Behavior of retail prices in common currency areas: The case of the Eurozone

Economic Modelling xxx (xxxx) xxx–xxx

Contents lists available at ScienceDirect

Economic Modelling journal homepage: www.elsevier.com/locate/econmod

Behavior of retail prices in common currency areas: The case of the Eurozone ⁎

Alex Nikolsko-Rzhevskyya, , Olena Ogrokhinab a b

Department of Economics, Lehigh University, 621 Taylor Street, Bethlehem, PA 18015, USA Department of Economics, Lafayette College, 730 High St, Easton, PA 18042, USA

A R T I C L E I N F O

A BS T RAC T

JEL: E31 F14 F15 F36

Does a common currency lead to price convergence? In this paper we both theoretically and empirically show that the effect of a common currency is ambiguous. First, we extend the Ganslandt and Maskus (2007) model of vertical pricing with parallel trade. Our innovation is to consider both domestic trade, where trading costs are relatively low, and international trade, where trading costs are relatively high. If trading costs decline, the model predicts price divergence in the former case, and price convergence in the latter case. Second, using the introduction of the euro as a natural experiment that reduced trading costs, we employ difference-in-differences estimation strategy to test the model's predictions. Our results show that while individual goods prices between countries converged by 2%, within-country prices diverged by 4%, supporting the predictions of our model.

Keywords: Price divergence Common currency Eurozone Difference-in-differences

1. Introduction It is generally assumed that adopting a common currency should lead to market integration within a currency union, primarily through a reduction in trade costs. As trade costs decline, arbitrage becomes more profitable, leading to increased quantities of trade and price convergence. However, in this paper we not only show that the effect of the falling trade costs on prices is ambiguous, but also provide a potential explanation. There could be several reasons why trade costs, which consist of transportation, transaction, local distribution, and other costs, decline following the adoption of a common currency. Among those reasons are increased competition and the loss of monopolistic power, as well as the elimination of foreign exchange risks. And indeed, accepting the euro has decreased trade costs within the Eurozone, as evidenced by the increased volume of trade, including parallel trade, between the Eurozone countries (Frankel, 2010; Glick and Rose, 2015).1 Yet, the empirical evidence on the effect of a single currency on prices, rather than trade volumes, is mixed.2 While some studies find a significant converging effect (Allington et al., 2005; Goldberg and Verboven, 2005; Glushenkova and Zachariadis, 2014), others find no effect at all (Engel and Rogers, 2004; Bergin and Glick, 2007a; Parsley

and Wei, 2008; Fischer, 2012; Ogrokhina, 2015). Regardless, significant price differences still remain even fifteen years following the introduction of the euro, presenting a challenge for economists and policymakers alike. Several theoretical papers have tried to reconcile the lack of price convergence or the presence of divergence with the integrative policies designed to promote trade. In other words, as trade continues to grow, why do we not see a corresponding decrease in international price dispersion? For example, Bergin and Glick (2007b) look at the nature of falling trade costs and hypothesize that only the reduction in per unit trade costs should have a direct impact on price-setting behavior. Or in another paper, using a model of vertical pricing where distributors are allowed to parallel trade, Ganslandt and Maskus (2007) show that as trading costs between two locations decline, a manufacturer with market power may decide to profit by artificially limiting competition, causing prices to diverge. In addition, several empirical papers have also addressed this issue. For example, Kyle (2011) explains the lack of price convergence across the EU countries by showing that manufacturers may choose to rely on non-price responses to reduce the arbitrage opportunities of parallel traders, while Dvir and Strasser (2014) show that manufacturers sustain the price differences by price discriminating not only between

⁎

Corresponding author. E-mail addresses: [email protected] (A. Nikolsko-Rzhevskyy), [email protected] (O. Ogrokhina). Parallel trade is defined as unauthorized reselling of goods without permission from the intellectual property right owner. Parallel trade is allowed within the EU due to the free movement of goods and services implemented to boost market integration. 2 See Rose (2016) for a good summary on the effect of the euro on trade. 1

http://dx.doi.org/10.1016/j.econmod.2017.09.005 Received 30 March 2017; Received in revised form 25 July 2017; Accepted 5 September 2017 0264-9993/ © 2017 Elsevier B.V. All rights reserved.

Please cite this article as: Nikolsko-Rzhevskyy, A., Economic Modelling (2017), http://dx.doi.org/10.1016/j.econmod.2017.09.005

Economic Modelling xxx (xxxx) xxx–xxx

A. Nikolsko-Rzhevskyy, O. Ogrokhina

differences (DID) framework, we estimate the effect of the euro on absolute relative prices—price differentials between two locations— separately for within-countries and between-countries. Because our data includes observations both pre- and post-euro and cover both Eurozone and non-Eurozone cities, the DID framework allows us to account for secular trends and separate the effect the euro had on relative prices. For the within-countries sample, where we assume trade costs were initially low, the introduction of the euro resulted in price divergence as relative prices increased between 12% and 4% depending on the specification. For the between-countries sample, where we assume trade costs were initially high, the introduction of the euro resulted in price convergence as relative prices fell by approximately 2%. Our data allows us to observe prices from two different outlets within the same city, and further test the predictions of the model by separately studying the price data collected from supermarkets and local convenience stores.6 First, we compare the within-countries response of supermarket prices and local convenience store prices following the introduction of the euro. We hypothesize that trading costs for supermarkets are lower than trading costs for local convenience stores since supermarkets have well developed distribution networks and can take advantage of economies of scale. Therefore, the model predicts more divergence for the supermarket prices than for local convenience store prices. Second, we compare the betweencountries response of supermarket prices and local convenience store prices following the introduction of the euro. In this case, the model predicts more convergence for the local convenience store prices than for supermarket prices, as already high between-country trade costs are even higher for local convenience stores than supermarkets. Empirical results confirm these predictions, and they are robust to various specifications. Our results emphasize the need for continuous study of both international prices between Eurozone countries as well as the domestic prices within Eurozone countries. If the observed price divergence is a result of economic distortions caused by market inefficiencies, then lasting price divergence will lead to inefficient allocation of resources and have negative implications not only for competitiveness, but will have a significant destabilizing effect on an already struggling monetary union. The rest of the paper is organized as follows: In the next section we build a theoretical model of vertical pricing with parallel trade, and set up an empirical specification. We discuss our data in Section 3 and proceed to testing the model in Section 4. Finally, the last Section 5 concludes.

