis quasi-experiment are presented in panel B of Table 4. The first row shows the
estimated coefficients for control firms (Ti D 0) and the second row the coefficients for
treated firms (Ti D 1). Consider first column (1), in which we present the results of the
estimation of the specification just described, in which the treatment group consists of
all firms whose precrisis main lender was bank C and the control group of all firms
whose main lender was bank A, B, or D. The results show that the coefficient of interest
is more than twice as large for treated firms as for control firms, but the difference
between the two subsamples is despite this not statistically significant. Hence, these
results provide some support for our conjecture, but the evidence is not conclusive.
Note, however, that although all treated firms had bank C as their main lender at
the outset of the crisis, some of them had active relationships with other banks as well,
and could therefore likely offset some part of the contraction in lending on the part of
bank C by increasing borrowing from their other banks. If so, some firms that were not
in fact affected by the negative loan supply shock would be included in the treatment
group, which would weaken the test. We therefore consider an alternative specification,
in which we drop all multiple-bank firms from the treatment group as well as from
the control group (the latter to avoid introducing differences in the characteristics of
treated firms and control firms). The results in column (2) show that the difference in
the estimated coefficients becomes even larger, 0.492 versus 0.192, when we consider
single-bank firms only; moreover, the difference is now statistically significant at the
5% level.
Finally, although bank D participated in the guarantee program and thereby largely
avoided the effects of the crisis on its funding costs, it is still possible that its lending
behavior was affected by the crisis. We therefore consider a specification that is
equivalent to that in column (2), except that the control group only includes firms whose
main precrisis lender was bank A or B. The results of this alternative specification,
reported in column (3), are, however, very similar to the results in column (2); hence,
whether or not borrowers of bank D are included in the control group matters little
for the results. In sum, the results of the analysis based on the Baltic crisis lends
further support to the claim that increases in liquidity costs is a mechanism underlying
the positive relationship between trade credit issuance and price changes during
the crisis.
20. Most firms have only one bank relationship at any given point in time and the main lender is then
simply the bank with which the firm has a relationship. For firms with multiple bank relationships, the
main lender is defined as the bank with the largest amount of lending to the firm at the outset of the crisis.
Firms with no banking relationships at all are excluded from the analysis.
26 Journal of the European Economic Association
This table reports results for estimations of equation (5) on various subsamples of firms for the period 2006–2009.
More specifically, the different columns report results for sample splits based on the change in buyer PDs (column
1); the change in sales in buyer industries (column 2); the change in sales in buyer industries instrumented
using the precrisis export share of each industry (column 3); the change in the sum of the claims underlying
the injunctions issued on behalf of each seller (column 4); and the change in the number of injunctions issued
on behalf of each seller (column 5). The changes are from 2007 to 2009 in all cases. In column (1), TiHigh is
equal to zero for firms in the bottom three deciles of the CustPDj.c/ distribution, and to one for firms in the
top three deciles. In remaining columns, TiHigh is equal to zero if Ti 0 and equal to one if Ti > 0. Reported
p-values correspond to one-sided tests, where the null hypothesis is that the estimates of ˇ are equal in each pair
and the alternative hypothesis that the coefficients are larger for firms with TiHigh D 1 in columns (1), (4), and
(5), and for firms with TiHigh D 0 in columns (2) and (3). The number of firms refer to the numbers of firms for
which TiHigh D 0 and TiHigh D 1, respectively. t-statistics calculated using robust standard errors clustered at the
firm-level are reported in parentheses. Significant at 5%; significant at 1%.
