1

Competition and Pass-Through: Evidence from the Greek Islands1

Christos Genakos2 and Mario Pagliero3

March 2018

PRELIMINARY AND INCOMPLETE

Abstract

Understanding how firms pass cost shocks through to prices under different competitive conditions

is of fundamental importance across many fields in economics. Yet, despite theoretical advances,

there is very little empirical evidence on this question. In this paper, we measure how pass-through

varies with competition in isolated markets of different size. Using daily pricing data from gas

stations, we study how changes in excise duty tax across petroleum products was pass-through to

retail prices across different Greek islands with a different number of competitors. We find that

pass-through significantly increases with competition.

JEL: H22, L1

Keywords: Pass-through; Tax incidence; Gasoline; Market structure; Competition

1 We would like to thank…. The opinions expressed in this paper and all remaining errors are those of the authors

alone. 2 Cambridge Judge Business School, AUEB, CEP and CEPR, E: c.genakos@jbs.cam.ac.uk 3 University of Turin, Collegio Carlo Alberto, and CEPR, E: mario.pagliero@unito.it

2

1. Introduction

A fundamental issue in economics is how firms pass cost shocks (taxes, exchange rates, input prices)

through to prices. The incidence and pass through of taxation are classic public policy issues (recent review

article in public economics). The pass-through of exchange rates and tariffs is important for its effects on

firm productivity and international trade (De Loecker and Koujianou-Goldberg, 2014). The pass through

of input prices is relevant in the analysis of oligopolistic markets (Weyl and Fabinger, 2013), price

discrimination (Aguirre, Cowan, and Vickers 2010), merger analysis (Jaffe and Weil 2013). The cost pass

through is also important for the policy debate in many industries, for example in the health (Cabral,

Geruso, and Mahoney 2015) and energy sectors (Fabra and Reguant 2014).

Theoretical analysis shows that competition is a key determinant of pass-through (Weyl and Fabinger,

2013). With regard to the empirical analysis, there is a well-established line of research exploiting

variability in costs to infer the magnitude of the pass through. Such variability may come from changes in

sales taxes (Barzel 1976, Poterba 1996, Besley and Rosen 1998, Marion and Muehlegger 2011), exchange

rate fluctuations (Campa and Goldberg 2005, Gopinath, Gourinchas, Hsieh, and Li 2011), or changes input

prices (Borenstein, Cameron, and Gilbert 1997, Genesove and Mullin 1998, Nakamura and Zerom 2010,

Miller, Osborne, Sheu 2015). However, these studies do not specifically focus on the impact of competition

on pass through and they provide little evidence on how pass through varies with competition.

In empirical studies, competition is generally measured by the number of competitors located within a

given geographical area around each firm (applications include the gasoline market). Variability in the

number of close competitors captures some important aspects of competition, but does not guarantee that

there are no significant substitution effects beyond the selected geographical area (for example, some

consumers may commute across geographical markets for work or family reasons, hence generating

substitution effects across geographically distant markets). A second feature of existing empirical studies

is that they are agnostic about the sources of variability in competition. While market structure is

3

recognized to be an endogenous outcome, there is little in the way of an explanation for why some firms

face more competition than others.

In this paper, we measure how pass through varies with competition in isolated markets of different

size. Our data come from the retail market for petroleum products (gasoline, unleaded gasoline, diesel,

heating gasoline) on small Greek islands, which clearly define local markets, as substitution effects across

islands are arguably zero.4 Greece is characterized by a large number of small islands. Due to natural

differences in size and their complex historical and economic development, some of these islands are so

small as to support only one gas station, while others support two, three or more. This provides us with a

natural environment, where the level of competition is exogenously determined. We then take advantage

of a policy change, whereby the Greek government altered the excise duty for gasoline products three

times during 2010. Interestingly enough, heating gasoline was excluded from these changes. Using daily

gas station data, we study how the pass-through of the excise duty tax varied across markets with a different

number of competitors, while using heating gasoline as a control group.

