Pricing decisions under …nancial frictions:
evidence from the WDN survey
Ioana A. Duca
José M. Monteroy
Marianna Riggiz
Roberta Zizzax
December 9, 2016
Abstract
We consider the customer market model with consumer switching costs and
capital-market imperfections developed by Chevalier and Scharfstein (1996) and extend it in two directions: we allow for demand persistence and for a procyclical
elasticity of demand by appealing to an increasing returns shopping technology.
This results in a countercyclical behavior of price-cost markups that is strengthened by our two additional mechanisms. We test our model’s predictions with a
subsample of European …rms participating in the 2014 version of the Wage Dynamics Network survey. In order to do this, we use a novel empirical approach developed
by Aakvik, Heckman and Vytlacil (2005) for estimating discrete choice models with
binary endogenous regressors and unobservables generated by factor structures. We
…nd that …rms subject to …nancial constraints had a signi…cantly higher probability of raising markups than in a counterfactual scenario without such constraints.
Moreover, the estimated partial e¤ects for the main variables in our model are in
overall accordance with the predictions from our theoretical framework.
JEL classi…cation: C25, C26, D22, L11
Keywords: markups, …nancial frictions, customer market
European Central Bank, Directorate Market Infrastructure and Payments, Market Infrastructure
Management Division
y
Banco de España, Directorate General Economics, Statistics and Research,Economic Policy Analysis
Division
z
Banca d’Italia, Directorate General for Economics, Statistics and Research, Economic Outlook and
Monetary Policy Directorate
x
Banca d’Italia, Directorate General for Economics, Statistics and Research, Economic Outlook and
Monetary Policy Directorate
1
Introduction1
1
In recent decades a substantial amount of theoretical and empirical research has addressed
the issues of how prices and markups vary over the business cycle and what are the driving
forces behind their movements.
The di¢ culty, shared by many empirical studies, of …nding signi…cant positive e¤ects
of demand on price margins 2 has urged economists to search for reasons why prices
are kept relatively high in times of low demand, and vice versa. This may occur because
…rms might be less able to collude in high-demand periods, generating "price wars" during
booms (Rotemberg and Saloner, 1986); because prices are sticky (as in the textbook new
Keynesian model); because of a procyclical entry of …rms (Jaimovich and Floetotto, 2008)
or, provided that consumers face high switching costs, because of a procyclical in‡ow of
new customers that can be captured using aggressive pricing behavior (Klemperer, 1995).
Beside these explanations, the countercyclical behavior of price margins has been
linked to the interaction between customer relations (in the spirit of Phelps and Winter,
1970) and …nancial constraints. The idea that markups might be countercyclical if …rms
are …nancially constrained and consumers face switching costs dates back to the works by
Gottfries (1991) and Chevalier and Scharfstein (1996; CS thereafter): intuitively, …rms
are more likely to be liquidity-constrained in periods of low demand, when they have
low cash ‡ow and greater di¢ culty in raising external funds. In this scenario, …rms
might prefer to set higher prices on their locked-in shoppers to boost short run pro…ts,
temporarily forgoing any e¤ort to gain market shares. Clearly, crucial to this mechanism
is the assumption that …rms have a degree of market power over their repeat-purchasers;
in this case, pricing decisions must be investment decisions in market shares, which need,
in a sense, available …nancial resources.
Our work addresses the role of …nancial frictions for markup formation in Europe
during the Great Recession (2010-2013). To this aim we use the third wave of the Wage
Dynamics Network (WDN) survey carried out in 2014 by the European System of Central Banks (ESCB), covering …rms from 25 ESCB countries and from a wide range of
sectors. The questionnaire, which consists almost exclusively of qualitative questions, is
particularly well-suited for the purpose of this paper, as …rms are asked directly how they
changed their price-cost margins over the period 2010-2013, together with questions re1
The views expressed in the article are those of the authors and do not involve the responsibility of
the European Central Bank, the Banco de España and the Bank of Italy. We thank Thomas Mathä and
Ladislav Wintr for providing us with data for Luxembourg.
2
See, for instance, Bils and Chang (2000) and Lundin et al. (2009) and the references therein.
2
lated to the evolution of competition, of demand for their products and to the di¢ culties
in obtaining external …nancing through the usual …nancial channels.
In order to discipline our understanding of the mechanisms underlying pricing decisions, we make use of the theoretical model developed by CS and extend it in two simple
ways with the goal of enriching its set of testable predictions. This allows to better exploit
the WDN questionnaire. First, we allow for some degree of demand persistence, in order
to study how changes in this factor a¤ect equilibrium prices and markup cyclicality. This
feature strikes as relevant given the exceptional persistence of the shocks a¤ecting the
European economy over 2010-2013 (ECB, 2014). A priori, the expected persistence of the
state of demand can be crucial when …rms set their markups. According to our theoretical model, higher expected persistence tends to magnify the e¤ects of …nancial frictions
on markup cyclicality, as it modi…es the trade-o¤ between forgoing short-run pro…ts and
gaining market share to increase long-run pro…ts. The questionnaire contains questions
that can serve as proxies for the perceived persistence of shocks to …rms’own demand,
which we use in our analysis.
Second, whereas CS model features a constant elasticity of demand, we allow for a
procyclical elasticity, so that …rms perceive stronger competition in expansions than during downturns. According to the theoretical model, a change in the degree of competitive
pressures ampli…es the e¤ects of …nancial frictions on markup cyclicality. As the elasticity
of demand falls in a downturn, the gain from a given price cut becomes smaller, which
reduces the bene…t from investing in market shares and, thus, long-run pro…ts 3 . On
the empirical side, we are able to exploit survey questions on the change in perceived
competition experienced in the …rms’main product market.
In sum, according to our model, we expect that when faced with a low demand environment, the likelihood of raising markups increases for …rms with limited access to
external …nance. Moreover, this countercyclical behavior is strengthened when …rms perceive the shock to demand to be highly persistent and the competitive pressures to have
gone down.
We test our model’s predictions with a subsample of European …rms participating in
the 2014 version of the WDN survey. Speci…cally, we use …rm-level data only for the seven
countries4 that replied to the question about the evolution of prices vis-á-vis total costs
over 2010-2013. Further, we use a novel empirical approach developed by Aakvik, Heckman and Vytlacil (2005) for estimating discrete choice models with binary endogenous
3
4
See Section 2 for evidence on the plausibility of assuming a procyclical elasticity of demand.
Czech Republic, Italy, Latvia, Luxembourg, Malta, Poland and Spain.
3
regressors and unobservables generated by factor structures. We …nd that …rms subject
to …nancial constraints had a signi…cantly higher probability of raising markups than in
a counterfactual scenario without such constraints. Moreover, the estimated partial effects for the main variables of interest in our model show that the likelihood of increasing
markups is procyclical on average, but it is less procyclical when competitive pressures
fall during downturns and when the shock to demand is expected to persist. If we take
the average partial e¤ects at face value, the negative e¤ect of an adverse demand shock on
the likelihood of raising markups would be roughly neutralized by the combined positive
e¤ect of a fall in competition and an increase in persistence. All in all, the empirical
results are partly in accordance with the predictions from our theoretical framework.
The paper is organized as follows. Section 2 relates our work to the literature. Section
3 lays out the theoretical framework and the testable predictions. Section 4 presents the
empirical strategy. Section 5 describes the dataset, while Section 6 discusses the results
and Section 7 concludes.
2
Related literature
The empirical relevance of …nancial constraints for markup formation has been the subject of di¤erent studies, using a variety of techniques. CS provide evidence from the
supermarket industry in the US suggesting that during regional and macroeconomic recessions, more …nancially constrained supermarket chains raise their prices relative to less
…nancially constrained ones. More recently, Asplund et al. (2005) test the theory in the
Swedish newspaper industry during the deep recession starting in 1990. Newspapers with
weak …nancial standings showed the highest increases in prices in the subscription market,
where switching costs are relevant, whereas …nancial standings could not explain prices
for advertising space, a market where buyers are less attached to a particular newspaper.
Kimura (2013) focuses on the post-bubble Japanese economy of the 1990s, where general
prices were fairly stable despite large ‡uctuations in the real economy, and relates this outcome to the countercyclical impact of …nancial positions on …rms’prices in sectors where
the customer market theory is more reasonable to be applied5 . Secchi et al. (forthcoming)
…nd that Italian exporters in the early 2000s tended to charge higher prices when facing
credit constraints, with a wider price premium for products and sectors where switching
5
Kimura (2013) shows that the countercyclicality in the pricing behaviour emerges only for large
…rms. He explains this result appealing to customer market theory: …nancial frictions do not a¤ect the
cyclicality of pricing decisions of small …rms, because their product brand is not well established in the
market and, consequently, they cannot lock-in customers.
4
costs are expected to be more relevant.
The debate on the role of …nancial frictions in corporate pricing policies gathered
pace in the context of the global …nancial crisis, as the extraordinary turmoil that swept
through …nancial markets during the Great Recession was accompanied by only a mild
decrease in in‡ation in most advanced countries. Gilchrist et al. (forthcoming) use a
micro-level data set, which contains good-level prices merged with the respondent …rms’
income and balance sheet data, to analyze how di¤erences in …rms’ internal liquidity
positions a¤ect their price-setting behavior during the recent …nancial crisis. Whereas
liquidity unconstrained …rms slashed prices in 2008, those with limited internal liquidity
signi…cantly increased their prices during the same period. Furthermore, these di¤erences
in price setting were concentrated in nondurable goods manufacturing, a sector where
the hallmark features of customer market theories –customer retention and acquisition
consideration– are very relevant. The hypothesis that changes in …nancial conditions
in‡uence the cyclical dynamics of prices is also upheld by Gilchrist and Zakrajsek (2015).
They show that prices in industries in which …rms rely more heavily on external …nance,
thus facing a higher likelihood of …nancing constraints, decline noticeably less in response
to economic downturns associated with a signi…cant tightening of …nancial conditions.