risk, but small and statistically insignificant in the group of firms for which the increase
5.2.2. Results Using Data on Injunctions Issued by the EA. The analysis in the
previous section relies on industry-level measures of counterparty risk, which by
construction are crude approximations for the true changes in risk facing individual
firms. We will therefore next consider a seller-level measure of risk, constructed using
data provided by the Swedish Enforcement Agency (EA), the government agency
tasked with coordinating the administrative process of bankruptcy resolution. One of
the responsibilities of the EA is to provide legal support to sellers for the management
of their unsettled trade credit claims. More specifically, a seller who holds an overdue
trade credit claim on a customer can request the EA to issue an injunction against the
customer; if the request is approved, the EA will notify the debtor for payment within
a fortnight and take further measures to enforce payment should the debtor persist in
dishonoring the claim. The data we have obtained from the EA consists of the complete
record of injunctions issued by the EA in response to applications submitted by the
universe of Swedish corporate firms from 2007 and onward. Hence, for each seller in
our sample, we observe all injunctions that were issued on its behalf in a given year,
as well as the sum of the claims they concern. The data on injunctions issued by the
EA thus allow us to identify sellers whose buyers are in financial trouble and who
therefore constitute particularly large credit risks. Hence, by computing the change in
the number of injunctions issued on behalf of a given seller between 2007 and 2009—or,
analogously, the change in the sum of the claims underlying the same injunctions—we
can identify sellers who faced particularly large increases in counterparty risk during
the crisis.21
The analysis based on the injunction data is structured as follows. For each seller,
we compute the change in the sum of the claims underlying the injunctions issued on
its behalf between 2007 and 2009, SumInji . We do this using the symmetric growth
rate formula, since the number of applications submitted in 2007 is zero for almost all
sellers in the sample.22 We then split the sample into two groups—one consisting of
firms with a growth rate strictly larger than zero and the other of firms for which the
growth rate is less than or equal to zero—and test the hypothesis that the relationship
between trade credit maturities and price changes during the crisis is stronger for firms
in the former group, that is, for sellers who faced a greater increase in counterparty
risk. We also conduct an analogous exercise based on the change in the number of
injunctions issued on behalf of each seller between 2007 and 2009, NoInji .
The results of the exercises based on SumInji and NoInji are reported in
columns (4) and (5), respectively, of Table 5. The findings are similar across the two
21. Applying for injunctions to settlement is normally the creditor’s last resort and typically occurs when
claims have been overdue for extended periods—several weeks, or longer.
22. For sellers with no application in either 2007 or 2009, the symmetric growth rate is not defined. We
assign a growth rate of zero to these firms.
Amberg, Jacobson, and von Schedvin Trade Credit and Product Pricing 29
6. Conclusions
Schwartz (1974) proposes that product prices include a trade credit price premium,
determined by the contracted loan maturity and an implicit interest rate, which, in turn,
is a function of the selling firm’s liquidity costs and the buying firm’s default risk.
This implies that changes in liquidity costs and counterparty risks cause changes in
product prices, and thus that trade credit issuance introduces a countercyclical element
into firms’ price-setting behavior. We test this proposition empirically using Swedish
manufacturing firm data and the 2008–2009 recession as a shock to liquidity costs and
counterparty risks.
Our analysis confirms that price adjustments are positively related to trade credit
issuance during the crisis: our baseline estimate indicates that the annual change
in the implicit trade credit interest rate was 20.9 percentage points higher during
the crisis period than in noncrisis years, which implies that a maturity difference of
20 days is associated with a relative annual price adjustment of 1.3 percentage points.
By exploring the mechanisms underlying the baseline result, we find evidence that
the association between trade credit issuance and price changes during the crisis is
stronger for firms that experienced larger increases in liquidity costs and counterparty
risks.
Our results relate to several strands of the literature. First, the findings contribute
to the trade credit literature by advancing the understanding of how trade credit is
priced. In particular, this paper is, to the best of our knowledge, the first to empirically
document the existence of implicit interest rates in product prices, which suggests that
trade credit is priced even in the absence of explicit interest rates. Second, our results
contribute to the macroeconomic literatures on the influence of financial conditions
on firms’ price setting, such as the traditional cost channel literature (e.g., Christiano,
Eichenbaum, and Evans 1997; Barth and Ramey 2001; Ravenna and Walsh 2006) and
the closely related literature studying how liquidity constraints affect price markups
in the presence of customer markets (Chevalier and Scharfstein 1996; Gilchrist et al.
2017).
References
Antràs, Pol and C. Fritz Foley (2015). “Poultry in Motion: A Study of International Trade Finance
Practices.” Journal of Political Economy, 123, 853–901.
Barth, Marvin J. and Valerie A. Ramey (2001). “The Cost Channel of Monetary Transmission.”
NBER Macroeconomics Annual, 16, 199–240.
30 Journal of the European Economic Association
Giannetti, Mariassunta, Mike Burkart, and Tore Ellingsen (2011). “What You Sell Is What You Lend?
Sveriges, Riksbank (2009). “The Riksbank’s funding survey, March 2009.” Technical report, Sveriges
Supplementary Data