Price Elasticity of Gasoline Smuggling: a Semi-Structural Estimation Approach

Transcription

1 Price Elasticy of Gasoline Smuggling: a Semi-Structural Estimation Approach Hamed Ghoddusi 1, Nima Rafizadeh 2, Mohammad H. Rahmati 3 Abstract We estimate the price elasticy of the demand for gasoline smuggling in Iran. For this purpose, we employ a detailed panel of monthly gasoline consumption data from 160 distribution hubs during the period We apply two different approaches which are diff-in-diff and panel data estimations. The results suggest that the foreign-to-home gasoline price ratios have a significant impact on the time-varying elasticy of demand. This finding supports the hypothesis that an increase in domestic gasoline prices will significantly reduce the demand for smuggling. In addion, as the distance of a region from s closest higherprice neighboring country diminishes, the price elasticy of demand for gasoline smuggling for that region declines as well. Finally, we find that when the ratio of foreign to domestic prices is higher, the gasoline demand sensivies to price in neighboring countries go up. Our work offers new insights regarding the welfare impact of removing fuel subsidies and modifying fuel pricing policies. Keywords: Gasoline Smuggling, Fuel Subsidy Removal, Price Elasticy, Gasoline Price Differential, Panel Data Estimation JEL Classification Codes: C23, F15, H77, Q41 This paper is based on the masters thesis of Nima Rafizadeh at the Sharif Universy of Technology supervised by the other two co-authors. 1 School of Business, Stevens Instute of Technology, Hoboken, NJ, USA. hghoddus@stevens.edu, webse: 2 Graduate Department of Management and Economics, Sharif Universy of Technology, Tehran, Iran. nima.rafizadeh90@gmail.com 3 Corresponding Author, Graduate Department of Management and Economics, Sharif Universy of Technology, Tehran, Iran. rahmati@sharif.edu Preprint submted to Energy Economics November 25, 2017

2 1. Introduction Smuggling is a common problem across international borders where variation in tax or subsidy policy might create a price differential between the source and destination regions. A notable example is the case of smuggling the heavily-subsidized gasoline from Iran to s neighboring countries, where the price of gasoline has typically been much higher than Iran. The lerature on smuggling 1 suggests that the prerequise for smuggling homogeneous goods (e.g. gasoline or beverages) is a price differential for the very same good in two areas. Since gasoline is very close to a homogeneous good, the substantial difference in the price of gasoline between Iran and s neighboring countries has created opportunies for an outbound smuggling of gasoline from Iran. For example, before the (partial) subsidy removal in 2010, the price of gasoline in Iran was on average one-fifteenth of the price in Turkey. Such a non-trivial price gap motivated individuals to earn income by the smuggling and transferring of gasoline to Turkey 2. The gasoline price in Iran has always been set by the government and is constant across all parts of the country. Following a national plan to reduce and eventually eliminate fuel subsidies, the government increased the nominal price of gasoline in three major events over the past couple of years. The new fuel pricing policy aims to set the domestic price at the worldwide equilibrium to narrow down the gap between domestic and foreign fuel prices. This policy was expected to migate or cut incentives for selling subsidized fuel to neighboring countries that typically have higher fuel prices. Capping the fuel smuggling was one of the key motives behind the fuel subsidy reforms (Victor (2009) and Dartanto (2013)). The other motives of this policy shift included controlling the ever increasing domestic demand and releasing the fiscal burden of the gasoline subsidies from the government. The policy has been implemented. For the past few years Iranian domestic fuel prices 1 For a more in depth understanding of smuggling, see Kanbur and Keen (1993), Trandel (1994), Haufler (1996), Wang (1999), Ohsawa (1999), Scharf (1999) and Nielsen (2001, 2002). 2 The smuggling could be carried out through illegal border crossing or via quasi-legal means such as filling the tanks of large trucks and pickups and whdrawing the gasoline right outside Irans border. 2

3 have been closer to the fuel prices in neighboring countries. However, the effectiveness of this policy experiment on reducing the extent of smuggled fuel has not been empirically examined. Our paper is the first to rigorously evaluate such outcomes. The objective of this paper is to use domestic variations in order to provide estimates of the price elasticy of the demand for gasoline smuggling in various regions of the country. We use the fact that gasoline prices in Iran are determined by a central government as a key identification factor. Our research uses administrative regional-level data (for 160 distribution hubs) during the years We utilize the heterogeney of the various provinces and their closest higher-price neighbors. As all of the provinces are forced to follow the same gasoline price, we estimate the elasticy of the demand for gasoline smuggling as a function of price differentials between Iran and neighboring countries, and the distance to the border. We offer multiple contributions to the existing research. First, we provide empirical evidence for the impact of changing government-set fuel prices to the demand for smuggling. To the best of our knowledge, this is the first attempt to provide such results in the context of a resource-rich country. Removing domestic fuel subsidies and bringing prices closer to the international levels have been the subject of recent heated debates. Due to various factors, several oil-rich countries have already started removing or reducing domestic fuel subsidies and narrowing the gap between domestic and international prices. However, there is not much lerature concerning ex-post empirical evaluation of the results of such policies. We fill this gap by offering empirical evidence from Iran, a vast and largely populated oil-rich country that undertook an ambious fuel subsidy removal program in Second, our paper contributes to the lerature of public finance and the demand for smuggling. Prior research has focused on various goods (especially luxury goods such as beverages and cigarettes). We provide evidence for gasoline, a homogeneous and essential good. Despe the typical reliable data collection issues in developing countries, we believe that 3

4 the granulary of our quanty and price observations in conjunction wh the particular instutional setups, which we explo for causal identification, enable us to provide a reasonably reliable estimate of the elasticy. We also offer multiple robustness tests to address or migate potential concerns. The paper is structured as follows. Section 2 presents the lerature review on the estimation of goods demanded wh regard to smuggling, cross-border shopping, and subsidizing fuel. Section 3 presents data. In Section 4 the empirical method used for the estimation of gasoline demand is discussed. Sections 5 and 6 deal wh results and robustness checks. Finally, Section 7 draws the conclusion. 2. Lerature Review 2.1. Cross-Border Shopping and Smuggling A seminal work in this area has been conducted by Leal et al. (2010) who provide a comprehensive review of the lerature on cross-border shopping as a result of tax differentials. According to Mikesell (1970), tax differences between various states in the Uned States induced an interest in cross-border shopping from the mid-1930s. The first applied research, however, was performed between the 1950s and 1970s. 3 In such works, the most general form for each good was calculated in a demand function wh the goods sold in each region as the dependent variable. Explanatory variables include the price and the relative taxation of the regarded goods in each region compared to s neighbors, the income in considered terrories, the cost of travel between neighbouring terrories, and a set of other standard control variables. Fox (1986) was the first paper to explicly incorporate variables related to transport costs into empirical models. More recent papers have also introduced new controls such as the number of Fridays in the month to explain the sales of alcohol (Asplund et al. (2007)), the stock of vehicles to explain the sales of gasoline (Banfi et al. (2005)), as well as 3 For instance, see Mikesell (1970, 1971), Maliet (1955), McAllister (1961) and Hamovch (1966). 4