Moreover, a weak balance sheet position in 2006 strongly in‡uenced the likelihood that a
…rm raised its prices above the industry average during the crisis. Using a panel of …rmlevel data, Montero and Urtasun (2014) …nd a signi…cant increase in estimated Spanish
…rms’price-cost markups since 2007. This …nding is explained through the high degree of
…nancial pressure faced by Spanish …rms, in terms of both high levels of corporate leverage
and tight …nancing conditions, on the background of an increase in the pace of business
destruction which probably resulted in a strengthening of surviving …rms’market power.
One of the extensions of the CS model that we consider is the possibility that the price
elasticity of demand is procyclical. The idea that this elasticity might display some cyclical
behavior has been long investigated in research that focuses on …rms’price setting policies
by using micro data with the aim of understanding key macroeconomic phenomena, such
as the countercyclicality of price markups and the inertial adjustment of prices to shocks.
In contrast to the Dixit-Stiglitz world of a constant elasticity, Kimball-style preferences
allow sellers to face a price elasticity of demand that is increasing in their goods’relative
price, and have emerged as the most suitable microfoundation for general equilibrium
macromodels to account for gradual and persistent real e¤ects of nominal shocks. In the
wake of an aggregate demand stimulus, a repricing …rm will temper its price increase since
this would result in a more elastic demand curve, so it takes longer for a demand shock
5
to fully pass through to the average price level.
Several channels could lead to a procyclical demand elasticity. First, recessions are
periods that typically entail a large increase in the pace of business destruction, together
with a marked sluggishness in business formation. Sbordone (2007) shows that a decrease
in the number of competitors (and hence in the number of traded goods) reduces the
steady-state value of the …rm’s elasticity of demand, altering the response of prices to
changes in the economic outlook. On this issue, Montero and Urtasun (2014) for Spain
and Riggi and Venditti (2015) for the euro area relate some changes in the dynamics of
markups, at the micro and macro level, respectively, to the cleansing e¤ect of recessions.
Another possible channel was put forward by Bils (1989), who uses a customer market
model in which the in‡ow of new customers more responsive to prices generates a higher
demand elasticity in booming periods. He provides indirect evidence on the cyclical
importance of new, more price-elastic consumers, at an aggregate level and for retailers
of electrical appliances.
Warner and Barsky (1995), in turn, propose a di¤erent mechanism related to consumers’behavior: retailers perceive their demand to be more elastic in the high demand
states, because in such periods consumers are more vigilant and better informed6 . This
would explain a well-known micro puzzle: the tendency for markdowns to occur when
shopping intensity is exogenously high, like in weekends or in the period prior to Christmas, as it allows sharing …xed costs of shopping across purchases. In fact, we rely on
this "increasing returns to scale shopping technology" in order to introduce a procyclical
demand elasticity in our theoretical model.
On the empirical side, Field and Pagoulatos (1997) provide empirical evidence of a
procyclical behavior of the price elasticity of demand for a panel of 38 US food manufacturing industries from 1972 to 1987. Riggi and Santoro (2015) …nd that whereas in the
pre-1999 period the price elasticity of demand was almost constant in Italy, after 1999 it
increased in the wake of a demand stimulus.
3
Theoretical framework
Our theoretical framework is based on the Klemperer (1995) model of competition with
consumer switching costs, extended to allow for liquidity constraints, as in CS. In this class
6
"Customers for whom it does not pay to search and travel very much when only one item is to be
purchased will invest more in information and transportation to obtain the lowest possible price when
purchasing a number of units of the same good or a number of di¤erent items for which search and travel
costs can be at least partly shared" (Warner and Barsky, 1995).
6
of models, …rms have a degree of market power over their repeat-purchasers, as consumers
have switching costs between similar products of competing …rms. This implies that …rms’
current market shares are valuable, as they determine …rms’future pro…ts, and that …rms
face a trade-o¤, in any period, between investing in market share by setting a low price
and extracting rents by setting a high price on their locked-in shoppers. In this customermarkets framework, CS show that price markups behave in a countercyclical fashion if
…rms are …nancially constrained, as …rms are more likely to be liquidity-constrained in
recessions and liquidity constrained …rms place a greater weight on short-run pro…ts than
on future pro…ts.
To derive some testable predictions on the cyclical behavior of markups, we start
with the two-period model of CS, in which consumers develop switching costs after their
…rst-period purchases, and we extend it in two ways.
First, in CS expected demand is 1 in the …rst period, while being normalized to 1
in the second one. In our model, instead, …rms attribute a certain probability to the
event that the …rst-period state of demand will persist in the future. This allows to study
how changes in the expected persistence of demand a¤ect equilibrium prices and, hence,
markups cyclicality.
Second, whereas CS’s model features constant demand elasticity, we allow for a procyclical nature. This is done by appealing to the "increasing return shopping technology", as in Warner and Barsky (1995): in our model the volume of shopping per
household increases (decreases) in booms (recessions) and the intention to buy a greater
(smaller) number of units during booms (recessions) leads households to bear higher
(lower) search/travel costs. Hence the elasticity of demand is higher in booms than it
is in recessions and …rms perceive stronger competition during expansions than in downturns. This allows to study the e¤ect of a change in competitive pressures on markups
cyclicality.
3.1
The model
There are two …rms k = A; B, who compete for two periods = 1; 2. The mass of
consumers is normalized to 1. They reside uniformly on the line segment [0; 1], with …rm
A located at 0 and …rm B located at 1. Each shopper has a reservation price R for one
unit of a good produced by A and B, at constant marginal cost c: Only one type of good
is bought and sold. In the …rst period consumers bear a transportation cost of t per unit
of distance traveled along the line to the …rm of their choice. These costs are zero in the
7
second period, but consumers develop switching costs, s, as a result of their …rst-period
purchases.
Each consumer exogenously purchases H or L < H units of the good per period; in
each period each customer buys the same quantity of goods. Firms set …rst-period prices
before they know the demand realization, i.e. before customers arrive to the store. For
each …rm, …rst-period demand can be high ( 1 = H ) with probability , or low ( 1 = L )
with probability (1
): The expected persistence of the …rst-period state of demand is
. In other words, is the probability that the …rst-period state of demand will persist
in the second period: P ( =
. The values of and are the same for both
1 )=
…rms.
In the second period, the market is "mature", as consumers’ switching costs have
already been built up: a fraction A
1 of the consumers has bought …rm A’s product in
= 1 and so each bears a switching cost s of buying from B; the complementary fraction
B
A
1 = (1
1 ) has previously purchased B’s product and has developed a switching cost
of buying from A. In this context Klemperer (1995) shows that in the second period
each …rm can charge the consumer’s reservation price R without fear of being undercut
by its rival. The intuition is that, provided that switching costs s are high enough, …rm
A cannot steal any of B’s customers unless it lowers its price a discrete amount below
B’s price (or, in the case of quantity competition, increases its output far enough that its
price falls to this extent). If A must charge the same price to all its customers, such a
large price cut gives up pro…ts on its own locked-in …rst-period customers by more than
it gains by attracting B’s consumers, so A does better to act as a monopolist against
its own customer base. Hence, the unique non-cooperative (Nash) equilibrium, in either
price or quantity competition, yields …rms’joint-pro…t-maximizing outcome.
Finally, in order to compete in this market …rms must invest an amount I at the
beginning of the …rst period.
Internally …nanced …rms
Let’s start by assuming that …rms are …nanced with internally generated funds. We denote
with pk the price charged by …rm k in period :
The second-period pro…ts for each …rm k depend on their …rst-period market shares
k
1 :
k
2
k k
1 ; p2 ; 2
= (R
8
c)
k
2 1
(1)
To evaluate the market shares in period 1, one must take into account that, given our
hypothesis and if pk1 1 + ty < R 1 , the location yi (with i = H; L) of the shopper who is
indi¤erent between A and B is:
pA
1
2t
pB
1
yi =
i
+
1
2
(2)
B
From (2), we get that market shares of …rm A ( A
1 ) and B ( 1 ), i.e. the fraction of
consumers that buy from A and B, respectively, in period 1 are given by:
pA
1
1
1
+ =1
2t
2
First-period pro…ts for …rm A can be written as:
A
1
A
1
=
pB
1
B
pA
1 ; p1 ;
1
= pA
1
A
1 1
c
B
1
(3)
( 1)
(4)
At the beginning of the …rst period, each …rm simultaneously and non-cooperatively
chooses prices, given its conjecture about its rival price, and before knowing the demand
realization (i.e. before the customers arrive to the store), to maximize total discounted
future pro…ts:
V A = pA
1
A
1 1
c
+ (R
1
c)
A
2 1
1
where we have assumed that the discount factor is 1 and 1 and 2 are …rm’s expectations
formulated at the beginning of time 1 for …rst and second period demand, respectively:
1
=
H
+ (1
)
(5)
L
and
2
=[
+ (1
) (1
)]
H
+ [(1
)
+ (1
)]
L
(6)
Maximizing with respect to …rst-period price, we obtain …rm A’s pricing reaction curve
as a function of …rm B’s price:
pA
1 =
pB
c
1
+
2
2
1
+
1
2
+
t
2
1
1
2
2R
(7)
1
implying that prices are strategic complements (i.e. …rm A’s optimal price is increasing
in its rival’s price). The symmetric equilibrium when both …rms are internally …nanced
9
is:
p1 =
t
+
1
+
1
2
2R
c
1
(8)
1
and the markup of price over marginal cost is:
m1 =
t
2
1
(R
@m1 7
;
@
The cyclicality of price margin can be measured by
is:
@m1
=
@
(
(R
c) (1
(9)
c)
1
)
(
H
+
L)
2
1
t
2
1
which, after some algebra,
)
(
H
L)
(10)
To gain some intuition, let us stress the di¤erence between the equilibrium markup
that emerges in our model (9) and the one in CS, which is m1CS = t (R c) :
1
First, in the CS framework, in a one-period setting, each …rm would charge a markup
t: In a one-period version of our model, instead, markup would be equal to t : This
1
di¤erence comes from having assumed that consumers wish to buy a di¤erent number
of units depending on being in a period of boom or bust. As a consequence, the travel
cost they are willing to bear varies with the number of goods they wish to buy. This
means that, when …rms expect high demand, they perceive greater competition for their
market area, i.e. a higher elasticity of demand a¤ecting pricing behavior. Note that
A
1 p1
: If we measure the way it varies with the
the demand elasticity is =
B
A
(p1 p1 ) 1 +t
A
tp1 ( H L )
@j j
cycle as
=
2 ; then the cyclicality of markups can be written as
@
[(pB1 pA1 ) 1 +t]
2
2
@m1
t
H
L
=
+
(1
)
(R
c)
:
2
@
1
1
Second, in a two-period setting price margins are lower by 2 (R c) in our framework
1
and by (R c) in CS. This di¤erence comes from having assumed a variable second-period
1
demand, whose expected level matters for …rms’ incentive to compete for …rst-period
market shares, on which they can later charge the monopoly price R.