We find that pass-through significantly increases with competition. Consistently with the results on

entry thresholds and the results of our price regressions, we find that the relation between competition and

pass through is highly nonlinear. A 10c increase in excise duties implies a xc increase in prices in

monopolistic islands, a xc increase in duopolistic islands, a xc increase in islands with three stations. We

do not detect any effect of increases in competition beyond y stations. [to be completed]

2. Theoretical background

Economic theory provides some general results on how competition and other variables interact in

determining the level of pass through. Following the conduct parameter approach of Genesove and Mullin

(1998), Weyl and Fabinger (2013) obtain an equation for the pass through in oligopolistic markets with n

4 Refueling a car by travelling to a different island is prohibitively expensive, and privately importing fuel in tanks

or similar containers is dangerous and illegal.

4

symmetrically differentiated firms (including, as special cases, monopoly, perfect competition, and most

commonly used oligopoly models). Denoting by 𝜖𝐷 = −𝑝

𝑞𝑝′ the elasticity of demand, they describe the

solution to the firm maximization problem by a conduct parameter 𝑝−𝑚𝑐(𝑞)

𝑝𝜖𝐷 = 𝜃, where 𝑚𝑐(𝑞) is the

marginal cost. 𝜃 captures the intensity of the competition among firms (𝜃 = 0 in a competitive market and

𝜃 = 1 in a monopolistic market). Independently of the specific model considered, the impact of an increase

in marginal cost (i.e., the pass through) on the equilibrium price is

𝜌 =1

1+𝜃

𝜖𝜃+

𝜖𝐷−𝜃

𝜖𝑆+

𝜃

𝜖𝑚𝑠

. (1)

The pass through 𝜌 depends on the conduct parameter 𝜃 and how it varies as the quantity produced

changes (𝜖𝜃 =𝜃

𝑞𝑑𝜃

𝑑𝑞

), but also on the determinants of the elasticity of demand 𝜖𝐷, the elasticity of the inverse

marginal cost curve 𝜖𝑆 (the elasticity of supply), and the curvature of the demand function 𝜖𝑚𝑠.5 In general,

the sign and magnitude of the pass through is ambiguous. Also the sign and magnitude of the effect of an

increase in the conduct parameter on the pass through is ambiguous.

The expression for 𝜌 greatly simplifies in some special cases, which highlight the role of the different

elements in the denominator of equation (1). The ratio 𝜖𝐷−𝜃

𝜖𝑆 links demand heterogeneity and pass through.6

This ratio is equal to zero if the marginal cost is constant. As we will argue in Section 2, there is direct

evidence in our data that marginal cost is constant at the firm level, at least in the short run, and for the

range of quantities typically sold by gas stations in our sample. This suggests that demand heterogeneity is

unlikely to play a big role in our application (we will come back to this in Section x).

5 𝜖𝑚𝑠 =

𝑚𝑠

𝑚𝑠′𝑞 , where 𝑚𝑠 is the negative of the marginal consumer surplus (𝑚𝑠 = −𝑝′𝑞). 𝜖𝑚𝑠 measures the

curvature of the log of demand (Fabinger and Weyl 2012). If demand is linear then 𝜖𝑚𝑠 = 1, if concave 𝜖𝑚𝑠 < 1, if convex 𝜖𝑚𝑠 > 1. 6 Note that if 𝜃 = 0, then 𝜌 =

1

1+(𝜖𝐷/𝜖𝑆), which is the classic formula for tax pass through in perfect competition.

5

A second interesting special case is when 𝜃 is constant. If 𝜃 is constant, then the term 𝜃

𝜖𝜃 is also equal to

0. The conduct parameter 𝜃 is a constant in a number of prominent models. For example, theta is equal to

1 in monopoly, equal to 0 in perfect competition and in the Bertrand model, equal to 1

𝑛 in the Cournot model.

