Jensen (2007) PDF - Quarterly Journal of Economics

THE Vol. CXXII August 2007 Issue 3 THE DIGITAL PROVIDE: INFORMATION (TECHNOLOGY), MARKET PERFORMANCE, AND WELFARE IN THE SOUTH INDIAN FISHERIES SECTOR* ROBERT JENSEN When information is limited or costly, agents are unable to engage in optimal arbitrage. Excess price dispersion across markets ca...

THE Vol. CXXII August 2007 Issue 3 THE DIGITAL PROVIDE: INFORMATION (TECHNOLOGY), MARKET PERFORMANCE, AND WELFARE IN THE SOUTH INDIAN FISHERIES SECTOR* ROBERT JENSEN When information is limited or costly, agents are unable to engage in optimal arbitrage. Excess price dispersion across markets can arise, and goods may not be allocated efficiently. In this setting, information technologies may improve market performance and increase welfare. Between 1997 and 2001, mobile phone service was introduced throughout Kerala, a state in India with a large fishing industry. Using microlevel survey data, we show that the adoption of mobile phones by fishermen and wholesalers was associated with a dramatic reduction in price dispersion, the complete elimination of waste, and near-perfect adherence to the Law of One Price. Both consumer and producer welfare increased. I. INTRODUCTION How do improvements in information impact market performance and welfare? Economists have long emphasized that information is critical for the efficient functioning of markets. For example, two of the most well-known results in economics, the First Fundamental Theorem of Welfare Economics (i.e., competitive equilibria are Pareto efficient) and the “Law of One Price” (LOP) (i.e., the price of a good should not differ between any two markets by more than the transport cost between them) rely heavily on the assumption that agents have the necessary price information to engage in optimal trade or arbitrage. These results * I thank two anonymous referees, Reuben Abraham, Christopher Avery, Satish Babu, Suzanne Cooper, Peter Cherian, Thomas DeLeire, Edward Glaeser, Sebastian James, C. M. Jolly, X. Joseph, Nolan Miller, C. K. Muhammad, Prakash Nair, Mai Nguyen, M. Philip, P. Philip, Lant Pritchett, V. Rajan, T. K. Sidhique, Joseph Thomas, and Richard Zeckhauser for valuable comments. © 2007 by the President and Fellows of Harvard College and the Massachusetts Institute of Technology. The Quarterly Journal of Economics, August 2007 879 Downloaded from https://academic.oup.com/qje/article/122/3/879/1879540 by Radboud Universiteit Nijmegen user on 18 September 2023 QUARTERLY JOURNAL OF ECONOMICS 880 QUARTERLY JOURNAL OF ECONOMICS 1. Perhaps ironically, Microsoft’s Bill Gates has been among the most prominent of such critics [Gates 2000]. Downloaded from https://academic.oup.com/qje/article/122/3/879/1879540 by Radboud Universiteit Nijmegen user on 18 September 2023 reflect some of the most fundamental functioning of and advantages to a market economy; when goods are more highly valued on the margin in one market than another, a price differential arises and induces profit-seeking suppliers or traders to reallocate goods towards that market, reducing the price differential and increasing total welfare in the process. In reality, however, the information available to agents is often costly or incomplete, as emphasized by Stigler [1961]. In such cases, there is no reason to expect excess price differences to be dissipated or the allocation of goods across markets to be efficient. Yet despite the fact that information is both central to economic theory yet so limited in reality, there are few empirical studies assessing the effects of improvements in information. Thus, questions such as how much market performance can be enhanced by improving access to information, how much society gains from such improvements, and how those gains are shared between producers and consumers remain largely unanswered. In this paper, we examine these questions by exploiting the introduction of mobile phones in the Indian state of Kerala as a natural experiment of improved market information. Beyond its prominent place in economic theory, the effect of information on market performance and welfare is also relevant to the debate over the potential value of information and communication technologies (ICTs) for economic development. Many critics argue that investments in ICTs should not be a priority for low-income countries, given more basic needs in areas such as nutrition, health, and education.1 However, this argument overlooks the fact that the functioning of output markets plays a central role in determining the incomes of the significant fraction of households engaged in agriculture, forestry, or fisheries production in low-income countries; for most of the world’s poorest, living standards are determined largely by how much they get for their output. Additionally, the functioning of these markets determines the prices and availability of food, fuel, and other important consumer goods. However, in most developing countries, markets are dispersed, and communications infrastructure is poor. Producers and traders often have only limited information, perhaps knowing only the price in a handful of nearby villages or the nearest town, so the potential for inefficiency in the allocation of goods across markets is great. By improving access to INFORMATION, MARKET PERFORMANCE, AND WELFARE 881 2. To cite just a few examples from popular media sources, such behavior has been observed in Thailand and the Philippines [Arnold 2001]; Kenya [England 2004]; Congo and South Africa [LaFraniere 2005]; Bangladesh and China [Alam 2005]; and even the case of fishermen in Kerala examined here [Rai 2001]. 3. During the period of study, most beach markets were open only from 5:00 to 8:00 A.M. Downloaded from https://academic.oup.com/qje/article/122/3/879/1879540 by Radboud Universiteit Nijmegen user on 18 September 2023 information, ICTs may help poorly functioning markets work better and thereby increase incomes and/or lower consumer prices. In fact, it has become increasingly common to find farmers, fishermen, and other producers throughout the developing world using mobile phones, text messaging, pagers, and the internet for marketing output.2 However, while there is some macrolevel evidence that ICTs promote economic growth [Roller and Waverman 2001], the microlevel evidence has been purely anecdotal. Thus, the case of mobile phones in Kerala will also allow us to examine whether ICTs can play a role in promoting welfare in developing countries; while much has been written about how the uneven spread of ICTs has created a “digital divide” between rich and poor countries, considerably less is known about the benefits such technologies can provide the latter. Fishing is an important industry in Kerala. For consumers, fish is a dietary staple [Kurien 2000]; over 70 percent of adults eat fish at least once a day, making it the largest source of many important nutrients, such as protein. Further, over one million people are directly employed in the fisheries sector [Government of Kerala 2005]. However, a significant limitation to fish marketing is that while at sea, fishermen are unable to observe prices at any of the numerous markets spread out along the coast. Further, fishermen can typically visit only one market per day because of high transportation costs and the limited duration of the market.3 As a result, fishermen