countries, but within countries as well. Although the behavior of prices between countries is a wellresearched topic, the dynamics of prices within a country is relatively under-studied.3 Since trade costs include all transport, transaction, and local distribution costs from the manufacturer to the consumer, the effect of the falling trade costs may manifest itself not only when studying between-countries prices, but also when looking at the behavior of prices at different locations within the same country. While the effect of removing trade and other barriers—including the adoption of a single currency—should be more pronounced on price differences between countries (due to presumably larger initial trade costs and a drastic decline in those costs after 1999), domestic withincountry trade should be affected as well. Some of the mechanisms could be a decrease in transportation and distributor costs as a result of increased competition from out-of-country transportation companies, as well as higher labor efficiency due to improved labor mobility. However, we were unable to find empirical evidence in the existing literature for or against within-country price convergence as the result of the single currency. To sum up, there appears to be an inconsistency when it comes to analyzing the behavior of prices following the introduction of a common currency. Some empirical studies find that prices converge, which is in line with most theoretical predictions, other studies find that prices diverge. As a result, these inconsistencies raise further questions. Our contribution to the literature is twofold. First, we extend the Ganslandt and Maskus (2007) model of vertical pricing with parallel trade, and show that the disparity of previous findings could be attributed to different levels of trade costs preceding a policy change. For the larger initial trading costs observed between countries, a reduction in these costs should lead to price convergence. For the smaller initial trading costs observed within countries, a reduction in the trading costs should result in price divergence. Second, using the introduction of the euro as a natural experiment, which reduced trade costs and boosted parallel trade (National Economic Research Associates, 1999), we empirically test and confirm the theoretical predictions of our model. The model with parallel trade is particularly relevant, because parallel trade exists in the EU under the principle of “community exhaustion”. This principle suggests that upon the first sale within the EU market, the holder of property rights (the manufacturer) transfers those ownership rights to the buyer.4 As a result, the manufacturer cannot control the movement of the good—either between EU countries or within an EU country—in order to prevent parallel traders (distributors) from importing this good from cheaper markets and reselling it in more expensive ones.5 Thus, the effect of the parallel trade is twofold. On one hand, parallel trade promotes competition at the retail level by reducing the ability of firms to set differential prices within the EU markets. On the other hand, this impediment to price-discriminate erodes manufacturer's profits and forces them to segment the markets by setting different wholesale prices. Consequently, depending on the initial level of trade costs, a further reduction in those costs can lead to price convergence—when the manufacturer does not segment the markets, or price divergence—when the manufacturer does segment them. To test the predictions of our model we use a proprietary dataset from the Economist Intelligence Unit (EIU) that includes individual good-level prices from both the Eurozone (treated) and non-Eurozone (control) cities for the period of 1990-2015. Applying the difference-in-

2. Model 2.1. Theoretical model Following Ganslandt and Maskus (2007), we set up a model of vertical pricing with parallel trade. Our model considers trade between different countries—international trade—as well as trade between different cities within the same country—domestic trade. Following the discussion above, as the result of a policy change such as the introduction of the euro in 1999, trading costs declined. However, while trading costs declined for both international and domestic trade, we assume that the initial level of costs was higher for the former relative to the latter. Let us assume there exists a manufacturer M which produces a unique good sold through exclusive distributors; i.e., one distributor per market. For simplicity, we assume the existence of only two markets, the central market A where manufacturer M operates and

3 Few studies emphasize the importance of studying regional within-country inflation differences. See for instance Beck et al. (2009); Vaona and Ascari (2012). 4 This principle also applies to the Eurozone countries which constitute a subsample of the EU countries. 5 Moreover, the volumes of parallel trade are considered to be significant. However, the European statistical agencies do not collect data on the actual volumes of parallel trade, because the authorities do not distinguish parallel trade from other trade flows.

6 While the EIU calls local convenience stores mid-priced stores, we will refer to them as “local convenience stores” for consistency and clarity.

2

Economic Modelling xxx (xxxx) xxx–xxx

A. Nikolsko-Rzhevskyy, O. Ogrokhina

manufacturer can only control the amount of goods supplied to markets A and B as well as the amount of competition in market B by setting profit maximizing w A and w B . We assume complete information, which allows the manufacturer to extract all economic profits from the distributors. Therefore, the manufacturer maximizes the following profit function:

which it easily oversees, and the distant auxiliary market B that M has no direct control over. Manufacturer M cannot exercise market power in market B because the manufacturer is not physically present there, the cost to enforce constraints is too high, or because of the regulatory differences between the markets. The two markets can be located either in different countries (e.g., France and Germany) or within the same country (e.g., Berlin and Frankfurt), creating the potential for parallel trade. We believe the assumption that the manufacturer has an exclusive distributor in each market is reasonable. Because market-specific services such as advertising can be quite expensive, free-riding might be a problem, and thus the manufacturer would prefer to limit the number of distributors. Also, the image of the brand and its services may diminish as the number of distributors increase. One example discussed in National Economic Research Associates (1999) and Ganslandt and Maskus (2007) is the Sony-BMG company that typically has a single distributor in each European country, and yet up to 25% of total music sales within the EU come from parallel trade. The exclusive distributors in markets A and B are called distributor 1 and 2, respectively. Distributor 1 operates in the central market A where manufacturer M is located. The demand function in the central market A is: A

A

Q =1−p .

max πM = Q Ap A + [Q Bp B − tq1B]

(3)

= q1Ap A + [(q1A + q2B )p B − tq1B].