5 the expendure on advertising by tobacco companies to explain the sales of tobacco (Coats (1995)). Regularly, the specifications take a multiplicative form, which favors subsequent logarhmic transformation and a direct interpretation of the estimated coefficients in terms of elasticy. An illustration of this is provided by the work of Walsh and Jones (1988). Their results suggest that the residents of the counties that border West Virginia choose to purchase outside their state in order to avoid paying higher taxes. Following the methodology proposed by Walsh and Jones (1988), Asplund et al. (2007) estimate how the monthly sales of alcohol in Sweden respond to prices in Denmark and Germany. They conclude that in urban areas close to the border, the elasticy of domestic demand wh respect to foreign prices is approximately 30%. These estimates are reduced to 20% (10%) when the distance is widened to 150 (400) kilometers. In sum, both distance and price differentials crically influence the size of cross-border shopping. Other works have tested the cross-border purchases of alcoholic drinks, such as Crawford and Tanner (1995) and Crawford et al. (1999) for the Uned Kingdom, Smh (1976) and Beard et al. (1997) for the Uned States, and Fleenor (1999) for Canada and the Uned States. There are also studies that focus on the cross-border shopping for tobacco in the Uned States where smuggling tobacco has been a tradional concern of authories (Fisher (2007)). The existence of cross-border shopping has been demonstrated empirically by, among others, Wertz (1971), Warner (1982), Coats (1995), Saba et al. (1995), Fleenor (1998), and more recently by Chiou et al. (2008), Lovenheim (2008) and Merriman (2010). The results obtained by the lerature consistently show that between 2% and 6% of the cigarettes consumed in the Uned States are smuggled. Wh regards to cross-border fueling, Banfi et al. (2005) employ a panel dataset to estimate the impact of gasoline price differences between the border regions of Swzerland and adjacent areas in Germany, Italy, and France on the demand for fuel in the Swiss border areas. Their results establish a strong association between cross-border shopping and price 5

6 differentials. If the price of gasoline in Swzerland were to be reduced by 10%, the demand for gasoline in Swiss border areas would have increased by almost 17.5% Subsidy on Fuel Oil-rich countries often fund fuel subsidies to distribute the resource rent among the people. It effectively lowers their transportation costs and increases gasoline s affordabily for the public. However well-intentioned, the policy will distort the gasoline market wh the resulting inefficiencies, not to mention external costs associated wh high gasoline consumption such as the emission of local pollutants, traffic congestion, and motor vehicle accidents (Davis (2014)). The World Bank recommends that developing nations abandon the gasoline subsidy (Loveless (2015)). The opportuny cost of such policy is high and rising (Coady et al. (2010)). The most common strategy for the removal of subsidies is to implement a phase-out policy. The government only partially cuts the subsidy at any one time. Indonesia has done so in 2000, 2002 and 2003 (Sindo (2014)) and Malaysia has also carried out such measures in 2010 and 2013 (Bridel and Lontoh (2014)). When resistance starts to build up due to negative side effects, the government will then back down on the removal plan. We call this policy a gradual elimination of subsidies or percentage cut, while lifting the whole subsidy is the ultimate goal of the policy. In this case, the government has greater flexibily. Several governments have also attempted a partial subsidy removal plan wh compensation benefs paid to the people. For instance, such a policy was implemented in Indonesia in 2005 and 2008 (WorldBank (2012)), as well as Iran in 2010 (Guillaume et al. (2011)). 3. Data and Instutional Context 3.1. Background on the Iranian Gasoline Market In Iran the national fuel price has always been set by the government in a fixed rate across all regions. In fact, there is no equilibrium market price or regional pricing in the country. A key instutional fact is that there are no private oil providers, and the government-affiliated 6

7 National Iranian Oil Products Distribution Company (NIOPDC) 4 is the sole importer and distributor of fuels. Historically, gasoline and diesel prices have been heavily subsidized in Iran, and the country had been experiencing one of the lowest average retail prices of fuels (in particular gasoline and diesel) in the world. There is no price difference between personal and commercial users of fuel; all vehicles pay the same price. The low price of fuels not only motivates outward smuggling, but also contributes to the ever-increasing domestic consumption. (Guillaume et al. (2011)) In a reaction to the rise in domestic consumption and the consequent increasing import of gasoline, the government implemented a comprehensive plan to migate the growth of fuel demand. Two major components of the plan were to increase the domestic prices (by removing the subsidies) and to ration the subsidized fuel per each vehicle. The rationed price was subsidized, yet was only offered in limed quanties per vehicle 5. The rationing plan was in place from June 2007 until May Following the national plan to reduce fuel subsidies, the government increased the nominal price of gasoline in three major events over the past years (June , December , and April ). The rationing policy required all consumers to use their so-called fuel card when pumping their vehicles at any of the fuel stations 6. Figure 1 presents the monthly retail prices for gasoline (in USD terms) in Iran and s neighboring countries between 2005 and The dollar price of gasoline for Iran is flat for a long period because both the nominal price of gasoline and the exchange rate were held fixed. After the major subsidy removal in 2010 the exchange rate was also floated; thus, the real Iranian gasoline prices have been fluctuating since then. 4 Webse: 5 Commercial vehicles including taxis and trucks were eligible to receive a larger quanty of rationed fuel. 6 For detail discussion of the policy in 2014 see Rahmati and Vesal (2017) 7

8 Figure 1: Gasoline Price in Iran and Its Neighboring Countries Fuel Smuggling Fuel smuggling is a major challenge in Iran. Unfortunately, official data on the volume of smuggled gasoline does not exist. However, unofficial and informal estimates suggest that the country is losing a significant volume of highly-subsidized fuels to s neighboring countries (especially Turkey) every year. For example, according to Iranian counter-smuggling authories the estimated value of the petroleum products smuggling in 2014 was about 3.8 billion USD 7. There are several methods for smuggling, including: fuel tank of personal vehicles, auxiliary tanks attached to trucks and pickups, small and mid-size boats, underground illegal pipelines, fuel tanker trucks, and even animals carrying barrels of gasoline. Anti-smuggling authories have been in work to prevent illegal cross border flows. The law suggests that the penalty for smuggling is three times of the market value of discovered goods. The Iranian police have setup stations along major smuggling routes and in areas close to the borders. Suspicious vehicles are regularly checked for ems such as auxiliary tanks and fuel containers. The stringency of anti-smuggling activies though varies across different provinces. Anecdotal evidence suggests that in regions wh a high rate 7 Source: 8

9 of poverty and unemployment, government officials tolerate small-scale (personal) smuggling functions because of insufficient social secury system. Thus, they sometime disregard small-scale activies by local residents and focus their efforts on the large-scale professional transportation of fuel Data Gasoline Consumption The administrative gasoline consumption data is directly obtained from the NIOPDC. The data set contains monthly observations of regular and premium gasoline consumption in 160 distribution hubs of NIOPDC between 2005 and We also collect official gasoline prices, which is one price in all distribution hubs. Hubs of NIOPDC are responsible for the logistics and distribution of various grades of gasoline to stations. There are 160 regional hubs, each covering a few urban and rural areas. Furthermore, since premium gasoline has a higher price than regular gasoline, and is considered a luxury good, we assume that the gasoline smuggling demand is mainly for the regular gasoline 9. In this paper, we estimate the demand for regular gasoline. Hereafter, whenever we talk about gasoline, we are referring to the regular one. The administrative gasoline sales data collection is managed through a fully automated nationwide system. All consumers are required to use their fuel card when buying gasoline at all the fuel stations. This data is then transmted to a central system. At the end of each day the aggregate consumption for each region is recorded. Relying on such an accurate data provides confidence that our consumption data is most likely free of measurement errors. 8 The province-level fixed-effects in our panel data specification attempt to partially capture the heterogeney in anti-smuggling activies. 9 We had explored the option to use the premium gasoline as a robustness test (or the placebo effect, as the other reviewer also suggested) from the beginning. However, after collecting and cleaning the data we realized a great deal of data recording problems, many missing observations, sparsy, outliers, and also a small sample size problem (the premium gasoline is not available in many locations). Moreover, the share of the premium gasoline is only 3.5% and mainly in large cies. Despe all these challenges, we report the estimations using the premium gasoline in Appendix B. As expected, due to the above reasons, the results are not significant. We relate the lack of significance to the small sample size as well as data problems. 9

