Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Implications of barriers to trade for exports of cultural goods and services Holloway, Isaac Robert 2012

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
24-ubc_2012_spring_holloway_isaac.pdf [ 1.61MB ]
Metadata
JSON: 24-1.0072656.json
JSON-LD: 24-1.0072656-ld.json
RDF/XML (Pretty): 24-1.0072656-rdf.xml
RDF/JSON: 24-1.0072656-rdf.json
Turtle: 24-1.0072656-turtle.txt
N-Triples: 24-1.0072656-rdf-ntriples.txt
Original Record: 24-1.0072656-source.json
Full Text
24-1.0072656-fulltext.txt
Citation
24-1.0072656.ris

Full Text

Implications of Barriers to Trade for Exports of Cultural Goods and Services by Isaac Robert Holloway B.Sc., The University of British Columbia, 2005 M.A., The University of British Columbia, 2006  A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF  DOCTOR OF PHILOSOPHY in THE FACULTY OF GRADUATE STUDIES (Business Administration)  THE UNIVERSITY OF BRITISH COLUMBIA (Vancouver) April 2012 c Isaac Robert Holloway, 2012  Abstract This dissertation contains three studies. Chapter 2 studies the effect of product quality on foreign entry using data on U.S. movie exports and direct and revealed measures of movie quality. In the model, fixed costs of entry mean only the more appealing movies will find it profitable to enter each country. Empirically, a one-standard-deviation increase in quality increases the probability of entry by 25-50%. Movies in culturally-laden genres are less likely to enter foreign markets and their probability of entry is more sensitive to quality. I exploit differences in the propensity to import different genre types to estimate a measure of cultural distance between countries. The cultural distance measure enters a gravity equation of merchandise trade significantly. Chapter 3 investigates the international diffusion of a new product. Products traditionally enter foreign markets sequentially. This paper proposes that part of the explanation is that firms want to learn about their products’ appeal before incurring the fixed costs of entry. Each successive release serves to update the firm’s expectations for future performance—and thus their decision to enter more markets. On a sample of U.S. movies, I find that a one-standard-deviation increase in the update, based on the previous round’s box-office “surprise”, is associated with a 25% increase in the probability of entry to a typical potential destination in the current round. Chapter 4 investigates Canada’s interprovincial and international trade in services. While modern technology has allowed for long-distance service provision, regulatory non-tariff barriers may constitute substantial hurdles for further trade liberalization. This chapter describes three exercises contributing to the analysis of Canadian service trade. Using a theoretically-motivated framework, I estimate provincial and national border effects, and track the effect over time that distance has had on international trade.  ii  Table of Contents Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  ii  Table of Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  iii  List of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  v  List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  vi  Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  viii  Dedication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  ix  1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  1  2 Foreign Entry, Quality, and Cultural Distance: Product-Level dence from U.S. Movie Exports . . . . . . . . . . . . . . . . . . . 2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2.1 Hierarchal Sorting . . . . . . . . . . . . . . . . . . . 2.2.2 Destination-Variety-Specific Demand Shocks . . . . . 2.2.3 Measuring Cultural Distance . . . . . . . . . . . . . . 2.3 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4 Empirical Results . . . . . . . . . . . . . . . . . . . . . . . . 2.4.1 Selection versus Random Entry . . . . . . . . . . . . . 2.4.2 Predicting Movie Imports: Quality and Geography . . 2.4.3 Movie Trade and Cultural Affinity . . . . . . . . . . . 2.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . .  iii  Evi. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  4 4 9 10 11 14 16 21 21 24 27 36  3 Learning via Sequential Market Entry: Evidence from International Releases of U.S. Movies . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.4.1 Firm Learning and the Decision to Enter . . . . . . . . . . . 3.4.2 Alternative Models . . . . . . . . . . . . . . . . . . . . . . 3.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  39 39 43 53 59 59 66 68  4 Distance and Border Effects for Canadian Trade in Services 4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . 4.2 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.3 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.4 Interprovincial Trade . . . . . . . . . . . . . . . . . . . . 4.4.1 Alternative Specifications . . . . . . . . . . . . . . 4.5 Distance and International Service Trade . . . . . . . . . . 4.6 National Border Effect . . . . . . . . . . . . . . . . . . . 4.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . .  . . . . . . . . .  69 69 71 72 74 77 86 90 94  5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  96  Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  99  iv  . . . . . . . . .  . . . . . . . . .  . . . . . . . . .  . . . . . . . . .  List of Tables Table 2.1 Table 2.2 Table 2.3 Table 2.4 Table 2.5  Probability of Exporting to a New Market . . . . . . . . Probability of Exporting: Alternative Quality Measures Probability of Exporting: Genre Effects . . . . . . . . . U.S. Bilateral Trade, 2002-2004 . . . . . . . . . . . . U.S. Bilateral Trade, 2002-2004 . . . . . . . . . . . .  . . . . .  . . . . .  . . . . .  . . . . .  . . . . .  27 28 29 35 37  Table 3.1 Table 3.2 Table 3.3 Table 3.4  Costs of Foreign Entry: Gone in Sixty Seconds Correlations of Residuals between Markets . . Probability of Exporting to a New Market . . . Probability of Exporting to a New Market . . .  . . . .  . . . .  . . . .  . . . .  . . . .  45 62 65 67  Table 4.1 Table 4.2 Table 4.3 Table 4.4 Table 4.5 Table 4.6 Table 4.7 Table 4.8 Table 4.9  Provincial Border and Distance Effects . . . . . . . . . . . . Province-Specific Border Effects . . . . . . . . . . . . . . . Nonlinear Distance Effects—OLS Estimates . . . . . . . . . Nonlinear Distance Effects—Poisson PML Estimates . . . . Nonlinear Distance Effects—Gamma PML Estimates . . . . Provincial Internal Trade Effects . . . . . . . . . . . . . . . Canada’s Top Service Trade Partners, 2004 . . . . . . . . . . Distance Effects for Canadian International Trade in Services International and Provincial Stacked Data . . . . . . . . . . .  . . . . . . . . .  . . . . . . . . .  77 78 81 82 82 85 88 91 93  v  . . . .  . . . .  . . . .  . . . .  . . . .  List of Figures Figure 2.1 Figure 2.2 Figure 2.3 Figure 2.4 Figure 2.5 Figure 2.6 Figure 2.7 Figure 2.8 Figure 2.9 Figure 2.10 Figure 2.11 Figure 2.12 Figure 2.13 Figure 2.14 Figure 2.15  Histogram of Netflix Ratings . . . . . . . . . . . . . . . . . . Histogram of IMDb Ratings . . . . . . . . . . . . . . . . . . Histogram of Number of Markets Entered . . . . . . . . . . . Histogram of U.S. Domestic Box-Office Revenue . . . . . . . Histogram of Production Budget . . . . . . . . . . . . . . . . Histogram of Number of Movies Imported . . . . . . . . . . . Minimum Quality vs. Number of Releases by Country in 2004 Mean Quality vs. Number of Releases by Country in 2004 . . Maximum Quality vs. Number of Releases by Country in 2004 Histogram of Hollywood Distance . . . . . . . . . . . . . . . Hollywood Distance by Destination Country . . . . . . . . . . Hollywood Distance versus Geographic Distance . . . . . . . Hollywood Distance versus Language Similarity . . . . . . . Hollywood Distance versus Genetic Distance . . . . . . . . . Hollywood Distance versus Religion Similarity . . . . . . . .  . . . . . . . . . . . . . . .  17 18 18 19 20 21 23 24 25 30 31 31 32 33 33  Figure 3.1 Figure 3.2 Figure 3.3 Figure 3.4 Figure 3.5 Figure 3.6 Figure 3.7 Figure 3.8 Figure 3.9 Figure 3.10 Figure 3.11  Intensive vs. Extensive Margins of Entry . . . . . . . . . . Domestic Revenue vs. Number of Markets Entered . . . . Mean Domestic Revenue by Number of Markets Entered . Domestic Box-Office Revenue vs. Production Budget . . . Median Delay vs. Production Budget . . . . . . . . . . . Histogram of Median Delay by Production Budget Quartile Entry Timing and Performance of “Anything Else” . . . . Entry Timing and Performance of “Grindhouse” . . . . . . Entry Timing and Performance of “50 First Dates” . . . . Histogram of Sequential Index . . . . . . . . . . . . . . . Sequential Index vs. Production Budget . . . . . . . . . .  . . . . . . . . . . .  44 46 46 47 48 49 49 50 50 52 53  vi  . . . . . . . . . . .  . . . . . . . . . . .  Figure 3.12 Average Round of Release by Country . . . . . . . . . . . . . . Figure 3.13 Average Round of Release vs. Correlation with U.S. Box-Office Figure 3.14 Average Round of Release vs. Correlation with U.S. Box-Office: Subsample of Movies Going to All Markets . . . . . . . . . . . Figure 4.1 Figure 4.2 Figure 4.3 Figure 4.4 Figure 4.5 Figure 4.6  Anticipating Distance and Border Effects . . . . . . . . . . . . Province-Specific Border Effects vs. GDP . . . . . . . . . . . . Coefficients on Internal Trade Dummies . . . . . . . . . . . . . Estimated “No BE” Internal Distances vs. Helliwell & Verdier Internal Distances: Linear Log Distance Specification . . . . . . Estimated “No BE” Internal Distances vs. Helliwell & Verdier Internal Distances: Quadratic Log Distance Specification . . . . Distance Effects for Canadian International Trade in Services . .  vii  54 54 55 75 79 83 86 87 92  Acknowledgments I am grateful to my supervisors, Keith Head and John Ries, for their encouragement and guidance during my whole Ph.D. program. This dissertation would not have been possible without their intellectual and financial support. I also appreciate helpful discussions with Hiro Kasahara, who served on my committee, and Chuck Weinberg, who helped me understand the movies business. I thank Barbara Spencer for expertly serving as my advisor and counsellor. Elaine Cho’s dedicated administrative work made my life easier for the past few years. Thank You. I appreciate generous funding from the Sauder School of Business, the University of British Columbia Four-Year-Fellowship program, as well as the Social Sciences and Humanities Research Council. My parents constantly inspire me to be a better person through their examples, and I thank them for giving me the foundation on which this achievement was built. I was supported in every sense of the word by my wife, Yuanyuan, and this degree belongs to her as much as it does to me.  viii  Dedication  To Charlie and Micah  ix  Chapter 1 Introduction This thesis comprises three essays on the effects of barriers to trade on the entry decisions and volume of international and regional transactions. Barriers to trade include physical costs, such as those related to the transportation of goods and/or people, government policy, such as tariffs and quotas, and cultural distance, such as linguistic and aesthetic differences. The first essay contributes to the literature on heterogeneous firms in international trade. In this literature, firms differ in the cost with which they can supply a product, or the quality of the product they supply. In a given environment, better firms are able to earn larger profits. In the presence of fixed entry costs, only the more able firms will find it profitable to enter, which provides implications for entry patterns and the composition of firms within each market. I develop a discrete-choice model of heterogeneous firms, in which consumers choose to purchase one unit if their valuation—a function of country characteristics and firm quality—is greater than the price. The model departs from the common monopolistic-competition framework, as it is geared towards the empirical application of motion picture exports, but maintains many of the empirical implications. Box-office revenues from each foreign market are a function of the size and wealth of the destination and the perceived quality of the movie. This latter component is decomposed into the average worldwide perceived quality and a countrymovie taste shock. Assuming that movie distribution firms decide to enter a foreign market if the forecasted revenues exceed the costs, the model provides an estimating equation for market entry. Estimation of this equation in the empirical section suggests a large role for movie quality, in addition to destination size and wealth, on the 1  pattern of international distribution. The simplest version of the model treats perceived quality as equal across destination markets and provides stark predictions: the highest-quality movie enters all destinations and the average movie quality in a market is decreasing in the number of movies entered. These predictions are tested graphically, along with simulations of random entry, in which country-movie taste shocks determine perceived quality. The data lie between the two models, supporting the modeling of country-movie taste shocks. The essay goes on to seek an interpretation of the taste shocks, and proposes cultural distance as an explanation for some of the variation. A measure of cultural distance is estimated, based on the difference in the propensity for each country to import movies in culturally-laden genres versus those in culturally-neutral ones. The assumption is that countries that are culturally distant from the United States will tend to draw lower affinities for movies in cultural genres than for movies in neutral genres: appreciation of action sequences is less tied to national culture than that of scenes depicting family values, relationships and humour. The cultural distance index—dubbed “Hollywood Distance”—is added to a standard gravity equation for U.S. merchandise trade, and is found to be significantly associated with lower bilateral trade, even after controlling for standard proxies of cultural and institutional linkages. The result is consistent with the notion that cultural distance acts as a significant barrier to international trade. Conversely, given that cultural distance acts a barrier to trade, the result validates Hollywood Distance as an economically meaningful index of cultural distance. The second essay considers how the presence of fixed entry costs impacts not only the long-run pattern of entry—established in the first essay—but firms’ dynamic decision-making of which markets to enter. International releases of U.S. movies are staggered, a fact which has several possible explanations. An implication of staggering is that firms have opportunities to revise their future entry decisions. Once a firm enters a given market, the entry costs are sunk. If it is not certain of its appeal, a firm risks making an irreversible investment yielding possibly little revenue. This essay builds on the model of Chapter Two by assuming firms are uncertain about their quality, but can make forecasts based on the production budget and other movie attributes. Firms use the information from each entered market to update their beliefs about the movie’s quality, and thus its revenues in other potential markets. A surprise  2  negative performance could then cause the movie not to enter a market that was previously expected to make small positive profits. The empirical section provides evidence suggesting that distributors do learn from past performances, by expanding entry on good news and not entering on bad news. As distributors move toward simultaneous release dates to combat international piracy, they lose the ability to learn across markets, a cost that should be weighed against the benefits of decreased piracy rates. The third essay investigates how trade in services between distinct political jurisdictions is subject to various frictions associated with geographic distance and borders. While modern technology has allowed for long-distance service provision, regulatory non-tariff barriers may constitute substantial hurdles for further trade integration. This chpater describes three exercises contributing to the analysis of Canadian service trade. Using a theory-grounded gravity framework, I estimate provincial and national border effects, and track the effect that distance has had on international trade in services. Provincial border effects are found to be substantially greater than their merchandise-trade counterparts, whereas the national border effect for services is found to be lower than past merchandise-trade estimates. The distance effect on service trade is significant and stable over time, suggesting little increased threat to jobs from off-shoring.  3  Chapter 2 Foreign Entry, Quality, and Cultural Distance: Product-Level Evidence from U.S. Movie Exports 2.1  Introduction  For over a decade international trade economists have been documenting and theorizing on the firm-level causes and effects of world trade. Early empirical studies show that exporters are different from purely domestic firms: they are larger and more productive (Bernard and Jenson, 1999 and 2004). Bernard, Eaton, Jenson and Kortum (2003) and Melitz (2003) introduce firm heterogeneity into trade theory, allowing for firms to differ in their productivity. The models provide a framework to understand the stylized facts documented earlier, and provide a new gain from freer trade: a compositional shift towards more productive firms. Subsequent theoretical and empirical studies have extended and tested the implications of these models. As noted by Melitz (2003), the parameter characterizing productivity in the heterogeneous firms models could alternatively be interpreted as quality. Thus, exporters’ products would be of higher quality than those of pure-domestics, and could therefore charge higher prices. Most of the existing empirical work on quality and trade uses unit or average prices to proxy for firm quality. Baldwin and Harrigan (2007) hypothesize and find evidence that average industry prices are increasing in distance, reflecting the compositional sorting implied by trade costs. Iacovone and Javorcik  4  (2008) find that Mexican firms increase their prices two years before they start exporting, and that exporters charge higher prices than non-exporters. This chapter studies the effect of heterogeneous quality on the export pattern of U.S. motion pictures. Movies are differentiated both horizontally and vertically. While reasonable people can disagree on which movie is better (horizontal differentiation), there is clear heterogeneity in popular appeal (vertical differentiation), as evidenced by the wide range of box-office returns. The movie industry makes an interesting laboratory for studying quality and trade patterns. Rather than quality being reflected in unit prices, as is assumed for many manufactured goods, movies tend to be priced identically within a geographic market, regardless of demand. Moreover, few products are judged, graded and rated with such zeal as movies. Hundreds of viewers express their opinions on the quality of movies on web sites every week. I make use of these direct measures of quality to study the foreign entry decisions of individual movies.1 This has the advantage that the data is at the product level. When a movie is released in an international market, we know that it is the exact same product, dubbing or subtitling notwithstanding. It is rare for firms to make a single product, and most firm-level measures of ability—productivity or quality— likely reflect a mix of the product line. Rather than adopt the standard Melitz-type model of heterogeneous firms, I model demand in a simple discrete choice framework. Like the Melitz-model, however, the simplest version of my model predicts a double hierarchy, whereby if a movie of a given quality is distributed in some market, d, it will go to all markets that are at least as attractive as d; and if a country of a given attractiveness imports a movie, m, it will import all movies that are at least as high quality as m. Eaton, Kortum and Kramarz (2008) investigate this type of hierarchy for French exporters and note that it fails to hold in the data. They introduce destination-firm-specific demand shocks to account for the discrepancy. Bernard, Redding and Schott (2009) similarly argue strongly for inclusion of idiosyncratic shocks. I study the extent to which the movies data conforms to the hierarchy and compare the “strict selection” hypothesis to that of random entry. I use a simulation technique to graphically demonstrate the poor fit to the hierarchy. I further investigate the role of quality in export decisions with regression analysis. The model predicts that a movie will be released in a given 1 Crozet,  Head and Mayer (2009) use expert ratings in their study of Champagne exports and  quality.  