(4)

The first term is the revenue obtained in market A and the second term is the revenue obtained in market B minus the costs of parallel trade. Using the demand function from Eq. (1), in the central market A distributor 1 solves the following maximization problem:

maxπ1A = q1A(p A − w A) = q1A{(1 − q1A) − w A}, q1A

(5)

which results in:

Q A = q1A =

1 (1 − w A). 2

(6)

The situation in market B is slightly more complicated. We have to maximize the profits of both distributors that will be supplying different quantities, q1B and q2B , at the same price, p B . Total quantity supplied to market B, Q B = q1B + q2B , at price level, p B , according to the demand function in Eq. (2), gives us the following profit-maximizing conditions:

(1)

Distributor 2 sells the goods in the auxiliary market B. The demand function in market B is:

Q B = 1 − b·p B = 1 − (1 − ϵb)·p B ,

w A, w B

⎧ B B ⎛ ⎞ ⎪ max Bπ B = q B(p B − w A − t ) = q B⎜ 1 − q1 − q2 − w A − t ⎟ 1 q ⎜ ⎟ 1 1 1 ⎪ b ⎪ ⎝ ⎠ ⎨ ⎛1 − q B − q B ⎞ ⎪ B B B B B 1 2 − w B⎟⎟ . (8) ⎪ max q2Bπ2 = q2 (p − w ) = q2 ⎜⎜ ⎪ b ⎝ ⎠ ⎩

(2)

where b = 1 − ϵb characterizes the relative demand elasticity. When 0 < ϵb⪡1 we say that demand is somewhat more elastic in market A than in market B. Identical goods are sold in both markets. However, since M cannot control the auxiliary market B, distributor 1 being the sole seller in the central market A can also sell in market B where distributor 2 is located. Hence, unlike distributor 2, distributor 1 has the option to parallel trade. In real life, this asymmetry is often present when products flow from a smaller to a larger market. Because the manufacturer has the ability to sell its goods in different markets, it then has incentives to set different wholesale prices based on market-specific demand functions. This allows the manufacturer to eventually extract more consumer surplus, thus boosting the manufacturer's profits. Therefore, manufacturer M discriminates between the markets and offers two different marketspecific pricing contracts C i = (wi , f i ), where wi is the wholesale price the distributor pays the manufacturer in market i, and f i is the onetime franchise fee. While these two-part tariffs avoid the downsides of double marginalization and extract all economic profits from both distributors, they cannot always limit parallel trade. As a result, even though in general C A ≠ C B , distributor 1 can sell the goods in either of the markets at the retail price it chooses, while distributor 2 can sell only in market B.7 In market A, distributor 1 faces no competition from possible parallel trade, i.e., Q A = q1A. In market B, due to the absence of restrictions on parallel trade, distributor 2 faces a Cournot-type competition from distributor 1, i.e., Q B = q1B + q2B . However, if distributor 1 wants to sell in market B, it faces additional marginal trade costs t ≥ 0 , which waste resources and eventually reduce profits for the manufacturer. Thus, the manufacturer dislikes competition between distributors. Ceteris paribus, M's incentives to limit parallel trade increase with t. Considering that the manufacturer is unable to control or monitor the quantities that distributors supply to consumers at the retail price pi , the manufacturer has to decide on the appropriate wholesale price wi based only on the quantities it sells to distributors. Therefore, the

(7)

Solving the system results in:

⎧ B ⎪ q1 = ⎪ ⎪ B ⎨ q2 = ⎪ ⎪ B ⎪Q = ⎩

1 (1 − 2bw A − 2bt + w Bb ) (9) 3 1 (1 + bw A + bt − 2w Bb ) (10) 3 1 q1B + q2B = (2 − b{w A + w B + t}). 3

(11)

Once the quantities in markets A and B have been determined and expressed as functions of wholesale prices, manufacturer M chooses w A and w B to maximize its profit πM defined in Eq. (3). This leads to the system of first order conditions:

⎧ ∂πM 1 1 2 2 2 2 = − w A + − bw A − tb − bw B + bt = 0 ⎪ ⎪ ∂w A 2 9 9 9 9 3 ⎨ ⎪ ∂πM = 1 − 2 bw A − 2 tb − 2 bw B − 1 tb = 0. (13) ⎪ B ⎩ ∂w 9 9 9 9 3

(12)

The solution to this system provides us with a pair of optimal wholesale prices:

⎧ w A * = 2bt (14) ⎪ ⎨ B 1 5 − t − 2bt , ⎪w * = ⎩ 2b 2

(15)

which implicitly assumes that both distributors sell positive quantities in both markets.8 By substituting the wholesale prices into the 8 As trading costs t increase, parallel trade becomes less attractive which can result in distributor 1 dropping out of market B, leaving distributor 2 the sole supplier there. Since 1 B . For w has to be positive, we can calculate the upper bound of trade costs: tmax = 2

5b + 4b

1

higher values of t we will have q1B = 0 and q2B = (1 − bw B ) . As distributor 2 supplies 2 only to B and distributor 1 supplies only to A, markets become segmented and the manufacturer sets wholesale prices in both markets equal to marginal costs, i.e., 1 1 w A = w B = 0 . This results in the following retail prices p A = and p B = .

7 We assume there is no parallel trade at the retail level, meaning that consumers and retailers cannot buy directly from other retailers and then parallel trade.

2

3

2b

Economic Modelling xxx (xxxx) xxx–xxx

A. Nikolsko-Rzhevskyy, O. Ogrokhina

2.2. Empirical model In this section, we test the implications of the model: as trading costs fall, within-country prices are expected to diverge, while betweencountries prices are expected converge. Since we are unable to observe trading costs, we use the introduction of the euro as a proxy for falling trading costs with an additional assumption that trading costs within a country are naturally smaller than trading costs between countries. First, we calculate the relative absolute prices d for within and between countries. Second, we analyze their behavior before and after 1999. We start by visually examining the series and then proceed to formal estimation using DID. Following the theoretical model, we define the price differences within and between countries as:

dAB, g, t = |lnpgA, t − lnpgB, t + lnst |,

(18)

pgA, t

where is the retail price for good g at time t in market (city) A, pgB, t is the retail price for the same good g at time t in market (city) B. st is the spot exchange rate between the two locations. Both cities—A and B— can be located either in two separate countries or within one single country (in that case st = 1). Relative prices are calculated for every good, year, and city-pair located within or between countries. Some countries adopted the euro in 1999 and some did not. We will call the former the treatment group (both A and B are in the Eurozone), and the latter the control group (both A and B are outside of the Eurozone). Using a control group is important because comparing before and after price differences within the Eurozone countries alone can be misleading. This comparison would not take into account other common factors unrelated to the introduction of the euro that could be affecting prices in Europe. For example, the internet made bargain hunting much easier, affecting relative prices in Eurozone and nonEurozone countries alike. To complete the DID setup, we need to define pre- and post-treatment samples. Because none of the European countries were using the euro before 1999, that period will be called the “before period”, while the time after 1999 will be called the “after period”. We will estimate the following specification:

Fig. 1. Behavior of retail prices as functions of trading costs. Notes: As trading and transportation costs t decline, price differential dp is decreasing for the “between” countries case, implying price convergence, and is increasing for the “within” countries case, implying price divergence. The model assumes that t is higher between countries than it is within countries. Additionally, we assume that t is higher for local convenience stores relative to supermarkets.

optimized market-specific quantities, (6) and (11), and then plugging them into the inverse demand functions (1) and (2), we solve for the following optimal retail prices:

⎧ A 1 * ⎪ ⎪ p = 2 + bt (16) ⎨ ⎪ p B * = 1 − 1 t , (17) ⎪ ⎩ 2b 2 where p A * < p B * if t = 0 given our assumption that b < 1. Our model predicts a somewhat different result than in Ganslandt and Maskus (2007). While both prices are functions of trade costs t, depending on the value of t, the effect falling trade costs have on relative prices between two locations will be different. Fig. 1 plots the behavior of retail prices as functions of trading costs t. If t* is defined as the level of 1−b trading costs when p A * = p B *, resulting in t* = b(2b + 1) , then for relatively high initial trading costs, t, such as t* < t < tmax , falling trade

dAB, g, t = αA, B + αg + γt + β ·treatmentAB·aftert + δ·XAB, g, t + ϵAB, g, t

(19)

=αA, B + αg + γt + β ·EURODID AB, t + δ · XAB, g, t + ϵAB, g, t ,

(20)

where αA, B are city A and B fixed effects, αg is a good specific fixed effect, γt is the time dummy that accounts for common shocks in all the countries. Since we are using city-pairs as a cross-section dimension, treatmentAB takes a value of one if both cities A and B are part of the Eurozone, and takes a value of zero if both cities A and B are not part of the Eurozone. For both Eurozone and non-Eurozone countries, aftert equals to one after 1999, and zero otherwise. Then, treatmentAB·aftert = EURO DID = 1 for the city-pairs AB if both A and B are in the Eurozone and time t is greater or equal to 1999. Eq. (19) is estimated separately for the two cases: withincountry estimation implies that both cities A and B are a part of the same country, and between-country estimation implies that cities A and B are in different countries. Matrix X includes relevant city- and country-level controls such GDP per capita and wages. GDPs enter the specification as a product of GDPs from two locations, as commonly used in the trade literature. We use country-level and region-level GDPs, depending on the specification. Considering that we are working with retail prices, the potential importance of the non-traded component needs to be taken into account. The non-traded component covers wages paid in the retail sector and at least partially contributes to distribution and marketing costs. To estimate the importance of wages in determining the relative price, we follow the approach of Crucini and Yilmazkuday (2014) who use hourly rates for domestic cleaning help as a proxy for wages in the service sector by treating this measure as non-traded services. Therefore we add to our specification the absolute difference in hourly rates for domestic cleaning help as a measure of absolute relative

costs will lead to a price increase in the auxiliary market B and a price decrease in the central market A, or price convergence, i.e., dp = |p A * − p B *| will decrease. If instead, trading costs are already low, 0 < t < t*, a further reduction in trading costs will result in price divergence, i.e., dp = |p A * − p B * | will increase. This non-linear relationship between price differential d and trading costs t presents a non-trivial challenge. Fortunately, the introduction of the euro provides an opportunity to test the predictions of our model. First, let's consider the within-country case. Because trading costs are defined as the sum of all costs such as transportation, transaction, and distribution costs, we can assume that trading costs are lower within a country than they are between countries since transaction costs are practically absent for within-country trade (due to a common currency). Therefore, the introduction of the euro—a major step to further reduce trade barriers within the EU—increased the competition from foreign transportation companies, thus lowering within-country trading costs t. As a result, we expect to find retail price divergence between cities within the same Eurozone country, where trading costs were low to begin with (0 < t < t*), as depicted in Fig. 1. This is a new result about the effect of a common currency on prices. Second, for the between-countries case, trading costs were higher since transaction costs were non-zero and transportation costs were higher between countries. Hence, after the introduction of the euro which reduced trading costs, we expect to see evidence of price convergence between different countries, since between-countries trading costs are higher than those within countries (t* < t < tmax ). 4

Economic Modelling xxx (xxxx) xxx–xxx

A. Nikolsko-Rzhevskyy, O. Ogrokhina

wages between two locations. If the observed price divergence/convergence was indeed brought by changes in trading costs (excluding distribution costs which are partially captured by changes in local wages), adding local wages to the estimation equation should have no effect on the estimated coefficients.

Table 1 Descriptive statistics.