10 Moreover, since the price is set by the central administrative authory, our observation of price is also accurate. Descriptive statistics of the gasoline consumption data are summarized in Table 1. The highest quanty of gasoline consumption is observed in Tehran in August The lowest one was in Savadkooh located in Mazandaran province (the northern strip of Iran) in January Gasoline Prices and the Distance from the Border The two key variables which are expected to have significant impacts on the elasticy of gasoline demand are: 1) the ratio of foreign to domestic price; 2) the distance from the closest higher-price border. Variations in these variables enable us to identify foreign and domestic demand in each region. Since we employ the price ratio, gasoline prices in Iran and s neighboring countries have all been expressed in terms of the Uned States Dollars. Table 1 illustrates that gasoline price in Iran is almost always less than s neighboring countries. 11 We assume that gasoline demand for domestic consumption and for smuggling use is made up from both the rationed gasoline price and the non-rationed price; hence, the average of the real price for rationed and non-rationed gasoline is used for domestic price. As for the distance from the border variable, each region s distance from s closest higher-price neighbour has been extracted meticulously both manually and wh the aid of Google Maps Further Socio-Economic Variables One of the limations of the current paper (as many other works in similar contexts) is the lack of granular socio-economic data to be used as perfect controls. In certain cases, the data series exist; however, their time frequency or spatial resolution does not f wh our 10 It should be noted that gasoline consumption data associated wh September 2007 has been ruled out due to inconsistency wh other information. The gasoline consumption in that particular month was inconsistent wh the observations from other months in that year, and also wh the similar months in the previous and following years. Thus, we treated as a possible data recording error. 11 It is possible that prices on the non-iranian side of the border are endogenous to the volume of smuggling. Ideally, one should use regional prices on the non-iranian side of the border to estimate the price ratios. However, reliable and long-spanning data on regional prices do not exist. Thus, we use the nationwide average gasoline prices of the country as a close proxy. If the size of smuggled fuel compared to the total fuel consumption of the region is small, the price effect will also be small and this concern will be alleviated. 10

11 Table 1: Summary Statistics of Key Variables Variable (Location) Type Measure Mean Max Min Gasoline consumption Regional hub/monthly 1000 L 13, ,621 9 Real price of rationed gasoline in Iran Country/Monthly Rial/L 2,329 4,630 1,188 Real price of non-rationed gasoline in Iran Country/Monthly Rial/L 4,826 8,102 1,606 Price of rationed gasoline in Iran Country/Monthly US Dollar/L Price of non-rationed gasoline in Iran Country/Monthly US Dollar/L Price of gasoline in Turkey (NW) Country/Monthly US Dollar/L Price of gasoline in Armenia (NW) Country/Monthly US Dollar/L Price of gasoline in Afghanistan (E) Country/Monthly US Dollar/L Price of gasoline in Pakistan (SE) Country/Monthly US Dollar/L Price of gasoline in Azerbaijan (NW) Country/Monthly US Dollar/L Price of gasoline in Iraq (W) Country/Monthly US Dollar/L Price of gasoline in Persian Gulf (S) Country/Monthly US Dollar/L Price of gasoline in Turkmenistan (N) Country/Monthly US Dollar/L Distance of each region from s the closest higher-price neighboring countries Region/Monthly Kilometers GDP per capa Province/Yearly Thousand Rials 683 3, Cars per capa Province/Seasonal Stock Motorcycles per capa Province/Seasonal Stock Unemployment rate Province/Seasonal GDP and gasoline price in Iran are based on real data in 2011 using consumer price index. Gasoline prices in neighboring countries have been transformed into monthly from their annual reports using gasoline price variation in the Persian Gulf FOB. NW, E, SE, W, S and N stand for northwest, east, southwest, west, south, and north, respectively. 11

12 un of analysis. Despe these expected limations, we did our best to collect the annual GDP, the number of the cars and motorcycles, and the seasonal rate of unemployment at the province level. We turn all variables to per-capa to make them comparable across different locations. 4. Empirical Method 4.1. Identification The key instutional feature of our identification strategy is the existence of an administratively set national price of gasoline. The non-varying prices across time and provinces migate the potential endogeney concern of equilibrium prices at the regional level. If the price was determined through market forces, or even through an administrative mechanism instuted at the regional level, one could argue that the demand for smuggling directly affects the price. We assume that the fuel prices in the neighboring countries are set based on their domestic considerations. Therefore, we can use the exogenous variations of gasoline prices in different countries, time, and distances as a useful variation to identify the elasticy parameters. Moreover, we explo the panel structure of the data and control for regional hub fixed effects. The central object of our analysis is to identify the price elasticy of smuggling. It expresses the sensivy of smuggling to changes in the relative price of gasoline in the foreign and domestic markets and will be defined explicly in the following section Overview of the Model In Iran, the government sets nationwide identical official gasoline prices for all regions. Noticeably, as discussed in Section 3 at each time t, the gasoline price inside the country is at least less than one of s neighboring countries, and smuggling would become profable. Two underlying assumptions are taken into account in this paper which support two different empirical approaches. First, according to the lerature presented in Section 2, by increasing the distance of a region from s neighboring countries, the tendency to smuggle is expected 12

13 to decrease. Therefore, is expected that the demand for gasoline smuggling is insignificant after a certain threshold. Under this assumption, the difference-in-difference approach is applicable because the smuggling behavior is different in bordering areas (treatment) compared wh central ones (control). However, if the demand for gasoline smuggling is accrued from the demand in all regions, then the difference-in-difference approach can no longer be applied. Therefore, is necessary to assume a structure for the demand to separate domestic from smuggling use. The structure we assume is that the propensy of the demand for smuggling is a linear function of the price ratios of the two countries and the distance from the closest higher-price border. Interestingly, we show that even wh different gasoline demand functions for each of the two countries, this structure allows us to identify the smuggling elasticy as a function of price ratios and distances. In other words, the structural model can be pinned down to a reduced form estimation; in a way that the elasticy can be estimated using reduced form coefficients. The coefficient of distance in this semi-structural approach provides an implic test for the underlying assumption behind the difference-in-difference approach. This method is explained in Section 4.4. For simplicy, region i which is located inside the country is shown by H i and s closest higher-price neighboring country at time t, is indicated by F Difference in Difference Approach From a theoretical perspective, aggregate gasoline consumption in a country responds when the price has changed in s neighboring countries. However, the effect of this channel is not significant in provinces and distribution hubs (H i ), which have a distance from F more than a threshold as denoted by d. We implicly assume that national factors (e.g. GDP growth, travel days, gasoline prices, etc) affect symmetrically all distribution hubs. Notwhstanding, the demand for smuggling only shows up in border hubs. Therefore, the foreign gasoline demand which stimulates smuggling can be identified through using a diff- 13