Based on (9) and (10), we can draw the following testable predictions:
1. Demand persistence and the level of price markups
7
As in CS we study the cyclicality of markups by di¤erentiating them with respect to : Indeed, high
values of can be interpreted as a boom while low values as a bust and the level of expected demand 1
is a monotonically increasing function of :
10
@m1
=
@
[2
1] (
L)
H
(R
c)
1
When the high demand state is more likely, higher demand persistence lowers price
@m
markups: if
> 21 , @ 1 < 0; by contrast when the high demand state is less likely,
@m
higher demand persistence raises price markups: if < 12 , @ 1 > 0. The intuition is
the following: when the state of demand is high (in booms), the more it is expected to
persist in the future, the stronger the relative convenience of investing in market shares by lowering current markups - to reap pro…ts in the future. By contrast, when the state
of demand is low (in recessions), the more it is expected to persist in the future, the lower
the relative convenience of investing in market shares - by lowering current markups - to
gain pro…ts in the future.
2. Markups cyclicality
(
@m1
=
@
L)
H
2
1
f(R
c) (
H
+
L ) (1
)
tg
Markups can be both procyclical > 0 or countercyclical ( < 0), depending on the
parameters of the model.
Markups might be procyclical, i.e. fall in recessions, because, as in CS, the fall in
current demand relative to future demand makes it more appealing to invest in market
shares by cutting prices (and increase monopoly pro…ts in the future when demand will
be relatively high), relative to charging a high price when demand is relatively low. The
opposite holds true during booms. However, the two additional channels that we have
considered weaken the procyclical behavior of markups: the procyclicality might be weakened and markups might even become countercyclical if the expected persistence of the
state of demand ( ) is high, or if competitive pressures fall (increase) strongly in recessions
(booms). Indeed:
2a. Demand persistence and markups cyclicality
@
<0
@
The higher the expected persistence of demand, the less procyclical (or the more
countercyclical) are price margins. Intuitively, when the low (high) state of demand is
expected to persist in the future, the relative convenience of lowering current markups to
reap pro…ts in the future, rather than in the present, is weaker (stronger).
11
2b. Changes in competitive pressures and markups cyclicality
@
<0
@
The more procyclical is the elasticity of demand, the less procyclical (or the more
countercyclical) are price margins. The intuition is the following: the more the elasticity
of demand falls in downturns, the smaller becomes the loss (gain) in demand size incurred
for a given price increase (decrease). This reduces the bene…t from investing in market
shares (by cutting prices) during recessions. Specularly, the more the elasticity of demand
increases in booms, the larger becomes the loss (gain) in demand size incurred for a given
price increase (decrease). This increases the bene…t from investing in market shares (by
cutting prices) during booms.
In sum, for non-…nancially constrained …rms, the following testable implications emerge:
a. When the high-demand state is more likely, higher demand persistence lowers the
level of price markups; by contrast when the high demand state is less likely, higher
demand persistence raises the level of price markups.
b. The markup of non-…nancially constrained …rms can be either procyclical or countercyclical;
c. It is less procyclical (or more countercyclical), the more the …rm expects the current
shock to demand to persist into the future;
d. It is less procyclical (or more countercyclical), the more the …rm perceives that
competitive pressures are falling during downturns.
Financially constrained …rms
We now extend the model to the case in which …rms need to raise I externally, allowing
for capital market imperfections. As in CS, the latter are introduced following Bolton
and Scharfstein (1990,1996) and Hart and Moore (1998): corporate cash ‡ow, while being
observable to the manager and to investors, cannot be observed by outside parties, i.e. it
is not veri…able, and hence contracts cannot be contingent on its realization. Furthermore
the manager can costlessly divert all of the cash ‡ow to himself.
As in CS the only way to get managers to pay out cash ‡ow is to threaten to liquidate
the …rm’s assets if they do not; liquidation is ine¢ cient as …rm’s assets are worth a fraction
12
< 1 of the remaining cash ‡ow if owned and managed by investors. As Hart and Moore
(1998) and Bolton and Scharfstein (1996) show, the optimal contract calls for a repayment
of D at date 1; if no payment is made, the investor has the right to seize the project’s
assets.
The manager has an incentive to pay out D if he has the cash to do so, because if he
fails to pay, the asset will be liquidated and he loses the ability to divert cash to himself
at date 2. If the assets have not been seized at date 1, then the manager will divert
all of the period 2 cash ‡ow to himself. From these assumptions, we get that incentive
k
compatibility requires D
2:
If the manager does not have enough cash ( k1 < D), then he would choose to pay
nothing and have the asset liquidated. His …rst-period payo¤ would be k1 .
As in CS, we assume that k1 ( L ) < D < k1 ( H ) ; consistently with the conjecture
that …rms are more likely to be liquidity-constrained in recessions.
Figure
k
In what follows we de…ne k1L
1 ( L ) as the …rst-period level of pro…t when demand
k
k
is low; 1H
1 ( H ) as the …rst-period level of pro…t when demand is high and 2=1H
and 2=1L as the expected second-period pro…t, conditional on having a high and a low
level of demand in the …rst period, respectively.
The investor will be willing to lend provided his expected payo¤s are nonnegative:
D + (1
) 2=1L I
0. Competition among investors ensures that the previous
13
condition is met with equality. Note that D is chosen taking into account product market
I (1 ) 2=1L
equilibrium that follows in periods 1 and 2. However, if this value of D, D =
is greater than 2=1H ; then the contract would not be incentive compatible and there would
be no feasible contract. We assume for the remainder that D
2=1H , that is, incentive
compatible contracts are feasible.
A
Firm A chooses pA
= [ A
1 to maximize the expected payo¤ over the two periods V
1H
A
A
B
D + 2=1H ] + (1
) 1L ; taking D and p1 as given.
@ A
@ A
@V A
2=1H
1H
=
[
+
] + (1
A
A
@pA
@p
@p
1
1
1
)
@ A
1L
@pA
1
(11)
De…ning expected demand in the second period conditional on having a high level of
demand in the …rst period 2=1H
) L ; from the …rst order condition we get
H + (1
that the symmetric equilibrium when both …rms are externally …nanced is:
1
p1 = c +
m1 =
where
2
H
+ (1
)
1
2=1H H
t
2=1H H
t
(R
(R
(12)
c)
(13)
c)
2
L.
The cyclicality of price margin when …rms are …nancially constrained is:
@m1
=
@
(R
c)
2=1H L
+ t(
L)
H
L H
2
(14)
or equivalently:
@m1
=
@
(R
c)
2=1H L
t
1
L H
2
(15)
We can draw the following testable predictions.
1. Demand persistence and the level of price markups
@m1
H
=
(R c) ( H
L) < 0
@
Higher demand persistence lowers price markups. The intuition is the following: when
…rms are …nancially constrained, demand persistence matters only if the …rst period state
14
of demand is high (otherwise, the assets are liquidated and second-period pro…ts go to
the investors). As in the unconstrained case, when the state of demand is high, the more
it is expected to persist in the future, the higher the relative convenience of investing in
market shares - by lowering current markups - in order to reap pro…ts in the future.
2. Markups cyclicality
=
(R
c)
2=1H L
+ t(
H
L)
L H
2
<0
The cyclicality of price margins when …rms are …nancially constrained is always negative. Intuitively, during recessions, price margins go up because …nancially constrained
…rms care less about the future; the increased probability of liquidation makes them prefer
extracting rents by setting a higher price rather than building market shares.
2a. Demand persistence and markups cyclicality
@
<0
@
The higher the expected persistence of demand, the more countercyclical are price
margins. The intuition is the following: when the low (high) state of demand is expected
to persist in the future, the relative convenience of lowering current markups to reap
pro…ts in the future, rather than in the present, is - all the more so - weaker (stronger).
2b. Changes in competitive pressures and markups cyclicality
@
<0
@
The more procyclical is the elasticity of demand, the more countercyclical are price
margins. The intuition is the following: the more the elasticity of demand falls in downturns, the smaller becomes the loss (gain) in demand size incurred for a given price increase
(decrease). This reduces the bene…t from investing in market shares (by cutting prices)
during recessions. Specularly, the more the elasticity of demand increases in booms, the
larger becomes the loss (gain) in demand size incurred for a given price increase (decrease). This increases the bene…t from investing in market shares (by cutting prices)
during booms.
In sum, for …nancially constrained …rms, the following testable implications emerge:
15
a. Higher demand persistence lowers the level of price markups.
b. The markup of …nancially constrained …rms is countercyclical;
c. It is more countercyclical the more the …rm expects the current shock to demand
to persist into the future;
d. It is more countercyclical, the more …rms perceive that competitive pressures are
falling during busts.