3. Data

Within - Eurozone - Non-Eurozone Difference in Means Between - Eurozone - Non-Eurozone Difference in Means

The price and wage data come from the dataset provided by the EIU. The original dataset contains annual prices from 1990 to 2015 for 123 cities in 79 countries. Since we are studying price differentials within countries, we use only the European subsample of the data that includes countries with more than one city per country. This leaves us with 4 Eurozone countries: Germany, France, Italy, and Spain and 2 non-Eurozone countries: the UK and Switzerland.9 For each of these countries, the dataset covers 300 prices for both traded and non-traded goods and closely represents an average European consumption basket.10 Because the original purpose of this dataset was to calculate the cost of living for expatriate executives who may be relocated to a different office in another part of the world, the EUI makes sure the goods used in the data collection are goods that can be found across different countries. The main assumption behind this dataset is it that expatriates are able to maintain the same lifestyle once they move to a different country or city. If the identical goods (brands) are not found, then a very close substitute is used. The EIU ensures that there are no differences in quality between the goods sold in different countries. Therefore, the data is very detailed in the description of each good.11 Since we consider the decline in trade costs as the primary mechanism for the observed price behavior, only traded goods are included in the sample, which leaves us with 120 goods. While the EIU collects the price data in both March and September, only September prices are released to the final users. In addition, the EIU collects data from two types of outlets: supermarkets and local convenience stores. Therefore, prices for identical goods are reported from two separate locations within each city. Because we use real GDP per capita as a control variable to control for the level of income in each location, country-level real GDP is collected from Eurostat. Since we are interested in studying withincountry relative prices, country-level GDPs won't control for the within-country city differences. Unfortunately, it is difficult to find the GDP for each city in the sample, so we use the next best option by including the GPDs at various NUTS (Nomenclature of territorial units for statistics) levels where each city is located. Regional data for France, Spain, Italy and the UK was collected from each country's official statistical agency. For Germany and Switzerland, we collected the data from the separate local statistical agencies our seven cities belong to.12

1990–1998

1999–2015

Difference

Mean

Median

Mean

Median

in Means

0.23 0.2 0.03

0.17 0.15 0.02

0.31 0.22 0.09

0.22 0.14 0.08

0.08 0.02 0.06

0.33 0.39 −0.06

0.27 0.33 −0.06

0.37 0.45 −0.08

0.3 0.39 −0.09

0.04 0.06 −0.02

Notes: Table presents mean and median values of absolute relative prices within Eurozone and non-Eurozone countries as well as between Eurozone and non-Eurozone countries before and after the introduction of the euro in 1999.

Paris and London. Second, the within and between groups are separated into Eurozone countries and non-Eurozone countries. Looking at the “within” panel, we can see that the absolute relative prices have increased for both Eurozone and non-Eurozone countries. However, the increase was significantly larger in the Eurozone compared to the non-Eurozone. While price differences within Eurozone countries increased by 0.08, prices differences within non-Eurozone countries increased by only 0.02, resulting in a net effect of 0.06 implying a (relative) price divergence , once we control for secular trends. For the “between” panel, we observe that price differences among Eurozone countries have increased following the introduction of the euro, indicating price divergence. The difference in mean relative prices is 0.04. However, in our sample, prices diverged even in non-Eurozone countries where absolute relative prices increased by 0.06 after 1999. Therefore, accounting for this increase and ignoring the influence of other factors, the introduction of the euro resulted in a net 0.02 decline in between-countries relative prices implying price convergence . To visually gauge the effect of the euro on prices, we plot price differences from 1990 to 2015 for the Eurozone and non-Eurozone countries. Fig. 2 (a) plots average price differences within Eurozone and non-Eurozone countries for all traded goods. Initially both series behaved very similarly. Shortly after 1999 price differences within the Eurozone countries start to rise, while they remain flat for the nonEurozone countries, picking up only after 2005. Regardless, price differences within non-Eurozone countries are always lower. Fig. 2 (b) plots between-country price differences for the Eurozone and non-Eurozone countries. The magnitude of price differences between countries is larger on average than within countries. The behavior of prices confirms our previous findings from Table 1. We can see that while Eurozone and non-Eurozone between-countries differences were similar before 1999, price differences for non-Eurozone countries started to rise after 1999, whereas the common currency in the Eurozone presumably kept price differences from going up. The two spikes in 1995 and 2011 are interesting and require further exploration. The 1995 spike was caused by a temporary appreciation of the British pound while the spike in 2011 is due to the Swiss franc appreciating prior to the introduction of the cap. Explicitly accounting for both of these outliers does not affect our general conclusion; in fact, excluding the period of British pound appreciation makes the results even stronger.

4. Results 4.1. Descriptive results Table 1 presents summary statistics for the absolute relative prices calculated as an absolute price difference between two locations averaged across cities and goods. To make sure the average is not driven by particularly large differences, median absolute relative prices are also reported. The table is organized in the following way: First, the results are reported for the within-country relative prices such as Berlin and Frankfurt, and between-countries absolute relative prices such as

4.2. Empirical results We start by estimating Eq. (19) for within-country city-pairs. Table 2 reports the DID estimates. The first column shows results for all goods. The EURO DID coefficient is positive and significant. In the most parsimonious specification, the absolute relative prices within the Eurozone countries increased by about 12% after the introduction of the euro relative to the control group countries. Because the non-

9

For the list of cities and countries refer to Table A1:. A detailed description of the data, including the exact list of goods, can be found in Engel and Rogers (2004). See Crucini and Shintani (2008) for a more detailed comparison of EIU data to price data collected by the Bureau of Labor Statistics. 11 Given that we are working with the European subsample of the dataset, majority of goods (brands) should be available in the European countries. 12 Refer to Table A1: for the list of cities, their NUTS regions and treatment status. 10