14 in-diff approach for hubs close and far from the borders (treatment vs. control). E[ln(D T h ) P h, P f, X ] = Π 0 + Π 1 ln(p h ) + Π 2 ln( P f ) + Π 3 ln( P f ) Dummy(d, d) + γ 3 X (1) P h P h Dummy(d, d) 1 if d d = 0 if d > d (2) Where D T h is the logarhm of the gasoline demand function in region i at time t that includes both domestic and smuggling uses. Domestic and foreign prices are denoted by (P h ) and (P f ), respectively. X is other covariates including regional hub fixed effects, month fixed effects, GDP per capal, vehicle per capa, and unemployment rate. Domestic use would not change by variation in foreign prices. Therefore, we can claim that after controlling for month fixed effects the elasticy of demand wh respect to foreign prices solely represents the elasticy demands for smuggling in each region i at time t as follows: 12 ɛ = ln(dt h) ln(p f ) = ln(dt h ln( P f ) ) = Π 2 + Π 3 Dummy(d, d) (3) 4.4. Semi-Structural Approach Consider a hypothetical isolated region i where there is no smuggling and s gasoline demand at time t is equal to D h. If smuggling takes place in this region, then s total demand would be D T h. Obviously, D T h D, h because smuggling would add up to domestic uses. However, the amount of demand related to smuggling is cannot be distinguished, so in order to identify the smuggling effect, we take the part of consumption in H i into account that is demanded by smuggling, i.e. D T h D h. Such demand is resulted from the higher price in F, and the smuggler demands for gasoline to gain a prof by selling across 12 Notice that this variation in time and spatial of elasticy stems from the change in d as the closest higher-price may be changed as gasoline prices vary independently in the neighboring countries. 14

15 borders. Accordingly, while H i is encountered wh smuggling, the total gasoline demand in H i at time t is influenced by two prices: P h and P f. When the smuggling in H i is zero, the amount of gasoline demand in H i at time t is: E[ln(D h ) P h, X ] = β 0 + β 1 ln(p h ) + γ 1 X (4) which is in fact the domestic gasoline demand requested by the consumer in H i and is only a price function in (P h ). Similarly, the amount of gasoline used by foreign consumers in F which is simply a function of price in F (P f ) is: E[ln(D f ) P f, Y ] = β 2 + β 3 ln(p f ) + γ 2Y (5) We define S as the influence ratio of gasoline demand in F from the lower price in H i. In fact, the S function is the likelihood in which each representative consumer in F influenced by the lower price in H i resorts to smuggling. We describe s aggregate as the ratio of the F demand influence of price in H i. Two basic presupposions are considered to obtain this ratio. First, S increases due to an increase in the marginal revenue derived from smuggling. Secondly, S decreases as long as gasoline smuggling costs increases from H i to F. The smuggling likelihood function for the consumer i who is located at distance d from the closest bordering region, which has lower prices than the region of the consumer self is defined as follows: S = φ + α ln( P f ) δ ln(d ) (6) P h Lovenheim (2008) uses similar probabily function for the demand of smuggling cigarettes, however he stated that the revenue from smuggling derived from the difference between the price logarhms in the region that purchase took place and residential prices. The cost of smuggling varies wh the distance logarhm from the border. Moreover, φ denotes the fixed cost sustained by each consumer for smuggling. 15

16 In Equation 6, wh an increase in P f, the amount of S increases as well, so we hypothesize that α would be posive. φ, which is the smuggling fixed cost, has an inverse relationship wh S. So, s coefficient s sign should be negative. Eventually, wh an increase in the distance of H i and F, the smuggling appears to become much more costly. Consequently, the demand influence of foreign price reduces, and the δ coefficient s sign in the function S is posive. The final step in the modeling of the total gasoline demand in H i that is D T h, is to extract a demand function that responds to smuggling in addion to domestic demand. As mentioned, when no smuggling activy is present, the demand available in F is obtained through Equation 5. However, when F is encountered wh smuggling, the F region is influenced by the price available in H i wh the likelihood S. Hence, the mathematical expectation for the demand function of region F is as follows: Where D fs E[ln(D T f ) P h, P f, Y ] = (1 S ) E[ln(D f ) P f, Y ] + S E[ln(D fs ) P h, Y ] (7) is the demand which is in F, but wh price P h. Thus, in the case of smuggling, compared to smuggling free suations, the mathematical expectation for the gasoline demand function in F alters as follows: E[ln(D T f ) P h, P f, Y ] E[ln(D f ) P f, Y ] = S (E[ln(D fs ) P h, Y ] E[ln(D f ) P f, Y ]) (8) Any changes in the mathematical expectation of the gasoline demand function is going to be responded wh smuggling from the resources inside H i. In other words, at time t, besides the mathematical expectation for demand available in H i for domestic consumption, the F region also requests gasoline from H i according to Equation 8. Hence, the overall demand function which exists for gasoline in H i at time t is as follows: E[ln(D T h ) P h, P f, X ] = E[ln(D h ) P h, X ] + S (E[ln(D fs ) P h, Y ] E[ln(D f ) P f, Y ]) (9) 16

17 By replacing Equations 4, 5, and 6 in the above equation, the reduced form of the total gasoline demand function in H i at time t is obtained as follows: E[ln(D T h ) P h, P f, X ] = β 0 + β 1 ln(p h ) φβ 3 ln( P f ) αβ 3 [ln( P f )] 2 + δβ 3 ln( P f ) ln(d ) + γ 1 X P h By defining the following equalies, the reduced form of the gasoline demand function is turned to Equation 11: P h P h (10) E[ln(D T h ) P h, P f, X ] = A 0 + A 1 ln(p h A 0 = β 0, A 1 = β 1, A 2 = φβ 3, A 3 = αβ 3, A 4 = δβ 3 ) + A 2 ln( P f ) + A 3 [ln( P f )] 2 + A 4 ln( P f ) ln(d ) + γ 1 X P h Using the structural model and the intuion behind s deep parameters, we can determine the coefficient sign of the reduced form model. According to the law of demand, price and the demand for gasoline should have an inverse relationship. 13 Hence, in Equations 4 and 5, β 1 and β 3 are negative. Now, the coefficient sign for A 1, A 2, A 3 and A 4 are extracted mathematically. Wh the aid of other coefficients stated in Equation 6 for S, the coefficient signs in the paper s presupposions are as follows: P h P h (11) β 1 < 0, A 1 = β 1 A 1 < 0 φ < 0, β 3 < 0, A 2 = φβ 3 A 2 < 0 α > 0, β 3 < 0, A 3 = αβ 3 A 3 > 0 δ > 0, β 3 < 0, A 4 = δβ 3 A 4 < 0 It is expected that wh an increase in gasoline prices in a country, s demand goes down. Hence, the A 1 coefficient sign, which became mathematically negative, is compatible wh our intuion. Similar to Equation 3, and based on Equation 11, gasoline demand elasticy 13 We assume that gasoline is not a Giffen good. 17

18 wh the proportion of foreign price to the domestic one in each i region at time t is as follows: ζ = ln(dt h) ln( P f ) P h = A 2 + 2A 3 ln( P f ) + A 4 ln(d ) (12) As appears, Equation 12 is the sum of three phrases. It is recognized that when the P h foreign price exceeds the domestic price, the impetus for gasoline smuggling appears. It was also stated earlier that irrespective of the price differential and the distance from the border in each region, smuggling has a fixed cost (A 2 ) which is almost the same in all the regions. Therefore, the coefficient sign is intuively expected to be negative. In addion, for fixed domestic prices and a specific location, as the price of foreign gasoline increases, then the demand would be more elastic to foreign prices. Therefore, the A 3 coefficient sign is anticipated to be posive. For the third phrase, can be said that according to earlier assumptions, the more distance from the closest higher-price neighboring country, the less the elasticy. That is, an increase of one percent in the ratio of the foreign price to the domestic price has less influence on gasoline demand in faraway areas from the border, and the A 4 coefficient sign is anticipated to be negative Discussion about variable of distance There might be a concern of the logarhm functional form we assumed in Equation 6 for distance. So, we also consider the linear relationship between smuggling demand as follows: E[ln(D T h ) P h, P f, X ] = A 0 +A 1 ln(p h )+A 2 ln( P f )+A 3 [ln( P f )] 2 +A 4 ln( P f ) (d )+γ 1 X (13) P h For robustness check, we also describe a dummy variable as in Equation 15 for the certain intervals of the border of the neighboring countries to capture non-lineary in demand by P h P h 18