5  market if the expected revenue obtained would be sufficient to overcome the fixed cost of distribution. Revenue is increasing in quality, cultural affinity and market size. Probability regressions confirm important roles for quality and market attractiveness in the international pattern of movie releases. This essay is also related to the literature on trade in cultural goods, and the measurement of cultural distance. Attempts to measure differences in culture have often relied on analysis of questionnaires.2 I develop a measure of revealed cultural distance by analyzing the pattern of trade in the motion picture industry. A movie is an example of a “cultural good”, the trade of which not only spreads ideas and values, but reflects the current state of shared preferences. Producers face extra costs to serving additional markets, including adapting to market, advertising campaigns, and distribution. Because viewers may discount the quality of a foreign cultural good, the potential size of the importing market is determined not only by destination-specific factors, such as population and wealth, but also by the bilateral cultural affinity between the origin and destination countries. I exploit data on movie genre to back out a cultural component from destination fixed effects. The procedure relies on the fact that the movie quality for some genres—such as comedy and drama—is discounted more heavily by foreign cultures than for other genres, like action or thriller. My cultural distance measure captures the extent to which cultural affinity diverges between the two genre-types. I am not the first to look at trade patterns as a measure of cultural distance. Disdier et al. (2010) use a gravity framework to estimate the “extra” aggregate bilateral trade in sectors that UNESCO deems to be cultural industries. They interpret bilateral excess trade as cultural affinity and find that it has explanatory power in a gravity equation for all merchandise trade. I use title-level data and a heterogeneous quality framework. This allows me to control for product quality in the empirical analysis. For example, the fact that a given country imported blockbusters like Titanic and Harry Potter is less likely to point to cultural affinity than if the country imported more obscure (less appealing) titles. My analysis exploits variation in the extensive margin of trade—whether a title is released in a given country—rather than the intensive volume of bilateral trade. The advantage of this approach is that it is less data intensive, and therefore more easily replicable for other industries, such as literature or music. 2 e.g.  Hofstede (1980), and more recently Maystre et al. (2009)  6  There are existing industry-specific studies of the trade patterns of cultural goods. Hanson and Xiang (2008) develop a version of the Melitz model for the motion picture industry. Their model includes a bilateral cultural discount applied to the popular appeal of the movie in the home country. Rather than estimating this parameter, they use proxies from the existing literature. The focus of the paper is on the nature of fixed exporting costs: whether the data is consistent with a model of a one-time global fixed cost, or whether export costs are export-market specific. Surprisingly, they find evidence that the global fixed-cost model is preferred. This stands is stark contrast to existing evidence for manufacturing trade, which suggests that adjustment in trade occurs primarily at the extensive margin (Helpman, Melitz and Rubinstein, 2008). Indeed, Hanson and Xiang find very little variation at the extensive margin in their sample of 46 importers. They argue that this may be due to a difference in the nature of fixed trade barriers between the sectors. For information services, like motion pictures, product delivery adheres to an international standard (movie theatres), and so the fixed costs of creating a marketing strategy and standard contract are only borne once, regardless of the number of importers. They use aggregate box-office receipts by importer for their measure of motion picture trade. My analysis of the title-level extensive margin of trade is thus complementary. My data also includes 97 importers, representing a larger fraction of the global movie trade, and shows wide variation at the extensive margin. Sin and Abramitzky (2010) use data on international book translations from UNESCO’s Index Translationum, and the “natural experiment” of the fall of communism in Eastern Europe, to estimate the extent to which communism slowed the trade in book translations—and hence information flows. They find a significant increase in translations between Western and former Soviet countries after the fall of communism, particularly in fields such as religion, philosophy and economics; whereas there was little impact on translation rates in the natural sciences. The differences in treatment effect on the two genre sets serve to reinforce the validity of their estimation procedure, as the result is intuitive. In my application, I rely on similar intuition for movie genres—supported by industry experts—to identify cultural distance based on the the extent to which importers’ affinity for the two subsets diverges. Ferreira and Waldfogel (2010) study the international trade in popular music. They find that trade volumes are higher between countries that are geographically proximate and those that share a common language. They also find that large coun-  7  tries’ music, like that of the United States, has not come to dominate world consumption shares. Rather, domestic music has grown its market share since 1960 in the 22 countries in their sample. Disdier, Head and Mayer (2010) look at how exposure to foreign media changes the cultural traits of the importing country. Thus, the emphasis is on the transmission of culture rather than the current level of affinity. Francois and van Ypersele (2002) and Olivier, Thoenig and Verdier (2008) provide theoretical treatments of international trade and cultural transmission. They use variations of an overlapping generations model where parents are altruistic but may not share their offspring’s preferences. Offspring’s preferences are a function of their parents’ and society’s, which in turn is affected by the consumption of domestic and imported goods. They find that free trade in cultural goods can be welfare decreasing for both importer and exporter. Rauch and Trindade (2006) model cultural production a` la Grossman and Helpman’s (1991) spillover technology: production of current cultural goods generates ideas that spill over to future cultural goods production. Crucially, current cultural production can draw on past foreign ideas. Free trade in ideas can thus reduce welfare in the long run due to cultural stagnation as all countries converge on a dominant culture.3 Maystre et al. (2009) build on this theory and use responses to the World Values Survey in their empirical analysis of trade and cultural transmission. They identify a hysteresis, whereby increased trade leads to closer cultural alignment, but subsequent reduced trade does not reverse this higher cultural affinity. My paper is silent on the issue of the dynamic effects of trade on culture (and culture on trade). Rather, I develop a methodology and index of cultural distance that may be useful in subsequent studies in this literature. The remainder of the chapter proceeds as follows. In section 2.2 I provide a theoretical model of the international trade in a cultural product. I present an overview of the movies data in section 2.3, while section 2.4 presents the results of three exercises. In section 2.4.1, I compare key statistics of the data to predictions from hierarchy and random entry models. In section 2.4.2, I estimate the average effect of quality on the propensity to enter foreign markets. In section 2.4.3, I estimate a measure of cultural distance and investigate its properties. Section 2.5 concludes. 3 Cowan  (2004) refers to a similar though more optimistic phenomenon as the cosmopolitan paradox: cultural trade leads to homogenization between countries but heterogeneity within countries.  8  2.2  Theory  The theory examined in this essay is based on a simple discrete choice model. The use of this framework is motivated by the application to the motion picture industry, in which consumers make the binary decision of whether or not to see a movie. Anderson, de Palma and Thisse (1992) show that the Constant Elasticity of Substitution (CES) utility function can be used to describe the preferences of a representative consumer only if each consumer also chooses a volume of consumption. Since this is not the case for movies, I depart from the standard CES model. The discrete choice framework is applicable to many industries where quantity is not relevant. Other cultural (or creative) industries, such as books and music, exhibit the same characteristic. The real-world institutional features of movie distribution are more complicated than depicted in the model. In particular, whereas I model decisions of movie entry as though they are independent from one another, each movie is in fact part of a portfolio belonging to a distribution company. There are six major studios in the U.S., all of which are vertically integrated with distribution subsidiaries.4 The third node in the vertical chain is the exhibitor, or movie theatre. Exhibitors are generally separate from studio-distributors—they are regulated as such in the United States— and receive a portion of box-office revenues. In addition, foreign distribution rights are licensed to foreign distributors in some countries. Given the two (or three) firms involved in taking a movie from one reel to screens around the world, there is scope for behaviour not captured by the model in this essay. For example, a studio might negotiate that a foreign distributor or exhibitor will take on a lower-quality movie as a condition to obtaining the rights to an attractive blockbuster. Such tied selling is beyond the scope of the present essay. This consideration notwithstanding, the forthcoming model presents a basic structure with which to analyze the effect of quality on entry decisions. In what follows I present increasingly refined versions of the model to guide the empirical work. 4 The six majors (and their parent companies) are Paramount (Viacom), Warner Brothers (Time Warner), Columbia (Sony), Walt Disney (Walt Disney), Universal (Comcast/General Electric), and 20th Century Fox (News Corporation).  9  2.2.1  Hierarchal Sorting  Individuals in country d purchase a variety of the product if their valuation of doing so is greater than the price, pd , which varies between countries but not within each country. Prices are taken as given in each market. The valuation of individual i from destination d consuming variety m is: vidm = k(β qm +Uidm )n(yd ),  (2.1)  where qm is the “true” universal quality of the variety, Uidm is the individual’s idiosyncratic utility, yd is the income per capita in country d and the functions k(·) and n(·) are increasing and could be destination-country specific. The parameter β adjusts for the scale on which quality is measured. Valuation is separable in per capita income, reflecting the higher willingness-to-pay in rich countries for any given quality level. Revenues from exporting to country d are given as the product of the price and the number of people who purchase the variety. This latter quantity can be expressed as the product of the total population and the proportion of the public who purchase: Rdm = pd Md P[vidm > pd ],  (2.2)  where Md is the population of country d and the proportion of the purchasing public is replaced by the probability that any of the (symmetric) individuals in the country will purchase. Plugging 3.2 into 3.3, Rdm = pd Md P[k(β qm +Uidm )n(yd ) > pd ] pd − β qm ] = pd Md P[Uidm > k−1 n(yd ) pd = pd Md (1 − P[Uidm < k−1 − β qm ]) n(yd )  (2.3)  If Uidm is distributed exponentially with parameter λ , then the above reduces to: λ +β qm −k−1  Rdm = pd Md e  pd n(yd )  (2.4)  Producers of a variety export to foreign market d if the revenues from doing so  10  are greater than the cost of entry. Fixed costs associated with market adaptation, advertising and distribution are captured by the term Fd and variable costs are negligible. Variety m is therefore exported to market d if profits are greater than zero: Πdm ≡ Rdm − Fd > 0. Since revenues are increasing in quality, a hierarchal order is predicted, whereby if a given variety is exported to market d then all varieties with higher quality are also exported to market d. If a variety is not exported to market d, than neither are any varieties with lower quality. The cutoff quality level, q∗d , for each market d can be found by setting profits equal to zero: Rdm − Fd = 0 λ +β q−k−1  pd Md e  pd n(yd )  − Fd = 0 q∗d =  1 Fd pd ln + k−1 −λ β pd Md n(yd )  (2.5)  This is a very strong prediction but is not specific to my model. Such “ability” sorting is present in any model with fixed costs of entry and where profits and revenues are increasing in ability.5 In particular, the Melitz (2003) model and related papers exhibit this feature. Eaton, Kortum and Kramarz (2008) demonstrate that such a hierarchy is at odds with the facts for French exports. It is easy to find counterexamples in the movies data of this paper. Eaton et al. (2008) introduce destination-varietyspecific demand shocks to the Melitz model to reconcile the theory with the facts. This “consumer tastes” term is also found to be important in Bernard, Redding and Schott (2009) and Crozet et al. (2009), among others.  2.2.2  Destination-Variety-Specific Demand Shocks  We can integrate the idiosyncratic shocks parsimoniously within the the existing model. Let ψdm equal the average idiosyncratic consumer utility for variety m over all individuals in destination d and define uidm to be the individual idiosyncratic utility from consumption in excess of the destination d average. Then Uidm can be 5 In  general, increasing profits does not imply increasing revenues since there may be costs associated with quality. If higher marginal costs are required for higher quality, and if prices are a function of marginal cost, then revenues may fall with quality—through movement along the demand curve— even as profits increase due to a higher price per unit. The relevant condition is a comparison of the elasticity of consumer demand with respect to quality versus the elasticity of marginal cost with respect to quality.  11  decomposed as Uidm = ψdm + uidm .  (2.6)  Plugging this expression into 3.4, we obtain Rdm = pd Md (1 − P[ψdm + uidm < k−1 = pd Md (1 − P[uidm < k−1  pd − β qm ]) n(yd )  pd − β qm − ψdm ]) n(yd )  Making the exponential distribution assumption about uidm produces the revenue equation: λ +β qm +ψdm −k−1  Rdm = pd Md e  pd n(yd )  ,  (2.7)  which breaks the monotonic relationship between quality and revenues, since although a variety may have high quality it may have a low destination-specific affinity, ψdm . Thus, high-quality varieties may not enter a market even when lower-quality varieties do so. The destination-variety-specific shocks are not observed by the econometrician, but suppose they are distributed according to the distribution function G(ψ). Then the probability that variety m is imported to destination d is given by the probability that revenues exceed fixed costs: λ +β qm +ψdm −k−1  pd n(yd )  > Fd ] pd Fd = P[ψdm > −λ − β qm + k−1 + ln ] n(yd ) pd Md pd Fd = P[ψdm < λ + β qm − k−1 − ln ], n(yd ) pd Md  P[Edm = 1] = P[pd Md e  (2.8)  where the last equation holds assuming G(·) is symmetric about zero. As the variance of ψdm collapses to zero, revenues approach those described by (3.5) and we obtain the hierarchal prediction: the probability of entry equals zero or one, depending on whether or not quality exceeds the cutoff in equation 2.5. As the variance of ψdm approaches infinity, the idiosyncratic component dominates the determination of revenue, and quality becomes irrelevant to the entry decision. The situation could then be modeled as random entry, with each variety equally likely to be released in any given market.  12  These observations suggest a test of the importance of quality versus idiosyncratic demand shocks. Under pure hierarchal sorting, a country that imports from N firms should import the N highest-quality varieties. Under random entry, the N varieties would be a random draw from the set of possible firms. This has implications for observable aggregate statistics: the minimum, mean, and maximum quality for each destination, d. The minimum-quality variety imported under pure sorting would simply be the Nth-highest quality firm. Since this value is decreasing in N, researchers have viewed a negative relationship between minimum quality (or productivity) and number of firms as evidence for quality sorting. However, under random entry, the expected minimum quality would equal the mean of the distribution for N = 1, and would approach the lower support of the quality distribution as N increases. The path of approach depends on the distribution of quality, but the prediction is again a negative relationship. The mean-quality variety imported under pure sorting would equal the top-quality firm for N = 1 and would be falling in the number of varieties imported. This is because as a country adds more varieties it is adding inferior firms. Each new addition drags the average down. Under random entry, the expected mean quality is the expected value of the quality distribution. It is independent of the number of varieties imported, although the variance of the mean quality falls with the number of varieties, due to the law of large numbers. The maximum-quality variety imported under pure sorting would be independent of the number of varieties: it would be constant and equal to the top-quality variety. Under random entry, the maximum quality would equal the mean of the distribution for N = 1 and would approach the upper support as N increases. Similarly to the minimum, the path of approach depends on the distribution of quality. In section 2.4 I compare the observed country-level statistics to the pure sorting predictions and a monte carlo simulation of random entry. By drawing from the empirical distribution I avoid imposing a parametric form for the distribution of quality. Equation 2.8 provides an estimating equation for the entry decision into foreign markets. The country-level-statistics exercise described above suggests that the role of quality in the entry decision is attenuated by possible destination-variety idiosyncracies. The coefficient on quality in the entry regression will give us a sense of how important quality is in shaping the pattern of international movie releases. Country  13  d’s size and required fixed costs enter in log terms on the right hand side. Country d’s price level and income per capita enter as more complicated functions. If prices are not available, estimating the effect of income per capita is particularly problematic. This is because prices are likely correlated with income, resulting in missing variable bias. The sign on prices is also not obvious. High prices would increase revenues for any given share of market coverage, but high prices would also lower the fraction of individuals choosing to purchase. For these reasons, I will estimate equation 2.8 using proxies for market size and fixed costs as a check, but the preferred specification will absorb country-specific variables with destination fixed effects. In what follows, I describe how we can use deviations from pure sorting to measure cultural distance between countries.  2.2.3  Measuring Cultural Distance  Destination-variety-specific demand shocks imply that a pure hierarchal sorting of foreign entry will not in general exist. Recall that these demand shocks were conceptually defined as country-level averages of individual idiosyncratic utility for each variety. In other words, the demand shocks reflect a central tendency of a country’s individuals’ tastes. Hofstede (2002) emphasizes that culture is not a fundamental that exists in its own right, but is a construct that might be useful in understanding aspects of the world. Culture reflects unobservable “mental programs”, but we can infer from observable behaviour—words or deeds—the presence of these mental programs and construct notions of culture accordingly. In the same article, Hofstede concedes that national culture is not easily measured, but that differences in culture can be obtained. This is the spirit in which I carry out the present exercise. Culture can mean different things in different contexts. The definition I will adopt is that culture is the aggregation of a society’s tastes for what is regarded as excellent in the arts.6 Differences in idiosyncratic demand shocks for cultural goods might thus be useful in measuring differences in culture between countries. If we had perfect measures of the country-level variables present in equation 2.8, we might infer from the destination component of the residuals of that estimation a measure of cultural affinity with the United States. Lacking such measures, destination fixed effects capture a mixture of effects relating to market size, wealth, fixed 6 This  definition is the author’s adaptation of the first definition in Random House (2010).  