4
Empirical strategy
In our theoretical model we distinguish between two types of …rms: those subject to …nancial frictions/constraints (i.e. being exposed to a “treatment”, f ci = 1), and those
internally …nanced or not subject to …nancial constraints (i.e. …rms in a “control”group,
f ci = 0). Table 1 collects a set of theoretical implications for both types of …rms in terms
@mj
, and
of the level of markups (mj ; j = 0; 1), the cyclical response of markups
j =
@
the sensitivity of this cyclical response with respect to the cyclicality of the elasticity of
@ j
@ j
, and with respect to the persistence of demand shocks
.
demand
@#
@
In other words, both the level (m) and the cyclical response of a …rm’s markup to
@
di¤erent economic factors
; ' = f# ; g depend on the regime/state the …rm is
@'
facing, which in turn is determined by the degree of …nancial frictions. We will follow
closely the empirical approach developed by Aakvik et al. (2005), who build up a discrete
choice model with unobservables generated by factor structures that …ts very well our
context, as it allows the responses to vary across states. Let us interpret that “nature”
assigns treatment f ci according to the following decision rule:
f ci = 1 if f ci = Zi
Uf ci
f ci = 0 if f ci = Zi
Uf ci < 1
0
(16)
where Zi is a vector of observed (by the econometrician) random variables and Uf ci is
an unobserved (by the econometrician) random variable; f ci is the net utility of “mother
nature” from assigning state 1 (i.e. experiencing …nancial constraints) to …rm i. For
16
each …rm i, we assume two potential outcomes m0i and m1i (whether a …rm decides to
raise its markups or not) corresponding, respectively, to the potential outcomes in the
untreated and treated states. Then we can assume that a linear latent index model (with
unobservables generated by a normal factor) generates the outcomes:
m1i = I(m1i =
1 xi
U1;i
0)
m0i = I(m0i =
0 xi
U0;i
0)
(17)
where I( ) is the usual indicator function. Notice that, in accordance with our theoretical model, we allow the response of markups to di¤er across regimes (i.e. 1 6= 0 ).
We do not observe m0i and m1i simultaneously, but instead:
mi = m0i if f ci = 0
(18)
mi = m1i if f ci = 0
We assume that the error terms are driven by the following factor structure:
Uf c i =
i
+ "f c i
U1i =
1 i
+ "1i
U0i =
0 i
+ "0i
(19)
where i , "f ci , "0i are jointly distributed N (0; I), and where I is the identity matrix
and we have imposed the normalization that V ar( i ) = V ar("j ) = 1 for j = f c; 0; 1.
0
, Corr(Uf c ; U1 ) = 1 =
These assumptions imply that Corr(Uf c ; U0 ) = 0 = p p
2 1 + 20
1
0 1
p
, and Corr(U1 ; U0 ) = 01 = p
. Given joint normality, this set
p p
2
2 1+ 1
1 + 20 1 + 21
of assumptions implies that the joint distribution (Uf c , U0 , U1 ) is known. Notice that this
framework allows us to control for selection on unobservables ( i ) related to the equation
for …nancial constraints.
17
The log-likelihood function for this model is:
ln(L) =
X
lnf
2 ( 1 xi ;
Zi ;
1 )g
f ci 6=0;mi 6=0
X
lnf
2(
lnf
2(
1 xi ;
Zi ;
1 )g+
f ci 6=0;mi =0
2 ( 0 xi ;
Zi ;
0 )g
f ci =0;mi 6=0
where
X
+
+
X
lnf
2(
0 xi ;
Zi ;
0 )g
f ci =0;mi =0
) is the cumulative distribution function of a bivariate normal distribution8 .
An important advantage of this latent variable model is that it can be used to estimate
mean treatment parameters from a common set of structural parameters. Let’s de…ne the
treatment e¤ect for a given …rm i as Di = m1i m0i , which is a counterfactual (you
cannot observe both outcomes at the same time). We are mostly interested in the e¤ect
of treatment on the treated (TT). In our context, TT would give the average e¤ect on the
probability of raising markups for a …rm subject to “treatment”(i.e. …nancial constraints).
For …rms with observed characteristics x it is de…ned as:
TT(xi )
E(Di jX = xi ; f ci = 1) = P r(m1i = 1jX = xi ; f ci = 1)
P r(m0i = 1jX = xi ; f ci = 1) =
2 ( 1 xi ;
Zi ;
1)
Uf c (
2 ( 0 xi ;
Zi ;
0)
(20)
Zi )
We can also be interested in the average e¤ect for a …rm chosen at random from
the population of WDN …rms. The expected e¤ect of the treatment (i.e. of …nancial
constraints) for a …rm with observed characteristics x is:
TE(xi )
E(Di jX = xi ) = P r(m1i = 1jX = xi )
P r(m0i = 1jX = xi ) =
U1 (
1 xi )
U0 (
(21)
0 xi )
We are also interested in averages over the support of X, such as average treatment
e¤ects (ATT and ATE) for the corresponding subgroups of the population, which can be
calculated by averaging the expressions for TT(x) and TE(x) over the observations in the
subgroups. For instance, the ATT for the whole subsample of “treated” (i.e. …nancially
8
The vector Zi may collect a set of instrumental variables, which are not strictly necessary, but help
in identifying the model.
18
constrained) …rms (Nf c ) can be calculated as:
Nf c
1 X
AT T =
TT(xi )
Nf c i=1
(22)
In short, we are in a context of a binary choice model with binary endogenous regressors
which is a bit more sophisticated than the bivariate probit or the probit model with sample
selection, as both biprobit and heckprobit assume equality of coe¢ cients in the outcome
equations for both treatment regimes9 .
5
Data
The empirical analysis is based on a unique cross-country dataset on European …rms’
price- and wage-setting behavior collected through and ad-hoc survey by the European
System of Central Banks in 2014 in the context of the third wave of the Wage Dynamics
Network (WDN). The questionnaire consists almost exclusively of qualitative questions
and includes three types of queries: (i) core harmonized questions, uniformly administered throughout the di¤erent countries to allow for international comparison; (ii) noncore questions (also harmonized across countries, but optional); and (iii) local questions,
administered only at the country level10 .
Our estimation sample is determined by the availability of the non-core question
(NC2.7b) on the evolution of prices vis-á-vis total costs over 2010-2013. Only 7 (out
of 25) countries responded to this question on price-cost margins, namely, Czech Republic, Italy, Latvia, Luxembourg, Malta, Poland and Spain. Moreover, we cleaned regulated
and non-market sectors, where pricing decisions are not very much driven by market forces
— such as electricity, gas and water, …nancial intermediation, public sector services and
arts— 11 . We also dropped …rms with less than 5 employees, as they were all concentrated
9
We use the Stata’s command switch_probit developed by Lokshin and Sajaia (2011) to estimate this
model as well as the ATT and the ATE.
10
See http://www.ecb.europa.eu/pub/economic-research/research-networks/html/researcher_wdn.en.html
for more details about the database.
11
Our theoretical model is based on the presumption that consumers develop switching costs after
their initial purchases, which provides …rms with a certain degree of market power over their customer
base. Thus, in our empirical exercise we would like to restrict the sample to …rms in industries which
are more prone to develop this type of “brand loyalty”. A priori, as argued by Motta (2004, pp 79-81)
and Klemperer (1995), one can realistically think that the existence of switching costs is a widespread
19
on Poland and Luxembourg. The distribution of …rms across countries/size strata and
industries is given in Table 2.
The main dependent variable in our estimation exercises is a dummy variable coded as
unity if the …rm raised markups, and zero elsewhere. To be more speci…c (see the Annex
for the precise wording of the main questions we relied upon), it equals one when …rms
replied that prices (as compared to total costs) increased either moderately or strongly
during 2010–2013.
Secondly, we include information on our main variables of interest, i.e. the dynamics
of demand, the evolution of the degree of competition, the volatility/uncertainty about
the level of demand, and the extent of …nancial constraints. We account for the dynamics
of demand (our cyclical variable) by introducing a dummy which is equal to one if the …rm
reported a negative evolution (strong/moderate decrease) of the domestic or foreign demand for its main product/service during 2010-2013. Regarding the level of competition,
we de…ne a dummy which equals one when …rms report a (strong/moderate) decrease in
competitive pressure on its main product/service (either on domestic or foreign markets),
compared to the situation before 2008. Additionally, we will proxy for the level of demand persistence through …rms’perception about volatility/uncertainty of their demand.
A higher volatility means that shocks are expected to be less persistent, as the likelihood
that there will be a future reversal of demand is higher. Thus, the dummy for the volatility of demand for the …rm’s main product/service is coded as one when the …rm reports
that volatility has not had a negative e¤ect on its activity during 2010-2013, because high
volatility is likely to be perceived as a negative factor12 .
Finally, and more importantly, we consider variables accounting for the impact of
credit availability on …rms’economic activity. In particular, the survey asked …rms to relate the di¢ culties in obtaining credit to the main purpose for which …nance was needed.
phenomenon across many industries. For this reason, we prefer to do a minimal cleaning and only drop
…rms belonging to regulated and non-market sectors where, as mentioned in the text, arguably pricing
decisions are not very much driven by competition and market forces.
12
A large and growing literature (see Bloom (2014) for a survey) points out that volatility is highly
countercyclical. In other words, recessions are periods of high volatility, and the latter may actually
signal a pessimistic future assessment rather than a positive one. Notwithstanding, in the WDN survey
about two thirds of …rms reporting that volatility/uncertainty had a strong negative e¤ect on their
activity considered this e¤ect transitory or at worst partly persistent (see the Annex for the wording of
the question). This supports our view that negative shocks that are volatile are more likely to be less
persistent.
20
Namely, they were asked to assign a ranking (“not relevant”, “of little relevance”, “relevant”, “very relevant”) to the events “Credit was not available”and “Credit was available
but conditions were too onerous” for …nancing the following activities: (i) working capital, (ii) new investment, and (iii) re…nance existing debt (rollover). Firms were de…ned
as …nancially constrained (dummy fc equal to one) if they replied “relevant” or “very
relevant”to any of the six questions.
Finally, we also account for a number of …rm-level characteristics “all of them 0/1
dummies“, such as country of origin, sectoral dummies (industry, trade and business services), …rm size (three dummies: for less than 50, between 50 and 199 and at least 200
employees), nationality of the ownership (mainly domestic or mainly foreign), degree of
autonomy (namely, whether the …rm is a subsidiary/a¢ liate or not) and organizational
structure (single- or multi-establishment …rm).