19 distance. Then, Equation 14 is applied for the gasoline demand in each i region at time t: E[ln(D T h ) P, h P f, X ] = η 0 + η 1 ln(p) h + η 2 ln( P f ) + η P h 3 [ln( P f )] 2 + P h (14) η 4 ln( P f ) Dummy(d P h, (d 0, d 1 )) η n+3 ln( P f ) Dummy(d P h, (d n 1, d n )) + γx 1 if d [d p, d q ) Dummy(d, (d p, d q )) = 0 if d / [d p, d q ) (15) 5. Result 5.1. Estimation Gasoline prices are the same across all Iranian regions and hubs at any given time period. Moreover, the price in each month is less than at least one of Iran s neighboring countries. Therefore, the assumptions presented in Section 4 for the modeling of the gasoline demand function are satisfied. As a result, we can estimate the coefficients presented in Section 4 using the data set introduced in Section 3. Moreover, in all the following approaches, standard errors for panel specifications are robust and clustered at the level of the cross-sectional un to allow for arbrary serial correlation Estimation Results of The Diff-in-Diff Approach The coefficient associated wh the log of price ratio ( ln( P f ) ) is a key indicator for the estimation of the elasticy of smuggling demand. The estimation results presented in Table 2 suggest that this coefficient is not significant in any of the various specifications (reported in various columns of Table 2). 14 The small and insignificant estimates of the coefficient of the log of price ratio ( ln( P f ) ) P h may stem from the inconsistent underlying assumption that smuggling took place wh higher 14 It should be noted that the domestic price (P h ) is constant across all hubs at a given time t. Hence, the domestic price is perfectly co-linear wh the monthly dummy and thus, can not be identified. We dropped from those regressions wh month fixed effects, and reported those results. 19

20 Table 2: Demand Estimation wh Diff-in-Diff Approach Dependent Variable: D (Gasoline consumption in the regions of Iran) Variables (1) (2) (3) (4) (5) ln( P f ) (0.0575) (0.0555) (0.0588) (0.0670) (0.111) ln( P f ) Dummy(d, 1) *** (0.0633) ln( P f ) Dummy(d, 50) (0.0288) ln( P f ) Dummy(d, 100) ** (0.0274) ln( P f ) Dummy(d, 200) (0.0238) ln( P f ) Dummy(d, 400) (0.0930) ln[(gdp per Capa) ] (0.143) (0.145) (0.146) (0.145) (0.145) ln[(cars per Capa) ] (0.0834) (0.0831) (0.0830) (0.0836) (0.0835) ln[(motorcycles per Capa) ] (0.0604) (0.0609) (0.0604) (0.0610) (0.0609) ln[(unemployment Rate) ] *** *** *** *** *** (0.0127) (0.0128) (0.0131) (0.0127) (0.0128) Constant 15.00*** 14.81*** 14.61*** 14.77*** 14.79*** (2.271) (2.277) (2.302) (2.295) (2.281) Region Fixed Effects Yes Yes Yes Yes Yes Month Fixed Effects Yes Yes Yes Yes Yes R-Squared Number of observations in each column: 16,552. Standard errors are robust and clustered by region to allow for arbrary serial correlation. *** and **: indicate 1% and 5% significance levels, respectively. Standard errors are in parentheses. 20

21 intensy in the border regions. In contrast, we find that the demand declines more in border regions when the rent from smuggling increases (the sign of the coefficient for the interaction of the log of price ratio and the dummy for distance when threshold equals to 1 km ( ln( P f ) Dummy(d, 1) ) is negative in Column 1 of Table 2). In other words, the border areas cannot be actually specified as the treatment group in response to exogenous domestic price increases. To validate whether there exists a variation in gasoline consumption between treatment and control in response to shocks, Figure 2 plots the average consumption for hubs in various distance thresholds similar to Columns 1-5 in Table 2. This figure, and the estimates in Table 2 indicate that if there are any differences between regions, though insignificant, there would appear to be less smuggling in areas near the border. Apparently, this finding contradicts the inial assumption that the smuggling reduces as the hub is located far from border. For seeing the frequency of distance of hubs from the closest higherprice border, we calculate the weighted average of the distance of hubs from the border for the period between 2005 and 2014, and report them in Table Table 3: Frequency of Hubs When Weighted Average of Distance from the Border is Equal or Less than d d (km) Frequency of Hubs Total number of hubs: Estimation of Semi-Structural Approach Table 4 shows the estimates of Equation 11 as the main finding of this paper. Column 1 reports the estimates of the panel model wh a limed number of fixed effects. There exist unobserved time-invariant hub specific factors that appear to be correlated wh price ratios. Excluding these fixed effects would bias the coefficients of Column 1 toward zero. Therefore, the coefficients of the log of price ratio ( ln( P f ) ), the log of quadratic price ratio ( [ln( P f P h )] 2), P h and the interaction of the log of price ratio and the log of distance ( ln( P f ) ln(d ) ) are 15 Refer to Appendix C to see the distribution of distance of hubs from the closest higher-price border. 21

22 Figure 2: Average of Gasoline Consumption Over Time for Various Thresholds Dotted lines are the average of gasoline consumption for hubs wh a distance less than the indicated threshold in each plot. The solid lines are the same averages for hubs in the other side of the threshold. All figures are normalized to their first observations. The vertical lines show major policy changes (significant price increases and rationing). Also, increases in the amount of nominal prices are in parentheses. 22

23 not significant. However, after adding month and region fixed effects we obtain promisingly significant results in the third column of Table 4. In the fixed effects estimation, heterogeneous intercepts are allowed for each hub. These intercepts capture the specific characteristics of each region such as the tendency to break the law (and enter the smuggling business), time invariant climate factors, natural barriers for smuggling, the stringency of anti-smuggling activies, outside options for employment, etc. We see that in comparison to the OLS method, the fixed effect estimation generates significant results. As described in Section 4 we would expect that the signs of the coefficients of the log of domestic price ( ln(p h ) ) and the log of price ratio ( ln( P f ) ) to be negative and the sign of the coefficient of the log of quadratic price ratio ( [ln( P f )] 2) to be posive. However, the sign of the coefficient of the interaction of the log of price ratio and the log of distance ( ln( P f ) ln(d ) ) is not in line wh our prior assumptions. The posive coefficient for the interaction of the log of price ratio and the log of distance ( P ln( f ) ln(d P h ) ) suggests that the smuggling elasticy rises as the distance to the border increases, keeping all other variables fixed. This finding inially appears to be puzzling, because we expect higher elasticy in areas close to the border. Importantly, the posive coefficient suggests that when the price increases in the neighboring country, those areas that are far from the border are more sensive to such price changes. In Appendix A, we have provided a demonstrative theoretical framework that highlight this matter The Specification of Distance We are also interested in investigating the effects of the distance level rather than s logarhm. For this purpose, Equation 13 is estimated by including cross-sectional fixed effects as well as time period fixed effects in the model. Results in the first column of Table 5 illustrate that in comparison to the distance logarhm specification, the significance of the coefficient of the distance level is reduced from the 1% level to the 10% level. It suggests that a non-linear relationship between the distance and smuggling activies is more intuive. We specify three dummies for the distance from the border to include hubs that are at 23