sell their catch almost exclusively in their local market. In addition, there is almost no storage (due to costs), and little arbitrage on land due to poor road quality and high transportation costs; ultimately, the quantity supplied to a particular market is determined almost entirely by the amount of fish caught near that market. Table I provides suggestive evidence of the resulting inefficiency. The table presents data for fifteen beach markets in northern Kerala, listed in north–south geographical alignment, on average fifteen kilometers apart. The first column provides the prevailing “beach price” (price paid to fishermen by wholesalers or retailers) for a kilogram of sardines on Tuesday, January 14, 1997, at 7:45 A.M., just before 882 AND EXCESS SUPPLY Kasaragod District Hosabethe Aarikkadi Kasaba Kanhangad Thaikadappuram Kannur District Puthiangadi Neerkkadavu Ayikkara Thalassery New Mahe Kozhikode District Chombala Badagara Quilandi Puthiyangadi Chaliyam AND TABLE I DEMAND IN FIFTEEN SARDINE BEACH MARKETS Price (Rs/kg) Excess buyers Excess sellers 6.2 4.0 0.0 7.2 9.7 0 0 0 0 11 0 0 4 0 0 8.7 6.9 8.4 4.3 6.2 2 0 1 0 0 0 0 0 0 0 9.9 0.0 9.8 0.0 6.4 15 0 12 0 0 0 11 0 6 0 Data from the Kerala Fisherman Survey conducted by the author. The first column contains the average 7:45– 8:00 A.M. price of sardines in each market on Tuesday, January 14, 1997, in rupees per kilogram. The markets are listed in north–south geographic alignment; starting from Hosabethe, the distance in kilometers between each market and the next is: 12, 14, 15, 15, 24, 15, 6, 14, 9, 8, 7, 15, 10, and 16. “Excess buyers” represents the number of buyers who leave the market without having purchased enough fish, and “excess sellers” is the number of fishermen who leave the market without selling their fish. the effective market closing. There is a great deal of price variation, with some markets having an effective price of zero (fishermen arrive to find all buyers have departed) while others range from 4.0 to as much as 9.9 rupees per kilogram (Rs/kg; $1 US ⬇ 36 Rs). Note in particular that Badagara has a price of zero while Chombala and Quilandi, both within fifteen kilometers, have prices of 9.9 and 9.8 Rs/kg, respectively. Since an average boat on this day was carrying 381 kg of fish and the fuel cost of traveling fifteen kilometers was about 205 Rs, a boat arriving at Badagara was forgoing as much as 3,400 Rs in profit. Columns (2) and (3) show this from another perspective, with data on the number of “excess buyers” (wholesalers/retailers who report having bought no fish because of high price or inadequate supply) and “excess sellers” (fishermen who arrive at a market and find no buyers and therefore dump their catch in the sea). The inefficiency is clear; while at Badagara there are eleven fishermen dumping their Downloaded from https://academic.oup.com/qje/article/122/3/879/1879540 by Radboud Universiteit Nijmegen user on 18 September 2023 PRICES QUARTERLY JOURNAL OF ECONOMICS INFORMATION, MARKET PERFORMANCE, AND WELFARE 883 Downloaded from https://academic.oup.com/qje/article/122/3/879/1879540 by Radboud Universiteit Nijmegen user on 18 September 2023 catch unsold, there are twenty-seven buyers within fifteen kilometers who are about to leave without purchasing any fish. Provided there are no other barriers to arbitrage, if fishermen had price information for all locations, the market should achieve an outcome where price dispersion is reduced, fish are allocated across markets more efficiently, waste is reduced or eliminated, and total welfare is increased (though how those gains will be shared between consumers and producers is ambiguous). Beginning in 1997, mobile phone service was gradually introduced throughout Kerala. Since most of the largest cities are coastal, many base towers were placed close enough to the shore that service was available twenty to twenty-five kilometers out to sea, the distance within which most fishing is done. By 2001, over 60 percent of fishing boats and most wholesale and retail traders were using mobile phones to coordinate sales. Thus, the case of Kerala provides an ideal setting for exploring the effects of information on market performance and welfare. Using microlevel survey data spanning this period, we find that price dispersion was dramatically reduced with the introduction of mobile phones; the mean coefficient of variation of price across markets (the standard deviation divided by the mean) declined from 60 –70 to 15 percent or less. In addition, there were also almost no violations of the Law of One Price once mobile phones were in place, compared to 50 – 60 percent of market pairs before. Further, waste, averaging 5– 8 percent of daily catch before mobile phones, was completely eliminated. Overall, the fisheries sector was transformed from a collection of essentially autarkic fishing markets to a state of nearly perfect spatial arbitrage. In addition, fishermen’s profits increased on average by 8 percent while the consumer price declined by 4 percent and consumer surplus in sardine consumption increased by 6 percent (though relative to average household expenditure, the latter effect is extremely small). The remainder of this paper proceeds as follows: Section II discusses a simple model that generates predictions for the effects of mobile phones on market performance. Section III discusses the data and empirical strategy. Section IV examines the effects of mobile phones on price dispersion, waste, and adherence to the LOP. Section V provides estimates of the welfare effects, and Section VI concludes. 884 QUARTERLY JOURNAL OF ECONOMICS AND WELFARE II.A. The Model Assume there are two towns along a coastline, each with an equal measure continuum of fishermen who leave in the morning and fish in the “catchment zone” near their town. Each fisherman’s catch is a random variable with an identical distribution across individuals, but there is positive correlation for fishermen within a catchment zone. Specifically, we assume that a fisherman’s catch depends on the density of fish, d, present in their catchment zone on a particular day, where each zone can be in either a high (H) or low (L) density state. The catch for fisherman i thus follows the distribution f( x i 兩d), where xi takes on values from zero to xmax. f( x i 兩d) satisfies the Monotone Likelihood Ratio Property, so that f( x i 兩H)/f( x i 兩L) is increasing in x, i.e., high catches are more likely in the high than low density state. For ease of exposition, we further assume that each zone has an equal probability of H and L each day, equal to one-half, and that these realizations are independent across zones. At the end of the day, there is a competitive fish market in each town, with many small buyers and sellers.4 We assume the aggregate demand curve P(Q) is identical for the two towns, where Q is the quantity supplied to the market, with P⬘(Q) ⬍ 0. The default option for each fisherman is to sell their catch in their local market. However, they could pay a transportation cost ␶ and sell in the other market (but they can only visit one market per day).5 On observing their own catch, each fisherman updates their assessment of the state of their catchment zone; a higher catch suggests the zone is more likely to be in a high density (low price) state and raises the possibility they could benefit from 4. In most studies of consumer search (see Stiglitz [1989] for a review), there are many sellers but only one at any particular location; consumers incur a cost for each price quote they wish to receive (i.e., each seller they visit). Each seller then knows that a consumer arriving at their store will only search for an additional quote if the expected price difference exceeds search costs, in effect creating market power for sellers. In the present case, search (by fishermen) is among competitive markets, each with many buyers and sellers, emphasizing the pure arbitrage value of information. In this way, our analysis differs from much of the theoretical and empirical literature on search. 