Table 3 contains descriptive statistics of the variables used in our empirical analysis
for treatment vs non-treatment status. It can be seen that there exist clear di¤erences
in the observable characteristics between treated and non-treated …rms. The former ones
are more likely to be small and medium sized (5 pp on average), younger (one year
on average), and operate in the manufacturing-construction sector (+1.3 pp) than nontreated units. Moreover, the share of …rms that are foreign-owned, a subsidiary or part
of a multi-establishment …rm is lower among treated …rms. Furthermore, these …rms are
more likely to report a fall in demand and a fall in the degree of competition, while they
are less probable to have a lower volatility. Finally, apparently there is a slightly higher
(unconditional) likelihood of raising their price-cost margins (more on this below) for
non-…nancially constrained businesses (27% vs 26%).
6
Results
We now brie‡y discuss the estimated coe¢ cient values from the selection and the outcome equations. Then, we report estimates of the mean treatment parameters (for the
population of WDN …rms) derived from those estimated coe¢ cients, as well as the partial e¤ects for the main regressors of interest. As noted above, under the normality and
factor structure assumptions no exclusion restrictions are required to identify the mean
treatment e¤ects. However, as recommended by Aakvik et al. (2005), we use a (plausible)
exclusion restriction in order to enhance identi…cation: …rm’s age, which besides being
21
an important determinant of the probability of being subject to …nancial constraints, it
complies with the fundamental requirement that is available for all the 7 countries 13 .
Hadlock and Pierce (2010) show that …rm size and age are particularly useful predictors
of …nancial constraint levels. They …nd that …nancial constraints fall sharply as young
and small …rms start to mature and grow 14 . Eventually, these relations appear to level
o¤. Moreover, they argue that an appealing feature of these variables is that they are
much less endogenous than most other usual proxies for …nancial constraints, such as a
…rm’s leverage and cash ‡ow. In sum, we expect …rm’s age to in‡uence its probability of
being …nancially constrained, but not to a¤ect the markup decision directly ntextemdash
only through its impact on …nancial constraintsntextemdash. Indeed, estimation results
reported in Table 7 below support the choice of age as a reasonable instrumental variable.
Estimated coe¢ cients and mean treatment parameters
Table 4 reports estimation results for the selection equation and the two outcome
equations. Starting with the selection equation, consistent with …gures in Table 3, there
is a clear evidence that there is non-random selection into “treatment”, i.e. …rms subject
to …nancial constraints di¤er signi…cantly from those not subject to such constraints with
respect to observable characteristics. For instance, those characteristics attached to a
lower probability of su¤ering …nancial problems are: experiencing a fall in volatility, or
not being a subsidiary nor a foreign-owned company. The coe¢ cient on our instrumental
variable (age) has the expected negative sign and is statistically signi…cant.
As regards regression results for the outcome equations (columns 2 and 3), those …rms
su¤ering a negative demand shock have lower chances of raising markups over the period
2010-2013, with a similar coe¢ cient in both states. This e¤ect is dampened if the demand
shock comes together with a fall in competition. The other statistically signi…cant coe¢ cients are those for a fall in competition (increasing the likelihood of raising markups), being a subsidiary (with opposite sign across states) and for foreign-owned …rms for f c = 0.
Finally, the LR test strongly rejects the null hypothesis of no selection bias due to unobservables (i.e. H0 : error terms are independent across equations) at conventional levels.
13
This instrument has also been used in a similar context by Gilchrist and Zakrajsek (2015).
The idea is that information asymmetries are likely to be especially large for young and newlyestablished …rms, because creditors have not had enough time to monitor such …rms and because such
…rms have not had enough time to build long-term relationships with suppliers of …nance (see inter alia
Coluzzi et al. 2015 and references therein).
14
22
Another important piece of information comes from the correlation among unobservables. The estimated correlation 0 suggests that the unobservables that promote …nancial
constraints are highly negatively correlated with the unobservables that promote higher
markups in the no-…nancial-constraints state. However, those unobservables are highly
positively correlated with the unobservables that promote higher markups in the …nancially constrained state (positive 1 ). Thus, …rms which are more likely to su¤er …nancial
constraints (i.e. with low values of Uf c ) are more likely to have higher treatment e¤ects;
hence, selection is positive for the outcome.
Additionally, one interesting issue that can be studied with our empirical framework is whether …rms experiencing a …nancial shock have a larger likelihood of raising markups than …rms non-…nancially constrained. In order to do this, we can use
expressions (20)-(22) above for computing the di¤erent mean treatment parameters associated with the estimated parameters just reported. We …nd (see …nal rows of Table
4) that the AT T = E(Djf c = 1) = 0.237, which suggests that a …rm subject to …nancial constraints had a 23.7 pp higher probability of raising markups compared with the
counterfactual scenario of …rms operating without …nancial constraints. Regarding the
AT E = E(D) = 0:021, which means that the e¤ect of a …nancial shock on the probability of increasing markups in a …rm randomly selected from the WDN population was negative, but small15 . Using the vocabulary from the program evaluation literature, these results suggest that selection into treatment (su¤ering a …nancial shock) was highly positive
for the outcome (increasing markups), which is consistent with the estimated correlation
coe¢ cients described in the previous paragraph. In comparison, as reported in Table 3,
the raw di¤erence in mean outcomes is -0.011 (i.e. E(m1 jf c = 1)E(m0 jf c = 0) = 0:011).
Thus, controlling for selection (both on observables and on unobservables) appears to have
a substantial impact on the point estimates for mean treatment e¤ects.
Estimated partial e¤ects of main variables
We can go a step further and compute the partial e¤ects of the main variables on the
probability of increasing price-cost margins in both states (f c = 0andf c = 1) and check
whether these e¤ects are consistent with the predictions of our theoretical model. To be
more speci…c, and for simplicity, assume that our empirical model can be written as
E(mjx1 ; x2 ; z; f c) = F (
1 x1
15
+
2 x2
+
12 x1 x2
+ z)
Bootstrapped standard errors for ATT and ATE are very large, which might be suggesting that we
need much larger sample sizes to accurately estimate this type of model.
23
where we have omitted subscripts and m = 1 if the …rm raises markups (0 otherwise);
x1 stands for the variable “fall in demand”and x2 = f“fall in competition”,“fall in volatility”g.
In order to test the predictions of our model, we want to estimate, …rst, the partial e¤ects
of our cyclical variable (fall in demand, i.e. x1 ):
E(mjx1 ;x2 ;z;f c)
x1
= F(
1
+
2 x2
+
12 x2
+ z)
And second, we would like to know how that partial e¤ect changes when either the
drgree of competition or the degree of persistence changes, i.e. the cross "derivative"16 :
(x1 ; x2 ) =
2 E(mjx
1 ;x2 ;z;f c)
x1 x2
Ai and Norton (2003) and Greene (2010) show that this expression is as follows:
(x1 ; x2 ) = [F (
1
+
2
+
12
+ z)
F(
2
+ z)]
[F (
1
+ z)
F ( z)]
Then asymptotic standard errors for the partial and interaction e¤ects can be calculated using the delta method. Notice that with this approach we can compute a partial
e¤ect and its associated standard error for each observation, i.e. we obtain a distribution
of partial e¤ects for our sample of …rms (Greene, 2010). Therefore, since it is di¢ cult to
create a single number that summarizes all the partial e¤ects, in Table 5 we report the
estimated average partial e¤ects across all observations, as well as the max/min e¤ects
and the max/min Wald statistics for the test of signi…cance for individual partial e¤ects.
Further, in order to give a ‡avor of overall statistical signi…cance, we also report the share
of partial e¤ects that are statistically signi…cant.
As shown in Table 5, the likelihood of increasing markups is procyclial on average
(negative partial e¤ect for “fall in demand”) in the case of both …nancially constrained
…rms and non-…nancially constrained ones, which in the former case would be inconsistent
with our theoretical model. The average partial e¤ect is around 0.18 in both cases, a large
magnitude. Moreover, the range of estimated partial e¤ects and the degree of statistical
signi…cance is somewhat similar across both types of …rms. In both categories of …rms, the
probability of raising markups is less procyclical when competitive pressures fall during
the downturn, which is consistent with our model. However, we …nd a lower degree of
statistical signi…cance (with an average around 30%) for these partial e¤ects than for the
previous ones (average around 90%). Finally, markups would be less procyclical in both
16
Strictly speaking we cannot speak about derivatives, but rather about discrete di¤erences, as we are
dealing with binary variables.
24
states the more the shock to demand is expected to persist (i.e. when volatility falls),
although the estimated interaction e¤ects are not statistically signi…cant across the whole
distribution of …rms. All in all, if we take the three average partial e¤ects at face value,
the negative e¤ect of an adverse demand shock on the likelihood of increasing markups
would be roughly neutralized by the combined positive e¤ect of a fall in competition and
an increase in the persistence of the shock.
Robustness
We have made two types of robustness exercises. First, we have introduced some
modi…cations in our preferred speci…cation. Second, we have estimated our baseline speci…cation using several alternative estimators. Starting with the former robustness check,
Table 6 reports the estimated partial e¤ects and mean treatment parameters for a set
of alternative speci…cations. The …rst two set of estimates (Panel A) consider separately
the baseline scenario for …rms in the manufacturing sector and in the services sector.
Although the overall message is similar, it is remarkable the higher degree of statistical
signi…cance in the case of manufacturing …rms, which would suggest that the model with
customer markets and switching costs with …nancial constraints …ts better this type of
…rms. Moreover, the ATT and the ATE are large and statistically signi…cant, while the
interaction of a fall in demand and in volatility (i.e. increase in persistence) has positive
and signi…cant e¤ect for (almost) all manufacturing …rms. Indeed, in this case, a combined scenario of a negative shock in demand with a fall in the degree of competition and
in the degree of uncertainty would deliver a highly countercyclical likelihood of raising
markups, consistently with our theoretical model. Results for …rms in the services sector
are in line with the baseline, although with a lower degree of signi…cance.