5

Economic Modelling xxx (xxxx) xxx–xxx

A. Nikolsko-Rzhevskyy, O. Ogrokhina

In column (2), we augment our basic specification with country level GDPs per capita. Compared to column (1), EURODID coefficient decreases from 12% to 9%, implying smaller divergence. The coefficient on real GDP is negative and significant. However, the inclusion of country-level GDPs may be misleading for a within-country specification since it does not account for differences across cities within the same country. Therefore, in column (3) we include a product of regional GDPs per capita allowing us to control for income differences between different cities within the same country. While the results remain in line with the previous findings, the point estimate of the EURO DID coefficient declines to 4.3%. In our last specification, column (4), we also control for wage differences between two cities as suggested by Crucini and Yilmazkuday (2014). We can see that including wages has no effect on the EURO DID coefficient once regional GDP differences are accounted for. Our price data is collected from two types of outlets: supermarkets and local convenience stores. Thus, for every location, we have a price observation for every good from two different outlets. We estimate our model separately with the data from each outlet. The results are reported in Table 3. Column (1) reports within-countries estimates for all goods, column (2) reports within-countries estimates for prices collected in supermarkets, and column (3) reports the estimates for convenience store prices. Point estimates imply that divergence is somewhat more significant for the supermarket data where relative prices increased by 5%, versus 4% and 3% for all goods and convenience stores, respectively. Greater price divergence across supermarket goods supports our theory that a reduction in trade costs (most notably seen for larger stores like supermarkets) led to price divergence, as evident from Fig. 1. Recall that our model predicts price divergence within countries, and price convergence between countries. We have just confirmed the first theoretical prediction. Next we employ the same price data and investigate whether the introduction of the euro resulted in a reduction of absolute relative prices between Eurozone countries. Table 3 columns (4) - (6) report the results. In column (4)—for all available goods—the introduction of the euro had a small but statistically significant converging effect on the between-countries relative prices. This finding confirms the intent behind the introduction of the euro: the single currency was supposed to promote competition and transparency, leading to price convergence. Unlike the within-countries case, the effect of wage differentials on between-countries relative prices is highly significant and positive. A reduction of wage differences across cities located in different countries leads to higher convergence of prices for traded goods. In columns (5) and (6), we separately estimate the effect of the euro on prices for supermarket goods and for local convenience stores. As before, separating goods by the type of outlet allows us to further confirm the predictions of our model. Since convenience stores have higher trading costs than large supermarkets (refer to Fig. 1), we see more price convergence for convenience store goods than for supermarkets.14 Why do we observe price divergence within countries while still seeing price convergence between countries? We hypothesize that the difference is due to varying trading costs. The theoretical model

Fig. 2. Average price differences between and within countries.

Table 2 Within-country city-pairs results. Dependent variable: Absolute price differential

EURODID

(1) 0.123***

(2) 0.090***

(3) 0.043***

(4) 0.041***

(0.009)

(0.009) −0.147*** (0.035)

(0.010)

(0.010)

−0.111*** (0.019)

CountryGDP

Yes Yes Yes 0.209

Yes Yes Yes 0.210

Yes Yes Yes 0.218

−0.114*** (0.019) 0.019 (0.012) Yes Yes Yes 0.218

82,641

73,932

52,435

52,435

RegionalGDP

Wage City FE Good FE Time FE Within-R 2 Observations

(footnote continued) time and city fixed effects. However, we choose to continue using city, good, and time fixed effects as it allows us to control for wage differentials which partially capture distribution costs, making the Euro coefficient β a better proxy for falling trade costs. Replacing fixed effects altogether with random effects was not a viable option as SarganHansen test rejects a random effects model at all conventional levels of significance. 14 The results are also not sensitive to the sample of goods included in the regression. Because one of the concerns could be that these results should hold for nonhomogeneous goods where manufacturers have market power, and should not hold for homogeneous good for which numerous substitutes are available. To test this, we have dropped food items—arguably the most homogeneous goods in our sample—from the estimation. Following the introduction of the euro prices diverged within the Eurozone countries by 5.2% and converged between the Eurozone countries by 4.2%. As expected, omitting the homogeneous goods from the sample made the diverging/converging effect of the introduction of the euro stronger. We thank the anonymous referee for this suggestion.

Notes: Robust standard errors in parentheses. ***, **, and * denote 1%, 5%, and 10% significance levels.

Eurozone countries serve as a control group, we can attribute this outcome to accepting a common currency. This confirms our preliminary findings. Even after controlling for location, good, and time fixed-effects, there is a clear within-country divergence of retail prices after the introduction of the euro.13

13

We have also tried combinations of city-time and good fixed effects as well as good-

6

Economic Modelling xxx (xxxx) xxx–xxx

A. Nikolsko-Rzhevskyy, O. Ogrokhina

Table 3 Within and between-countries city-pairs results broken down by outlet type. Within country

EURODID

Wage RegionalGDP City FE Good FE Time FE Within-R 2 Observations

Between countries

AllGoods

Supermarket

ConvinienceStore

AllGoods

Supermarket

ConvinienceStore

(1) 0.041***

(2) 0.050***

(3) 0.032**

(4) −0.014***

(5) −0.009

(6) −0.022***

(0.010) 0.019 (0.012) −0.114*** (0.019) Yes Yes Yes 0.218

(0.015) 0.010 (0.019) −0.092*** (0.031) Yes Yes Yes 0.228

(0.015) 0.024 (0.016) −0.139*** (0.027) Yes Yes Yes 0.189

(0.005) 0.077*** (0.005) 0.020* (0.011) Yes Yes Yes 0.111

(0.007) 0.061*** (0.008) 0.023 (0.018) Yes Yes Yes 0.089

(0.007) 0.098*** (0.008) 0.006 (0.016) Yes Yes Yes 0.095

52,435

23,787

24,869

271,947

123,312

128,863

Notes: Robust standard errors in parentheses. ***, **, and * denote 1%, 5%, and 10% significance levels. Dependent variable: absolute price differential. First we present the results within Eurozone countries. Second, we present the results between Eurozone countries. Columns (1) and (4) present the results for absolute price differentials for all available goods. Columns (2) and (5) present the results based on prices collected from supermarkets. Columns (3) and (6) present the results based on prices collected from local convenience stores.

Table 4 Within and between-countries city-pairs robustness check by various subsamples. Within country

EURODID RegionalGDP

Wage

Between countries

All

Exclude 1999

Exclude 2008–2015

Exclude 1999

Exclude 2008–2015

(1) 0.043***

(2) 0.051***

(3) −0.013***

(4) −0.019***

(5) −0.016***

(0.011) −0.113*** (0.020) 0.017 (0.012)

(0.009) −0.044** (0.022) −0.011 (0.013)

(0.005) 0.020* (0.011) 0.077*** (0.005)

(0.005) −0.039** (0.016) 0.073*** (0.006)

(0.005) 0.005 (0.010) 0.086*** (0.005) 0.059***

EURODID × WITHIN

Yes Yes Yes 0.219

Yes Yes Yes 0.219

Yes Yes Yes 0.111

Yes Yes Yes 0.127

(0.006) −0.154*** (0.006) Yes Yes Yes 0.121

50,050

37,586

254,729

209,471

324,382

WITHIN City FE Good FE Time FE Within-R2 Observations

Notes: Robust standard errors in parentheses. ***, **, and * denote 1%, 5%, and 10% significance levels. Dependent variable: absolute price differential. The results are based on prices for all available goods. Columns (1) and (2) present the results within Eurozone countries. Columns (3) and (4) present the results between Eurozone countries. Column (5) presents the results for all available prices and cities within and between Eurozone countries.