5. In practice, it is rarely possible to visit more than one market per day because markets are open for only a few hours (and travel for boats loaded with fish is time consuming and expensive). Because overnight storage by fishermen, traders, or consumers is prohibitively expensive, fish must be consumed the day they are caught. Markets close early because fish sold later would not have enough time to travel the supply chain from beach to consumer. Downloaded from https://academic.oup.com/qje/article/122/3/879/1879540 by Radboud Universiteit Nijmegen user on 18 September 2023 II. INFORMATION, PRICE DISPERSION, INFORMATION, MARKET PERFORMANCE, AND WELFARE 885 THEOREM 1. When each fisherman observes only their own catch, there exists a Bayes–Nash equilibrium where 1. there is a threshold x(␶), with x⬘(␶) ⱖ 0, such that all fishermen with catch greater than this value sell in the nonlocal market and all those below sell in the local market, 2. price dispersion between the markets exceeds (per unit) transportation costs when the markets are in opposite states (the prices are equal when they are in the same state), and 3. there is a threshold, ␶*, above which all fishermen always sell in their local market. The proof is in the Appendix. Theorem 1 is intuitive. When fishermen observe only their own catch, those with the highest catches switch to the nonlocal market both because they assess a higher likelihood of being in an H state and because their high catch yields a greater expected gain in profits for a given expected price difference. Fishermen with lower catches either believe it is more likely they are in a low-density (high price) state, or recognize that even if they are in a high-density state, fishermen with greater catches will switch markets and reduce the equilibrium expected price difference to where it is no longer profitable for them to switch, given their small catch. For the marginal fisherman who switches markets, the expected equilibrium price difference equals the (per unit) transportation cost, ␶ /x. Since fishermen do not know the state of either zone with certainty, arbitrage is less than the full-information optimum, and the equilibrium price differential exceeds transportation costs. As transportation costs increase (or it becomes more difficult to predict a zone’s state from one’s own catch) there will be less switching and greater price dispersion in equilibrium. In the extreme, there may be no switching because even for the fisherman with the highest catch, the expected gain is less than the transportation costs. 6. We assume x⬘[P(Q L ) ⫺ P(Q H )] ⬎ ␶, 0 ⬍ x⬘ ⬍ x max, i.e., in the default state there are profitable arbitrage opportunities. 7. We assume fishermen are risk neutral, since in practice this is a high frequency (daily) repeated game and smoothing income or consumption over such short intervals is relatively easy. Downloaded from https://academic.oup.com/qje/article/122/3/879/1879540 by Radboud Universiteit Nijmegen user on 18 September 2023 selling in the other market.6 The fishermen’s problem is to maximize profits by choosing where to sell their fish.7 886 QUARTERLY JOURNAL OF ECONOMICS THEOREM 2. There exists a Bayes–Nash equilibrium where 1. there is a threshold x(␺) such that all fishermen with catch greater than this value purchase search (and switch markets when the zones are in opposite states) and 2. a reduction in ⌿ weakly reduces price dispersion between the markets. In the Appendix, this theorem is proven for the case where ␶ ⱖ ␶*, since in practice there was no arbitrage before mobile phones were available (as shown later).8 As before, fishermen with the greatest catches are more likely to believe they are in a high density zone and thus may gain by switching. They are therefore more likely to purchase search.9 And although it entails an additional cost for potential arbitrageurs, introducing the search technology makes it possible for arbitrage to occur despite the fact that it would not otherwise because when search costs are sufficiently small, the threshold catch for purchasing search is lower than the threshold for engaging in “blind” switching. Search allows fishermen to learn the state of both zones with certainty and thereby avoid unprofitable switching (transportation costs incurred when both zones turn out to be in the same state, and transportation costs plus lower revenue when the blind arbitrageur guesses incorrectly and switches from an L to an H market). Search is purchased up to the point where the expected gain from arbitrage (net of transportation costs) equals the cost of search. And thus as the cost of search declines more fisherman 8. When ␶ ⬍ ␶*, i.e., there would be some switching even without the search technology, Theorem 2 continues to hold but only when search costs are below a threshold, ⌿*(␶). If search costs are high relative to transportation costs, two cases can arise: (1) no fishermen purchase search, but those with the highest catches switch anyway (as in Theorem 1 or 2) fishermen with the highest catches switch without purchasing search and fishermen with catches in an intermediate range below this buy search and switch only when the zones are in opposite states. 9. In a repeated game where the search technology is a durable good like a mobile phone, fishermen purchase search when the discounted stream of expected gains from switching markets over the life of the technology exceeds the cost. Variation in the stream of expected gains can arise through heterogeneity in average catch (such as due to boat size or fishing gear) or arbitrage costs [due to the type of engine or boat (construction material or hull shape, for example) being used]. The basic conclusions of the model continue to hold. Downloaded from https://academic.oup.com/qje/article/122/3/879/1879540 by Radboud Universiteit Nijmegen user on 18 September 2023 We now introduce a search technology, where for a cost, ⌿, fishermen can learn the catch in both zones. The fisherman’s problem now is whether to purchase the technology and where to sell their catch. INFORMATION, MARKET PERFORMANCE, AND WELFARE 887 II.B. Welfare Effects Beyond reducing price dispersion, increased arbitrage due to search will also result in a net welfare gain. Figure I shows the basic analytics of the welfare change under the assumption of perfectly inelastic supply (which we show approximates the Kerala case). The figure shows consumer and producer surplus when one zone is in an H state and the other is in an L state, with and without arbitrage. In the L zone, consumers gain A⫹B while producers lose A and gain C when X fish caught in the H zone are added to the market. These changes