24 Table 4: Demand Estimation wh The Semi-Structural Approach Dependent Variable: D (Gasoline consumption in the regions of Iran) Variables (1) (2) (3) ln(p h ) *** (0.176) (0.0489) ln( P f ) *** *** (0.312) (0.0464) (0.0793) [ln( P f )] * *** (0.0882) (0.0115) (0.0144) ln( P f ) ln(d ) *** *** (0.0556) ( ) ( ) ln[(gdp per Capa) ] (0.180) (0.103) (0.143) ln[(cars per Capa) ] (0.267) (0.0719) (0.0826) ln[(motorcycles per Capa) ] (0.176) (0.0552) (0.0595) ln[(unemployment Rate) ] *** *** (0.156) (0.0116) (0.0125)) Constant 11.36*** 16.33*** 14.97*** (3.349) (1.483) (2.257) Region Fixed Effects No Yes Yes Month Fixed Effects No No Yes R-squared Number of observations in each column: 16,552. Standard errors are robust and clustered by region to allow for arbrary serial correlation. ***, ** and *: indicate 1%, 5% and 10% significance levels, respectively. Standard errors are in parentheses. 24

25 a maximum of 100km from the border, between 100km and 200km, and beyond 200km. Only the estimate associated wh the first dummy is significant, suggesting that smuggling activies are sensive to the distance at the border regions. Consistent wh the earlier findings the sign is negative. One plausible explanation is that smugglers who are living close to the border have a lower price elasticy compared to the professional smugglers that benef from pumping of hubs in areas far from the border. The cost of smuggling also increases as the distance from the border increases. Thus, the demand for smuggling in central areas is probably more sensive to the price differentials Disaggregated Elasticy Estimations After reporting the aggregate estimates of elasticy (for all time periods and over all regional hubs), we focus on the elasticy over time and across different regions. We conduct a system estimation for individual hubs based on the specification in Equation Time-Varying Elasticy Figure 3 presents the weighted average over all regions of the time-varying price elasticy of smuggling demand for gasoline. Four crical incidents are worth noticing. A good match between the timeline of real world events and our estimated values provide strong anecdotal support to the validy of our estimations. The first incident (fuel rationing) takes place around July 2007 and is related to a pure quanty effect. We observe that prior to the enforcement of the fuel rationing policy (July 2007), the level of smuggling elasticy is substantial. However, right after the start of the rationing policy (by obliging the consumers to use their electronic fuel card) a significant drop (almost to zero) is observed. 16 Some of the model-predicted smuggling elasticy values are negative (particularly in hubs that are close to the border.) These few negative values should be interpreted wh some caution. As the theoretical model of Appendix B highlights, in the presence of fixed legal and transportation costs the smuggling only takes place when the price differential is larger than a certain threshold. The estimated intercept for the main equation is negative; thus, the predicted values of model are only valid when the price differential is sufficiently larger. 25

26 Table 5: Demand Estimation wh The Semi-Structural Approach, Alternative Distance Estimation Dependent Variable: D (Gasoline consumption in the regions of Iran) Variables (1) (2) ln( P f ) ** (0.0685) (0.0696) [ln( P f )] *** *** (0.0152) (0.0161) ln( P f ) (d ) * ( ) ln( P f )*Dummy(d, (1, 100)) *** (0.0315) ln( P f )*Dummy(d, (100, 200)) (0.0289) ln( P f )*Dummy(d, (200, 560)) (0.0224) ln[(gdp per Capa) ] (0.144) (0.144) ln[(cars per Capa) ] (0.0825) (0.0833) ln[(motorcycles per Capa) ] (0.0598) (0.0601) ln[(unemployment Rate) ] *** *** (0.0125) (0.0131) Constant 14.89*** 14.77*** (2.282) (2.280) Region Fixed Effects Yes Yes Month Fixed Effects Yes Yes R-Squared Number of observations in each column: 16,552. Standard errors are robust and clustered by region to allow for arbrary serial correlation. ***, ** and *: indicate 1%, 5% and 10% significance levels, respectively. Standard errors are in parentheses. 26

27 Figure 3: Weighted Average of monthly Elasticy of Gasoline Smuggling Demand in Iran The solid line is the estimate of the smuggling elasticy and the dotted line is one standard error of the elasticy computed by bootstrap (50 replications). The vertical lines show major policy changes (significant price increases and rationing). Also, increases in the amount of the nominal price are in parentheses. Finally, the dashed line is corresponding to the secondary vertical axis and is the gasoline price in Iran. The second interesting event window happens around mid-2008, when due to the drop in the global crude oil prices, gasoline prices in the neighboring countries also declined (while the domestic prices were kept constant). Third, the domestic gasoline prices experienced a more than two-fold increase in December We see that the elasticy of gasoline smuggling has reached s lowest bound in the past ten years, and that the average of monthly gasoline smuggling elasticy in the country is almost zero in the late Finally, starting in mid-2012, a significant deprecation of the Iranian currency reduced the dollar value of domestic gasoline prices and created a new wave of gasoline smuggling. Smuggling activies became profable again and a slow upward trend in elasticy is observed which lasts until May Two observations are worth noting. First, in order to estimate smuggling, the fact that the major variation over time comes mainly from price differentials between Iran and neighboring countries actually works to our advantage. In the absence of smuggling, we would not expect the fuel prices in a neighboring country to affect the sales of gasoline in Iran; whereas, we observe a significant sensivy of our estimated parameter to the prices in 27

28 neighboring countries. Second, the time-varying elasticy estimates in Figure 3 are entirely estimated off of cross-sectional variation in the pairing of hubs to the neighboring countries. Thus, the correct matching of the smuggling source and the destination country is crucial for obtaining correct results. If the pairing is not reflective of actual smuggling patterns, this will introduce measurement error in the foreign price (P f ), which would attenuate the estimated smuggling elasticy Location-Specific Elasticy We also report estimates of smuggling elasticy at the provincial level. One expects that the elasticy must be higher in locations closer to major illegal export distinctions (e.g. Turkey in the case of our study). The weighted average of price elasticy of the demand for gasoline smuggling in each province (for the period 2005 and 2014) is shown in Figure 4. Gasoline smuggling intensy is distributed heterogeneously wh respect to the distance from the border and the price differential wh the closest higher-price neighbor country. The highest elasticy is estimated for West Azerbaijan (WA) wh an approximate elasticy of This province is the source for a large-scale smuggling to Turkey and Iraq. Razavi Khorasan (RK) and North Khorasan (NK) provinces wh the elasticy of and come next. It might be interesting to know that the highest monthly elasticy of gasoline smuggling demand in Iran (0.14) belongs to Xoy, a small cy located in West Azerbaijan that is 45 kilometers away from Turkey s border. Note that in the model, the elasticy of smuggling is determined by the sum of two effects: 1) the distance to the border (- up to a point), 2) the relative price at the closest higher-price border (always +). What happens is that fuel prices are much higher in Turkey than in other countries; thus, the relative price effect dominates the (decreasing) distance effect and makes the elasticy for provinces close to Turkey very high. A meticulous question that arises after seeing Figure 4 is why are not enforcement efforts reallocated from low-smuggling hubs to high-smuggling hubs? First of all, we are not sure if the authories can easily identify those hubs (our study might help them in this regard). 28