Removing sampling weights (Panel B, …rst columns) from the estimation does not alter
our results, while replacing sampling with employment weights (Panel B, latter columns)
yields similar results as well, except for the estimated ATE, which turns out to be positive,
though non-signi…cant. In both cases, we …nd a lower degree of statistical signi…cance in
the partial e¤ects for the interaction between a fall in demand and a fall in competition17 .
17
A di¤erent picture emerges from the exercise in which we drop the variable “age” from the equation
for …nancial constraints, results available upon request. In this case, the estimated partial e¤ects are
quantitatively and qualitatively similar to the other speci…cations, but never statistically signi…cant,
while the ATE and the ATT are highly negative and statistically signi…cant. This result highlights the
importance of …nding a relevant instrumental variable for …nancial constraints and, indirectly, of markups,
because otherwise the interpretation of the model’s results can be substantially altered regarding the e¤ect
of …nancial constraints on the likelihood of increasing markups.
25
In Table 7 we compare our estimated treatment parameters with the estimates produced from some alternative linear estimators to assess whether our results are driven
mainly by the highly nonlinear approach we follow. In particular, we use the linear IV
approach suggested by Angrist (2001) or Angrist and Pischke (2009). Thus, we impose a
linear probability model for the outcome equation assuming that treatment (i.e. f c) only
shifts its intercept, and use the variable age as instrumental variable in a 2SLS regression.
Results in Table 7 indeed show that the estimates based on linear 2SLS that use as IV
the variable age are very close to the Aakvik et al. (2005) approach, which is reassuring.
Moreover, when we split the sample in two (between …rms with f c = 0 and those with
f c = 1) and estimate a simple linear probability model by OLS, we also obtain similar
results. Finally, we also report at the bottom of the …rst column some statistics to assess
the quality of our instrumental variable. Firm’s age is highly signi…cant in the …rst stage
regression, and even though the F statistic is slightly below 10, the Wald statistic for the
Cragg-Donald test of weak instruments would pass the test for a 15% rejection rate (test
size). All in all, …rm’s age seems to be a reasonable instrumental variable.
7
Conclusions
In this paper we have used the third wave of the WDN survey to investigate the role of
…nancial frictions in …rms’ price-setting behavior in some European economies over the
period between 2010 and 2013.
In order to rationalize our results, we start from the model by Chevalier and Scharfstein
(1996), who …rst made the point that the interaction between customer markets and
…nancial frictions might lead to a countercyclical behavior of price margins. We extend
this model to allow for some degree of persistence in demand shocks and for a procyclical
behavior of the elasticity of demand. According to the theoretical model, price markups
behave in a countercyclical fashion if …rms are …nancially constrained, as they are more
likely to be liquidity-constrained in recessions and, thus, place a greater weight on shortrun pro…ts than on future pro…ts. This performance is further strengthened when the state
of demand is perceived to be highly persistent and the more the competitive pressures fall
during busts, as it diminishes the convenience of lowering current markups to reap future
pro…ts.
We take our model to the data using a subsample of European …rms participating in
the 2014 version of the WDN survey. Moreover, given the characteristics of the database,
26
which consists mostly of qualitative variables, we use a novel empirical approach developed
in the context of the treatment e¤ects literature by Aakvik et al. (2005) for estimating
discrete choice models with binary endogenous regressors and unobservables generated by
factor structures. We …nd that …rms subject to …nancial constraints had a signi…cantly
higher probability of raising markups than in a counterfactual scenario without such
constraints. Moreover, the estimated partial e¤ects for the main variables of interest in
our model show that the likelihood of increasing markups is procyclical on average, but it
is less procyclical when competitive pressures fall during downturns and when the shock
to demand is expected to persist. If we take the average partial e¤ects at face value, the
negative e¤ect of an adverse demand shock on the likelihood of raising markups would
be roughly neutralized by the combined positive e¤ect of a fall in competition and an
increase in persistence. All in all, the empirical results are partly in accordance with the
predictions from our theoretical framework.
In sum, our …ndings suggest that the extreme degree of …nancial turbulence experienced by the European economy over the period 2010-2013 could be a relevant factor
behind the evolution of in‡ation in some countries. This would also be consistent with
the available evidence for Japan, the US, Spain or Italy. Thus, the evidence presented in
this paper supports the view that any attempt to understand price-cost margins in the
aggregate must account for …rms’di¤erential access to capital markets over time. The interaction between a …rm’s balance sheet position and its pricing decision is an interesting
avenue for future research.
Annex A: questions posed in the WDN survey and used in the estimates
To derive high_mup:
How did the following factors evolve in your …rm during 2010-2013? Please choose one
option for each line
1=Strong decrease; 2=Moderate decrease; 3=Unchanged; 4=Moderate increase; 5=Strong
increase
[...]
Prices (as compared to total costs)
[...]
To derive fc:
With regard to …nance, please indicate for 2010-2013 how relevant were for your …rm
each one of the following events? Please choose one option for each line. Note: credit
here refers to any kind of credit, not only bank credit
27
1=Not relevant; 2=Of little relevance; 3=Relevant; 4=Very relevant
Credit was not available to …nance working capital
Credit was not available to …nance new investment
Credit was not available to re…nance debt
Credit was available to …nance working capital, but conditions (interest rate and other
contractual terms) were too onerous
Credit was available to …nance new investment, but conditions (interest rate and other
contractual terms) were too onerous
Credit was available to re…nance debt, but conditions (interest rate and other contractual terms) were too onerous
To derive low_dem:
How did [. . . ] demand for your main product evolve during 2010-2013? Please choose
one option for each line
1=Strong decrease; 2=Moderate decrease; 3=Unchanged; 4=Moderate increase; 5=Strong
increase
[. . . ]
Domestic demand for your main product/service
Foreign demand for your main product/service
To derive low_volat:
How did the following factors a¤ect you …rm’s activity during 2010-2013? Please
choose one option for each line
1=Strong decrease; 2=Moderate decrease; 3=Unchanged; 4=Moderate increase; 5=Strong
increase
[. . . ]
Volatility/uncertainty of demand for your products/services
[. . . ]
For those factors which a¤ected your …rm strongly, were the e¤ects transitory, partly
persistent or long-lasting for 2010-2013?
[. . . ]
Volatility/uncertainty of demand for your products/services
Where:
1 = Transitory
2 = Only partly persistent
28
3 = Long-lasting
[. . . ]
To derive low_comp:
Compared to the situation before 2008, how has the competitive pressure on your main
product domestic and foreign markets changed in the period 2010-2013? Please choose
one option for each line
1=Strong decrease; 2=Moderate decrease; 3=Unchanged; 4=Moderate increase; 5=Strong
increase
Domestic market
Foreign market
29
Annex B
To evaluate the market shares in period 1, one must take into account that a shopper
located at y, who is going to buy i (with i = H; L) units of the good, is indi¤erent
between store A and store B if
pA
1
i
+ ty = pB
1
i
+ (1
(1a.)
y)t
From (1a.) one gets that the location yi ; with i = H; L of the shopper who is indi¤erent
between A and B; is:
pB
1
yi =
pA
1
2t
i
+
1
2
(2a.)
B
1 ),
From (2a.), we get that market shares of …rm A ( A
1 ) and B (
consumers that buy from A and B, in period 1 are given by:
A
1
pA
1
2t
pB
1
=
1
+
1
=1
2
i.e. the fraction of
B
1
(3a.)
Internally …nanced …rms
At the beginning of the …rst period, each …rm simultaneously and non-cooperatively
chooses prices, given its conjecture about its rival prices, before knowing the demand
realization, to maximize total discounted future pro…ts:
V A = pA
1
A
1 1
c
1
+ (R
c)
A
2 1
1
First order condition is the following:
A
1 1
1
+ pA
1
c
@
1
A
1
1
@pA
1
+ (R
c)
@
2
A
1
1
@pA
1
=0
(4a.)
That is:
1
"
pB
1
pA
1
2t
1
1
+
2
#
pA
1
c
2t
2
1
(R
c)
After some algebra we get (7), (8) and (9) in the main text.
@m1
The cyclicality of price margin can be measured by
@
30
2 1
2t
=0
(5a.)
@m1
=
@
Taking into account 1 =
+ (1
) (1
2 = [
t @
2
@
(R
1
2
1
1
+ (1
)
)] H + [(1
H
@
@
L
1
c) @
@
2
1
and
) + (1
=
H
2
L,
)]
@
@
1
(6a.)
one gets that:
(7a.)
L
and
@
@
2
= (2
1) (
(8a.)
L)
H
Hence (6a.) can be written as:
=
(
L)
H
t + (R
2
1
c) (2
1)
1
2
(9a.)
substituting 1 and 2 into (9a.) we obtain (10) in the main text.
From (3a.) one gets:
@
=
"
(pB1
pA
1
)
1
2t
+ 12
#
@pA
1
pA
1
(
pB
1
pA
1
2t
pA
1 1
pB
pA
1 +t
1
1
=
(
)
=
1
+ 12
(10a.)
)
The cyclicality of demand elasticity is:
@j j
@
=
pA
1(
H
] pA1 1 (pB1 pA1 ) @@ 1
2
[(pB1 pA1 ) 1 +t]
B pA
A
B pA (
L )[(p1
L)
1 ) 1 +t] p1 1 (p1
1) H
=
2
B pA
p
+t
[( 1 1 ) 1 ]
A
p (
)
= t B1 HA L 2 > 0
p
p
+t
[( 1 1 ) 1 ]
=
pA
1
@ 1
@
[(pB1
pA
1)
1 +t
(11a.)
Using (10a.) and (11a.) into (6a.), markup cyclicality in a symmetric equilibrium can
be rewritten as:
=
t
1
(R
2
1
c) @
@
31
2
1
2
@
@
1
(12a.)