can be viewed as a net gain of B⫹C and a transfer of A from producers to consumers (because 10. Fish retailers in Kerala report that saturation points affect their decisionmaking; there is a limit to how much fish they are willing to buy because they know that only a certain number of customers are likely to come to their market on a given day, and there is a limit as to how much any customer will buy, even at arbitrarily low prices. Downloaded from https://academic.oup.com/qje/article/122/3/879/1879540 by Radboud Universiteit Nijmegen user on 18 September 2023 purchase it and engage in arbitrage when the markets are in opposite states, thereby reducing price dispersion. The model is easily extended to include waste (as observed in Table I). Waste arises because of saturation points in demand; while consumers purchase more fish on days when the price is low, there is a limit to how much they will purchase on any given day, especially since fish cannot be stored.10 Thus, if the maximum quantity demanded at each town is less than the total catch when a zone is in the H state, there will be waste in a market whenever the corresponding catchment zone is in state H and there is no arbitrage. Lower search costs reduce waste by facilitating arbitrage when the zones are in opposite states. It should be noted that while we have modeled it here as a problem of costly information, excess price dispersion or a lack of arbitrage may arise for other reasons, such as constraints on trade. For example, fishermen may collude to punish buyers who purchase from nonlocal fishermen, buyers may collude to punish fishermen who sell outside their local market, or there may be interlinked transactions, such as when a fisherman receives credit from a buyer and in exchange must always sell to them (as seen in Giné and Klonner [2002] and Platteua [1984]). In these cases, reduced search costs would not lead to more arbitrage unless it affected the ability to sustain such constraints. However, in the region of study, fishermen reported no such constraints on fish marketing during this period. Downloaded from https://academic.oup.com/qje/article/122/3/879/1879540 by Radboud Universiteit Nijmegen user on 18 September 2023 FIGURE I Changes in Welfare Associated with Arbitrage QUARTERLY JOURNAL OF ECONOMICS 888 INFORMATION, MARKET PERFORMANCE, AND WELFARE 889 11. For example, with a linear demand curve, P ⫽ a ⫺ bQ, the percent increase in welfare from arbitrage is given by Xb(Q H ⫺ Q L ⫺ X)/(a(Q H ⫹ Q L ) ⫺ .5b(Q 2L ⫹ Q 2H )). If a ⫽ 10, b ⫽ .1, QL ⫽ 1, and QH ⫽ 9, the gain ranges from 12 percent when one fish is arbitraged to 27 percent when four fish are arbitraged (though we must subtract transportation costs). 12. Consider the case with zero search costs and perfect information; in equilibrium, the price difference between the markets is ␶ /x̃, where x̃ is the catch of the marginal fisherman who switches. Then the area of rectangle C above P(QH – X) (i.e., the top point of the quasi-trapezoid E⫹F) is (X/x̃)␶. Note that (X/x̃) is greater than the total number of fishermen who switch markets since all fishermen who switch will have catch at least as great as the marginal switcher. Thus, this area alone (and thus C⫺(E⫹F) alone) is greater than total transportation costs incurred (␶ times the number of fishermen who switch). A similar argument holds when search costs are added. Downloaded from https://academic.oup.com/qje/article/122/3/879/1879540 by Radboud Universiteit Nijmegen user on 18 September 2023 the QL fish caught in that zone are now sold at a lower price than if there were no arbitrage). In the H zone, consumers lose D⫹E, while producers gain D and lose F, representing a net loss of E⫹F and a transfer of D from consumers to producers (since the QH –X nonarbitraged fish now sell for a higher price). The net change in total welfare is the difference in the two quasi-trapezoids, (B⫹C)⫺(E⫹F) or 兰QQ LL⫹x P(Q)dQ ⫺ 兰QQHH⫺x P(Q)dQ. Provided the demand curve has a negative slope everywhere between QL and QH, the net change is always positive because the two quasitrapezoids have the same base, while P(QL⫹X) is always greater than P(QH –X), by at least the transportation cost of the marginal switcher. The difference reflects the increase in welfare from moving X fish from where on the margin they were valued less (the high catch, low price market) to where they were valued more (the low catch, high price market). These gains can be substantial, especially when the no-arbitrage price difference is large.11 Further, the net gain will exceed total search and transportation costs.12 Finally, while we used consumer surplus to measure welfare, Hicksian compensated demand curves can be substituted for the Marshallian curves in Figure I; since the former are always downward sloping, the same prediction of a net gain in welfare holds for other measures of welfare. The size and direction of the net transfer from consumers to producers, D⫺A, as well as the net gain for each group, (C⫺A)⫹(D⫺F) for producers and (A⫹B)⫺(D⫹E) for consumers, will depend on the shape of the demand curve (in particular, the price elasticities of demand at the initial quantities) and the amount of arbitrage. Thus, how the net welfare gain is shared between the two groups, and whether, in fact, one group gains while the other loses 890 QUARTERLY JOURNAL OF ECONOMICS III. DATA AND EMPIRICAL STRATEGY The data for this paper come from surveys in Kerala’s three northern districts, Kasaragod, Kannur, and Kozhikode. We conducted a weekly survey of 300 sardine fishing units15 throughout the region of study on Tuesdays of every week from September 3, 1996, to May 29, 2001. We first chose fifteen of approximately thirty-five beach markets (which also serve as the ports or “landings” for the fishing units) throughout the districts, selected so that there was one market, on average, every fifteen kilometers. Within each landing, we made a census of all sardine fishing units and then randomly chose ten large (twenty-eight feet or above) and ten small units. Interviewed in the afternoon regarding that morning’s 13. Synthesizing earlier work by Waugh [1944] and Oi [1961], Massell [1969] argued that consumers lose and producers gain when price is stabilized at its arithmetic mean if supply shocks drive price variability, and vice versa for demand shocks. However, this result relies on the assumption of linear supply and demand curves. 14. Though if all consumers engaged in such substitution, there would be no price variation even without arbitrage. If everyone tried to consume on low price days, the increased demand would drive up the price, and vice versa on high price days. Demand shocks would perfectly offset supply shocks; in equilibrium the price today must equal the expected price tomorrow; though heterogeneity or limited substitution could generate equilibrium price variation. 