14  costs, etc. It is not obvious how to extract a cultural component from this. Suppose the variety-space of the product in question could be partitioned into two subsets. For one subset, consumption value is dependent on cultural context and for the other it is not. For the culturally-dependent set of varieties, idiosyncratic consumer utility will tend to be low for individuals in countries that are culturally distant from the United States. For the case of movies, this subset could be movies in the genres comedy and drama, whereas the compliment subset is made up of the action, adventure and thriller genres.7 Equation 2.6 decomposes idiosyncratic consumer utility into country-level and individual-specific terms. We can further decompose this term by defining ηdg as the average demand shock in a country over all varieties in a genre-type, and writing the decomposition: Uidm = ηdgm + ψˆ dm + uidm , (2.9) where ψˆ dm is the destination-variety demand shock in excess of the genre shock. Revenues become λ +β qm +ηdgm +ψˆ dm −k−1  Rdm = pd Md e  pd n(yd )  ,  (2.10)  and the probability of entry becomes P[Edm = 1] = P[ψˆ dm < λ + β qm + ηdgm − k−1  Fd pd − ln ]. n(yd ) pd Md  (2.11)  Consider estimation of this equation using the two different genre types: varieties in culturally-dependent genres and varieties in culturally-neutral genres. Label the destination fixed effects FXdC and FXdN for the two samples, respectively. Then, d for g ∈ {C, N}, FXdg = (ηdg − k−1 (pd /n(yd )) − log pdFM , and thus FXdN − FXdC = d ηdN − ηdC . The variation in this measure across destinations reflects the difference in affinity that countries exhibit for the different samples. This measure equals zero if both sets of varieties carry equal affinity, but increases as the degree of affinity for culturally-dependent varieties lags that for culturally-neutral varieties. I define this difference as country d’s “cultural distance” from the origin country. In effect, differencing the fixed effects from the two samples strips away all of the destinationspecific influences that are common across the genre-types and leaves only the dif7 The  precise partition used will be explained in the empirical section.  15  ference in destination-genre affinity. The identifying assumption is that for countries that are culturally close to the origin country, discounting of culturally-dependent genres will be small; whereas culturally distant countries will discount culturallydependent genres more heavily than neutral genres. I carry out this exercise on the sample of U.S. movies in section 2.4.  2.3  Data  The movie data for this study comes from the International Movie Database (IMDb), an international project that catalogues countless movie trivia on line. I extracted the full set of titles, release dates by country, countries of origin, and userratings. I am therefore able to tell, for any given movie, where it was produced, where it was released, and how users of imdb.com rated it on a score from one to ten. The total number of titles covered in the database is 437,041, produced in 53 exporter countries and released in 115 importer countries. I remove from the sample all movies that were not released theatrically outside of film festivals. A key movielevel attribute in this study is the perceived quality, or popular appeal. Because the population of IMDb contributors may not be representative of the universe of potential movie-goers, it is useful to have an alternative measure of quality. To this end, I exploit a database of movie titles and ratings provided by the commercial site NetFlix.com. NetFlix allows customers to rent movies on line, then sends them a copy of the chosen DVDs by mail. The NetFlix ratings are based on customer feedback. The data set became available to the public when Netflix announced a contest open to the machine-learning community. The data consists of every rating by each individual for a sample of 17,000 movies. For my purposes I took the average rating given to each movie. These ratings suffer from the same potential sampling problems as the IMDb ratings, but offer an independent measure nonetheless. I focus on viewer ratings rather than “expert” film critics’ since the logic of selection is based on expected sales volumes. The quality I am interested in derives from consumer preferences and not conceptual art. Moreover, the Netflix service was only available in the United States during the sample period, and thus the quality ratings reflect home-country preferences. Intersecting the IMDb data set with the Netflix films reduces the sample considerably, to a total of 6,413 distinct titles. After restricting the sample to U.S.-produced movies released between 1995 and 2004, the sample size  16  Figure 2.1: Histogram of Netflix Ratings  drops to 1,604. The correlation coefficient between the two measures of quality is 0.52. Theory tells us that higher-quality movies should be released in more destinations, all else equal. Indeed, for the intersected sample, the correlation coefficients for the number of countries in which the movie is released and the two ratings are 0.29 and 0.17 for NetFlix and IMDb, respectively. Figures 2.1–2.3 provide histograms of the Netflix and IMDb ratings and the number of destinations entered. The quality distributions are relatively symmetric, but the corresponding distribution of entry is severely skewed right. This is not very surprising given the theory. If quality translates into revenue exponentially, and expected revenue determines entry, then we should expect to see a skewed distribution of entry as long as fixed costs of entry are important. I compliment the two ratings-based quality measures with two financial measures. The first is the U.S. domestic box-office revenue. This measure roughly tells us how many people actually went to see the movie in the United States. Since people often act on recommendation and word-of-mouth when choosing a movie, domestic revenues indicate how well-received the movie was at home. To the extent that decision-  17  Figure 2.2: Histogram of IMDb Ratings  Figure 2.3: Histogram of Number of Markets Entered  18  Figure 2.4: Histogram of U.S. Domestic Box-Office Revenue  makers delay foreign entry, revenues may directly influence foreign-entry strategy. In any case, the measure is likely correlated with studios’ expectations. This measure also coincides with that of Khandelwal (forthcoming), who defines quality as market share, given equal prices. The second financial measure of quality is the movie’s production budget. Assuming that a higher investment produces a better product, this measure should be a signal of quality. De Vany (2004) finds that (expensive) star power is a good predictor of movie success. Kugler and Verhoogen (2008) also posit and find supporting evidence that more expensive inputs lead to higher quality in their sample of Columbian firms. Both of these financial variables come from the web site www.thenumbers.com. U.S. box-office data was available for 1,236 of the 1,604 movies in the sample. Budget data was available for just 802 movies. Histograms of these variables are presented in Figures 2.4 and 2.5. Figure 2.6 shows the histogram for the number of movies entering each country. Most countries do not release many movies. Of the 97 countries in the sample, 42 released fewer than 50 movies. Of these, 35 released fewer than 20 and 30 countries  19  Figure 2.5: Histogram of Production Budget  released fewer than 10 movies. The distribution of release numbers for countries releasing more than 50 movies is far less skewed, with just as many releasing between 400 and 700 as released between 100 and 400. The top five importers were Spain (1063), Australia (1060), United Kingdom(1044), France(1024), and Germany(981). For the entry regressions, I use data on GDP, population, bilateral distance, and other typical gravity covariates described in the empirical section. All of this data was obtained from CEPII. Information on movie genre is contained in a separate file in the International Movie Database. After merging the data, the sample size is reduced to 877 movies. Of these, 459 are coded as comedy or drama. The set of countries that imported a positive number of both sets of movies is reduced in size from 97 to 86.8 8 The  following countries had zero imports in at least one of the two subsets of movies: Bahamas, Bosnia-Herzegovina, Cuba, Faroe Islands, Ghana, Iran, South Korea, Macau, Nepal, Syria, Tanzania.  20  Figure 2.6: Histogram of Number of Movies Imported  2.4  Empirical Results  The empirical enquiry proceeds in three steps. First, I compile country-level statistics describing the distribution of movie qualities in each market. I use a graphical simulation to contrast pure hierarchal sorting against random entry, and investigate where the data fits between these two extremes. Second, I estimate the foreign entry equation to discern how important movie quality, and destination size and distance are to the probability that a given U.S. movie will be released in a given destination country. Third, I repeat the estimation of the entry equation for two different genre-types, and interpret the difference between the destination fixed effects as a measure of cultural distance. I then compare this measure to other indexes of cultural distance and test its explanatory power in a gravity equation of bilateral trade.  2.4.1  Selection versus Random Entry  The simplest model of section 2.2 makes strong predictions about which movies are shown where. If all countries agreed on which are the best films, then we should see a hierarchal sorting in the release pattern. The best movie would go to the most 21  destinations, and countries that are attractive enough to import more would select down the list in order of quality. Implications for the minimum, mean, and maximum quality observed in each market were derived in section 2.2. Chen and Moore (2009) study the entry decisions of French foreign direct investors, and document a negative relationship between the number of firms investing and the minimum productivity of these firms. They claim that this is evidence of sorting, but their analysis omits a critical factor: purely random entry will also lead to a negative slope. The expected minimum value taken from N draws will always be decreasing in N. In order to distinguish the selection effect from random entry, I add two components to the analysis. The first is to look at statistics other than the minimum. Under randomness, the mean quality will not vary with the number of releases; whereas, under selection the mean quality would be decreasing. The maximum quality will be increasing under randomness for the same reason the minimum is decreasing. However, under selection the best movie is released in all countries and hence the maximum does not vary with the number of releases. The second addition to the analysis is to simulate random entry by drawing from the empirical distribution of movie quality. For each number of movies released (a country-level variable ranging from one to 150), I draw that many times from the empirical quality distribution. I take the relevant statistic (min, mean, or max) and then repeat a thousand times, saving the 5th and 95th percentiles and the mean over the thousand repetitions. Figures 2.7- 2.9 plot the results along with the actual data for the year 2004 and the pure hierarchal sorting predictions. Inspection of Figure 2.7 shows that the actual minimum quality values lie above those predicted by random entry, almost tracing out the upper bound of the 95th percentile. This is evidence in favour of selection, since under selection, a country importing few movies will tend to release the better ones. Comparing the data to the pure-sorting predictions, however, shows that idiosyncratic demand shocks pull the minimum-quality values toward randomness. Figure 2.8 similarly shows limited support for the selection model. The mean quality movie decreases in the number of releases, indicating that countries importing more movies are on average adding lower quality films. As the simulation confirms, random entry would predict no relationship (a horizontal slope) between mean quality and number of releases. As with the minimum quality, the data points mostly lie close to the 95th percentile line, and are far lower than pure sorting would predict.  22  Figure 2.7: Minimum Quality vs. Number of Releases by Country in 2004  Figure 2.9 illustrates the data and simulation for the maximum quality movie in each market. Under selection, we would expect every country to release the topranked movie, and hence there should be no relationship (a horizontal slope). The simulation confirms that under random entry, the maximum quality increases in the number of releases. The data for this exercise lie entirely within the 90% confidence interval for random entry. Negative shocks appear to keep the top-ranked movie out of many of the foreign markets; however, three-quarters of the markets import one of the top-three ranked movies in the sample. The evidence suggests that quality selection is taking place, but that moviedestination-specific demand shocks are also important drivers of the distribution pattern. In the following section, I test whether the Netflix measure of movie quality has a systematic impact on the probability of a movie’s import. In section 2.4.3, I attribute a component of the idiosyncratic demand shocks to country-genre preferences, and interpret differences in preferences as cultural distance.  23  Figure 2.8: Mean Quality vs. Number of Releases by Country in 2004  2.4.2  Predicting Movie Imports: Quality and Geography  The model in section 2.2 predicts that the probability of entry is increasing in the size and wealth of the destination market and the quality of the movie, and decreasing in the magnitude of fixed costs required to enter the market. Entry is stochastic from the point of view of the analyst since we do not observe the destination-moviespecific demand shocks. Table 2.1 reports results of estimating equation 2.8 on a sample of 1,604 U.S. movies and 97 destination countries. The movies were released in the U.S. over the ten-year period between 1995 and 2004. The binary variable Entrydm is coded as one if destination d had imported movie m by the end of 2009. Since we are restricting attention to cinematic releases, the lag between the last U.S. release date and the end of 2009 should be sufficient to allay any concerns of censorship. Indeed, the last observed entry in the sample occurred in 2005. The left panel of Table 2.1 displays results for Probit estimation; the right panel is for the linear probability specification. Columns 1 and 3 omit destination fixed effects  24  Figure 2.9: Maximum Quality vs. Number of Releases by Country in 2004  and therefore allow for estimation of destination-level determinants of entry. Data on gross domestic product is unavailable for Cuba and Faroe Islands so they are omitted from these regressions. Fixed costs are proxied by (log) geographic distance and a set of dummy variables indicating whether the destination country shares with the U.S. a common border, common official language, common colonial origins, a free trade agreement, or a strict currency union. These variables are commonly used in the trade literature as covariates in the gravity equation, to proxy for variable and/or fixed trade costs. Columns 2 and 4 include destination fixed effects and the focus is on the impact of quality on the extensive margin. In the model, qm represents the average perceived quality globally. Over the sample period, Netflix conducted business only within the U.S. and only through DVDs sent in the mail. As long as the worldwide perceived quality is an affine transformation of the U.S. perceived quality, applying the U.S. measure of quality is consistent with the model. The Netflix quality rating has been normalized by subtracting its mean and dividing by its standard deviation. The coefficient can thus be directly interpreted as the increase in the probability of entry associated with a 25  one standard deviation increase in the movie’s rating. The coefficients suggest that increasing the quality of a movie by one standard deviation raises the probability of entering an “average” destination by about four-and-a-half percentage points. This compares to an overall probability of entering a foreign market of about 17%. Population and per capita GDP are also associated with more entry, as predicted. Surprisingly, geographic distance has a very small effect on the extensive margin, and the effect is positive for the Probit model. This contrasts with the elasticity of -0.2 for merchandise trade, reported by Helpman, Melitz and Rubinstein (2008). The other proxies for (low) fixed costs are similarly surprising. Sharing a border or currency union with the U.S., or speaking English, are all associated with a lower probability of entry. Participating in a free trade agreement has the expected positive sign and may signal closer economic ties more generally than simply reduced tariffs. These coefficients should be viewed with some suspicion, given that trade is considered from only a single source. For example, only Canada and Mexico share a border with the United States, and one of Canada’s official languages is English. This turns the common-border dummy into a de facto Mexico fixed effect. The Netflix ratings are just one measure of movie quality. Alternatively, we could use the ratings found in the International Movie Database; or look at “revealed” popular appeal, as measured by domestic box office revenues in the U.S.; or consider the cost of production, assuming that a higher investment leads to a better product. The latter of these measures is the only true ex ante measure of quality. To the extent that international releases lag U.S. domestic release, the revenue-based measure may also directly inform decision-makers. In any case, it is likely correlated with the decision-makers’ ex ante beliefs about the movie’s appeal.9 Table 2.2 reports estimates from the destination fixed effects models for these three alternative quality measures. The variables have been standardized from their original scales. Financial data is more patchy than the viewer ratings. The sample size of movies is reduced to 1,236 for U.S. revenues and 802 for budget data. Both variables are from the web site, www.the-numbers.com. The IMDb user rating is a weak predictor of movie entry, with a one standard deviation increase in the rating associated with about a two-and-a-half percentage point increase in the probability of foreign entry to an average market. Domestic revenues and budgets are associated with a seven to ten percentage point increase in the prob9 The  dynamics of movie diffusion internationally are explored in Chapter 3.  26  Table 2.1: Probability of Exporting to a New Market  Probit (1) (2) Entrydm Entrydm .047*** .046*** (.004) (.006) logPopd .058*** (.002) logGDPPCd .109*** (.004) logDistd .016*** (.002) Borderd -.056*** (.006) Englishd -.012*** (.002) FTAd .039*** (.005) CUd -.031*** (.005) Destination FE No Yes Year FE Yes Yes N 152,380 155,588 R2 .237 .354 Model : Depvar: Net f lixm  LPM (3) Entrydm .044*** (.003) .067*** (.001) .117*** (.002) -.007*** (.002) -.113*** (.010) .0009 (.002) .032*** (.006) -.039*** (.003) No Yes 152,380 .199  (4) Entrydm .043*** (.003)  Yes Yes 155,588 .300  Note: Statistical significance of 1% is indicated by ***. Pseudo-R2 reported for Probit specifications. Probit coefficients are reported as average partial effects. Robust standard errors are adjusted for clusters in movies.  ability of entry. A one standard deviation increase in these measures increases the probability that a movie will enter a typical foreign market by 50%.  2.4.3  Movie Trade and Cultural Affinity  In this section I consider how countries’ different propensities to import U.S. movies might be interpreted as cultural differences. To identify a cultural component from the destination fixed effects, I estimate the entry equation of two samples of movies. Each movie in the IMDb is assigned one or more genre. Typically, movies  27  Table 2.2: Probability of Exporting: Alternative Quality Measures  Model : Depvar: Imdb Rating  (1) Entrydm .0243*** (.005)  U.S. Revenue  Probit (2) Entrydm  (4) Entrydm .0241*** (.003)  .107*** (.003)  Budget Destination FE Year FE N R2  (3) Entrydm  Yes Yes 155,588 .339  Yes Yes 119,213 .469  LPM (5) Entrydm  (6) Entrydm  .0886*** (.003) .0711*** (.004) Yes Yes 76,190 .448  Yes Yes 155,588 .292  Yes Yes 119,892 .398  .0756*** (.004) Yes Yes 76,190 .441  Note: Statistical significance of 1% is indicated by ***. Pseudo-R2 reported for Probit specifications. Probit coefficients are reported as average partial effects. Robust standard errors are adjusted for clusters in movies. Bahamas and Faroe Islands omitted from budget regressions since they predict failure perfectly for the reduced sample.  