In the monopolistic competition model of Atkeson and Burstein () theta is also a constant. Price competition

with symmetrically differentiated products implies that 𝜃 = 1 − 𝐴, where 𝐴 ≡ − ∑𝜕𝑞𝑗

𝜕𝑝𝑖/

𝜕𝑞𝑖

𝜕𝑝𝑖 𝑗≠𝑖 is the

aggregate diversion ratio, which is a constant if the demand is linear in prices.7 The conduct parameter is

assumed to be a constant in most empirical applications based on the conduct parameter approach (… ).

Finally, an important determinant of the pass through is the demand curvature 𝜖𝑚𝑠. Most empirical

studies are based on linear demand specifications, which directly imply that 𝜖𝑚𝑠 = 1 (references). A few

studies assume different demand specifications that imply different curvature (…), but there is little

empirical evidence on the sign and magnitude of 𝜖𝑚𝑠. Hence, it is difficult to put a priori restrictions on the

sign and magnitude of 𝜖𝑚𝑠.

Of course, if the marginal cost were constant, 𝜃 were constant, and demand were linear, then 𝜌 =1

1+𝜃

and an increase in the conduct parameter would lead to lower pass through. The first assumption is realistic

in our application and can be met in a number of industries. The second is often considered a reasonable

simplification in empirical studies, not putting restrictions on the intensity of competition. However, the

third is difficult to justify without specific evidence on the second derivative of the demand function. We

see this as the main obstacle in deriving theoretical predictions on the impact of competition on pass through

in our setting. Therefore, we take an agnostic view of the sign and magnitude of the impact of competition

7 The relation between the conduct parameter and the number of firms n illustrates the sense in which an increase in the number of firms leads to more competition. In empirical papers, which typically deal with specific industries, the number of firms is typically used as a proxy for the intensity of competition (references …).

6

on pass through and estimate this effect in a difference in difference framework exploiting exogenous

variability in sales taxes.8

3. Industry background and data

Oil is the main energy source in Greece. In 2010 it accounted for 52% of the country’s total primary energy

supply, which is substantially higher than the average in most other advanced countries (36% in 2010).9

Two companies operate in the Greek refining industry: Hellenic Petroleum has three refineries, while Motor

Oil Hellas has one. Hellenic Petroleum controls 72% of the wholesale market.10 There are ten oil terminals

in Greece, seven of them are located in the Attica area (Athens) and three in the Salonica area. In 2010,

there were 20 fuel trade companies operating in the retail market, the largest of which were EKO (a

subsidiary of Hellenic Petroleum), Shell, BP, Avin Oil (100% subsidiary of Motor Oil), and Jet Oil. The

retail network is large. On average, Greece has 1 station per 1400 inhabitants, while the EU average is 1

per 3800 inhabitants.

EU member states are obliged to impose a minimum set of energy taxes, but each member state has

significant freedom in setting tax rates.11 There are two main taxes that are imposed on energy products:

excise duties, which is a unit tax rate (€-cents per liter), and the Value Added Tax (VAT), which is a

percentage tax. In this paper, we only focus on changes in excise duties.

In May 2010 the inability of the Greek government to borrow new funds from the international markets,

led to financial support from euro area Member States and the International Monetary Fund. The Greek

government also increased excise duties on fuel in an attempt to increase tax revenues. Excise duties on

fuel were raised three times during 2010. Each of these three tax changes was announced and implemented

8 We will come back to the determinants of pass through in Section x, in which we discuss some additional and robustness results. 9 International Energy Agency, Energy Policies of IEA Countries, 2011 review. 10 The Greek government owns 35.5% of Hellenic Petroleum, but no shares in Motor Oil Hellas. 11 EU guideline 2003/96/EU.

7

within a very short period of time. Table 1 shows that the increase was very different across products. On

average, gasoline increased by xx while diesel increased by xx. Remarkably, excise duties remained

unchanged for heating gasoline.12.