15. A unit may contain more than one boat, as with ring seine units that use several boats and nets to encircle fish. Downloaded from https://academic.oup.com/qje/article/122/3/879/1879540 by Radboud Universiteit Nijmegen user on 18 September 2023 in response to increased arbitrage, is a priori ambiguous.13 In general, the gains for consumers will be smaller (or even negative) when demand is less price elastic. However, it is possible for both groups to gain, especially if arbitrage also reduces waste. In analyzing the welfare effects of commodity price stabilization via storage, Newbery and Stiglitz [1981] and Wright [2001] emphasize the direct benefits of reduced price risk, including possible supply responses. However, later we will argue that these issues are not relevant for the present case. Perhaps the most significant aspect of welfare omitted so far is the consequence for consumers of reduced price variability. Consumers may prefer prices that vary day to day because they can engage in intertemporal substitution, waiting to consume only on days when prices are low.14 However, consumers also gain from less variable prices because they can have smoother consumption and because they do not need to incur costs to visit markets to find out if prices are low since the price is stable and predictable. The net effect for consumers of more stable prices is therefore ambiguous. INFORMATION, MARKET PERFORMANCE, AND WELFARE 891 16. The flat part of the graph in region I was caused by long-term contracts among the first adopters. (Such contracts were not required during other periods.) 17. By contrast, adoption among the general population was less than 5 percent during this period. Downloaded from https://academic.oup.com/qje/article/122/3/879/1879540 by Radboud Universiteit Nijmegen user on 18 September 2023 market, each fishing unit was asked for the amount of fish caught, market of sale, quantity sold, sale price, time of sale, costs, and whether they used a mobile phone. Fishermen were also asked for wind and sea conditions (calm, mild, severe) and approximate fishing location (indicated on a map). Mobile phone service first became available in Kerala on January 1, 1997. However, due to high investment costs and uncertainty about demand, service was introduced gradually throughout the state, rather than all at once. For the three districts we consider, service became available first in Kozhikode (Kozhikode city, effective January 29, 1997), followed by Kannur (Kannur city on July 6, 1998, and Thalassery on July 31, 1998) and then Kasaragod (Kasaragod city and Kanhangad on May 21, 2000). Figure II shows the timing of mobile phone service availability, where the area of study is divided into three regions based on service provision; each region also contains five markets from our survey. While mobile phone service was not explicitly planned to accommodate fishermen, the cities listed above are coastal, so with a service radius of about twenty-five kilometers for each mobile phone tower, service became available for much of the range in which sardine fishing occurs (ten to thirty kilometers from the shore). Mobile phones spread widely among fishermen and buyers. Figure III provides data on adoption by fishermen in each of the three regions. The vertical lines represent the dates at which service became available in each region (weeks twenty-three, ninety-eight, and 198 in our sample). In each case, adoption increased rapidly before reaching a plateau after a few months.16 The ultimate penetration level is high, ranging from 60 –75 percent across the regions.17 The phones were widely used for fish marketing; while almost all sales before mobile phones were conducted via beach auctions, fishermen with phones, often carrying lists with the numbers of dozens or even hundreds of potential buyers, would typically call several buyers in different markets before deciding where to sell their catch, in essence conducting a virtual auction, and committing to a price while at 892 QUARTERLY JOURNAL OF ECONOMICS sea.18 In general, phones were bought by the largest boats first, since they faced the largest potential gains to arbitrage and were also more likely to be able to afford the phones, which were initially expensive (as much as $100 US). Our empirical analysis compares how changes in the outcomes of interest (price dispersion, waste, and welfare) correspond to the staggered introduction of mobile phones across the regions. We can break the sample into four time periods: period 0 18. Both fishermen and buyers report that it is extremely rare for a negotiated deal at sea to be broken later, largely due to the need to establish a credible reputation. Downloaded from https://academic.oup.com/qje/article/122/3/879/1879540 by Radboud Universiteit Nijmegen user on 18 September 2023 FIGURE II Spread of Mobile Phone Coverage in Kasaragod, Kannur, and Kozhikode Districts Downloaded from https://academic.oup.com/qje/article/122/3/879/1879540 by Radboud Universiteit Nijmegen user on 18 September 2023 FIGURE III Mobile Phone Adoption by Fishermen Data from the Kerala Fisherman Survey conducted by the author. 893 INFORMATION, MARKET PERFORMANCE, AND WELFARE 894 QUARTERLY JOURNAL OF ECONOMICS ៮ I,1 ⫺ Y ៮ I,0兲 ⫺ 共Y ៮ II,1 ⫺ Y ៮ II,0兲 共Y (1) and ៮ I,1 ⫺ Y ៮ I,0兲 ⫺ 共Y ៮ III,1 ⫺ Y ៮ III,0兲. 共Y (2) Similarly, for the addition of mobile phone service to region II, we can compare ៮ II,2 ⫺ Y ៮ II,1兲 ⫺ 共Y ៮ I,2 ⫺ Y ៮ I,1兲 共Y (3) and ៮ II,2 ⫺ Y ៮ II,1兲 ⫺ 共Y ៮ III,2 ⫺ Y ៮ III,1兲. 共Y (4) Finally, for region III, we can compare, ៮ III,3 ⫺ Y ៮ III,2兲 ⫺ 共Y ៮ I,3 ⫺ Y ៮ I,2兲 共Y (5) and ៮ III,3 ⫺ Y ៮ III,2兲 ⫺ 共Y ៮ II,3 ⫺ Y ៮ II,2兲. 共Y (6) To control for other factors that may influence market outcomes, we estimate, 冘 ␤ Region ⫹ 冘 ␤ Period II Y r,t ⫽ ␣ ⫹ 3 r r r⫽I p p p⫽1 冘 冘␤ II ⫹ 3 r_ p Regionr ⴱ Periodp ⫹ ␥Zr,t ⫹ εr,t , r⫽I p⫽1 where Z is a set of control variables that may affect the extent of arbitrage, including wind and sea conditions and the price of fuel. This strategy eliminates fixed differences across the regions and Downloaded from https://academic.oup.com/qje/article/122/3/879/1879540 by Radboud Universiteit Nijmegen user on 18 September 2023 (weeks one to twenty-two), when no region had mobile phones; period 1 (weeks twenty-three to ninety-seven), when only region I had mobile phones; period 2 (weeks ninety-eight to 197), when regions I and II had mobile phones; and period 3 (weeks 198 – ៮ r,p 249), when all three regions had mobile phones. Letting Y represent the average value of the outcome of interest in region r ៮ in region I between in period p, we can examine the change in Y periods 0 and 1, i.e., before versus after the introduction of mobile phones in the region, relative to the change over the same periods for regions II and III, i.e., 895 INFORMATION, MARKET PERFORMANCE, AND WELFARE Percent of fishermen who fish in local catchment zone Region I Region II Region III Percent of fishermen who sell in local catchment zone Region I Region II Region III Number of fishing units Region I Region II Region III TABLE II CHANGES AND IN FISH MARKETING BEHAVIOR Period 1 Period 2 Period 3 Period 0 (pre-phone) (region I adds phones) (region II adds phones) (region III adds phones) 0.98 (0.003) 0.99 (0.002) 0.98 (0.002) 0.99 (0.001) 0.98 (0.001) 0.98 (0.001) 0.98 (0.001) 0.99 (0.01) 0.98 (0.001) 0.98 (0.002) 0.99 (0.001) 0.99 (0.001) 1.00 (0.00) 1.00 (0.00) 1.00 (0.00) 0.66 (0.005) 1.00 (0.00) 1.00 (0.00) 0.63 (0.005) 0.64 (0.004) 1.00 (0.00) 0.62 (0.006) 0.58 (0.006) 0.70 (0.005) 83 69 53 85 74 55 85 75 54 89 75 56 Data from the Kerala Fisherman Survey conducted by the author, using fishermen’s self-report of fishing location and market of sale. The catchment zone for each town is the area of sea defined by lines extending out to sea at the midpoint between a town and its nearest neighbors