will be assigned more than one genre; for example, drama-musical, or romancecomedy. It is well known in the industry that drama and comedy movies do not travel as well overseas as action, adventure, or thriller films.10 I code a movie as culturally dependent if comedy or drama is listed among its genres, but action, adventure and thriller are not. A film such as Jackie Chan’s Rush hour (1998) is considered an action-comedy-thriller-crime movie, according the IMDb’s genre data. I would therefore not code it as culturally dependent. This definition is more conservative than simply counting everything that is partly comedy or drama. Table 2.3 reports estimates of the entry equation for a linear probability fixed effects model. The effect of Netflix rating is robust to the reduced sample size. Culturally-dependent movies are less likely to enter an average destination, as pre10  Film distributor Hammad Zaidi (2010) writes that, “When it comes to comedies, romantic comedies, dramas, coming-of-age films, personal stories, family films...you have a better chance of winning the lottery than you do of enjoying healthy sales internationally...The reason that most genres don’t work overseas is because their content is specifically designed to work within the country they were made. For example, in comedies, what’s funny in Los Angeles may not be funny in Zimbabwe and in romantic comedies, whats romantic in Nashville may be offensive in China.”  28  Table 2.3: Probability of Exporting: Genre Effects  Model : Depvar: Netflix Rating  (1) Entrydm .050*** (.001)  (2) Entrydm .048*** (.001) -.040*** (.002)  Yes No Yes 75,422 .021  Yes No Yes 75,422 .024  Comedy-Drama Nf × Com-Dra Destination FE Dest-Genre FE Year FE N R2 (overall)  (3) Entrydm .043*** (.002) -.041*** (.002) .011*** (.002) Yes No Yes 75,422 .024  (4) Entrydm .048*** (.001)  No Yes Yes 75,422 .021  Note: Statistical significance of 1% is indicated by ***. Robust standard errors in parentheses.  dicted. A comedy or drama would have to be about one standard deviation higher in quality in order to have the same probability of entry as its culturally-neutral counterpart. Moreover, the interaction term suggests these movies are about 25% more sensitive to quality. In column 4 I include destination-genre fixed effects. Differencing these fixed effects by destination will give the measure of cultural distance from the United States. I call this measure the country’s “Hollywood distance” because it is estimated using U.S. movies. Recall that it is not strictly a measure of the country’s Hollywood (dis)affinity, however, because it is based on the difference in propensity to import genre-types. Any general affinity toward U.S. movies is thus wiped out. There is an extra degree of freedom in estimating a full set of fixed effects in addition to a constant. An arbitrary restriction must therefore be imposed. I choose to normalize so that the minimum value of Hollywood distance is zero. A potential issue of using the linear probability model is that predicted probabilities could lie outside the [0, 1] interval. For column 4 of Table 2.3, there are 28 observations (out of 75,422) with a predicted probability greater than one, all of which are associated with entry. More worryingly, there are 12,930 observations with a predicted probability less than zero, all but 82 of which are associated with no entry. The range of predicted probabilities is [−0.20, 1.34]; the percentage of out-of-bounds 29  Figure 2.10: Histogram of Hollywood Distance  observations is 17.2%.11 Figure 2.10 provides a histogram of Hollywood distance. There is a large group of countries with a relatively low Hollywood distance, and the number of countries falls as the distance increases. Figure 2.11 plots a bar graph of each of the countries’ Hollywood distance. Reassuringly, Canada has the lowest distance. Perhaps not surprisingly, the Gulf nation of Kuwait is by far the most “distant”. Casual inspection suggests a link between the Hollywood distance index and geographic proximity to the United States. Indeed, a scatter plot of Hollywood distance versus geographic distance—provided in Figure 2.12—confirms that, with the exception of Mexico, geographically proximate countries are also culturally close to the United States. We can also compare Hollywood distance to other indices of cultural distance or similarity. Figure 2.13 plots Hollywood distance against an index of language similarity constructed by Dyen et al. (1992). The language-similarity index equals the proportion among two hundred words that share a common origin, for sixtythree Indo-European languages. For instance, “mother” in English and “mere” ` in 11 As a robustness check, I also estimated a probit model with fixed effects, which potentially suffers  from the incidental parameters problem. Results were similar to those reported.  30  Figure 2.11: Hollywood Distance by Destination Country  Figure 2.12: Hollywood Distance versus Geographic Distance  31  Figure 2.13: Hollywood Distance versus Language Similarity  French share a common origin, whereas “head” and “t ete” ˆ do not. We might expect a negative correlation between the two indices, but a cluster of countries in the bottomleft corner break this pattern, leaving no discernible relationship between Hollywood distance and language similarity. Similarly, we can compare Hollywood distance to genetic distance. The fixation index, FST , measures the degree of genetic diversity due to allele frequency differences among populations (Cavalli-Sforza et al., 1994). Figure 2.14 plots Hollywood distance against Genetic Distance. As with language similarity, we do not find a correlation between the two indices. Finally, Figure 2.15 plots Hollywood distance against religion similarity, defined as the probability that an individual from one population will share the same religion as a randomly chosen individual from the other population. Again, we find no relationship. While Hollywood distance is correlated with geographic distance, it does not appear to correlate with other measures of cultural distance, related to language, genetic drift, or religion. But does Hollywood distance capture an economically meaningful component of culture? To investigate further, I include it as a covariate in a standard 32  Figure 2.14: Hollywood Distance versus Genetic Distance  Figure 2.15: Hollywood Distance versus Religion Similarity  33  gravity equation of U.S. international trade in goods. If Hollywood distance captures cultural distance between the U.S. and its trading partners, and if cultural distance is an impediment to trade in goods, then we expect Hollywood distance to be negatively related to the value of bilateral trade. I obtained industry-level bilateral exports between the U.S. and each of the 82 countries in my movies sample for the years 2002 to 2004. The data is from The Center for International Data at UC Davis. I merge the export data with the CEPII gravity covariates and Hollywood distance. Disdier et al. (2010) match HS industry descriptions to a UNESCO definition of cultural goods. I use their list of HS cultural sectors to code each of the industries in the sample as “cultural” or “not cultural”, and aggregate the industry trade volume data for each partner-year according to this variable. Table 2.4 displays the results of running different specifications of a gravity equation. The first three columns include traditional bilateral variables, such as distance, GDP and institutional linkages. Coefficients on these variables are in line with the existing gravity literature.12 The elasticities of distance and GDP with respect to trade volumes are approximately equal to one in magnitude. All of the other controls have significant and predictable signs, apart from colony—indicating a common colonial relationship—and currency union, which are imprecisely estimated, and contiguity, which enters with a negative coefficient. The coefficient on the “cultural” dummy indicates that trade volumes are much lower for cultural industries. In column 2, Hollywood distance enters with a positive but insignificant coefficient, indicating that the measure of cultural distance captured by the procedure of this paper does not add to our understanding of the overall level of trade. In column 3, the interaction between Hollywood distance and the cultural dummy is included. The coefficient is again positive but insignificant. Column 4 reports the gravity equation with importer- and exporter-year fixed effects, for which results are unchanged. Column 5 reports a specification with the logarithm of the ratio of cultural to non-cultural trade volumes on the left-hand side. The coefficient on Hollywood distance is then a continuous analogue of a difference-in-difference estimator. The result is unchanged. The specifications reported in Table 2.5 investigate the robustness of the (non-) results. All of the specifications except those of column 4 are run omitting the linkage variables of Table 2.4. Columns 1 and 2 show that when these determinants of trade 12 See  Disdier and Head (2008).  34  Table 2.4: U.S. Bilateral Trade, 2002-2004 (1) lvalue -1.042∗∗∗ (0.129)  (2) lvalue -1.057∗∗∗ (0.129)  (3) lvalue -1.058∗∗∗ (0.128)  lgdp o  1.086∗∗∗ (0.0292)  1.079∗∗∗ (0.0322)  1.079∗∗∗ (0.0325)  lgdp d  1.088∗∗∗ (0.0253)  1.081∗∗∗ (0.0284)  1.081∗∗∗ (0.0286)  contig  -0.710∗ (0.280)  -0.717∗∗ (0.277)  -0.717∗ (0.281)  comlang off  0.518∗∗ (0.174)  0.506∗∗ (0.174)  0.506∗∗ (0.173)  colony  0.275 (0.216)  0.286 (0.215)  0.286 (0.216)  rta  1.104∗∗∗ (0.133)  1.122∗∗∗ (0.134)  1.122∗∗∗ (0.133)  custrict  0.283 (0.159)  0.300 (0.163)  0.301 (0.158)  comleg  0.539∗∗ (0.175)  0.569∗∗ (0.182)  0.570∗∗ (0.182)  cultural  -5.727∗∗∗ (0.0752)  -5.727∗∗∗ (0.0752)  -5.729∗∗∗ (0.0750)  0.0350 (0.0566)  -0.0404 (0.0501)  963 0.897  0.154 (0.0965) 963 0.897  ldist  std hollywood  std hollyXcult N adj. R2  963 0.897  Robust standard errors in parentheses ∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001  35  (4) lvalue  (5) lratio  -5.713∗∗∗ (0.0631) 0.153 (0.0970) 0.153 (0.0972) 979 0.926  487 0.010  are not controlled for, Hollywood distance is in fact associated with less bilateral trade. The coefficient is reported in standardized form, so that column 1 indicates that a one-standard-deviation increase in Hollywood distance is associated with an 11% decrease in bilateral trade. The interaction in column 2 is not significant, but including it raises the magnitude of the effect of Hollywood distance to 19%. In columns 3-5, observations involving Kuwait are omitted. This is because Kuwait is a large outlier for Hollywood distance and might be driving the results. In column 3, the interaction term is significant and positive. The magnitude of the Hollywood distance effect for non-cultural trade flows is increased to 26%, and the total effect for cultural goods is not significantly different from zero. Column 4 investigates results when the gravity linkage variables are included, but Kuwait observations are dropped. In this case, the Hollywood distance effect for non-cultural industries returns to 10% in magnitude, and cultural-goods trade actually increases significantly in Hollywood distance. Column 5 confirms the difference of effect using the log ratio “differencein-difference” specification omitting Kuwait observations. The results of Table 2.5 show that when Kuwait—a large outlier—is not included in the regression, Hollywood distance has additional explanatory power in a traditional gravity equation. This result is consistent with interpreting Hollywood distance as an index of cultural distance, with larger values associated with either higher costs of trade or lower congruities of demand. Curiously, cultural goods are traded more intensively as Hollywood distance increases. This suggests that cultural goods trade could be determined more by a desire to experience new products, or by comparative advantages in production, rather than lower transaction costs due to closer cultural understanding. The result highlights the complex nature of the interaction between culture and trade.  2.5  Conclusion  In this chapter I developed a simple model of international trade in a heterogeneous unit-demand product. Foreign revenues are rising in variety quality and destination-country size and wealth. Fixed costs of entry imply that only varieties that are appealing enough will be exported. Using a direct measure of movie quality, I test how well U.S. movie exports adhere to this hierarchy. Graphical techniques suggest that selection is important, but also that destination-movie-specific demand  36  Table 2.5: U.S. Bilateral Trade, 2002-2004 (1) lvalue -0.852∗∗∗ (0.0890)  (2) lvalue -0.852∗∗∗ (0.0887)  (3) lvalue -0.852∗∗∗ (0.0889)  (4) lvalue -1.050∗∗∗ (0.127)  lgdp o  1.119∗∗∗ (0.0285)  1.119∗∗∗ (0.0288)  1.115∗∗∗ (0.0274)  1.071∗∗∗ (0.0307)  lgdp d  1.122∗∗∗ (0.0251)  1.122∗∗∗ (0.0254)  1.122∗∗∗ (0.0253)  1.078∗∗∗ (0.0282)  cultural  -5.725∗∗∗ (0.0811)  -5.727∗∗∗ (0.0808)  -5.684∗∗∗ (0.0793)  -5.687∗∗∗ (0.0728)  std hollywood  -0.113∗ (0.0539)  -0.187∗∗∗ (0.0523)  -0.258∗∗∗ (0.0539)  -0.101∗ (0.0510)  0.150 (0.103)  0.313∗∗∗ (0.0893)  0.315∗∗∗ (0.0781)  ldist  std hollyXcult  contig  -0.689∗ (0.280)  comlang off  0.498∗∗ (0.173)  colony  0.291 (0.214)  rta  1.123∗∗∗ (0.132)  custrict  0.314∗ (0.155)  comleg  0.578∗∗ (0.182) 952 0.902  N adj. R2  963 0.880  963 0.880  Standard errors in parentheses ∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001  37  952 0.884  (5) lratio  0.300∗∗∗ (0.0598)  482 0.042  shocks play a large role in foreign entry decisions. I use direct and revealed measures of movie quality to look for a systematic role for movie quality in export decisions. Estimates suggest that a one-standard-deviation increase in quality from the average leads to a three-to-eight percentage-point increase in the probability of entry. This compares with an overall probability of 17% in the sample. I exploit data on movie genre to estimate a measure of cultural distance (“Hollywood distance”) between destination countries and the U.S. This measure of cultural distance is correlated with geographic distance, but uncorrelated with prominant other indices of cultural distance. Hollywood distance is associated with lower bilateral trade volumes between the U.S. and its trading partners, but higher trade volumes for cultural goods. The result might be due to the combination of two opposing forces. On the one hand, cultural distance increases transaction costs as it is associated with lower trust. On the other hand, people are interested in consuming cultural goods from exotic places.  38  Chapter 3 Learning via Sequential Market Entry: Evidence from International Releases of U.S. Movies 3.1  Introduction  New products are almost never released simultaneously around the world.1 Typically, a product will be released domestically and then spread geographically over time. There are a variety of reasons that might lead firms to delay the global distribution of a new product. For instance, firms may choose to delay if domestic success contributes to a positive “buzz”, which could be harnessed for international marketing efforts (Elberse and Eliashberg, 2003; McCalman, 2005). Alternatively, if cash-poor firms have private information about the quality of their products, they may not be able to convince creditors to finance international expansion without the proof of robust domestic sales (Chaney, 2005; Manova, 2010; Minetti and Zhu, 2011). Or firms may be uncertain about their own products’ profitability (Akhmetova, 2010; Albornoz et al., 2010; Eaton et al., 2011), and thus each subsequent entry could serve to add information about the product’s appeal, allowing the firm to update expectations in potential subsequent markets—and avoiding costly entries where they are likely to fail. 1 See,  for example, Gatignon et al. (1989) and Ganesh et al. (1997) for marketing research on international diffusion by multinationals.  39  At the heart of all three of these explanations is a lack of information, either on the part of consumers, financiers, or the firms themselves. The recent papers cited above exemplify the increased interest in the role of firm learning on the sequential nature of foreign entry. In this chapter, I use the motion picture industry to investigate the phenomena of sequential entry. I document facts about the spatial and temporal patterns of theatrical releases of U.S. movies in key international markets, and propose a model of firm learning. Taking into account the ex ante variability of performance within each market as well as the correlations of performance across foreign markets, I find that a one-standard-deviation increase in the update to expected box-office revenues, based on the previous round of entry, is associated with a 25% increase in the probability of entry to a typical potential destination in the current round. Robustness checks suggest the result is due to genuine learning, as opposed to an identified potential missing-variables problem. An additional alternative explanation is particularly well-suited to the movies industry. If some factors of production are reusable then staggering entry is a costsaving exercise. The prints on which films are encoded are expensive, costing about two thousand dollars each.2 To the extent that they can be reused in multiple countries, delaying international release dates could allow for significant savings. Distributors could release in big markets first and then spread to the smaller markets as prints become available. The marketing effort of the stars—often in the form of local talkshow appearances—is also reusable, and of course by its nature cannot take place simultaneously in distant locales. These tangible examples from Hollywood serve to illustrate more general issues, such as managerial attention to a product release. The alternative explanations of staggered entry do not preclude the strategy that firms use sequential release dates to learn about their product quality. It is possible that all of these concepts are at play. The aim of this paper is to show that, even if sequential entry is due to other factors, distributors do learn from their experiences, and act on them by entering more markets on good news and fewer markets on bad news. There are also forces that would contribute towards simultaneous release dates. Discounting of the future implies firms would rather realize their profits sooner than later. More importantly, delayed foreign release increases the potential for lost sales due to piracy. Distributors face a trade-off between strategically delaying entry, and moving to “day-and-date” simultaneous releases to combat international piracy. If 2 Finney  (2010).  40  firm learning is an important aspect of staggered release schedules, then the costs of simultaneous international entry could be larger than previously thought. This essay is related to the nascent literature on exporting and firm learning. Eaton et al. (2011) (EEKKT, hereafter) observe in Columbian transaction-level data that many firms export small amounts and have short tenure as exporters. These facts appear to be inconsistent with the dominant theory of fixed-exporting costs introduced by Melitz (2003). If fixed costs are important, there should be a minimum scale required of exporters in order to break even. EEKKT reconcile the facts with fixed-exporting costs by introducing a search and learning model of trade, in which exporters are initially uncertain of their products’ appeal in the foreign market. They estimate their model using the U.S.–Columbia transaction data. Thus, the focus is on learning over time in a bilateral setting. By contrast, the present study considers learning across foreign markets. Albornoz et al. (2010) (ACCO, hereafter) similarly write a model in which exporters are uncertain about their profitability of exporting. Theirs is a stylized model of two countries and two periods, and emphasizes the value of information from entering one market at a time. The empirical predictions derived from the model are indirect indications of firm learning. ACCO find supporting evidence from a census of Argentinean manufacturers that, conditional on survival, growth rates are highest between the first and second periods in the first export market of the firm. This pattern is consistent with a firm that adjusts its supply upon receiving good news, assuming that the firm can learn almost perfectly after one period. The empirical section of the present study directly tests whether firms respond to past performances by including a Bayesian-derived updating term in an entry regression. Moreover, correlations between markets are not perfect, and differ from one country-pair to another. Akhmetova (2010) introduces a model in which new exporters can choose a “testing technology”, which allows them to export without paying fixed costs, but at a marginal cost that is convex. A second technology exhibits linear marginal costs but requires the payment of a one-time fixed cost of entry. Firms observe noisy signals about their demand during the testing stage. The model endogenizes the length and intensity of the testing stage and—like Arkolakis (2008) and EEKKT—the size of the second-stage entry cost. In choosing the amount of investment in each period, the exporter takes into account the expected revenues that will obtain, but also the value of the information that is learned.  