Describe the contracts between gas stations and oil companies. Make the point that the marginal cost

of the gas stations is constant in the short run and for the relevant range of quantities sold by individual

gas stations.

3.1. Data

We collected two complementary datasets. First, we obtained daily station-level retail prices for all

available gasoline products across Greek islands during 2010. Second, we obtained socioeconomic and

geographic characteristics of each island from the Hellenic Statistical Authority. The data on prices was

officially collected through a reporting system run by the Greek Ministry for Development and

Competitiveness. Retail prices of different products are recorded on a daily basis by the managers of each

petrol station and entered into a data base. The data contains information on five different gasoline products:

unleaded 95, unleaded 100, super (or leaded gasoline), diesel and heating gasoline. Table 2 reports summary

statistics on prices.

The reporting system was first introduced in 200x and only by the end of 2010 all islands and stations

were covered (provide references). The expansion of the reporting system implies that the number of gas

stations reporting their prices is constant for xx islands, increases over time on xx islands, and decreases

in only xx. Industry reports show that entry and exit was essentially zero in this period.

In the analysis, we focus on the number of competitors at the end of the sample period as the relevant

measure of competition at the island level. Still, all our results are not affected when using the number of

12 Heating gasoline is chemically equivalent to the Unleaded 95 gasoline and it is sold by the same gasoline stations

throughout the country. It is treated with a lower excise duty as it is considered a necessity, since the vast majority of

houses in Greece use heating gasoline instead of gas or electricity during the winter.

8

competitors reporting their prices at each reform in the excise duty or, alternatively, focusing only on

islands in which the number of reporting firms did not change.

4. Preliminary evidence

4.1. Competition and market size

Islands greatly vary in size, number of gas stations, and other characteristics (GDP per capita,

population, number of tourist arrivals). Table 3 provides summary statistics for variability across islands in

number of stations, population, GDP per capita, and number of tourist arrivals (N=54, mean, median,

quartiles). The distribution of islands by population and number of gas stations are very skewed, with a

large number of small islands and a few large ones. The smallest islands (with population between x,000

and y,000) typically have only one gas station (examples include x and y), but the number of stations

significantly increases with population (Figure 1). The largest islands have a population of over x,000 and

as many as x00 gas stations. Crete is an outlier, with as much as 1/x th of the Greek population. Figure 1

includes all islands, Figure 2 includes islands below the 95th percentile, Figure 3 below the median.

We estimate the entry thresholds of B&R. The results, reported in Table x, show that larger markets (mainly

in terms of population) support more firms. Entry thresholds, reported in table x and described in Figure x,

show that differences in the number of firms correspond to differences in competition across islands. The

logic behind this approach is that the number of competitors on a given island is the result of an entry game.

In equilibrium, larger islands can sustain more competitors, each of them enjoying smaller markups

(Breshnahan and Reiss). The non linear relation between entry thresholds and number of firms is the result

of the non linear increase in competition as the number of firms grows. As first noted by B&R, competition

is increasing fast for small number of competitors. The effect then levels off after about x firms.

4.2. Competition and prices

9

Prices significantly vary across islands. Figure 6 shows the distribution of the average price for one of

our products (unleaded 95) across different islands (the same results hold for the other products, the

corresponding figures are reported in the appendix). The distribution is /is not very dispersed and

highly/moderately skewed. The standard deviation is x percent of the mean, suggesting/ not suggesting a

large dispersion of prices across islands.

Figure 7 shows a very strong correlation between the number of stations and average price of gasoline.

This correlation holds when we specifically focus on small islands in Figures 8 and 9. The same results

hold for the other products (see Appendix).