to the north and south. Regions and periods are as defined in the text. Standard errors in parentheses. common trends or changes over time in factors that affect all three regions equally, such as changes in state fisheries policy or boat, engine, or storage technologies. The identifying assumption is that in the absence of the introduction of mobile phone service, there would have been no differential changes in the outcomes across these regions. We discuss potential challenges to this assumption in detail in Section IV. Tables II and III demonstrate the identification strategy. Table II shows that prior to the introduction of service (period 0), in all three regions fishermen both fished and sold their catch almost exclusively within their local catchment zone.19 However, once mobile phones are introduced in region I, while all fishermen 19. Catchment zones are defined as the area of sea closest to each fishing village (i.e., a line extending out to sea at the midpoint between a village and the nearest town to the north or south). Downloaded from https://academic.oup.com/qje/article/122/3/879/1879540 by Radboud Universiteit Nijmegen user on 18 September 2023 MOBILE PHONE INTRODUCTION 896 QUARTERLY JOURNAL OF ECONOMICS Max–min spread (Rs/kg) Region I Region II Region III Coefficient of variation (percent) Region I Region II Region III Waste (percent) Region I Region II Region III AND TABLE III WASTE IN KERALA SARDINE MARKETS Period 0 (pre-phone) Period 1 (region I adds phones) Period 2 (region II adds phones) Period 3 (region III adds phones) 7.60 (0.50) 8.19 (0.44) 8.24 (0.47) 1.86 (0.22) 7.30 (0.29) 7.27 (0.27) 1.32 (0.10) 1.79 (0.19) 7.60 (0.25) 1.22 (0.44) 1.57 (0.16) 2.56 (0.34) .68 (0.07) .62 (0.04) .69 (0.09) .14 (0.01) .55 (0.04) .57 (0.04) .08 (0.01) .12 (0.01) .54 (0.03) .07 (0.01) .08 (0.01) .14 (0.02) 0.08 (0.01) 0.05 (0.01) 0.07 (0.01) 0.00 (0.00) 0.04 (0.01) 0.06 (0.01) 0.00 (0.00) 0.00 (0.00) 0.06 (0.01) 0.00 (0.00) 0.00 (0.00) 0.00 (0.00) Data from the Kerala Fisherman Survey conducted by the author. Period and regions are as defined in the text. The max–min spread is the difference between the highest and lowest 7:30 – 8:00 A.M. average price on a given day among the five markets making up each region, in year 2001 Rs/kg. The coefficient of variation is the standard deviation of the 7:30 – 8:00 A.M. average price on a given day across the five markets within each region divided by the mean 7:30 – 8:00 A.M. average price for each region. Waste refers to the percent of fishermen who report not selling their catch. Standard errors in parentheses. there continue to fish in their own catchment zone, about onethird now sell their catch outside their local market. By contrast, all fishermen in regions II and III continue to sell in their local market. However, similar patterns of change in marketing are seen in these other regions once they receive mobile phone service in periods 2 and 3. Overall, the introduction of mobile phones leads to the onset of a significant amount of arbitrage, with 30 – 40 percent of fishermen on average selling outside their local market on any given day, from an initial situation of near autarky. Using the same strategy, Table III considers changes in Downloaded from https://academic.oup.com/qje/article/122/3/879/1879540 by Radboud Universiteit Nijmegen user on 18 September 2023 PRICE DISPERSION INFORMATION, MARKET PERFORMANCE, AND WELFARE 897 Downloaded from https://academic.oup.com/qje/article/122/3/879/1879540 by Radboud Universiteit Nijmegen user on 18 September 2023 market outcomes. Since prices may vary within a market over the course of the morning, in order to construct a measure of price dispersion we need the prevailing price in each market at a particular point in time. Since in our small sample we do not have a sale at exactly, say, 7:45 A.M. in each market on every day, we instead take the average price for all sales occurring within a time interval; in particular, for most of our analysis we use the average 7:30 – 8:00 A.M. price, which represents the market closing price (though the results are robust to using alternative times). We assign price based on time of sale rather than time of exchange, i.e., prices for sales via beach auction are assigned to the time of auction, whereas sales via mobile phone are assigned to the time when the sale was arranged, not when the fish were delivered. Provided buyers offer the same price at a point in time in an auction as they would if a fisherman called at that time (even though the fish arrive later), price at time of sale is the most appropriate measure for examining price dispersion since it is the price a fisherman with a phone, who could choose among different markets, would be offered at that time. Finally, a price of zero was assigned when a catch was not sold. The top panel of Table III shows the max–min price spread, the difference between the highest and lowest 7:30 – 8:00 A.M. price across the five markets in each of the three regions defined earlier. Prior to the introduction of mobile phones, there were large price differences across markets, with the average max–min spread within a region ranging from 7.6 to 8.2 Rs/kg. However, when phone service was introduced in region I in period 1, the mean spread declined to 1.86 Rs/kg, while declining only slightly in the other two regions. Similarly, when region II received phone service in period 2, the mean spread declined to 1.79 Rs/kg while increasing slightly in region III and declining in region I. Finally, the addition of phones to region III resulted in a similar, though slightly smaller, decline. The second panel shows similar patterns for a more commonly used measure of dispersion, the coefficient of variation (the standard deviation divided by the mean) of the 7:30 – 8:00 A.M. price across the five markets within each region. In the initial period, price dispersion is high, with the standard deviation within a region 62– 69 percent of the mean price in that region. But in each region, once mobile phones are added this measure declines dramatically, to 14 percent or less. In line with the discussion in Section II.B, the fact that price dispersion is so 898 QUARTERLY JOURNAL OF ECONOMICS Downloaded from https://academic.oup.com/qje/article/122/3/879/1879540 by Radboud Universiteit Nijmegen user on 18 September 2023 large before mobile phones suggests the net welfare gain from arbitrage is likely to be substantial. The third panel of the table considers the incidence of waste, measured as the percent of fishing units that do not sell their catch. In the initial period, the incidence of waste is high, with 5 to 8 percent of fishermen unable to sell their catch on an average day. But once mobile phones were introduced to region I, the incidence of waste declined to zero, while declining only slightly in regions II and III. As earlier, similar changes are seen when mobile phones are introduced in regions II and III. The elimination of the significant amounts of waste initially found in the markets suggests not just greater potential welfare gains from arbitrage but also raises the possibility that consumers and producers may both gain on net. To see these effects even more clearly, Figure IV presents price series for the average 7:30 – 8:00 A.M. price for one kilogram of sardines in each of the fifteen markets over the sample