41  This essay is most similar to ACCO, in that it investigates whether firms sequentially add export destinations as a result of a learning strategy. The “firms” in this study are motion pictures, however, and thus the industry context is quite different. Foreign motion-picture revenue is trade in services (of the actors, directors, editors, etc.), as opposed to the manufacturing trade that was the focus of the other papers. Moreover, movies are cultural products, and thus subject to possibly wide differences in appeal across markets. A necessary condition for firm learning is a positive correlation of appeal in this dimension. ACCO allow for less-than-perfect correlation across markets in an appendix, but focus on the case of single-period (perfect) learning. Furthermore, the life cycle of an individual movie is much shorter than most traded goods. In any given market, a movie will typically play for 4–6 weeks; in the sample of this paper, movies had entered 95% of their ultimate markets within 12 months. Because of these features, firms only make one entry decision per market; there is no scope for intra-market learning. A key feature of this essay is that I track individual products (movies) and not just firms that may sell multiple products. This is important because the theory relates to the level of demand at the product level. A positive market response to one product may not translate to a firm’s other offerings, so it is cleaner to have product-level data. Moreover, the product does not improve over time, as a firm’s productivity might, and thus results are not subject to that potential confounding factor. Secondly, in the empirical section I directly test whether past surprises in revenue affect the current probability of further entry. This contrasts with ACCO, who indirectly test learning by separating first-year exporters from experienced (all other) exporters. I am also able to demonstrate the validity of the exercise by showing that future (unrealized) surprises are not nearly as salient as past surprises in explaining entry, which would have indicated that firms do know their true quality—and anticipate the “surprises”—but stagger for other reasons. The results suggest an additional cost to piracy. If distributors move toward simultaneous release in order to thwart pirates, they lose the value of learning-by-staggering, and thus may incur substantial fixed costs even if the movie turns out to have low appeal. As discussed in the opening paragraph, there is a confluence of factors that inform the international entry strategy of new products. The marketing literature has identified many of these issues for motion pictures,3 but little attention has been 3 See  Elberse and Eliashberg (2003) and Eliashberg, Elberse and Leenders (2006) and the refer-  42  given to the idea that distributors might be using a learning strategy to avoid bad investments. Neelamegham and Chintagunta (1999) propose a Bayesian model of box-office forecasting, in which projections for each movie-destination are updated as new information becomes available. Crucially, however, they treat the decision of whether or not to release a movie in a given country as exogenous. In contrast, I focus on how firms update their expectations in order to inform the entry decision. The chapter proceeds as follows. In section 3.2 I describe the data set and document features of the spatial-temporal pattern of entry for U.S. movies. In section 3.3 I derive a model of firm learning that guides the empirical specifications that follow. The regressions suggested by the model are estimated in section 3.4, along with tests for alternative explanations and other robustness checks. The conclusion summarizes the main findings.  3.2  Data  Ticket sales revenue by country were collected from the web site boxofficemojo.com. The full sample includes all U.S. movies that were shown in at least one of the other markets considered over the period 2002-2008.4 Production budget data was taken from the web site the-numbers.com and is available for 761 of the movies. These financial statistics are augmented with categorical variables, including the main genre—also taken from the-numbers.com—and MPAA rating, from imdb.com. Figure 3.1 graphs for each country the number of movies imported versus the average box-office revenue in the market. Countries with the most entries are Englishspeaking (Australia and UK) or large (Germany). Nevertheless, Spain is the secondmost-preferred foreign destination. The markets with the least number of entries are culturally-distant Hong Kong and Japan, both of which also have thriving domestic film industries. In general, countries with higher average revenues also import more films. Japan is a large outlier, as it is the second-least-preferred market but is the highest performer for the movies that do enter. This might be due to large entry costs for the Japanese market. With large entry costs (e.g. print and advertising costs), only the higher-grossing movies will find it profitable to enter, thus increasing the average ences therein. 4 The other 13 markets are Argentina, Australia, Czech Republic, France, Germany, Hong Kong, Italy, Japan, Netherlands, New Zealand, Norway, Spain, and United Kingdom.  43  Figure 3.1: Intensive vs. Extensive Margins of Entry  box-office revenue in the country and decreasing the total number of entrants. For the other countries in the sample, it seems that the market size effect dominates this selection effect. All movies tend to do better in bigger markets, so even though this means low-quality movies may enter, average revenues increase in the number of entrants because both are driven by the large market. To get a sense for the variation in entry costs, we can turn to anecdotal evidence. Epstein (2005) breaks down the costs of foreign entry for the movie, Gone in Sixty Seconds (2000). The figures are presented in Table 3.1. Advertising costs are given by country for seven important foreign markets, and other costs are provided in aggregate form. The table confirms Japan is a high-cost market, with advertising costs more than twice as high as the next most costly country. The table also demonstrates how costly foreign entry is. More than half the box-office revenue of $129.5 million is taken by exhibitors, and after paying expenses, the total profit left for the distributor, Disney, is just $18 million. Epstein (2005) also provides the example of Clint Eastwood’s Midnight in the Garden of Good and Evil, which earned $3.1 million at the foreign box office, but cost $6 million in foreign prints and advertising. He notes  44  Table 3.1: Costs of Foreign Entry: Gone in Sixty Seconds JPN DEU GBR Advertising 6.5 3.1 2.5 Prints Dub/Subtitle Shipping Foreign Taxes Curr. Conv. For. Trade Assoc. Total Costs Exhibitors’ Share Total Deductions For. Box Office Foreign Profit All figures in millions of U.S. dollars Source: Epstein (2005)  FRA 1.4  AUS 1.1  ESP 1.0  ITA 0.9  Other 8.7  Total 25.2 5.7 0.82 0.46 5.0 0.27 0.12 38.0 73.5 111.5 129.5 18.0  that many releases lose money abroad. Figure 3.1 shows that there is wide variation at both the extensive and intensive margins of entry. To understand which films are traveling abroad, consider Figure 3.2, which plots the number of markets a movie enters against its domestic box-office revenue. Better-performing movies tend to enter more foreign markets, and the relationship is tighter for the non-comedy-drama genres. While there is considerable noise at the disaggregated movie level, Figure 3.3 shows that, when movies are aggregated according to the number of markets they entered, there is a monotonic relationship between the number of markets and the mean domestic revenue. Figures 3.2 and 3.3 tell us about the long-run pattern of entry. Not surprisingly, the more appealing a movie is to audiences—as measured by domestic box-office returns—the more countries that movie is likely to enter. This result is natural if destination-specific fixed costs are important. Then only those movies that are likely to earn enough at the box-office to offset these costs will enter a foreign market. As long as returns are correlated across the domestic and international markets, we expect the positive relationship that we observe. Studios and distributors can’t forecast precisely how well a movie will perform before it is released, but they can make reasonable predictions. Domestic revenues are correlated with the production budget, as shown in Figure 3.4. Apart from a few low-budget surprise successes, the budget does quite well in predicting performance,  45  Figure 3.2: Domestic Revenue vs. Number of Markets Entered  Figure 3.3: Mean Domestic Revenue by Number of Markets Entered  46  Figure 3.4: Domestic Box-Office Revenue vs. Production Budget  with an R2 of 0.43.5 If firms are using a sequential entry strategy because of uncertainty about the profitability of their products in each market, it seems less likely that they would do so when the production budget (and thus, expected profit) is high. Indeed, big-budget blockbusters are far more likely to have near-simultaneous release dates internationally. The median delay to foreign release since the U.S. premiere is decreasing in the size of the budget (Figure 3.5), despite the fact that bigbudget movies enter more markets. Considering the quartiles of production budget in turn, the histograms of median delay are increasingly skewed towards early release (Figure 3.6). This might be a reaction to movie piracy, but it also demonstrates the confidence that firms have in recouping their entry investments. Distributors of smaller productions might test the popularity of their products one market at a time, expanding with good news or limiting distribution if profitability is allusive. To illustrate, Figure 3.7 charts the entry dates and performance (relative to initial prediction) of the 2003 Woody Allen production, Anything Else. With a budget of $18 million, 5 The  linear fit displayed in the figure is calculated by omitting the movies with a budget of less than $100,000. The OLS slope-coefficient is 0.84; with the low-budget movies included it is 0.74.  47  Figure 3.5: Median Delay vs. Production Budget  it sits in the second quartile of the sample. In the first month, the movie disappointed in the U.S. but was a surprise success in Italy. The film then spread to France and Spain where it also played well. Three months later it was released in Argentina, to a neutral performance, before stumbling in Northern and Eastern Europe. It did not enter the remaining five markets in the sample. Figure 3.8 illustrates the entry timing and performance of the 2007 Quentin Tarantino horror, Grindhouse. Its budget of $45 million puts it in the third quartile, but poor performance in all four of its first markets appears to have limited further international releases. Apart from the Italian release of Anything Else, which occurs in the same month as that of the U.S., Figures 3.7 and 3.8 exemplify a pure sequential entry pattern. By this I refer to the fact that each country was entered in a seperate month. To illustrate another entry pattern, consider Figure 3.9, which plots the entry dates and performance for the 2004, seventy-five-million-dollar comedy, 50 First Dates, starring Adam Sandler and Drew Barrymore. Here we see a mix between sequentiality and simultaneity, with multiple countries entered in each month. To capture the degree to which a movie is released according to a sequential entry strategy, I compute  48  Figure 3.6: Histogram of Median Delay by Production Budget Quartile  Figure 3.7: Entry Timing and Performance of “Anything Else”  49  Figure 3.8: Entry Timing and Performance of “Grindhouse”  Figure 3.9: Entry Timing and Performance of “50 First Dates”  50  a “sequential index” for each movie. First, I partition the release dates into months since the U.S. release (with month 1 indicating the month of the U.S. release). Any country that is entered in the same month as another is considered to be entered simultaneously with that other country. For most movies, there are gaps in the month in which new entry occurs. For example, a movie might go to two markets in month one, three markets in month two, but then only enter its last market in the fifth month. I refer to months in which the movie does enter new markets as “rounds” of release, so for this hypothetical movie, the fifth month would be considered round three. The sequential index, Z, is computed as follows: Z=  Nrounds − 1 , Nmarkets − 1  (3.1)  where Nrounds is the number of rounds of release and Nmarkets is the total number of markets entered for each movie. The index gives the ratio of the number of extra rounds taken to the number of foreign markets entered. Thus, if the movie enters ten countries and takes ten rounds of release to do so, the fraction is one, and this movie is characterized by pure sequential entry. If the movie entered all ten countries in one round, the fraction would be zero, indicating pure simultaneous entry. Interior values indicate the degree to which the movie followed a sequential entry strategy. For example, consider a movie that entered five markets. If it did so in three rounds, the index is (3 − 1)/(5 − 1) = 0.5, reflecting the fact that a mix of simultaneous and sequential entry is observed. Figure 3.10 provides a histogram of the sequential index. It shows a large spike at one, reflecting the fact that more than one hundred of the movies exhibit pure sequential entry. Just twenty of the movies were released according to a pure simultaneous strategy (all within a month of the U.S. release). The remainder fall somewhat symmetrically around a value of one half, with a small spike around the 0.2 mark. We have established that movies with large production budgets tend to diffuse internationally more quickly. The longer delay for low-budget movies could be due to sequential entry, or it could occur if they are delaying all their foreign releases for some time, and then entering them all simultaneously. To investigate which movies are indeed entering sequentially, Figure 3.11 plots the sequential index against the production budget, along with a Lowess smoother. The figure confirms that lowbudget movies do in fact employ a greater degree of sequential entry than their big-  51  Figure 3.10: Histogram of Sequential Index  budget counterparts. Moreover, the effect is not driven by differences in budget by genre type, although comedies and dramas do show a higher propensity for sequential entry at any given budget level. This makes sense given their lower appeal in foreign markets. In addition to looking at which types of movies tend to be sequentially released, we can investigate the order in which countries tend to be entered. The simplest measure of this is the average round of release among a country’s imported movies. Figure 3.12 plots this value for each of the 13 foreign markets. Note that the four destinations with the lowest average round of entry are all English-speaking. Figure 3.13 plots the average round of entry against the correlation between the countries’ boxoffice revenues and those of the U.S.. On average, countries with closer agreement to the U.S. are entered in earlier rounds than countries with less agreement. Both the average rounds of entry and the box-office correlations with the U.S. could be affected by selection. As a robustness check, Figure 3.14 reproduces the plot where only information from the 202 movies that entered all 14 markets is used. The relationship between average round of entry and correlation with the U.S. persists, and appears  52  Figure 3.11: Sequential Index vs. Production Budget  stronger. Australia and New Zealand become earlier markets than Hong Kong for this sample, while France, Germany and Spain move up relative to the Netherlands and Norway. The data patterns illustrated in this section are consistent with the idea that firms that are uncertain about their export profitability may use sequential entry to learn and update expectations. The next section introduces a model that encompasses this idea and provides estimating equations to take to the data. Section 3.4 presents the methodology for operationalizing the model, and reports results of directly estimating the effect of a performance surprise on the probability of further entry.  3.3  Theory  The model draws on Albornoz et al. (2010), which posits a role for learning by manufacturing exporters. In that paper, there is uncertainty about the intercept of the linear demand and the marginal cost of exporting. In this paper, there is uncertainty about the quality of the movie. Consider a risk-neutral firm making entry decisions in  53  Figure 3.12: Average Round of Release by Country  Figure 3.13: Average Round of Release vs. Correlation with U.S. Box-Office  54  Figure 3.14: Average Round of Release vs. Correlation with U.S. Box-Office: Subsample of Movies Going to All Markets  K segmented markets. To enter any of the markets, indexed by d, firm m must incur a per-destination fixed cost of Fdm , corresponding to print and advertising costs. Movies are heterogeneous in their appeal, which is not directly observable even by their distributors. The appeal of any given movie also varies between markets, due to country-specific idiosyncracies in taste. Holloway (2010) introduces a discrete choice model in which revenues for different varieties of a product (e.g. different movies) depend on country- and variety-specific terms multiplicatively, in addition to the multiplicative idiosyncratic factor. The derivation is as follows: Individuals in country d purchase a variety of the product if their valuation of doing so is greater than the price, pd , which varies between countries but not within each country. Prices are taken as given in each market. The valuation of individual i from destination d consuming variety m is: vidm = k(β qm + ψdm +Uidm )n(yd ),  (3.2)  where qm is the “true” universal quality of the variety, ψdm is the country-variety 55  taste shock, Uidm is the individual’s idiosyncratic utility, yd is the income per capita in country d and the functions k(·) and n(·) are increasing and could be destinationcountry specific. The parameter β adjusts for the scale on which quality is measured. Valuation is separable in per capita income, reflecting the higher willingness-to-pay in rich countries for any given quality level. Revenues from exporting to country d are given as the product of the price and the number of people who purchase the variety. This latter quantity can be expressed as the product of the total population and the proportion of the public who purchase: Rdm = pd Md P[vidm > pd ],  (3.3)  where Md is the population of country d and the proportion of the purchasing public is replaced by the probability that any of the (symmetric) individuals in the country will purchase. Plugging 3.2 into 3.3, Rdm = pd Md P[k(β qm + ψdm +Uidm )n(yd ) > pd ] pd ] = pd Md P[β qm +Uidm + ψdm > k−1 n(yd ) pd = pd Md P[Uidm > k−1 − β qm − ψdm ] n(yd ) pd = pd Md (1 − P[Uidm < k−1 − β qm − ψdm ]) n(yd )  (3.4)  If Uidm is distributed exponentially with parameter λ , then the above reduces to: λ +β qm +ψdm −k−1  Rdm = pd Md e  pd n(yd )  λ −k−1  eqm  (3.5) pd n(yd )  Define the attractiveness of country d as Ad ≡ pd Md e and let Qm ≡ ψ dm and Ψdm ≡ e . We can then express the revenue equation succinctly as Rdm = Qβm Ad Ψdm .  (3.6)  Taking the logarithm of equation (3.6) gives the linear equation, rdm = β qm + ad + ψdm ,  56  (3.7)  where lower-case letters represent logarithmic terms and ψdm ∼ N(0, σψ2 ). Although the quality of the movie is not known ex ante, there are known imperfect proxies. A firm might make reasonable predictions about future revenues in each of the prospective markets by substituting the known proxies in for unknown quality, and using historical data to estimate country fixed effects—to substitute for ad —and the parameter β . That is, if the firms know the “law of revenues”, they can substitute in their quality proxies to make initial predictions about potential revenues in each of the markets. In particular, suppose that quality, qm , is a function of the logarithm of the movie’s budget, bm = ln Bm : qm = αbm + ξm ,  (3.8)  where ξm ∼ N(0, σξ2 ). Substituting equation (3.8) into (3.7), and replacing ad by a set of destination fixed effects gives6 rdm = β αbm + ad + νdm .  (3.9)  where νdm = β ξm + ψdm . Firms form beliefs for each market according to equation (3.9).7 The normality 2 ), assumptions on ξ and ψ—and thus on ν—imply a normal prior: rdm ∼ N(µdm1 , σd1 where µdm1 = β αbm + ad 2 = σν2d . σd1  (3.10) (3.11)  For clarity of exposition, let us assume there are only three destinations, A, B, and C. All movies enter market A but the firms can choose whether or not to enter B and C. After the first period, firms update their expectations about log revenues, rdm , in the remaining potential markets, d ∈ {B,C}, using realized revenues in market A. 2 ) with According to Bayes’ Law, rdm2 ∼ N(µdm2 , σdm2 6 By  an abuse of notation, I am calling the destination-specific constant (fixed effect), ad , which is not equal to the conceptual ln Ad . 7 That is, firms know the value of the compound parameter β α and the destination-specific constants.  57  σd1 (rAm − µAm1 ) σA1 2 2 = σd1 (1 − ρAd )  µdm2 = µdm1 + ρAd  (3.12)  2 σdm2  (3.13)  where ρAd is the correlation between νAm and νdm , σd1 is the square root of σν2d , and (rAm − µAm1 ) ≡ νAm is the difference between the realized and expected log revenues for movie m in country A. These Bayesian updating formulas provide intuition for how predictions in future potential markets depend on the surprises observed in entered markets. The surprises are tempered by the degree of correlation across the two countries, and the degree of variation within each of the countries. The posterior variance is always decreased after new information is attained, but again the amount of precision gained depends on the correlation between the markets involved.8 The model abstracts from the informational value of entering and is thus a model of “passive” learning. The movie will enter market d if the expected profit from doing so is positive. Recall that a fee of Fdm is required for movie m to enter market d. Assume that Fdm = Kd ζdm , where Kd is destination-specific but ζdm is moviedestination-specific and is unobservable to the econometrician, though known to the firms. Letting Edmt denote the indicator function for entry of m into d in period t, the probability that movie m will enter destination d is9 P[Edmt = 1] = P[Et [Rdm − Fdm ] > 0] = P[Et [Rdm ] > Kd ζdm ] = P[eµdmt +  2 σdmt 2  > Kd ζdm ]  2 σdmt > logKd + logζdm ] 2 σ2 = P[logζdm < µdmt + dmt − logKd ] 2 2 σdmt = P[logζdm < µdm,t−1 + sdm,t−1 + − logKd ], 2  = P[µdmt +  8 In  (3.14)  practice, movies enter more than one country per period. To aggregate the surprises in each of the entered markets, a matrix version of Bayes’ Law is required. It is introduced in section ??. 9 If  log(X) ∼ N(µ, σ 2 ) then X ∼ Log-N(µ, σ 2 ) and EX = e(µ+  58  σ2 ) 2  .  where sdm,t−1 is the update based on last period’s performance. Note that the variance terms become movie specific after the first period. This is because not all movies enter countries in the same order. Since the updated variance depends on the correlation coefficient between the country in consideration and the country of last entry, the variance terms will differ if the country of last entry differs. In the simple three-country example, country B for one movie may be country C for another.  3.4  Results  The model predicts that surprises in box-office revenue in previous markets affect the probability of entry into potential future markets. I test this prediction using regression analysis and consider alternative explanations for the results.  3.4.1  Firm Learning and the Decision to Enter  First it is necessary to construct the appropriate variables, in particular the update to movie m at time t. Recall from the model that, initially, distributors form expected revenues for each potential destination based on movie characteristics such as the budget, and destination characteristics such as the country’s historical expenditure on movies. I form ex ante predicted revenues for each movie-destination pair by regressing ex post actual (log) revenues on the movies’ (log) budget and destination fixed effects. I allow the coefficient on budget to differ across the destinations and augment the equation with interactions between country dummies and genre and MPAA-rating dummies: ln Rdm = αd ln Bm + {FEd } × {GENREm } + {FEd } × {MPAAm } + {FEd } + εdm . (3.15) 1 I then set the first-period predicted log revenues, µdm , equal to ln Rdm . Time periods are based on the month since the U.S. release. Although I have data on the precise day on which a movie was released in any given market, it is impractical to use days as the unit of time. Using daily time periods would introduce a lot of noise since there may be many idiosyncratic reasons for releasing on one day rather than the next. Recalling that the benefit to “pulling the plug” on a release is the saved fixed costs, the incentive to do so decreases as the period between learning  59  that the movie will not make money in the market and the release date narrows. This is because advertising costs are sunk once they are spent. Similarly, adding a new market based on good performance would take time to organize and promote. I set the unit of time to be a month (30 days), but robustness checks show the qualitative conclusions are unaffected by changing this window. For most movies, there are gaps in the month in which new entry occurs. For example, a movie might go to two markets in month one, three markets in month two, but then only enter its last market in the fifth month. According to the updating theory, there is no explanation for the delay in entry to the fifth month. The same information was available in the third and fourth months. Indeed, the model is about information sets and not time. Accordingly, I collapse the data to the level of information sets—or rounds of entry—rather than keep all possible months for each movie. Thus, for the hypothetical movie above, only observations corresponding to months one, two, and five would be kept. A final set of observations represents the round after the last new entry has taken place. It is important to include since it is informative that none of the remaining potential markets imported the movie in this last information set. This procedure highlights the fact that we are not trying to explain the magnitude of the delay to foreign release, but rather to test whether new information affects the decision to release. I compute the expected revenues and surprises for each destination-movie-round triple using an iterative procedure. The updating equations of section 3.3 apply if only one country is entered per period. In practice, many movies enter multiple countries per round and the surprises from each entered country must be aggregated to form the update for each remaining potential market. To do this we can employ the matrix versions of the Bayesian updating equations. Denote the set of countries entered in period t − 1 by Y and the set of remaining potential destinations X.10 The updating equations become: t−1 µXt = µXt−1 + Σt−1 XY ΣYY  ΣtXX =  −1  t−1 t−1 −1 Σt−1 XX − ΣXY ΣYY  10 These  rY − µYt−1  (3.16)  Σt−1 , XY  (3.17)  sets of course depend on the movie, m, and the period, t, but the subscripts are omitted for convenience of exposition. Note that the set of destinations entered before t − 1 is irrelevant to the calculations since information from these entries is already incorporated into the t − 1 prior.  60  where µXt and µYt are vectors of predicted log revenues going into period t for the sets X and Y , respectively, rY is the vector of realized log revenues in Y , ΣtXX and t ΣYY are variance-covariance matrices, and Σt−1 XY is a cross-covariance matrix. All initial variance and covariance elements are calculated from the residuals, εdm , from equation (3.15). Table 3.2 provides correlation coefficients of εdm for each country pair. On the main diagonal, the variance of the residuals within each country is reported. The upper triangle reports Pearson correlation coefficients, which describe the strength of the linear relationships, and are directly related to the covariances between countries. The country-pair with the highest correlation is Australia–New Zealand, at 0.768, followed by Australia–United Kingdom (0.744) and Netherlands–Norway (0.743). The country-pair with the lowest correlation is France–United States, at 0.396. The next three lowest correlations also involve the United States, paired with Spain and Italy (each at 0.402) and Japan (0.414). In general, the correlations point to regional and colonial groupings: there are high correlations among Northern European and North American markets (U.S. statistics include box-office revenue in Canada); Mediterranean European countries exhibit high correlation among themselves; the market in most agreement with Japan is Hong Kong; Argentina’s ties to Spain and Italy are reflected, although its second highest correlation is surprisingly with Hong Kong. The lower triangle reports Kendall’s tau, which provides a non-parametric measure of concordance of the ranking of movies for each country pair. The same general patterns of association are uncovered through this alternative measure.  61  Table 3.2: Correlations of Residuals between Markets  62  ARG  ARG 1.32  AUS CZE DEU ESP FRA GBR HKG ITA JPN NLD NOR NZL USA 0.526 0.537 0.554 0.660 0.557 0.496 0.592 0.583 0.448 0.518 0.550 0.511 0.475  AUS  0.398  2.53  0.507  0.703  0.633  0.628  0.744  0.532  0.521  0.452  0.698  0.630  0.768  0.635  CZE  0.385  0.415  1.37  0.619  0.492  0.554  0.585  0.583  0.515  0.440  0.572  0.618  0.554  0.479  DEU  0.403  0.573  0.472 2.74  0.715  0.701  0.719  0.528  0.665  0.510  0.695  0.677  0.647  0.546  ESP  0.478  0.474  0.363  0.523  2.60  0.736  0.613  0.487  0.707  0.460  0.589  0.576  0.477  0.402  FRA  0.416  0.489  0.416  0.541  0.535 2.98  0.608  0.510  0.730  0.502  0.645  0.653  0.560  0.396  GBR  0.387  0.625  0.444  0.575  0.455  0.514 2.40  0.540  0.559  0.513  0.714  0.666  0.733  0.614  HKG 0.430  0.380  0.412  0.383  0.335  0.362  0.401  1.27  0.459  0.591  0.533  0.523  0.530  0.445  ITA  0.447  0.428  0.399  0.500  0.565  0.584  0.450  0.338 3.28  0.468  0.556  0.599  0.497  0.402  JPN  0.347  0.357  0.356  0.382  0.337  0.385  0.373  0.419  0.372 2.26  0.441  0.454  0.469  0.414  NLD  0.400  0.541  0.459  0.569  0.444  0.507  0.563  0.393  0.431  0.365 1.84  0.743  0.656  0.629  NOR  0.415  0.500  0.480  0.521  0.421  0.487  0.527  0.397  0.445  0.366  0.569 2.01  0.637  0.584  NZL  0.372  0.638  0.431  0.491  0.368  0.451  0.560  0.398  0.380  0.359  0.491  0.496 1.13  USA  0.334  0.540  0.365  0.455  0.337  0.358  0.502  0.349  0.330  0.341  0.488  0.439  Pearson correlation coefficients on upper triangle; Kendall’s tau on lower triangle; variances on main diagonal  0.582  0.469 1.48  To provide a sense of how surprises in one market translate into updates in other markets, consider the following example. Among all first-round releases that outperform expectations, the median multiplicative surprise is 2.42. This means that for releases with “positive” surprises, the median release’s box-office revenue was 2.42 times the predicted revenue. The median box-office revenue in the first round is $5.7M. Taking the ratio of the two gives a typical predicted revenue of $5.7M/2.42 = $2.36M, leaving a typical “surprise” in absolute terms of $5.7M − $2.36M = $3.34M. Following are sample calculations for the multiplicative updates in United Kingdom (GBR), Spain (ESP), and Japan (JPN) from a $3.34M surprise in Australia (AUS), assuming a realized revenue of $5.7M (the calculation involves the multiplicative factor; the other numbers simply provide context):  Raus σd · ρd,aus · ln σaus Predaus √ 2.40 Ugbr = exp √ · 0.744 · ln(2.42) 2.53 √ 2.60 Uesp = exp √ · 0.633 · ln(2.42) 2.53 √ 2.26 U jpn = exp √ · 0.452 · ln(2.42) 2.53 Ud = exp  = 1.90 = 1.76 = 1.46  The predicted revenue would increase by a factor between 1.5 and 2 in each of the countries. The update increases less than proportionately with the correlation coefficients. This implies that, while more emphasis is placed on abnormal performances in highly-correlated countries, the variation of updates across countries is somewhat more compressed than the variation in correlation coefficients. Table 3.3 reports the main result of the study. Each specification estimates the probability that a movie enters a destination in a given round, conditional on the movie not being released there previously. The table reports standardized marginal effects at the mean, so that it presents the change in the probability of release induced by a one-standard-deviation increase of the variable in question. The first column estimates the degree to which current expected revenue affects the decision to release.  63  The first round of releases is excluded from the regression because this specification acts as a benchmark for the other columns, which include lagged variables. The coefficient implies that a one-standard-deviation increase in the (log) predicted revenue increases the probability of entry in the current period by 14.8 percentage points, compared to an average probability of 21.3%. As predicted, expected revenue makes a big difference in the decision to release a movie in a given country. The second specification examines the constituent parts of the expected revenue, namely the expected log revenue in the previous round plus the update from the previous period. If firms do not adapt their entry strategies based on information learned in period (t − 1) then we should not expect the coefficient on the update to be significant. In fact, the coefficient implies that a one-standard-deviation increase in the previous round’s update is associated with a 5.0 percentage-point increase in the probability of entry. This is an increase of more than 23% over the average probability of entry. Column 3 includes interactions between the update and dummies for the first period and all other periods. This specification checks whether firms are learning only after the first round of entry, or whether subsequent entries also affect entry decisions. The coefficients on the interactions are nearly identical, suggesting that learning is ongoing. Columns 4 and 5 investigate whether the effect of a surprise in a movie’s performance depends on whether the surprise is positive or negative. Column 4 indicates that the increase in the probability of entry due to positive news is more than twice as large in magnitude as the decrease in the probability due to negative news. This is likely due to how the expected revenues are distributed around the entry cutoffs. The result suggests that there is a larger mass of expected revenues within one standard deviation of update below the cutoffs than there is above. Column 5 breaks down the effect of positive and negative updates by quartile of lagged expected log revenue. Negative updates become more salient as the quartile increases. In fact, observations in the first quartile are unaffected by negative updates. Since these observations are unlikely to be associated with a release at all, the negative news does not have an impact. Positive updates have the greatest salience for observations in the middle quartiles. It is in this range that surprise good performances are most likely to push expectations above the entry thresholds.  64  Table 3.3: Probability of Exporting to a New Market model depvar pred. ln R  (1) released 0.148∗∗∗ (0.006)  (2) released  (3) released  (4) released  lag pred. ln R  0.144∗∗∗ (0.006)  0.144∗∗∗ (0.006)  0.148∗∗∗ (0.006)  lag update  0.050∗∗∗ (0.003)  per. 2 lag update  0.0514∗∗∗ (0.004)  per. > 2 lag update  0.0489∗∗∗ (0.005)  negative update  0.0321∗∗∗ (0.004)  positive update  0.0730∗∗∗ (0.007)  (5) released  -0.191∗ (0.080) lag pred. Q2 (d) -0.0624 (0.080) lag pred. Q3 (d) 0.0604 (0.080) lag pred. Q4 (d) 0.168∗ (0.080) neg. update × pred Q1 0.0089 (0.012) neg. update × pred Q2 0.0372∗∗∗ (0.009) neg. update × pred Q3 0.0402∗∗∗ (0.007) neg. update × pred Q4 0.0474∗∗∗ (0.009) pos. update × pred Q1 0.0501∗∗∗ (0.010) pos. update × pred Q2 0.0695∗∗∗ (0.010) pos. update × pred Q3 0.0671∗∗∗ (0.010) pos. update × pred Q4 0.0542∗∗∗ (0.012) N 24251 24251 24151 24251 24151 pseudo R2 0.129 0.131 0.131 0.132 0.121 Standardized average partial effects; robust standard errors are adjusted for clusters in movies (d) for discrete change of dummy variable from 0 to 1 ∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001 lag pred. Q1 (d)  65  3.4.2  Alternative Models  There is an alternative explanation for the main result that “surprise” performances affect further entry. It is probable that firms have information about the quality of their movies that is not captured by the first-stage regression of equation (3.15). In the extreme, they could know the quality perfectly, in which case any deviation from their expectations would be entirely due to idiosyncratic moviedestination demand shocks. Movies with seemingly big positive surprises would enter more countries in subsequent periods because they are good movies. Distributors would know this from the start and could be delaying for reasons other than learning. The methodology of this paper would erroneously attribute the correlation between “surprises” and entry to learning. To see whether this is driving the results, we can use the fact that this alternative hypothesis implies that firms can anticipate the surprises from future rounds. If no learning was taking place, substituting the update from the current period (which isn’t observed before current-period entry decisions are made) should produce similar results to including the lagged update. If the significance of the lagged update is due entirely to learning, then the current-round update should not enter significantly. The first column of Table 3.4 estimates the effect of current expected log revenues on the probability of entry. The difference from column 1 of Table 3.3 is that first-round observations are included in the regression. This is to act as a benchmark for the specification of column 2, which includes the current predicted log revenue and the update derived from current entries. Column 2 indicates that a one-stand-deviation increase in the current update increases the probability of entry by 0.85 percentage points. This suggests that firms do have some information not accounted for in the initial forecast equation, but the estimated effect is about one-sixth of the estimate for lagged updates, which is reproduced in column 3 for convenience. The fact that the effect of lagged updates is so much stronger than current (unrealized) updates suggests that we should not abandon the learning hypothesis. In column 4, the variance of the prior distribution of log revenues is included. Recall from equation (3.14) of section 3.3 that we expect the variance to enter positively. This is because the logarithm of the expected revenue is the expected log revenue plus one half the variance of log revenue if revenue is distributed log-normally. Thus the combination of the assumption of risk-neutral preferences on the part of firms and log-normally distributed conditional revenues provides the hypothesis that 66  Table 3.4: Probability of Exporting to a New Market model depvar pred. ln R  (1) released 0.166∗∗∗ (0.006)  (2) released 0.166∗∗∗ (0.006)  (3) released  (4) released  lag pred. ln R  0.144∗∗∗ (0.006)  0.120∗∗∗ (0.005)  lag update  0.0504∗∗∗ (0.003)  0.0422∗∗∗ (0.003)  24251 0.131  -0.165∗∗∗ (0.010) 24251 0.150  0.00852∗∗ (0.003)  current update  variance ln R N pseudo R2  34144 0.134  34139 0.135  Standardized average partial effects Robust standard errors adjusted for clusters in movies ∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001  the variance term should enter positively. Column 4 of Table 3.4 shows that high-variance observations are actually less likely to be associated with entry. This suggests that firms may in fact be risk averse. Goettler and Leslie (2004) note that industry-insiders claim to treat risky movies differently in their study of cofinancing in the motion picture industry. In that paper, it is the studios’ decisions of how to finance the movies’ production that is at issue. The authors point out that risk-averse behaviour is not expected for publically-owned firms, since shareholders can diversify risk through their portfolio. Lambart (1986) argues that risk-averse behaviour could be explained by agency issues with riskaverse managers. Adding risk aversion to the present model would only strengthen the desire of distributors to learn their movie’s quality before committing to foreign entry. I therefore leave such considerations outside the model. Furthermore, recall that the model of section 3.3 does not explicitly take into account the informational value of entry. Entering a high-variance country would be less valuable in this respect  67  because the firm could not infer as much about quality as it could for a low-variance country. This is because surprises could be due to large idiosyncratic shocks rather than high or low quality. Thus, the informational component of the value of entry would lead to firms favouring low-variance markets, as the empirical results suggest.  