Price regressions in Table x show a strong correlation between average prices of petroleum products and

number of firms, controlling for a number of other characteristics of the islands. The results in column x

include dummies for each number of stations. The coefficients, reported in Figure x, are consistent with a

nonlinear effect of competition on prices, for all the five products in our sample. This is clearly consistent

with the results in Figure x on entry thresholds. The effect of competition on prices is largest moving from

monopoly to duopoly, and then to markets with three and four firms. It then levels off for markets with x

or more firms.

5. Competition and pass through

5.1. Identification and empirical methodology

Our approach is a classic diff in diff exploiting differential changes in excise duties across petroleum

products. We start by estimating the following model

𝑃𝑘𝑖𝑠𝑡 = 𝜌𝑇𝑎𝑥𝑘𝑡 + 𝜆𝑖 + 𝜆𝑠 + 𝜆𝑘 + 𝜆𝑡 + 𝜆𝑖𝑠 + 𝑒𝑘𝑖𝑠𝑡 (2)

in which 𝑃𝑘𝑖𝑠𝑡 denotes the retail price of product k, on island i, in gas station s, on day t within a 5 day

window around the tax change, 𝑇𝑎𝑥𝑘𝑡 is the excise duty. The coefficient 𝜌 captures the tax pass through.

The identifying assumption is 𝐸(𝑒𝑘𝑖𝑠𝑡|𝑋 = 0), where X is a matrix of fixed effects. This condition is

10

reasonably met in our diff in diff setting and our application reasonably fits within the diff in diff framework

(see Section 2 for the industry features that allow using this approach). The common trend assumption is

reasonably met as different petroleum products tend to follow similar pricing patterns. Differential changes

in excise duties provide identification.

We then compare the estimated pass through on islands with different number of competitors.

Empirically, we focus on the interaction between taxes and competition and estimate the model

𝑃𝑖𝑠𝑘𝑡 = 𝜌(𝑛𝑖)𝑇𝑎𝑥𝑘𝑡 + 𝜆𝑖 + 𝜆𝑠 + 𝜆𝑘 + 𝜆𝑡 + 𝜆𝑖𝑠 + 𝑒𝑖𝑠𝑡 (3)

where the pass through 𝜌(𝑛𝑖) can be a parametric function of n, for example 𝜌(𝑛𝑖) = 𝑏0 + 𝑏1𝑛𝑖 + 𝑢𝑖, but

can also be estimated non parametrically for each value of the number of firms by replacing 𝜌(𝑛𝑖) =

∑ 𝐼(𝑛𝑖 = 𝑗)𝑁𝑗=1 + 𝑢𝑖, with 𝐸(𝑢𝑖|𝑛𝑖) = 0.13

5.2. The impact of competition on pass through

We use a five day window before and after each event to isolate the short run effects of tax changes.

Figure 5 presents some preliminary evidence on the differential effect of tax across products. The solid

lines represent local linear regressions separately estimated before and after the tax change for each product.

We can see that the tax had a clear impact on all products, whereas heating gas remains constant. These

graphs do not control for any island or station characteristics, which we incorporate below.

Regression results

First, we control for day fixed effects.

13 In principle, having enough observations, one could even estimate the pass through for each gas station on each island. Identification is obtained by comparing the impact of different changes in excise duties for the different petroleum products.

11

Second, we control for station-product fixed effects, where a product is defined as the combination of brand

name and type of fuel, for example BP-Unleaded 95. In that way we control both for the type of fuel but

also for each brand name non-parametrically.

Table 3 reports the results of the average pass-through for each excise change separately.

Table 4 groups all excise changes together.

Table 5 analyses the effect of competition for each excise incidence semi-parametrically by introducing

different indicator variables for each number of competitors.