Financially constrained …rms
Firm A chooses pA
1 to maximize the expected payo¤ over the two periods
VA = [
A
1H
A
2=1H ]
D+
A
1L
(13a.)
)
@ A
1L
@pA
1
(14a.)
+ (1
)
@ A
@V A
@ A
2=1H
1H
= [ A +
] + (1
A
@p1
@p1
@pA
1
taking D and pB
1 as given.
Taking into account that:
A
1H
A
1L
A
2=1H
B
pA
1 ; p1 ;
H
= pA
1
c
H
A
1
(
H)
(15a.)
B
pA
1 ; p1 ;
L
= pA
1
c
A
L 1
(
L)
(16a.)
= (R
A
1
(
A
1
c) (
H) =
(
L)
+ (1
H
=
pB
1
pA
1
2t
pB
1
pA
1
2t
)
H
L
A
1
L)
(
(17a.)
H)
+
1
2
(18a.)
+
1
2
(19a.)
we get the following …rst-order condition:
@V A
=
@pA
1
[
A
1
H
(
+(1
(pA1 c)
H)
)
h
2
H
(R c)(
2t
A
L 1 ( L)
H +(1
2t
pA
1
c
)
i
2
L
2t
L) H
]+
=0
(20a.)
After some algebra, de…ning expected demand in the second period, and conditional on
having a high level of demand in the …rst period 2=1H
) L ;we get:
H + (1
@V A
@pA
1
=
H[
+(1
)
(pB1
pA
1)
2t
(pB1
L
(pA1 c)
H
+ 21
pA
1) L
2t
(R c) 2=1H
]+
2t
H
2t
+
32
1
2
pA
1
c
L
2t
=0
(21a.)
=
h
2
H
t
+ (1
pA
1
)
2
L
t
i
pB
1
2
c
2
+
h
h
pB
c
= 1 +
2
2
2
H
t
2
H
t
+ (1
(R
H
2
H
2
i
2
) tL pA
1 =
i
2
+ (1
)
c)
+ (1
L
H
t
2=1H
2
L
)
+
(R c) 2=1H
2t
t
2
+ 12 [
H
1
2
H
+ (1
)
2
L
+ (1
) L]
(22a.)
(23a.)
From the previous equation, we get (12) and (13) in the main text.
2
) 2L : Markup cyclicality is:
Let de…ne
H + (1
@m1
@
=
t
@ 1
@
=
t
(
H
[
L)
t
(
@
t@1
2
H +(1
2
H
1
2
t
)
2
L
2
L
(
1
2
]
[
2
H
)
H (R
2
L
)
H +(1
c)
H (R
2
2=1H
H (R
) L ]( 2H
2
L
c)
)
2
2=1H L
2
H
2
c)
2=1H
2
H
2
L
)
(R
=
(24a.)
=
Rearranging terms we get (14) and (15) in the main text.
33
(
c)
2
2=1H L
2
Table 1: Model’s predictions
Internally financed firms (f ci = 0)
θ2
tϑ∗
(R − c)
−
∗
η
(θ
θ1
θ1
θ1
H − θL )
2
2
(θ − θ )
tϑ∗
∂m0
=
+ (1 − α) H 2 L (R − c) ≶ 0
λ0 =
∂µ
η∗ θ1
θ1
t
∂λ0
=
<0
∂ϑ∗
η∗ θ1
2)
(θ2 − θL
∂λ0
= −(R − c) H
<0
∂α
θ1
m0 =
t
−
θ2
Externally financed firms (f ci = 1)
(R − c) = −
µθH θ2|1H
tη ∗ (θH − θL )
−
(R − c)
2 + (1 − µ)θ 2 )
2 + (1 − µ)θ 2
ϑ∗ (µθH
µθ
L !
H
L
θH θL
tϑ∗ θ1
(R − c)θH θ2|1H −
<0
2
2 2
η∗
µθH + (1 − µ)θL
∂λ1
tθ1 θH θL
=
<0
2 + (1 − µ)θ 2 2
∂ϑ∗
η ∗ µθH
L
2θ
θL
∂λ1
H
= −(R − c)
2 (θH − θL ) < 0
2
∂α
µθ + (1 − µ)θ2
m1 = −
λ1 =
∂m1
=−
∂µ
H
L
Table 2: Sectoral breakdown (based on NACE)
Country name
CZ
ES
IT
LU
LV
MT
PL
Manufacturing
Construction
Trade
Business services
498
506
553
54
100
32
415
85
0
15
129
69
9
124
156
600
229
113
173
26
300
272
851
290
165
203
72
305
519
608
573
457
146
139
100
46
855
431
220
91
1,047
492
381
235
Size (#workers)
5-19
20-49
50-199
200+
34
Table 3: Descriptive statistics for treated/non-treated firms
Means (std.dev.)
Increase in markups = 1
Sector (Manufacturing = 1)2
Size (SME = 1)
Age
Ownership (Foreign = 1)
Subsidiary = 1
Structure (Multi-establ.=1)
Fall in demand = 1
Fall in competition = 1
Fall in volatility3 = 1
Treated firms
(fc=1)1 = 2853
0.260
0.413
0.897
20.97
0.168
0.160
0.190
0.541
0.080
0.604
(0.439)
(0.492)
(0.303)
(14.48)
(0.374)
(0.367)
(0.392)
(0.498)
(0.271)
(0.489)
1
Non-treated firms
(fc=0) = 3131
0.271
0.399
0.843
21.78
0.232
0.283
0.250
0.437
0.060
0.671
(0.445)
(0.490)
(0.363)
(14.78)
(0.422)
(0.451)
(0.433)
(0.496)
(0.238)
(0.470)
Treated firms are those which replied to any of the questions C2.3a-f that difficulties in getting credit were “Relevant” or “Very
relevant”.
2 Includes Construction firms.
3 The evolution of volatility is deduced from the impact of volatility/uncertainty of demand on a firm’s activity (question C2.1b).
Those which replied a positive effect are assumed to experience lower volatility (i.e. higher persistence) of demand shocks.
35
Table 4: Descriptive statistics for treated/non-treated firms
Selection equation
Coeff.
(s.e.)
Non-treatment
Treatment
outcome (fc=0) outcome (fc=1)
Coeff.
(s.e.)
Coeff.
(s.e.)
Fall in demand = 1
−0.086
(0.203)
−0.738
(0.260)∗∗∗
−0.713
(0.324)∗∗
Fall in competition = 1
−0.11
(0.500)
−0.408
(0.182)∗∗
−0.804
(0.326)∗∗∗
Fall in volatility = 1
−0.603
(0.183)∗∗∗
−0.119
(0.248)
−0.211
(0.274)
Fall in (demand & compet.)
0.294
(0.566)
0.256
(0.269)
1.142
(0.489)∗∗
Fall in (demand & volat.)
0.376
(0.253)
0.291
(0.325)
0.593
(0.381)
Ownership (Foreign = 1)
−0.565
(0.164)∗∗∗
0.608
(0.160)∗∗∗
−0.202
(0.263)
Subsidiary = 1
−0.456
(0.165)∗∗∗
0.386
(0.139)∗∗∗
−0.579
(0.201)∗∗∗
Structure (Multi-establ.=1)
Age
{ρ0 , ρ1 }
LR test independent errors
(H0 : ρ0 = ρ1 = 0)
0.073
(0.109)
0.026
(0.191)
−0.166
(0.046)∗∗∗
−0.981
(0.018)∗∗∗
-
0.203
(0.208)
0.983
(0.032)∗∗∗
χ2 (2) = 28.28 (p-val = 0.000)
ATT
0.237
(0.230)
ATE
-0.021
(0.129)
Obs.
5147
Industry, size and country dummies included but not reported. Robust standard errors in parentheses. Sampling weights. *, **,
*** denote statistical significance at the 10%, 5%, and 1% level, respectively. Bootstrapped (200 replications) standard errors for
ATT and ATE.
36
Table 5: Average partial effects (APE) of main regressors on P(m = 1|f c, X)
Panel A. P(m = 1|f c = 0, X)
APE
(%stat.sig.)
Min.
effect
Max.
effect
Min. Wald test
stat. (p-value)
Max. Wald test
stat. (p-value)
Fall in demand = 1
-0.189
(90.6%)
-0.667
-0.011
0.000
0.493
Fall in (demand
& compet.)
0.121
(30.0)%
-0.130
0.344
0.005
0.960
Fall in (demand
& volatility)
0.094
(0.0)%
0.016
0.215
0.343
0.510
Panel B. P(m = 1|f c = 1, X)
APE
(%stat.sig.)
Min.
effect
Max.
effect
Min. Wald test
stat. (p-value)
Max. Wald test
stat. (p-value)
Fall in demand = 1
-0.183
(86.6)%
-0.532
-0.010
0.000
0.508
Fall in (demand
& compet.)
0.126
(32.5)%
0.025
0.385
0.000
0.660
Fall in (demand
& volatility)
0.098
(0.0)%
0.007
0.229
0.343
0.450
APE: averaged across all observations. In parentheses we report the share of APE that are statistically significant at a 10%
level. Min/Max effects and Min/Max Wald tests based on values computed for each observation. Standard errors estimated by
the Delta method.
37
Table 6: Average Partial Effects for alternative specifications
Panel A. Manufacturing vs Market Services
Baseline: Manufacturing
P(m = 1|f c = 0, X)
Fall in demand = 1
Fall in (demand
& compet.)
Fall in (demand
& volat.)
Baseline: Mkt. Services
P(m = 1|f c = 1, X)
P(m = 1|f c = 0, X)
P(m = 1|f c = 1, X)
APE
(% stat.sig.)
Min./Max.
Effect
APE
(% stat.sig.)
Min./Max.
Effect
APE
(% stat.sig.)
Min./Max.
Effect
APE
(% stat.sig.)
Min./Max.