3.5  Conclusion  A small but growing body of work has suggested that manufacturing firms learn about their export profitability through exporting. This chapter adds to that literature by considering a new type of product. Motion picture distributors can only release movies once in any destination country, and have no scope for intra-market learning. On the other hand, potentially large idiosyncratic differences in taste across markets mean distributors can not infer too much from any given success or failure. Nonetheless, I show that distributors do appear to adjust their entry strategies based on prior-market performance. The correlation between past “surprises” and entry decisions could be due to omitted factors in the initial forecasts, but results show that while this is likely a factor, unrealized surprises have a much smaller effect on entry decisions than past surprises, leading to the conclusion that learning is indeed taking place. As firms move toward the “day and date” simultaneous release strategy to fight international piracy, they lose the ability to use the information from prior markets, adding to the cost of piracy.  68  Chapter 4 Distance and Border Effects for Canadian Trade in Services 4.1  Introduction  In 2006, the governments of British Columbia and Alberta signed the Trade, Investment, and Labour Mobility Agreement (TILMA), a stronger, more rules-based follow-up to the 1994 pan-Canadian Agreement on Internal Trade (AIT). Both agreements were designed to foster deeper provincial integration, with particular emphasis on the harmonization of standards and regulations. The AIT has been described as too conservative in its language, without providing proper recourse in the event of non-compliance.1 In this respect, the TILMA is more ambitious and more controversial. It calls for increasing harmonization and mutual recognition on the part of the B.C. and Alberta governments, and allows for fines of up to $5,000, to be awarded by a special panel. The TILMA explicitly targets barriers to trade in services; it contains an appendix of professions for which greater harmonization in licensing and standards is required. Not surprisingly, the new agreement has opened old debates as to the extent of the barriers to interprovincial trade. Critics ask whether such an all-encompassing agreement is warranted if barriers are small and limited in scope. But supporters of the TILMA argue that past approaches to integrate the provincial economies have failed, and claim billions of dollars to be at stake. Interestingly, there has been little 1 See  Trebilcock and Schwanen (1995), for example.  69  empirical research on the question of barriers to service trade. In section 4.4, I employ provincial service trade data to estimate the effect of crossing provincial borders on the volume of service transactions. Helliwell and Verdier (2001) estimate border effects for interprovincial merchadise trade. They find small effects for the Atlantic provinces and Saskatchewan, but insignificant border effects for the four largest provinces. I find significant effects for all provinces, but similarly find the economically smaller provinces to incur larger effects. I apply the fixed-effects specification advocated by Feenstra (2002), which may correct for possible biases in the traditional gravity model. I find the average provincial border effect to be about 18, suggesting that, all else equal, a province trades with itself 18 times more than it does with other provinces. Hillberry and Hummels (2008) analyse shipments within the United States at the zip-code level, and find that distance has a very large attenuation effect over small distances. Accounting for this nonlinear effect of distance is enough to fully explain the state border effect. I apply their estimates of small-distance attenuation to Canadian provincial trade to investigate whether they can explain the provincial border effect. Advances in information and communications technology (ICT) may account for much of the growth in international service trade. To the extent that modern business practices increasingly exploit this technology, we may expect the effect of distance on trade flows to diminish over time. But if the physical transport of service products is literally responsible for the decline in trade flows over distance, then e-commerce should render the service-trading world truly flat. Geographic distance is a proxy not only for transport costs, however, but also for informational shortcomings. Distant places may be separated by language, culture, and tastes, all of which could contribute to the attenuation of bilateral transactions. In their study of distance effects for internet commerce, Blum and Goldfarb (2006) find that for taste-dependent digital products, such as music and games, distance is important, but for non-tastedependent products, like software, the distance effect is insignificant. These findings reinforce the notion that distance plays an informational role. The degree to which the off-shoring of services is a threat to Canadian jobs depends in large part on the ability of firms to circumvent the disadvantages of distant provision. In section 4.5, I analyse the evolution of the distance effect for Canadian commercial service trade with fifty-nine trading partners. Contrary to the theory that the increasing use of information and communications technology has rendered distance  70  impotent, I find a stable and significant distance effect. At the national level, Canada has played an important role in the literature on merchandise-trade border effects, in part because of the availability of good-quality provincial data. McCallum (1995) uses data on trade between provinces and states to estimate a large (Canadian) border effect for Canada-U.S. trade. Helliwell (1996) examines the effect of the Quebec-U.S. border. Anderson and van Wincoop (2003) use Canada-U.S. data to demonstrate their critique of traditional gravity-style models. Service-trade barriers have been much less extensively analysed. Helliwell (1998) estimates border effects for trade in services between Canada and the U.S. indirectly. First, he computes the ratio of interprovincial to international trade in services and goods separately. Then he takes the ratio of the two ratios to form a measure of the relative domestic focus of service trade vis-`a-vis merchandise trade. Multiplying this measure by the estimated merchandise-trade border effect, Helliwell obtains a crude estimate of the service-trade national border effect. In section 4.6, I combine the provincial and international data sets to estimate the effect of Canada’s national border on the provision of services. The result suggests that Canada trades about 3.5 times more within its border than across it. Before getting to the empirical analysis, I present motivating theory behind my estimation specifications in section 4.2, and a description of the data in section 4.3. Section 4.7 concludes.  4.2  Theory  The gravity model for trade in goods began as a convenient and successful empirical specification (Deardoff, 1984). Tinbergen (1962) and Linnemann (1966) are early applications, but since then a multitude of micro-foundations has emerged. Anderson (1979) began the work on theoretical underpinnings, which was subsequently buttressed by Bergstrand (1985, 1989), building on the ‘new trade theory’ of Helpman and Krugman (1985). An equivalent specification was developed by Deardoff (1995) from a Heckscher-Ohlin framework. Eaton and Kortum (2002) develop a Ricardian model that incorporates geography and technology as determinants for trade. Additional work has looked into issues of appropriate econometric specification based on the theory. Anderson and van Wincoop (2003) point to possible missing-variable bias associated with often over-looked ‘multi-lateral resistance’ terms. Feenstra (2002)  71  advocates for the use of country fixed-effects to control for these unobservables, a practice anticipated by Harrigan (1996). Trade in services is inherently different than trade in goods, particularly due to the immediate-consumption nature of many service products. Common to all theoretical developments of the merchandise-trade gravity equation is the assumption that labour is immobile across countries/regions. With service trade, it is precisely this labour that is being traded, not embodied in goods a` la Vanek (1968), but directly. Head, Mayer and Ries (2009) develop micro-foundations for a gravity equation for service trade. Their derivation builds on Eaton and Kortum (2002), and follows the exposition of the Head-Ries (2008) model of bilateral FDI stocks. The model posits that No “candidates” in the origin country compete for Sd service positions in the destination. The productivity of each candidate, zo , is an independent draw from a Fr´echet distribution: F(z) = exp(−z/κo )−θ , where θ is an inverse measure of productivity variation and κo is a location parameter. The model assumes that the position goes to the candidate that can perform the service for the lowest unit-cost, inclusive of delivery. Denoting the value of exports from o to d by Vod , they formulate an equation of the form: Vod = exp[FXo + FMd − θ ln(1 + τod )] (4.1) where τod refers to the productivity-adjusted cost in service hours of transporting the service from o to d. This formulation of transport costs is analogous to the conventional “iceberg” assumption for trade in goods, and τod is presumed to be increasing in distance. FXo and FMd are exporter and importer fixed-effects, which stand in for country-specific unobservables such as size (No and Sd ) and wage rates.2 Taking the logarithm of equation (4.1) provides the familiar linear-in-logs gravity specification.  4.3  Data  The source for interprovincial trade data is Statistics Canada’s provincial inputoutput tables.3 The data is constructed from a variety of sources, such as surveys and administrative statistics. The primary method is to ask producers where their goods and services are sold. 2 See 3 The  Head, Mayer and Ries (2009) for details. following description draws on G´en´ereux and Langen (2002).  72  I employ data for trade in manufacturing goods as a comparison to the service trade results. Manufacturing data comes primarily from the Annual Survey of Manufacturers. The information is compiled at the establishment level, and the resultant trade patterns are applied to all commodities shipped by the establishment. The obvious limitation of this approach is that where a manufacturer ships multiple goods, the trade patterns are constrained not to vary across goods. To alleviate this problem, domestic supply and demand controls, as well as wholesale activities are used to adjust the figures. Another important limitation is the question of whether the shipment destination is the location of consumption. To the extent that shipping networks distort the true destination of consumption, measured trade patterns will not reflect economic reality. To mitigate this problem, the Wholesale Trade Commodity Survey by Origin and Destination is used. This survey provides information on the origin and destination of sales of wholesalers, and therefore acts as a bridge between first and final destination. Service trade data is divided into the following categories: Transportation and Storage, Communications, Professional, Financial, Wholesale Margins, Retail Margins, and Travel. Transportation and Storage reflects the value of transportation services of moving people and goods between provinces, and is primarily provided by carriers. Grain storage is based on the flows of the major grain commodities. Communications covers telecommunications, postal services, radio and television broadcasting, and cable programming. Professional services includes sectors ranging from architectural, to computers, to travel agents to advertising. Financial services includes banking, insurance and real-estate. Wholesale margins are derived from the Wholesale Trade Commodity Survey by Origin and Destination, while Retail margins are based on a survey providing the proportion of out-of-province sales to smalland medium-sized retailers. Interprovincial trade in travel and tourism is provided by the Canadian Travel Survey, which contains information on the province of origin of travelers, and the province of expenditures for various travel-related services. When international and interprovincial trade flows have been estimated, they are entered into a structural accounting framework in order to reconcile trade patterns with provincial domestic supply and demand. The following three accounting identities are central to the framework: (i) In each province and for each commodity, total domestic supply must equal the sum of international exports, interprovincial exports, and domestic sales; (ii) In each province and for each commodity, total domestic  73  demand must equal the sum of international imports, interprovincial imports, and domestic sales; and (iii) The sum of international exports and imports by province must equal their national counterparts. Canada’s international service-trade data is broken into the categories Government, Travel, Transportation, and Commercial Services. I focus on Commercial Services because this category relates to normal business activity that is the focus of trade agreements and off-shoring debates. The data set covers the period 1993-2004, and gives trade volumes for Canadian imports and exports (payments and receipts) with fifty-nine trading partners. Data for interprovincial and international trade are obtained from CANSIM, for the years 1997-2004, and 1993-2004, respectively. Provincial internal distance measures are taken from Helliwell and Verdier (2001), and distances between provinces are measured as the great-circle distance between each of the capital cities. Populationweighted distances between Canada and its trading partners were generously provided by John Ries, along with other bilateral cultural/institutional variables employed in the international analyses of sections 4.5 and 4.6. Before applying any econometric machinery, it is worthwhile to have a look at the data. Figure 4.1 plots GDP-deflated commercial service trade against distance. The variable “commercial services” is constructed by adding the totals for the service categories Financial, Professional, and Communications. I then construct the GDPdeflated measure by averaging the ratio of bilateral trade in commercial services to the product of each partner’s GDP over the years of the sample. The averaging simply serves to unclutter the plot; scaling by GDP controls for economic size. The resulting figure supports the notion that trade falls with distance. Intra-provincial trade is larger than would be predicted by a linear extrapolation of cross-border trade, suggesting a significant border effect.  4.4  Interprovincial Trade  The primary aim of this section is to estimate the degree to which commercial service trade within Canada is constrained by provincial borders. There is no single category for commercial services in the provincial data, so I aggregate three categories that coincide with those services included in the international commercial services data: financial, communications, and professional services.  74  Figure 4.1: Anticipating Distance and Border Effects  Recalling that τod represents trade costs, I follow the conventional assumption that ln(1 + τod ) is linear in log geographic distance, lnDod , and add a dummy variable, INTRAod , that takes the value one when o = d. Thus INTRAod turns on only for intra-provincial trade. Adding an error term, uod , to account for possible omitted variables (assumed to be independent of distance and the distribution of provincial borders) yields: ln(1 + τod ) = δ ln Dod + β INTRAod + uod  (4.2)  Since I am using a panel data set, province-specific fixed effects are time-varying. I make this adjustment to (4.1), and add a vector of year dummies. Substituting (4.2) into the modified version of (4.1) yields: Vodt = exp[FXot + FMdt + YEARt − θ δ ln Dod + θ β INTRAod + θ uodt ] = exp[FXot + FMdt + YEARt − θ δ ln Dod + θ β INTRAod ]ηodt  (4.3)  where ηodt ≡ exp(θ uod ) has an expectation of one. The estimating equation can then 75  be obtained by taking logs of both sides of (4.3): lnVodt = FXot + FMdt + YEARt − θ δ ln Dod + γINTRAod + ln ηodt .  (4.4)  The coefficient of interest is γ ≡ θ β . Note that eγ quantifies how many more times an average province trades with itself relative to another province, all else equal. I refer to this measure as the (provincial) ‘border effect’. If provincial borders had no effect on trade, the border effect would equal unity. Equation (4.4) is the base model. Table 4.1 displays results from four regression models. Columns 1 and 2 estimate ‘na¨ıve’ gravity equations (equation (4.5) below) for commercial services and goods, respectively. The na¨ıve specification simply posits a log-linear relationship between trade flows and distance, GDP and per capita GDP of each trading pair. Some authors include population in place of per capita GDP, but this does not affect the fit of the model since the same information is included. I append the equation with year dummies to account for any year-specific shocks. Column 3 corresponds with the fixed-effects model of equation (4.4), while column 4 estimates the same model for trade in goods. lnVodt = α ln Dod + β1 ln GDPot + β2 ln GDPdt + β3 ln GDPPCot + β4 ln GDPPCdt +γINTRAod + YEARt + µodt .  (4.5)  Table 4.2 similarly displays service trade results for both na¨ıve (column 1) and fixed-effects (column 2) models. In both specifications the intraprovincial trade dummy has been interacted with provincial dummies, allowing for province-specific border effects. As we can see in Figure 4.2, the economically smaller provinces tend to suffer from larger border effects. This phenomenon is predicted by Anderson and van Wincoop (2003), who prove the result as a consequence of their theory. It is an intuitive result that can be explained as follows. Suppose region A has an output of 10 and region B has an output of 90. Ignoring any distance effects, in the absence of a border friction region A would export 9 to region B and keep 1 for itself and region B would export 9 to region A and keep 81 for itself. Now suppose the border impedes trade by a fraction 1/9. Then region A exports 8 and keeps 2 and region B exports 8 and keeps 82. The ‘border effect’ is defined as the multiple by which a region trades with itself more than would be expected absent the border friction. For region A this  76  Table 4.1: Provincial Border and Distance Effects  Model : Depvar: intcpt  (1) (2) lcomserv lgoods 3.78 -0.28 (2.34) (2.07) intra 3.13*** 1.31*** (0.11) (0.11) distance -0.73*** -1.15*** (0.03) (0.03) origin GDP 1.11*** 0.84*** (0.03) (0.03) destination GDP 0.66*** 0.77*** (0.03) (0.02) origin GDP per capita -0.51*** 0.91*** (0.19) (0.15) destination GDP per capita 0.60*** 0.19 (0.16) (0.17) N 672 672 R2 0.918 0.918 RMSE .652 .592  (3) lcomserv 23.24*** (0.33) 2.92*** (0.12) -0.81*** (0.03)  (4) lgoods 28.42*** (0.37) 1.13*** (0.10) -1.22*** (0.03)  672 0.951 .569  672 0.937 .582  Note: Statistical significance of 1%, 5%, and 10% are indicated respectively by ***, **, and *.  is 2/1 = 2. For region B it is 82/81 ≈ 1.012. Since trade is a smaller part of region B’s economy, the border friction does not make as much of a difference. Of course this fact is well-known to trade negotiators from small and large countries.  4.4.1  Alternative Specifications  There is a potentially important caveat to these estimates of provincial border effects. The problem lies in the nature of service products and the relationship between small distances and the geography of borders. To make the point concrete, consider the real estate sector. Many real estate transactions occur between people living in the same city. A family may outgrow its condominium and decide to buy a house, for example. For this type of transaction, it is likely that the family would retain a local real estate agent, who knows the local market. This is an example of an extreme distance effect, since it is the fact that the local real estate agent is close-by and has local 77  Table 4.2: Province-Specific Border Effects  Model : Depvar: intcpt  (7) lcomserv 2.09 (2.44) BC 2.44*** (0.09) ALTA 2.20*** (0.12) SASK 3.44*** (0.08) MAN 3.29*** (0.08) ONT 1.46*** (0.11) QUE 1.93*** (0.11) NB 3.65*** (0.10) NS 3.64*** (0.09) PEI 4.47*** (0.14) NFLD 4.27*** (0.12) distance -0.77*** (0.03) origin GDP 1.21*** (0.03) destination GDP 0.75*** (0.03) origin GDP per capita -0.53*** (0.19) destination GDP per capita 0.60*** (0.16) N 672 R2 0.934 RMSE .587  (8) lcomserv 25.06*** (0.31) 1.71*** (0.14) 2.12*** (0.12) 3.49*** (0.08) 2.94*** (0.10) 0.49*** (0.09) 2.74*** (0.11) 3.42*** (0.15) 3.65*** (0.08) 4.26*** (0.17) 3.28*** (0.12) -0.88*** (0.03)  672 0.97 .45  Note: Statistical significance of 1%, 5%, and 10% are indicated respectively by ***, **, and *.  