29 Second, the hubs far from the border are typically located in urban areas (and connected to many individual gas stations). In practice is almost impossible to identify a smuggler from a leg commercial or individual buyer of the fuel. Figure 4: Weighted Average Gasoline Smuggling Elasticy By Province Legend shows the intervals and number of provinces in each. Numbers in neighboring countries are average gasoline price per ler in terms of US dollar. Results are from WA, KR and NK stand for West Azerbaijan, Razavi Khorasan and North Khorasan, respectively. 6. Discussion and Robustness Checks In this section we offer multiple robustness checks by changing crical specifications of the model As a robustness test, we included the quadratic price ratio term interacted wh the log of distance in the semi-structural approach, but the results did not change in a significant way. 29

30 6.1. Effect of Octane To properly interpret the price differentials between Iran and s neighboring countries, one should consider the octane level of gasoline produced and sold in these markets. Table 6 reports the usual gasoline octane level in the countries presented in our study. We see that except for Turkey, all other countries have an equal or a lower octane number compared to Iran. Though not always but typically a higher octane may imply a better type of gasoline, if were true, the smuggled gasoline from Iran could be considered as a premium good in those destinations. Addionally, the Iranian gasoline should be sold at a discount in the Turkish market. Thus, the estimate for the elasticy in the Turkish market might be biased downward. Table 6: Gasoline s Octane Rating in Iran and Its Neighboring Countries Country Iran Turkey Pakistan Azerbaijan Iraq Persian Gulf Octane Rating Unemployment Rate The first robustness check considers the role of the unemployment rate. The negative sign of the estimated unemployment rate coefficient indicates that the gasoline demand is reduced by an increase in the unemployment rate. This influence might be different for regions closer to the border. Raising the unemployment rate in border areas leads to a reduction in income, so may increase the incentive for the local population to smuggle gasoline. Therefore, unlike the inner provinces, a higher rate of unemployment in border provinces may indeed increase the recorded sales of gasoline in that region. To test for this alternative hypothesis, we conduct an analysis which is demonstrated in the first column of Table 7. We define a dummy variable to represent provinces closer to the border and add an interaction term wh the unemployment rate to the regression equation. If a higher local unemployment indeed encourages smuggling, we expect to see a posive sign for the coefficient of the interaction term. However, the negative sign of the estimated 30

31 term suggests that gasoline demand diminishes wh the increase of the unemployment rate even in regions near the border. Therefore, we conclude that eher the local demand effect dominates that of the other variables, or that the province-level unemployment is not a significant determinant of the incentive for smuggling Reverse Smuggling Among Iran s neighboring countries, Turkmenistan has always had a lower gasoline price throughout the time span of this study. Thus, a (reverse) smuggling route from Turkmenistan to Iran is conceivable. To gauge the effect of this possibly contaminating channel, we re-estimate the model after dropping certain cies. These cies include cies closed to Turkmenistan and cies that are closer to Afghanistan, but are also close enough to Turkmenistan as well. The results reported in the second column of Table 7 illustrate that the significance level and the coefficients size of the previous estimations are not different Domestic Price: Rationed or Non-Rationed? Following what we have stated in the data section, we assume that the demand for domestic consumption and smuggling are fulfilled through both rationed and non-rationed gasoline 18. Hence, as an addional robustness check, the average price for rationed and non-rationed gasoline is used. Moreover, the rationed (i.e. heavily subsidized) gasoline price is used for domestic demand and the non-rationed gasoline price (less subsidized) is applied for foreign demand estimation. The third column of Table 7 illustrates that the coefficients have not changed compared to previous estimations. Moreover, the coefficients still are significance at the same level, and only the significance of the coefficient for the log of price ratio ( ln( P f ) ) is reduced from the 1% level to 5% level in comparison to the previous estimation. 18 Except for a short period of time in 2007, the rationing did not put any quanty restrictions on the consumer. It was possible to purchase any quanty of gasoline beyond the subsidized ration by paying a higher official price. This policy of dual prices lived for a few years and then was abandoned in 2015 toward a unified subsidy-free policy. 31

32 6.5. Measurement Errors in Covariates Because we control for both region fixed effects and time fixed effects, aggregate variables like GDP per capa and unemployment rate are demeaned. Thus, their deviations help to identify their coefficients. It is likely that if these deviations are just measurement errors then including them in our estimation cuts the significance of other variables. 19 In order to investigate this concern, we estimate the model by ignoring these two control variables. It is observed in the fourth column of Table 7 that in comparison to the previous estimation results, all the three important coefficients for the price elasticy of smuggling demand for gasoline have preserved their coefficient signs and significance levels Non-lineary in Distance Another concern is that the socio-economic condions may be non-linear by distance so the way we model them in Equation 11 undermines these effects. In particular, the nonlineary in the socio-economic condions of hubs that are located very close to the border compared wh hubs in border regions may derive our results. Hence, as the last robustness check, we om the areas wh a distance greater than d and estimate the model wh the remaining data. In Table 8, we consider the following amounts for d: 50km, 100km, 200km, 300km, and 400km. Results in Table 8 indicate that the coefficient of interaction of the log of price ratio and the log of distance ( ln( P f ) ln(d ) ) is not significant for the areas wh a distance of less than 100km. In the areas wh greater distances, however, all three important coefficients in the elactisy for smuggling are significant at the 1% level, and wh the expected signs. 7. Conclusion This paper investigates the price elasticy of the demand for gasoline smuggling in Iran. Toward this goal, a panel of monthly gasoline consumption data has been used that contains 19 Fisher and Schlenker (2012) highlight the same issue in the context of climate change 32

33 Table 7: Semi-Structural Approach Estimation - Robustness Check Dependent Variable: D (Gasoline consumption in the regions of Iran) Variables (1) (2) (3) (4) ln( P f ) *** *** ** *** (0.0791) (0.101) (0.0699) (0.0781) [ln( P f )] *** *** *** *** (0.0144) (0.0154) (0.0109) (0.0148) ln( P f ) ln(d ) *** *** *** *** ( ) ( ) ( ) ( ) ln[(gdp per Capa) ] (0.142) (0.157) (0.143) ln[(cars per Capa) ] (0.0824) (0.0924) (0.0829) (0.0809) ln[(motorcycles per Capa) ] (0.0590) (0.0625) (0.0600) (0.0607) ln[(unemployment Rate) ] ** *** *** (0.0132) (0.0124) (0.0126) ln[(unemployment Rate) ] *Dummy(d, 1) *** (0.0685) Constant 15.31*** 14.75*** 15.25*** 17.75*** (2.262) (2.452) (2.556) (1.288) Region Fixed Effects Yes Yes Yes Yes Month Fixed Effects Yes Yes Yes Yes Observations 16,552 14,305 16,552 16,552 R-squared Standard errors are robust and clustered by region to allow for arbrary serial correlation. ***, ** and *: indicate 1%, 5% and 10% significance levels, respectively. Standard errors are in parentheses. 33

34 Table 8: Semi-Structural Approach Estimation - Distance Robustness Check Dependent Variable: D (Gasoline consumption in the regions of Iran) Variables (1) (2) (3) (4) (5) d 50 d 100 d 200 d 300 d 400 ln( P f ) ** *** *** *** *** (0.101) (0.106) (0.106) (0.0884) (0.0799) [ln( P f )] *** *** *** *** *** (0.0246) (0.0273) (0.0190) (0.0157) (0.0141) ln( P f ) ln(d ) *** *** *** (0.0185) (0.0157) (0.0103) ( ) ( ) ln[(gdp per Capa) ] (0.131) (0.195) (0.174) (0.156) (0.147) ln[(cars per Capa) ] ** (0.0718) (0.131) (0.117) (0.0923) (0.0834) ln[(motorcycles per Capa) ] 0.257*** (0.0984) (0.0873) (0.0695) (0.0620) (0.0597) ln[(unemployment Rate) ] *** *** * ** *** (0.0164) (0.0165) (0.0164) (0.0129) (0.0131) Constant 11.02*** 10.77*** 13.08*** 14.68*** 14.84*** (1.518) (3.367) (2.810) (2.461) (2.320) Region Fixed Effects Yes Yes Yes Yes Yes Month Fixed Effects Yes Yes Yes Yes Yes Observations 4,346 7,325 10,899 14,694 16,112 R-squared Standard errors are robust and clustered by region to allow for arbrary serial correlation. ***, ** and *: indicate 1%, 5% and 10% significance levels, respectively. Standard errors are in parentheses. 34