price convergence for the between-countries city-pairs. In addition, we exclude the recent financial crisis taking into account that price dynamics after 2008 were perhaps determined by factors unrelated to adopting the euro. Therefore, we re-estimate the previous specifications for 1990-2007 only. Within-country results are reported in column (2) and between-countries results are in column (4). The results remain in line with our previous findings, although they are somewhat more pronounced for the within-country city-pairs— following the introduction of the euro the within-country absolute relative prices increased by over 5%. The higher point estimate for the EURODID coefficient for this shorter sample suggests that withincountry price divergence slowed down after the financial crisis. For the between-countries city-pairs we observe stronger results if we omit crisis years. We see more evidence of price convergence and the EURODID coefficient approaches −2%. In column (5), we combine all city-pairs—within and betweencountries—and estimate the main specification augmenting it with a “WITHIN ” dummy and an interaction term between EURO DID and WITHIN . The estimation results agree with our descriptive observation

predicts that when trading costs are low (and become even lower), prices start to diverge as manufacturers try to maintain profits by segmenting the markets. For high trading costs, a similar change results in price convergence. While trading costs within countries are undoubtedly lower than those between countries, adopting the euro has likely lowered both. Therefore, due to this asymmetric U-shaped effect, prices within each country start to diverge, while prices between countries start to converge. 4.3. Robustness checks In order to test our results for consistency, we conduct several robustness checks. First, we test our results using different subsamples. This robustness check is presented in Table 4. Specifications (1) and (3) report within- and between-countries results for all goods when we drop the “transition year” of 1999, the first year the euro was adopted. Omitting the transition year has a minor effect on the within-country estimates, and almost no effect on the between-countries estimates. We continue seeing price divergence for the within-country city-pairs and 7

Economic Modelling xxx (xxxx) xxx–xxx

A. Nikolsko-Rzhevskyy, O. Ogrokhina

predictions and obtain additional insight. First, we observe the divergence of prices for within-countries city-pairs and convergence of prices for between-countries city-pairs. Second, assuming supermarkets have lower trading costs than local convenience stores, prices from supermarkets exhibit more divergence for within-country citypairs, and prices from convience stores exhibit more convergence for between-country city-pairs as can be seen in Fig. 1.

Table 5 Within and between-countries city-pairs robustness check by various subsamples broken down by outlet type. Within country

EURODID RegionalGDP

Wage Within-R2 Observations

EURODID RegionalGDP

Wage Within-R2 Observations City FE Good FE Time FE

Between countries

Exclude 1999

Exclude 2008–2015

Exclude 1999

(1) 0.053***

(2) (3) Panel A: Supermarkets 0.054*** −0.008

−0.014*

(0.017) −0.090*** (0.032) 0.008 (0.019) 0.229

(0.013) −0.072** (0.036) −0.060*** (0.021) 0.224

(0.007) −0.051** (0.025) 0.052*** (0.009) 0.096

(0.008) 0.027 (0.018) 0.062*** (0.008) 0.090

Exclude 2008–2015 (4)

22,693 17,059 115,427 Panel B: Local Convenience Stores 0.033** 0.050*** −0.022***

−0.028***

(0.016) −0.138*** (0.027) 0.022 (0.016) 0.191

(0.013) −0.019 (0.031) 0.025 (0.020) 0.192

(0.007) 0.002 (0.016) 0.098*** (0.008) 0.095

(0.007) −0.017 (0.023) 0.101*** (0.009) 0.114

23,743 Yes Yes Yes

17,836 Yes Yes Yes

120,730 Yes Yes Yes

99,266 Yes Yes Yes

5. Conclusion In our paper we develop a model of vertical pricing with parallel trade, and, using the introduction of the euro as a natural experiment that reduced trading costs, we empirically test the model's predictions. Our model considers trade between different countries as well as trade between different cities within the same country. We assume that trading costs are higher between countries than within a country. Given that within-country trading costs are low to begin with, the model predicts that a reduction in those costs would have a diverging effect on relative prices as manufacturers pursue other ways to segment markets. However, between-countries trading costs are higher, so a reduction in these trading costs should not have a large effect on manufacturers’ profits, leaving little incentive to pursue market segmentation. According to the model, this results in a small convergence of prices due to increased arbitrage. Our empirical analysis confirms both of these predictions. These findings are concerning because they indicate that falling trade costs followed by higher volumes of trade do not necessarily lead to greater market integration. Unfortunately, the effect is ambiguous and requires further investigation. While markets between EU countries may benefit from falling trade costs, markets within EU countries might become more segmented. Often, within-country price dispersion is used as the benchmark for between-countries price dispersion. As a result, the appearance of market integration within the Eurozone can be driven by rising within-country price divergence, and not increasing between-country price convergence. The issue of price convergence and divergence is particularly important given the “one-size fits all” monetary policy. First, consider the relationship between relative prices and inflation and how it can affect the monetary policy decisions of the European Central Bank (ECB). The ECB targets average inflation across the Eurozone countries. However, the convergence of prices observed between the Eurozone countries might not necessarily imply convergence of inflation rates between these countries. For price levels to converge, prices in one country might be growing faster than prices in other country. This could result in divergence of inflation rates rather than convergence. Therefore, the ECB needs to pay particular attention to what drives these prices levels. Second, market segmentation observed at the within-country level adds to the problem since it can significantly hinder business cycle synchronization which is one of the prerequisites for the optimum currency areas. This can create a new set of challenges when deciding on the appropriate monetary policy that would work for all Eurozone countries. Additionally, because inflation is measured using mean or median prices within a country, using such a measure could be grossly misleading if within-country prices diverge as illustrated in the paper. These effects exists at least for a number of goods considered, with parallel trade being one of many potential explanations. For these reasons we believe studying the underlying causes of price divergence observed at the country level should be a priority for policy makers.