Table 5 shows that more intense competition leads to higher pass through … the effect is non linear …

5.3. Additional results and robustness

5.3.1. Covariates in 𝝆(𝒏𝒊)

One potential limitation of this approach is that it does not account for the possible correlation between

competition and other market characteristics potentially affecting pass through. The main concern is that

the number of firms 𝑛𝑖 is correlated with determinants 𝑋𝑖 of the demand elasticity 𝜖𝐷(𝑋𝑖). For example,

income per capita might impact pass through indirectly through the entry process (hence 𝑛𝑖) and also

directly through the demand elasticity 𝜖𝐷(𝑋𝑖) in equation (1). However, if the marginal cost is constant, as

argued in Section x, then 𝜖𝐷(𝑋𝑖) drops from equation (1).

From an empirical point of view, model (3) can be easily extended to include the covariates 𝑋𝑖,

𝑃𝑖𝑠𝑘𝑡 = 𝜌(𝑛𝑖, 𝑋𝑖)𝑇𝑎𝑥𝑘𝑡 + 𝜆𝑖 + 𝜆𝑠 + 𝜆𝑘 + 𝜆𝑡 + 𝜆𝑖𝑠 + 𝑒𝑖𝑠𝑡, (4)

where 𝜌(𝑛𝑖, 𝑋𝑖) = 𝑏0 + 𝑏1𝑛𝑖 + 𝑏2𝑋𝑖 + 𝑢𝑖 or, alternatively, 𝜌(𝑛𝑖, 𝑋𝑖) = ∑ 𝐼(𝑛𝑖 = 𝑗)𝑁𝑗=1 + 𝑏2𝑋𝑖 + 𝑢𝑖, and

𝐸(𝑢𝑖|𝑋𝑖, 𝑛𝑖) = 0.

12

Constant marginal cost implies that the estimated impact of competition on pass through (𝑏1) is not

expected to change with the introduction of the determinants of demand elasticity 𝑋𝑖. Moreover, coefficients

in 𝑏2 are not expected to be significantly different from zero. Table x reports the estimated coefficients of

model (4). We do not find significant changes in the estimated impact of competition on pass through and

there is no significant impact of the covariates on the estimated pass through.

In principle, however, covariates 𝑋𝑖 could also enter (1) in other ways. For example, the conduct parameter

theta might not only be a function of 𝑛𝑖, but also of covariates 𝑋𝑖, which could then enter the expression for

rho even if the marginal cost is constant. While this is possible, the empirical literature has typically

assumed a constant conduct parameter theta, which rules out this second channel. Finally, 𝑋𝑖 could be

correlated with the concavity of the demand function. Hence, covariates could enter (1) through 𝜖𝑚𝑠. While

there is no direct empirical evidence on the impact of 𝑋𝑖 on the concavity of the demand function (nor a

clear theoretical argument for why specific variables could affect it), one cannot exclude that some 𝑋𝑖 might

have some systematic impact on the concavity of the demand function.

The small effect of observed island characteristics on the magnitude of the pass through (and the impact of

competition) are consistent with the observation of constant marginal cost and a small or negligible effect

of covariates through the conduct parameter and the concavity of the demand function.

5.3.2. Other robustness results

1) drop the islands for which we observe <50% of stations

2) same gas stations throughout

2) only for unleaded 95

13

6. Concluding Remarks

[to be completed].

References (incomplete)

Card, David (1992), “Using Regional Variation to Measure the Effect of the Federal Minimum Wage”,

Industrial and Labor Relations Review, Vol. 46, pp. 22-37.

Card, David and Alan Krueger (1994), “Minimum Wages and Employment: A Case Study of the Fast-Food

Industry in New Jersey and Pennsylvania”, American Economic Review, Vol. 84, pp. 772-784.

Card, David and Alan Krueger (2000), “Minimum Wages and Employment: A Case Study of the Fast-Food

Industry in New Jersey and Pennsylvania: Reply”, American Economic Review, Vol. 90, pp. 1397-1420.

Carranza, Juan Esteban, Clark, Robert and Jean-François Houde (2013), “Price controls and market

structure: Evidence from gasoline retail markets”, Journal of Industrial Economics, forthcoming.