35 observations for 160 distribution hubs of the National Iranian Oil Products Distribution Company (NIOPDC) between 2005 and The aggregate estimation results of the elasticy of monthly gasoline smuggling demand shows the significant effect of gasoline price ratios between Iran and s neighboring countries. On the other side, as the distance of a region from s closest higher-price neighboring country diminishes, the price elasticy of demand for gasoline smuggling for that region declines as well. Our paper takes the first step to provide reliable quantative evidence to support the effect of domestic price adjustments on the tendency to smuggle fuel. Our results suggest that adjusting domestic fuel prices to bring them closer to the international equilibrium can generate considerable fiscal savings. The key point of our study is that smuggling acts as a variable component of government budget. The higher the level of domestic prices, the lower is the demand for smuggling (thus, the lost subsidy to neighboring countries) as well as the domestic demand for gasoline. Both channels reduce the subsidy burden. The removal of subsidy basically replaces a variable level of subsidies wh a time-invariant (in nominal terms) cash transfer. This paper can be extended in multiple directions. First, the unique method presented in the current paper can also be used to estimate the elasticy of domestic demand for gasoline. Second, our results do not include the public finance aspects of the problem. Future work can provide estimates of government subsidies and the fiscal saving and welfare implications of the price adjustment. Our results can also be combined wh more granular socio-economic variables at the household level to examine the distributional effects of timevarying smuggling activies on households involved wh those activies. Our model estimates the changes in smuggling as a response to changes in the relative price. However, is not possible to estimate the baseline magnude of smuggling using this model (because one can only observe the differential response). We added some unofficial figures to the introduction of the paper but there is no reliable estimate of the total volume of the smuggled gasoline. An important follow-up research can focus on estimating the 35

36 magnude of the smuggled fuels. Acknowledgment We are grateful for comments received following the presentation of an earlier draft of this paper at Sharif Universy of Technology and the members of NYC s Dark Coffee Economics Discussion Group. We also would like to thank Djavad Salehi-Isfahani, Mohammad Vesal, Sorena Rahi and two anonymous reviewers for their very helpful comments on an earlier draft. The authors are, of course, responsible for all errors and omissions. 36

37 Appendix A Relationship Between the Elasticy and the Distance We present a stylized theoretical model of smuggling in order to derive the relationship between the elasticy of the demand for smuggling and the distance from H i to the border of F. Consider an economy wh representative smugglers in each region. We solve the utily maximization problems of the representative smugglers to derive the equilibrium outcome of the smuggling intensy for a given level of price differentials (between home and foreign countries), probabily of apprehension, level of penalty, and finally the distance from the border. The utily function of a smuggler in region i and at time t who smuggles q amount of gasoline from H i to F can be wrten as follows: U (q ) = (1 λ ) ( (P f P h )q Φ(q, d ) ) Zλ q (16) where, λ is the probabily of the apprehension and penalizing of the smuggler by the police, Φ(q, d ) is the cost of transporting q uns of gasoline from H i to F, and Z is a fixed cost that the smuggler is obliged to pay per un of gasoline when he is penalized. It should be noted that in each region i at time t, the prices, the distance and the probabily may be have different values, and hence, the utily of the smuggler can vary over time and space. We define P = P f P h and then wre the F.O.C in the form of (1 λ ) P = (1 λ )Φ 1 (q, d ) + λ Z. We use the implic function theorem to obtain the elasticy: ɛ = ln(q ) ln( P ) = P (q ) = P 1 q P q Φ 11 (q, d ) (17) In order to derive the relationship between the elasticy of the demand for smuggling 37

38 and distance, we use the following derivative to obtain the relationship: ɛ d = ɛ q q d = ( P q 2 Φ 11(q, d ) + P Φ 111(q, d ) ( Φ11 (q, d ) ) 2 )( ) Φ 12 (q, d ) Φ 11 (q, d ) (18) If one assumes that Φ 11 (q, d ) > 0, then if Φ 12 (q, d ) < 0, we conclude that ɛ d > 0, which is what we obtained from our estimates. This model corresponds to the notion that those who are far from the border, are more sensive to price changes. Appendix B Estimation of the Demand for Premium Gasoline The premium gasoline is the higher qualy version of the fuel wh a higher octane. It is sold wh a significant price premium (between 10% - 30% higher) in the Iranian market. To gain insights from a possible alternative estimation result, we repeat the semi-structural estimation procedure by replacing the regular gasoline price and quanty data wh those of the premium gasoline. A summary of key results in this regard are reported in Table 9. The results suggest that none of the coefficients associated wh the determinants of smuggling are significant when the estimation is done using the premium gasoline. There are multiple plausible explanations for this negative (i.e. insignificant) result. First, the premium gasoline is more expensive than the regular gasoline; however, the qualy of the gasoline in the destination is not verifiable and will be sold at the same price as the conventional one. Therefore, the price differentials between Iran and neighboring countries will be smaller for the premium gasoline, resulting in a lower incentive to smuggle this type of gasoline when the conventional type is available. Also, note that in the case of apprehension, the smuggler will lose a larger value of the confiscated good. Secondly, premium gasoline is mainly offered in larger cies far from the border. Therefore, is more difficult to obtain a larger volume of the premium type in hubs near the border. Finally, the total market share of the premium gasoline is very small (only 3.5% of the total gasoline market). We also have noted data recording problems, missing observations, outliers, and finally small sample size problems in the premium gasoline data. As a result, the results of this estimation are likely 38

39 Table 9: Demand Estimation wh The Semi-Structural Approach Dependent Variable: D (Premium gasoline consumption in the regions of Iran) Variables (1) (2) (3) ln( P f ) 0.630* (0.377) (0.106) (0.144) [ln( P f )] (0.0917) (0.0396) (0.0425) ln( P f ) ln(d ) (0.0721) (0.0179) (0.0180) ln[(gdp per Capa) ] 0.770** (0.300) (0.254) (0.309) ln[(cars per Capa) ] *** (0.348) (0.127) (0.144) ln[(motorcycles per Capa) ] (0.235) (0.105) (0.0918) ln[(unemployment Rate) ] ** *** (0.213) (0.0456) (0.0349) Constant *** (5.204) (3.437) (4.303) Region Fixed Effects No Yes Yes Month Fixed Effects No No Yes R-squared Number of observations in each column: 9,461. Standard errors are robust and clustered by region to allow for arbrary serial correlation. ***, ** and *: indicate 1%, 5% and 10% significance levels, respectively. Standard errors are in parentheses. 39

40 to be subject to small sample and measurement errors. Appendix C Distribution of the Distance We plot the histogram for the weighted average of the distance of hubs from the border for the period between 2005 and 2014 in Figure 5. Figure 5: Histogram Plot of Weighted Average of Distance of Hubs from Border The numerical values above the bars show the frequency of hubs in each bin. 40