95,025

Notes: Robust standard errors in parentheses. ***, **, and * denote 1%, 5%, and 10% significance levels. Dependent variable: absolute price differential. Panel A presents the results based on prices collected from supermarkets. Panel B presents the results based on prices collected from the local convenience stores.

from Table 1—price differences within countries are smaller than price differences between countries, as evident from the negative WITHIN coefficient of −0.154. Point estimates of the EURO DID coefficients are in line with our main specification. The EURO DID coefficient is negative and significant, indicating that when WITHIN = 0 (implying that we are looking at between-countries city-pairs only), the absolute relative prices declined by 1.6%. The interaction term EURODID × WITHIN is positive and significant. The introduction of the euro increased relative prices for within-country city-pairs by 5.9% compared to betweencountries city-pairs, implying a net effect of 4.3%. As a second robustness check presented in Table 5, we estimate the same specifications (1) through (4) as in Table 4, but split the sample into prices from supermarkets and prices from local convenience stores. In Panel A, we report the results for the supermarket prices. For the withincountry city-pairs, columns (1) and (2), the introduction of the euro increased relative prices by more than 5%. For the between-countries city-pairs, columns (3) and (4), the results depend on the specification. The EURO DID coefficient is either negative (when the financial crisis is omitted) or insignificant (when we drop the “transition year”). In Panel B, we report the results for the relative prices from local convenience stores. In columns (1) and (2) the within-countries estimates for EURO DID are slightly lower than they are for supermarkets, i.e., the introduction of the euro resulted in less divergence for within-country convenience store prices than supermarket prices. In columns (3) and (4), for the betweencountries city-pairs, the effect is negative and significant, implying convergence of about 2.5% on average. Splitting the sample by within/between city-pairs and by the type of outlet where prices were collected allows us to confirm our theoretical

8

Economic Modelling xxx (xxxx) xxx–xxx

A. Nikolsko-Rzhevskyy, O. Ogrokhina

Appendix

Table A1 List of all available European cities by treatment status. Country

NUTS

Treatment status

Paris Lyon Berlin Dusseldorf Frankfurt

France France Germany Germany Germany

1 1 1 1 1

Hamburg Munich Rome Milan Madrid Barcelona Zurich Geneva London Manchester

Germany Germany Italy Italy Spain Spain Switzerland Switzerland UK UK

Ile-de-France Rhone-Alpes Berlin Dusseldorf Frankfurt am Main, Kreisfreie Stadt Hamburg Munchen, Kreisfreie Stadt Lazio Lombardia Comunidad de Madrid Cataluna Zurich Geneva Inner London Greater Manchester

1 1 1 1 1 1 0 0 0 0

Notes: NUTS (Nomenclature of territorial units for statistics) specify the region(state) where each city is located. Treatment status indicates whether a country is a Eurozone member.

Frankel, J., 2010. The Estimated Trade Effects of the Euro: Why Are They Below Those from Historical Monetary Unions among Smaller Countries?. in 'Europe and the Euro', NBER Chapters. National Bureau of Economic Research, Inc., 169–212. Ganslandt, M., Maskus, K.E., 2007. Vertical distribution, parallel trade, and price divergence in integrated markets. Eur. Econ. Rev. 51 (4), 943–970. Glick, R., Rose, A.K., 2015. Currency Unions and Trade: A Post-EMU Mea Culpa. CEPR Discussion Papers 10615, C.E.P.R. Discussion Papers. Glushenkova, M., Zachariadis, M., 2014. Law-of-One-Price Deviations Before and After the Euro: The case of Cyprus. Cyprus Econ. Policy Rev. 8 (2), 61–85. Goldberg, P.K., Verboven, F., 2005. Market integration and convergence to the law of one price: evidence from the European car market. J. Int. Econ. 65 (1), 49–73. Kyle, M., 2011. Strategic responses to parallel trade. B. E. J. Econ. Anal. Policy 11 (2), 1–34. National Economic Research Associates, 1999. The Economic Consequences of the Choice of Regime in the Area of Trademarks. Final Report for dg xv of the European Commission. National Economic Research Associates, Londo. Ogrokhina, O., 2015. Market integration and price convergence in the european union. J. Int. Money Financ. 56 (C), 55–74. Parsley, D., Wei, S.-J., 2008. In search of a euro effect: big lessons from a Big Mac Meal? J. Int. Money Financ. 27 (2), 260–276. Rose, A.K., 2016. Why do estimates of the emu effect on trade vary so much? Open Econ. Rev., 1–18. Vaona, A., Ascari, G., 2012. Regional inflation persistence: evidence from Italy. Reg. Stud. 46 (4), 509–523.

References Allington, N.F., Kattuman, P.A., Waldmann, F.A., 2005. One market, one money, one price? Int. J. Cent. Bank. 1, 3. Beck, G.W., Hubrich, K., Marcellino, M., 2009. Regional inflation dynamics within and across euro area countries and a comparison with the United States. Econ. Policy 24, 141–184. Bergin, P.R., Glick, R., 2007a. Global price dispersion: are prices converging or diverging? J. Int. Money Financ. 26 (5), 703–729. Bergin, P.R., Glick, R., 2007b. Tradability, productivity, and international economic integration. J. Int. Econ. 73 (1), 128–151. Crucini, M.J., Shintani, M., 2008. Persistence in law of one price deviations:evidence from micro-data. J. Monet. Econ. 55 (3), 629–644. Crucini, M.J., Yilmazkuday, H., 2014. Understanding long-run price dispersion. J. Monet. Econ. 66 (C), 226–240. Dvir, E., Strasser, G.H., 2014. Does Marketing Widen Borders? Cross-Country Price Dispersion in the European Car Market, Boston College Working Papers in Economics 831, Boston College Department of Economics. Engel, C., Rogers, J.H., 2004. European product market integration after the euro. Econ. Policy 19 (39), 347–384. Fischer, C., 2012. Price convergence in the EMU? Evidence from micro data. Eur. Econ. Rev. 56 (4), 757–776.

9