period, with markets grouped by the regions defined earlier based on when mobile phones were introduced. The graph shows that before any region had mobile phones, the degree of price dispersion across markets within a region on any given day is high, and there are many cases where the price is zero (i.e., waste). However, within a few weeks of mobile phones being introduced in region I, there is a sharp and striking reduction in price dispersion. Prices across markets in the region rarely differ by more than a few rupees per kilogram on any day, compared to cases of as much as 10 Rs/kg prior to the introduction of mobile phones. In addition, the prices in the various markets rise and fall together and the week-to-week variability within each market is much smaller, since catchment zone-specific quantity shocks are now spread across markets via arbitrage. Further, there are no cases of waste in this region after phones are introduced. By contrast, price behavior in regions II and III appears largely unchanged after phones are introduced in region I. However, after mobile phones are introduced in region II, prices again become much less dispersed across markets on any given day, less variable within markets over time, and waste is ultimately eliminated, whereas region III again remains unchanged. Finally, the same pattern holds once region III adds phones. This figure demonstrates clearly the extent to which the changes in price dispersion and waste were large and sudden, with timing that corresponds Downloaded from https://academic.oup.com/qje/article/122/3/879/1879540 by Radboud Universiteit Nijmegen user on 18 September 2023 FIGURE IV Prices and Mobile Phone Service in Kerala Data from the Kerala Fisherman Survey conducted by the author. The price series represent the average 7:30 – 8:00 A.M. beach price for average sardines. All prices in 2001 Rs. INFORMATION, MARKET PERFORMANCE, AND WELFARE 899 900 QUARTERLY JOURNAL OF ECONOMICS IV. RESULTS: MARKET PERFORMANCE IV.A. Price Dispersion and Waste Before turning to the full regression specification allowing for separate treatment effects for each region, for ease of presentation we first pool the treatments and estimate, Y r,t ⫽ ␣ ⫹ ␤ 1Period1 ⫹ ␤2 Period2 ⫹ ␤3 Period3 ⫹ ␤IRegionI ⫹ ␤IIRegionII ⫹ ␤A Phoner,p ⫹ ␥Zr,t ⫹ εr,t where Phoner, p is a dummy variable equal to one in all periods p in which region r has mobile phone access. Table IV presents the results, which largely mirror those in Table III. The first column shows that the max–min spread across the markets within a region is reduced by 5 Rs/kg on average when mobile phones are added to that region. These changes represent a substantial reduction, since the mean spread prior to the introduction of mobile phones was 7– 8 Rs/kg. Column (2) shows the results for the coefficient of variation are again large, with the addition of mobile phone service associated with a reduction of 38 percentage points in the standard deviation relative to the mean. Finally, column (3) shows that waste is reduced by 4.8 percentage points when mobile phones are introduced. Thus, overall, the regression results confirm that the addition of mobile phones was associated with a large and dramatic reduction in price dispersion and waste. Factors affecting the profitability of arbitrage generally have the expected sign for the various market outcomes, with worse wind/sea conditions20 and higher fuel prices, both of which increase transportation costs, generally associated with greater price dispersion. However, in all cases the effects are small, and we cannot reject the hypothesis that these factors have no effect on the outcomes. The lack of statistical significance may be due to the fact that nearly half the sample consists of period*zone observations where there was no mobile phone coverage and thus no arbitrage, so factors affecting transportation costs would not be expected to influence price dispersion. We therefore estimate regressions where we interact these variables with the indicator for 20. Since wind and sea conditions are highly collinear, we add the two into a single index, varying from zero to six. Downloaded from https://academic.oup.com/qje/article/122/3/879/1879540 by Radboud Universiteit Nijmegen user on 18 September 2023 closely to the three distinct dates when mobile phone service was introduced in each particular region. 901 INFORMATION, MARKET PERFORMANCE, AND WELFARE OF Phone Region I Region II Period 1 Period 2 Period 3 Fuel cost Wind/sea index TABLE IV MOBILE PHONE SERVICE ON PRICE DISPERSION POOLED TREATMENTS WASTE: (2) (3) (5) (6) (1) Max–min spread Coefficient of variation Percent have waste (4) Max–min spread Coefficient of variation Percent have waste ⫺5.0 (0.27) ⫺0.92 (0.26) ⫺0.46 (0.21) ⫺0.89 (0.29) ⫺1.1 (0.32) ⫺1.2 (0.40) 0.02 (0.12) 0.086 (0.051) ⫺.38 (0.03) ⫺.06 (0.03) ⫺.04 (0.02) ⫺.12 (0.04) ⫺.17 (0.04) ⫺.19 (0.04) .01 (0.01) .001 (0.004) ⫺0.048 (0.004) ⫺0.007 (0.005) ⫺0.011 (0.004) ⫺0.017 (0.008) ⫺0.019 (0.008) ⫺0.022 (0.009) 0.001 (0.002) ⫺0.002 (0.002) ⫺5.3 (2.9) ⫺0.94 (0.26) ⫺0.46 (0.21) ⫺0.84 (0.29) ⫺1.0 (0.33) ⫺1.2 (0.40) ⫺0.13 (0.19) ⫺0.03 (0.06) 0.25 (0.14) 0.19 (0.08) ⫺.41 (0.32) ⫺.06 (0.03) ⫺.04 (0.02) ⫺.12 (0.03) ⫺.16 (0.04) ⫺.19 (0.04) ⫺.02 (0.02) ⫺.01 (0.01) .026 (0.014) .021 (0.008) ⫺0.047 (0.06) ⫺0.006 (0.005) ⫺0.011 (0.005) ⫺0.016 (0.008) ⫺0.018 (0.008) ⫺0.021 (0.009) 0.003 (0.005) ⫺0.003 (0.003) ⫺0.003 (0.006) 0.003 (0.005) 747 747 74,700 747 747 74,700 Phone*fuel cost Phone*wind/sea index Number of observations AND Data from the Kerala Fisherman Survey conducted by the author. The variable Phone is assigned a value of one for all dates in which a region has mobile phone service available. All prices are in 2001 Rs. Standard errors, clustered at the village level for columns (3) and (6), in parentheses. whether the region had mobile phones. In columns (4) and (5), both interaction terms are statistically significant for the max–min price spread and the coefficient of variation, with the expected signs; higher fuel costs and worse wind/sea conditions increase price dispersion when there is arbitrage in a region. And, as expected, we cannot reject the hypothesis that these variables have no effect on dispersion when mobile phones are not available in a region. In column (6), the wind/sea and fuel interaction terms still have no effect on waste since there is no waste after mobile phones are introduced. As stated earlier, we can exploit the variation in the timing of introduction of mobile phones across the three regions by estimating regressions with separate treatment effects. Table V presents the estimated effects of mobile phones on the market outcomes for each Downloaded from https://academic.oup.com/qje/article/122/3/879/1879540 by Radboud Universiteit Nijmegen user on 18 September 2023 EFFECTS 902 QUARTERLY JOURNAL OF ECONOMICS TABLE V MOBILE PHONES ON MARKET OUTCOMES: SEPARATE TREATMENTS OF Estimated effects of adding phones to region I (a) Using region II as the control group (Y I,1 ⫺ Y I,0 ) ⫺ (Y II,1 ⫺ Y II,0 ) ⫽ ␤ RI_P1 ⫺ ␤ RII_P1 (b) Using region III as the control group (Y I,1 ⫺ Y I,0 ) ⫺ (Y III,1 ⫺ Y III,0 ) ⫽ ␤ RI_P1 Estimated effects of adding phones to region II (c) Using region I as the control group (Y II,2 ⫺ Y I,1 ) ⫺ (Y I,2 ⫺ Y I,1 ) ⫽ ␤ RII_P2 ⫺ ␤ RII_P1 ⫺ ␤ RI_P2 ⫹ ␤ RI_P1 (d) Using region III as the control group (Y II,2 ⫺ Y II,1 ) ⫺ (Y III,2 ⫺ Y III,1 ) ⫽ ␤ RII_P2 ⫺ ␤ RII_P1 Estimated effects of adding phones to region III (e) Using region I as the control group (Y III,3 ⫺ Y III,2 ) ⫺ (Y I,3 ⫺ Y I,2 ) ⫽ ␤ RI_P2 ⫺ ␤ RI_P3 (f) Using region II as the control group (Y III,3 ⫺ Y III,2 ) ⫺ (Y II,3 ⫺ Y II,2 ) ⫽ ␤ RII_P2 ⫺ ␤ RII_P3 Max–min spread Coefficient of variation Waste ⫺4.8 (0.68) ⫺.46 (0.07) ⫺0.064 (0.005) ⫺4.8 (0.68) ⫺.42 (0.07) ⫺0.060 (0.005) ⫺5.8 (0.43) ⫺.39 (0.05) ⫺0.039 (0.003) ⫺4.9 (0.43) ⫺.36 (0.05) ⫺0.038 (0.003) ⫺4.9 (0.48) ⫺.38 (0.05) ⫺0.055 (0.004) ⫺4.7 (0.48) ⫺.35 (0.05) ⫺0.054 (0.004) Data from the Kerala Fisherman Survey conducted by the author. The table reports the estimated effects of mobile phones on market outcomes separately for each of the three regions using the combinations of coefficients listed in small type, based on the full regression results in columns (1)–(3) in Table X. Standard errors, clustered at the village level for column (3), in parentheses. All prices in 2001 Rs. of the three regions, which, in most cases, are a combination of the coefficients from the full regressions (presented in the Table X). The results are broadly similar to those for the pooled regressions. Estimators (1) and (2), the impact on region I of adding phones between periods 0 and 1, reveal that the max–min spread across markets was reduced by 4.8 Rs/kg when compared to either region II or III. For region II (estimators (3) and (4)), the effects are slightly larger, 4.9 and 5.8 Rs/kg, than for region I; using region I as a control group results in a higher estimate than using region III, but we cannot reject the hypothesis that the two effects are equal. Finally, the effects in region III are similar to those in region I. And as with the other two regions, we cannot reject the hypothesis that the estimated effects are equal for the two comparison groups. Overall, the estimates show some variation in the magnitude of the effects across the regions, ranging from 4.7 to 5.8 Rs/kg for the max–min price spread, 35 to 46 percentage points for the coefficient of variation and 3.8 to 6.4 percentage points for waste. However, for both the max– Downloaded from https://academic.oup.com/qje/article/122/3/879/1879540 by Radboud Universiteit Nijmegen user on 18 September 2023 ESTIMATED EFFECTS INFORMATION, MARKET PERFORMANCE, AND WELFARE 903 IV.B. The Identifying Assumption The identifying assumption for the empirical strategy is that, had it not been for the introduction of mobile phone service, there would have been no differential changes in the market outcomes across these regions over this period. We discuss three potential areas of concern. First, in attributing all the differential changes in market outcomes to the addition of mobile phones, we are assuming that there were no pre-existing differential trends in market outcomes across these regions and that no other factors that could also have influenced these outcomes changed differentially across the regions. Figure IV revealed that the changes in market outcomes were sharp and sudden and correspond closely to the distinct points of introduction of mobile phone service in each region. And the fact that no other large changes in price dispersion are observed except around these three distinct points suggests that differential changes in other factors are unlikely to have caused any significant fraction of the changes in price behavior attributed to mobile phones, since it is very unlikely that these other factors would have differentially changed at the same three specific dates at which each region received mobile phone service, but not at any other time. The sharp and sudden changes also make it unlikely that differential trends across the regions explain much of the differential changes in outcomes (common trends are controlled for). More formally, in regressions for the market outcomes using only the observations before mobile phones were available in any region (period 0) and including a linear time trend, region indicators, and time*region interactions, both the trend and interaction terms are small and not statistically significantly different from zero (results not shown). The same holds for regressions using only regions II and III, with data from periods 0 and 1 (before either region had mobile phones).21 21. However, we cannot rule out differential trends arising only around the same time mobile phones were introduced in each region. Downloaded from https://academic.oup.com/qje/article/122/3/879/1879540 by Radboud Universiteit Nijmegen user on 18 September 2023 min spread and the coefficient of variation, we cannot reject the hypothesis that the effects are equal for all pairwise comparisons of region*control group; for waste, the effects are statistically significantly smaller for region II than for either regions I or III due to the fact that waste was lowest there prior to the introduction of mobile phones. Overall, the results confirm that the introduction of mobile phones was associated with a large and dramatic reduction in price dispersion and waste, with broadly similar effects across the regions. 904 QUARTERLY JOURNAL OF ECONOMICS 22. There is not a concern, however, regarding nonrandom placement since the initial plan of mobile phone providers and the ultimate outcome was to cover the entire coast, not just select areas. 23. Though we have to assume that phone companies did not accurately forecast in advance differential changes in these other factors, there is no evidence that there were any specific periods of large, sharp differential changes in, say, economic growth in these regions over this period, much less predictable changes. Downloaded from https://academic.oup.com/qje/article/122/3/879/1879540 by Radboud Universiteit Nijmegen user on 18 September 2023 A second concern is that the timing of service across the regions was nonrandom.22 According to the mobile phone providers, the order of placement of service was determined by the size of the potential market, i.e., the population of the main city in each region. While the effects of fixed factors that differ across regions like population size are controlled for in the regressions, and while we saw no evidence of differential trends across the regions, we may be concerned that the timing of introduction of service in a particular region was delayed or sped up in response to other factors that could also affect market outcomes. For example, rapid economic growth could have caused firms to speed up the delivery of mobile phone service because of the potential increased demand and separately could also have improved fish market outcomes, such as by increasing overall demand and reducing waste. This would result in a close correspondence between the introduction of mobile phones and changes in market outcomes, without the former having caused the latter. As with the first concern, from Figure IV alone we consider this possibility unlikely, since we do not see any differential trends, or any large changes in the price series at any points in time other than when phones were introduced, and it is unlikely that changes in these other factors happened to occur at these three specific points in time (but no other time). Further, since mobile phone service takes a long time to set up, if the timing of service was responding to changes in factors (like economic growth) that w