Effect
−0.184
(35.1%)
0.197
(37.2%)
0.251
(99.9%)
-0.791/0.214
−0.135
(35.2%)
0.157
(38.0%)
0.203
(99.7%)
-0.479/0.148
−0.193
(57.9%)
0.115
(1.1%)
0.043
(0.0%)
-0.604/-0.027
−0.191
(52.6%)
0.117
(4.2%)
0.045
(0.0%)
-0.750/-0.010
0.024/0.506
0.022/0.665
ATT
0.000/0.252
0.000/0.385
-0.164/0.333
-0.003/0.107
0.244
(0.148)*
0.323
(0.127)**
2096
ATE
38
No . observations
-0.032/0.535
-0.005/0.193
0.239
(0.289)
-0.033
(0.139)
3051
Panel B. Sampling weights
No sampling weights
P(m = 1|f c = 0, X)
APE
Min./Max.
Fall in demand = 1
Fall in (demand
& compet.)
Fall in (demand
& volat.)
ATT
ATE
No . observations
See notes to Table 5.
Employment weights
P(m = 1|f c = 1, X)
APE
Min./Max.
P(m = 1|f c = 0, X)
APE
Min./Max.
P(m = 1|f c = 1, X)
APE
Min./Max.
(% stat.sig.)
Effect
(% stat.sig.)
Effect
(% stat.sig.)
Effect
(% stat.sig.)
Effect
−0.164
(95.9%)
0.069
(0.0%)
0.032
(0.0%)
-0.426/-0.046
−0.132
(95.9%)
0.058
(0.0%)
0.038
(1.5%)
-0.221/-0.037
−0.164
(82.7%)
0.125
(0.0%)
0.053
(0.0%)
-0.366/-0.013
−0.184
(87.5%)
0.142
(0.0%)
0.060
(0.0%)
-0.395/-0.013
0.035/0.164
0.009/0.129
0.239
(0.235)
0.056
(0.135)
5147
0.023/0.077
0.010/0.070
0.043/0.170
-0.004/0.200
0.196
(0.358)
0.323
(0.282)
5147
0.044/0.184
-0.014/0.214
Table 7: Results for alternative estimators: linear models
2SLS
OLS
fc = 0
OLS
fc = 1
0.207
(0.342)
-0.284
(0.092)***
0.294
(0.128)**
0.183
(0.115)
-0.341
(0.096)***
0.096
(0.069)
0.237
(0.111)**
-0.243
(0.119)**
0.449
(0.176)***
0.125
(0.149)
R2
0.107
0.135
0.126
Age (First stage)
First-stage F-stat1
-0.065
(0.022)***
8.80
Cragg-Donald Wald test2
10.49
No . observations
5147
2694
2466
fc (= ATT)
Fall in demand = 1
Fall in (demand & compet.)
Fall in (demand & volat.)
1
2
For excluded instrument (AGE)
Stock and Yogo’s (2005) critical values for a rejection rate of at most 10% and 15% are 16.38 and 8.96, respectively.
39
References
[1] Aakvik, A., J. J. Heckman, and E. J. Vytlacil (2005), "Estimating Treatment Effects for Discrete Outcomes When Responses to Treatment Vary: An Application
to Norwegian Vocational Rehabilitation Programs", Journal of Econometrics, 125,
15-51.
[2] Ai, C. and E. C. Norton (2003), "Interaction Terms in Logit and Probit Models",
Economics Letters, 80, 123-129.
[3] Angrist, J. D. (2001), "Estimation of Limited Dependent Variable Models with
DummyEndogenous Regressors: Simple Strategies for Empirical Practice", Journal
of Business & Economic Statistics, 19, 2-16.
[4] Angrist, J. D. and J.-S. Pischke (2009), "Mostly Harmless Econometrics: An Empiricist’s Companion ", Princeton university press.
[5] Asplund, M., R. Eriksson and N. Strand (2005), "Prices, Margings and Liquidity
Constraints: Swedish Newspapers 1990-1996", Economica, 72(286), 349-359.
[6] Bils, Mark (1989) ,"Pricing in a customer market." The Quarterly Journal of Economics", 104(4), 699-718.
[7] Bils, M. and Y. Chang (2000), "Understanding How Price Responds to Costs and
Production" Carnegie-Rochester Conference Series on Public Policy, 52, 33-77.
[8] Blanchard, O. J. (2009), "The State of Macro", Annual Review of Economics, Annual
Reviews, 1(1), 209-228.
[9] Bloom, N. (2014), "Fluctuations in Uncertainty", The Journal of Economic Perspectives, 28, 153-176.
[10] Bolton P. and D. S. Scharfstein (1990), "A theory of Predation Based on Agency
Problems in Financial Contracting", American Economic Review, 80(1), 93-106.
[11] Bolton P. and D. S. Scharfstein (1996), "Optimal Debt Structure and the Number of
Creditors", Journal of Political Economy, 104(1), 1-25.
39
[12] Chevalier J. A. and D. S. Scharfstein (1996), "Capital-Market Imperfections and
Countercyclical Markups: Theory and Evidence", The American Economic Review,
86(4), 703-725.
[13] Coluzzi, C., A. Ferrando and C. Martinez-Carrascal (2015), "Financing obstacles
and growth: an analysis for euro area non-…nancial …rms", The European Journal of
Finance, 21 (10–11), 773–790.
[14] D’Amuri, F., Fabiani, S., Sabbatini, R., Tartaglia Polcini, R., Venditti, F., Viviano,
E. and R. Zizza (2015), “Wages and prices in Italy during the crisis: the …rms’
perspective”, Questioni di Economia e Finanza (Occasional Papers) 289, Bank of
Italy, Economic Research and International Relations Area.
[15] Field, Martha K., and Emilio Pagoulatos (1997), "The cyclical behavior of price
elasticity of demand", Southern Economic Journal, 64(1): 118-129.
[16] Gilchrist S., R. Schoenle, J. Sim, E. Zakrajsek (forthcoming), "In‡ation Dynamics
During the Financial Crisis", American Economic Review.
[17] Gilchrist S. and E. Zakrajsek (2015), "Customer Markets and Financial Frictions:
Implications for In‡ation Dynamics", Proceedings: Economic Policy Symposium,
Jackson Hole, Federal Reserve Bank of Kansas City.
[18] Gottfries, N. (1991), "Customer Markets, Credit Market Imperfection and Real Price
Rigidity", Economica, 58, 317-323.
[19] Greene, William (2010), "Testing hypotheses about interaction terms in nonlinear
models", Economics Letters 107, 291-296.
[20] Hadlock C. J. and J. R. Pierce (2010), "New Evidence on Measuring Financial Constraints: Moving Beyond the KZ Index," Review of Financial Studies, 23 (5), 19091940.
[21] Hart O. and J. Moore (1998), "Default and Renegotiation: A Dynamic Model of
Debt," Quarterly Journal of Economics, 113(1), 1-41.
[22] Jaimovich N. and M. Floetotto (2008), "Firm Dynamics, Markup Variations and the
Business Cycle", Journal of Monetary Economics, 55(7), 1238-1252.
[23] Kimura, T. (2013), “Why Do Prices Remain Stable in the Bubble and Bust Period?”
International Economic Journal, 27(2), 157–177.
40
[24] Klemperer, P. (1987), "Markets with Consumer Switching Costs", Quarterly Journal
of Economics, 102(2), 375-394.
[25] Klemperer, P. (1995), "Competition When Consumers Have Switching Costs - An
Overview with Applications to Industrial Organization, Macroeconomics, and International Trade," Review of Economic Studies, 62(4), 515-539.
[26] Lokshin, Michael, and Roger B. Newson. (2011), "Impact of interventions on discrete
outcomes: Maximum likelihood estimation of the binary choice models with binary
endogenous regressors", Stata Journal 11.3, 368-385.
[27] Lundin, M., N. Gottfries, C. Bucht and T. Lindström (2009) "Price and Investment
Dynamics: Theory and Plant-Level Data," Journal of Money, Credit and Banking,
41(5), 907-934.
[28] Montero J. M. and A. Urtasun (2014), “Price-cost markups in the Spanish Economy:
a microeconomic approach,”Banco de España Working Papers - 1407.
[29] Motta, M. (2004), “Competition Policy: Theory and Practice”, Cambridge University Press, Cambridge, UK.
[30] Phelps, E.S. and S.G. Winter Jr.(1970), "Optimal Price Policy under Atomistic Competition," in E.S. Phelps et al., Microeconomic Foundations of Employment and In‡ation Theory, New York: Norton.
[31] Riggi, M. and S. Santoro (2015), "On the Slope and the Persistence of the Italian
Phillips Curve", International Journal of Central Banking, 11(2), 157-197.
[32] Riggi, M. and F. Venditti (2015), "Failing to forecast low in‡ation and Phillips curve
instability: a euro area perspective", International Finance, 18(1), 47-68.
[33] Rivers, D. and Q. H. Vuong (1988), "Limited information estimators and exogeneity
tests for simultaneous probit models", Journal of Econometrics, 39, 347-366.
[34] Rotemberg, J. J., and G. Saloner (1986), "A Supergame-theoretic Model of Price
Wars During Booms," American Economic Review, 76(3), 390-407.
[35] Sbordone, Argia (2009), "Globalization and In‡ation Dynamics: the Impact of Increased Competition", in International Dimensions of Monetary Policy, edited by
Jordi Gali and Mark Gertler, NBER, University of Chicago Press, 2009.
41
[36] Secchi, A., F. Tamagni and C. Tomasi (forthcoming), "Export price adjustments
under …nancial constraints", Canadian Journal of Economics.
[37] Stock J. and M. Yogo (2005), "Testing for Weak Instruments in Linear IV Regression," In: Andrews DWK Identi…cation and Inference for Econometric Models. New
York: Cambridge University Press pp. 80-108.
[38] Warner E. J. and R. B. Barsky (1995), "The Timing and Magnitude of Retail Store
Markdowns: Evidence from Weekends and Holidays", Quarterly Journal of Economics, 110(2), 321-352.
[39] Wooldridge, J. M. (2002), “Econometric analysis of cross section and panel data”,
The MIT Press, Cambridge, Massachusetts, London, England.
42