78  Province‐Specific Border Effects vs. GDP  Provincial Border Effect  80 PEI  70 60 50  NS  40  NB 30 Nfld 20  10  Sask Man  Quebec  BC  Ontario  Alberta  0 0  100000  200000  300000  400000  500000  GDP in Millions CDN$  Figure 4.2: Province-Specific Border Effects vs. GDP  information that drives the decision. Whether provincial regulations deny access to out-of-province agents is largely immaterial. The problem is that if many of these transactions occur in cities within the provinces, all of the commission fees will add to the within-province service trade values. This will tend to push up the estimate of provincial border effects even though it is really a consequence of distance. Put another way, the linear-in-logs specification implies a constant elasticity of trade with respect to distance, but the elasticity may be greater at shorter distances. To disentangle the effects of distance and borders, it is necessary to have point-topoint data on trade within provinces. With data on trade between cities we could be confident that each transaction is indeed occurring over the given distance. With the provincial data, an average distance is given for transactions within each province,  79  600000  whereas the actual distances are far from homogeneous. While I don’t have the data to make this correction for provincial trade, I can apply the analogous method of using provincial and international service trade data to estimate the national border effect. I do this in section 4.6. Hillberry and Hummels (2008) use data on shipments at the zip-code level in the United States, and find distance to matter much more for small distances. They find the state border effect loses significance when they estimate a quadratic-in-logs specification. In this subsection I estimate the provincial-trade gravity equation allowing for quadratic and cubic terms in log distance. Table 4.3 displays results for the OLS specification with province-year fixed effects. Columns 1 repeats the linear-in-logs specification, while columns 2 and 3 allow for second- and third-order polynomials in log distance, respectively. The coeffcient on the variable intra (a dummy for intraprovincial trade) drops from 2.924 to 2.167 when log distance enters quadratically. This implies the estimated border effect is more than cut in half, from 18.6 to 8.7. The cubic specification results in an estimated border effect of exp(2.29) = 9.9, and is not statistically different from the estimate of column 2. These results suggest that restricting the elasticity of trade with respect to distance to be constant for all distances may give misleading estimates of other variables of interest, in this case, the provincial border effect. Santos Silva and Tenreyro (2006) point out that the standard linear-in-logs specification of the gravity equation may be biased if the error term is heteroskastic. They advocate for estimation in multiplicative form, using the poisson pseudo-maximum likelihood (PPML) estimator. Table 4.4 reports results for linear, quadratic, and cubic functions of log distance using the PPML estimator, and Table 4.5 reports analogous results for the gamma pseudo-maximum likelihood estimator (GPML). For PPML the conditional variance is assumed to be proportional to the conditional mean; for GPML the conditional variance is proportional to the square of the conditional mean. Figure 4.3 plots the coefficient on the intra-provincial trade dummy for each specification. We see that for the linear specification, the implied border effect is slightly smaller under PPML. More striking, the coefficient drops considerably once the quadratic log distance term is entered, but is not affected much by the addition of the cubic term.4 The Gamma PML estimates are insignificantly different from those 4 Quadratic  specifications naturally imply a turning point, after which predicted trade is increasing in distance. For the given estimates, the turning points are very low: about 2,500 Km (for PPML) and  80  Table 4.3: Nonlinear Distance Effects—OLS Estimates  Model : Depvar: intcpt intra ldistance  (1) lcomserv 23.237*** (0.331) 2.924*** (0.116) -0.808*** (0.029)  (2) lcomserv 33.235*** (1.224) 2.167*** (0.152) -3.623*** (0.328) 0.197*** (0.023)  672 yes 0.951 .569  672 yes 0.956 .539  ldistance squared ldistance cubed N Province-Year Fixed Effects R2 RMSE  (3) lcomserv 48.418*** (3.388) 2.292*** (0.143) -11.344*** (1.658) 1.461*** (0.267) -0.067*** (0.014) 672 yes 0.958 .524  Note: Statistical significance of 1% indicated by ***. Robust standard errors in parantheses.  of OLS. The Poisson PML estimate with the third degree polynomial in log distance implies a border effect of just 2.59 (i.e. provinces trade 2.59 times more internally than with other provinces, all else equal). This is vastly smaller than the figure of 18.3, estimated from OLS with linear log distance. Mathematically, the border effect measures the degree to which observed trade deviates from predicted trade, controlling for distance (and importer and exporter fixed effects). Thus, the border effect estimate is very sensitive to how distances are measured. Head and Mayer (2010) review many of the approaches taken to approximate internal distance, and suggest another that takes into account the effect of distance on trade internally. Given that no “true” internal distance exists, it may be interesting to turn the question around and ask: Given the actual trade values, what internal distances would be consistent with no border effect? The results could then be compared to geography-based measures of internal distance, like those of about 8,000 Km (for OLS and GPML). Adding the cubic term helps to solve this problem. None of the three estimates with cubic log distance contain regions in which trade is increasing in distance.  81  Table 4.4: Nonlinear Distance Effects—Poisson PML Estimates  Model : Depvar: intcpt intra ldistance  (1) comserv 22.078*** (0.528) 2.748*** (0.167) -0.431*** (0.067)  (2) comserv 46.282*** (2.086) 1.094*** (0.222) -7.103*** (0.549) 0.460*** (0.037)  672 yes -5.74e+10  672 yes -4.62e+10  ldistance squared ldistance cubed N Province-Year Fixed Effects Log pseudolikelihood  (3) comserv 106.031*** (8.307) 1.052*** (0.226) -34.842*** (3.631) 4.691*** (0.533) -0.212*** (0.026) 672 yes -4.03e+10  Note: Statistical significance of 1% indicated by ***. Robust standard errors in parantheses.  Table 4.5: Nonlinear Distance Effects—Gamma PML Estimates  Model : Depvar: intcpt intra ldistance  (1) comserv 23.451*** (0.239) 3.137*** (0.083) -0.786*** (0.028)  (2) comserv 33.520*** (1.022) 2.316*** (0.126) -3.586*** (0.276) 0.195*** (0.019)  672 yes -13617.56  672 yes -13609.66  ldistance squared ldistance cubed N Province-Year Fixed Effects Log pseudolikelihood  Note: Statistical significance of 1% indicated by ***. Robust standard errors in parantheses.  82  (3) comserv 40.639*** (2.813) 2.369*** (0.124) -7.212*** (1.364) 0.789*** (0.220) -0.031*** (0.012) 672 yes -13608.84  3.5  3  Coeff. on Internal Trade Dummy  2.5  2  1.5  OLS PPML GPML  1  0.5  0 0  1  2  3  4  Degree of log Distance Polynomial  Figure 4.3: Coefficients on Internal Trade Dummies  Helliwell and Verdier (2001). To answer this question, I perform the following exercise. First, I set lnDod = 0 whenever o = d, and define a set of indicator variables, {δd }, each equal to one when trade is internal to province d, respectively, and zero otherwise. I then run the following gravity specification:  10  lnVodt = θ ln Dod +  ∑ βd δd + γot FXot + αdt FMdt + µt YEARt + εodt .  (4.6)  d=1  For a given o, d, and t with o = d, the equation reduces to: lnVodt = θ ln Dod + γot + αdt + µt + εodt .  83  (4.7)  For o = d, the equation becomes: lnVddt = βd + γdt + αdt + µt + εddt .  (4.8)  βd is a non-parametric control for factors affecting within-d trade. Thus, assuming no provincial border effect, it can be interpreted as θ ln Ddd . Dividing by θ and exponentiating gives the estimate of the internal distance that implies no border effect: βd Ddd = exp . (4.9) θ Column 1 of Table 4.6 presents results. Column 2 reports results for the specification with quadratic log distance:  lnVodt = θ1 ln Dod + θ2 (ln Dod )2 +  10  ∑ βd δd + γot FXot + αdt FMdt + µt YEARt + εodt . d=1  In this case, no border effect implies βd = θ1 ln Ddd + θ2 (ln Ddd ternal distance:   −θ1 − θ12 + 4θ2 βd . Ddd = exp  2θ2  )2 .  (4.10) Isolating in-  (4.11)  Columns 3 and 4 of Table 4.6 report results for the Poisson PML specification, while columns 5 and 6 report those for Gamma PML. To help interpret these results, Figures 4.4 and 4.5 plot the estimated internal distances for each province— calculated from the point estimates of Table 4.6 and equations (4.9) and (4.11), respectively—against the the internal distances used in Helliwell and Verdier (2001). With the exception of Ontario, the linear specifications require extremely small internal distances to explain away the border effect. The quadratic-in-log-distance specifications all lead to very similar estimates of internal distances. The implied distances are smaller than those of Helliwell and Verdier (2001), but not so much as to be unrealistic. If internal trade is mostly confined to transactions between parties within cities—or between parties in nearby cities—then the estimated border effect may well be artificially inflated by the use of large internal distances.  84  Table 4.6: Provincial Internal Trade Effects  Model : Depvar: Estimator intcpt  (1) lcomserv  (2) lcomserv OLS 25.062*** 32.522*** (0.310) (1.617) alta -2.554*** -9.993*** (0.229) (1.603) bc -2.815*** -10.194*** (0.268) (1.618) sask -1.212*** -8.733*** (0.198) (1.614) man -1.388*** -8.950*** (0.216) (1.635) ont -4.193*** -11.804*** (0.224) (1.634) que -1.886*** -9.396*** (0.212) (1.618) nb -0.756*** -8.086*** (0.256) (1.616) ns -0.627*** -7.922*** (0.195) (1.553) pei 1.019*** -6.212*** (0.237) (1.569) nfld -1.437*** -8.963*** (0.235) (1.638) ldist -0.876*** -3.004*** (0.028) (0.456) ldist squared 0.150*** (0.032) N 672 672 R2 /Log PL 0.97 0.971  (3) (4) (5) (6) comserv comserv comserv comserv PPML GPML 21.041*** 37.001*** 23.780*** 32.792*** (0.171) (1.339) (0.215) (1.520) 0.074 -15.834*** -2.585*** -11.461*** (0.196) (1.331) (0.216) (1.496) 0.115 -15.649*** -2.827*** -11.640*** (0.185) (1.315) (0.249) (1.506) 1.627*** -14.333*** -1.109*** -10.119*** (0.170) (1.339) (0.189) (1.515) 1.374*** -14.671*** -1.317*** -10.393*** (0.154) (1.335) (0.197) (1.536) -1.477*** -17.440*** -4.077*** -13.168*** (0.157) (1.331) (0.211) (1.531) 0.504*** -15.239*** -1.976*** -10.925*** (0.139) (1.315) (0.213) (1.503) 2.271*** -13.342*** -0.905*** -9.664*** (0.170) (1.317) (0.231) (1.503) 2.278*** -13.525*** -0.510*** -9.263*** (0.164) (1.312) (0.189) (1.467) 3.597*** -12.001*** 0.751*** -7.829*** (0.211) (1.321) (0.184) (1.452) 1.867*** -14.092*** -1.380*** -10.468*** (0.181) (1.339) (0.219) (1.549) -0.492*** -4.862*** -0.865*** -3.411*** (0.022) (0.364) (0.026) (0.425) 0.298*** 0.179*** (0.025) (0.030) 672 672 672 672 -5.438e+09 -4.360e+09 -13593.51 -13590.19  Note: Statistical significance of 1% indicated by ***. Robust standard errors in parantheses. Log pseudolikelihood reported for PPML and GPML.  85  140  120  Estimated Internal Distance  100  80 OLS-lin PPML-lin GPML-lin 60  40  20  0 0  50  100  150  200  250  H&V (2001) Internal Distance  Figure 4.4: Estimated “No BE” Internal Distances vs. Helliwell & Verdier Internal Distances: Linear Log Distance Specification  4.5  Distance and International Service Trade  Before combining the international and interprovincial data to estimate a national border effect, I analyse the distance effect for Canada’s international trade in commercial services. The level and trend of the distance effect could have implications on the threat that Canada faces from the off-shoring of services. The data set for international service trade is not a full matrix of exports between each of the trading partners, but rather focuses on Canadian imports and exports. Each observation thus corresponds with a given year and partner country, and a dummy equal to one when Canada is the importer. Table 4.7 lists Canada’s top fifteen importing and exporting partner countries for 2004, along with their value and share of Canadian service trade. The United States 86  250  Estimated Internal Distance  200  150 OLS-quad PPML-quad GPML-quad 45 degrees 100  50  0 0  50  100  150  200  250  H&V (2001) Internal Distance  Figure 4.5: Estimated “No BE” Internal Distances vs. Helliwell & Verdier Internal Distances: Quadratic Log Distance Specification  is by far the most important trading partner, making up over 70% of Canada’s service imports in 2004. Of course, the United States is both the largest and the closest trading partner, so its dominance is not surprising. Japan is the second most important exporter to Canada, but this may be explained by its economic size outweighing any negative effects of distance. Conspicuous among the top partners are Bermuda and Barbados, which both have small economic size and are not much closer to Canada than Europe. Both countries attract substantial business registration through favourable tax rules, which likely explains these outliers.5 The theory of the last section largely translates over to the current environment, 5 Indeed,  one prominent tax-payer to both of these nations during the sample period is Canada Steamship Lines, which was owned by former Canadian prime minister Paul Martin.  87  Table 4.7: Canada’s Top Service Trade Partners, 2004 IMPORTER U.S.A BERMUDA U.K. BARBADOS GERMANY FRANCE SWITZERLAND SWEDEN JAPAN AUSTRALIA CHINA NETHERLANDS MEXICO IRELAND BRAZIL  VALUE 21,433 1,609 1,528 1,136 881 716 661 416 376 362 338 338 327 327 218  SHARE .644 .048 .046 .034 .026 .022 .020 .013 .011 .011 .010 .010 .010 .010 .007  EXPORTER U.S.A. JAPAN BARBADOS U.K. BERMUDA GERMANY FRANCE NETHERLANDS SWITZERLAND AUSTRALIA HONG KONG IRELAND NORWAY ITALY BELGIUM/LUX.  VALUE 27,591 1,862 1,710 1,373 1,083 1,079 554 493 392 190 143 125 120 119 113  SHARE .720 .049 .045 .036 .028 .028 .014 .013 .010 .005 .004 .003 .003 .003 .003  Note: Exports and Imports expressed in millions of CDN dollars. “Share” is the ratio of the importer or exporter’s trade value over total imports or exports, respectively.  as we now consider trade at the country level as opposed to provincial. Recall that equation (4.4) was derived by assuming trade costs took the form of equation (4.2). In this formulation, the only explicit bilateral determination of trade is distance. Distance acts partly as a proxy for informational inadequacies, but it may be possible to account for more of these by controlling for certain cultural or institutional relationships between countries. Indeed, this has become standard practice in the gravity literature. In this section I therefore include four other determinants of trade costs, namely whether the countries share a common legal system, a common official language, common colonial origins, or participate in a regional trade agreement (RTA). Another difference between the current environment and that of equation (4.2) is that now Canada is always one of the trading partners, and so each observation is not of the form origin-destination-year, but rather partner-year. Membership in an RTA is potentially time-variant, so we must add a time subscript to the trade costs, resulting in the following: ln(1 + τ pt ) = δ ln D p + β Culture pt + u pt  88  (4.12)  where Culture pt is the vector [Comleg p Lang p Colony p RTA pt ] and β is the associated vector of coefficients. Comleg p is equal to one when the partner shares a common legal system with Canada (common law); Lang p is equal to one when the partner shares an official language with Canada (English or French); Colony p is equal to one when the partner shares a colonial relationship with Canada (British Empire); and RTA pt is equal to one if the the partner and Canada were part of a regional trade agreement in year t. The following estimating equation can then be derived analogously to equation (4.4) by substitution and taking logs:  lnVpt = PT pt + YEARt − θ δ ln D pt − θ β Culture pt + Canimp pt + ln η pt .  (4.13)  where PT pt represents partner-year fixed effects and Canimp pt is a dummy equal to one when Canada is the importer. The inclusion of ‘Canimp’ simply relaxes the equality constraint on the intercept for cases where Canada is importing or exporting. As discussed in the introduction, it would be interesting to see how the distance effect has evolved over time. The sample period 1993 − 2004 is a particularly interesting period given the great advances and adoption of communications technology. There are countering forces to the notion that distance must be decreasing in importance. Many nations have increasingly relaxed their Foreign Direct Investment policies, or have begun actively seeking such investment. To the extent that foreign subsidiary earnings are not reflected in trade data, there could be a real or perceived increased importance of distance on trade if companies are choosing to locate subsidiaries in distant countries. The Canadian data is based on Balance of Payments, which does not include commercial presence (mode three) service provision. I estimate year-specific distance effects by interacting log distance with year dummies for all but the first year. The equation is of the form: lnVpt = PT pt +YEARt +α ln dist pt +αt Y EARt ∗ln dist pt −θ βCulture pt +Canimp pt +ln η pt . (4.14) Since distance coefficients are typically negative, we can define the ‘distance effect’ as the negative of the distance coefficient. The coefficient on ln dist pt then gives (the negative of) the base-line distance effect for 1993, while the subsequent distance  89  coefficients (α2 , α3 , etc) give (the negative of) the change in distance effect from the base year to the corresponding current year. To calculate the distance effect for each year we simply add each year’s coefficient to that of the base year and multiply by negative one. Equation (4.14) is the preferred specification, but it is interesting to compare its output with that of other models. One may observe that because each partner-year combination is represented by just two observations (one for Canadian imports, and one for exports), that the parter-year dummies take up many degrees of freedom relative to the number of observations. If we instead hold partner fixed-effects constant, we gain degrees of freedom at the expense of possible mis-specification. We may then note that two sources of partner-specific year effects are changes in GDP and per capita GDP. I therefore also append Equation (4.14) with partner countries’ (log) GDP and per capita GDP and estimate it with fixed partner dummies. Finally, I estimate a ‘na¨ıve’ gravity equation (with the inclusion of ‘Culture’ and year dummies). Table 4.8 displays the results from each of the four models. Column 1 corresponds to the partner fixed effects model while column 2 corresponds to the appended na¨ıve model. Column 3 displays results for the combined fixed effects and GDP/per capita GDP model and column 4 shows the preferred time-varying partner fixed effects approach. The distance effects over time for each of the four models are plotted in Figure 4.6. Interestingly, The na¨ıve model corresponds very well with that of the time-varying partner fixed-effects model, whereas the two models involving fixed partner dummies estimate much larger distance effects. It is particularly unclear why the addition of GDP and per capita GDP should push the distance effects of the fixed partner-dummies model higher, rather than bringing them more in line with the time-varying partner-dummies approach. In all four models the distance effects are significant and stable. In the preferred specification the effects are similar to or slightly higher than past estimates of merchandise-trade distance effects.6  4.6  National Border Effect  In this section I combine the two data sets in order to estimate the effect on commercial services of Canada’s national border. The idea is that interprovincial trade 6 Disdier and Head (2008) perform a meta-analysis of studies that estimate merchandise-trade distance effects, and find the average estimate to equal 0.9.  90  Table 4.8: Distance Effects for Canadian International Trade in Services  Model : Depvar: intcpt  (1) lservices 35.549*** (1.559)  distance by year  (2) lservices 15.389*** (2.253)  (3) lservices 38.145*** (5.842)  (4) lservices 26.669*** (2.185)  See Figure 4.6  canada importer dummy 1 for common legal system 1 for common official primary language 1 for pair ever in colonial relationship 1 for regional trade agreement in force  -0.763*** (0.038) 2.117*** (0.114) 1.668*** (0.099) -0.060 (0.162) 0.046 (0.146)  partner GDP partner GDP per capita N R2 RMSE  1372 0.867 .704  -0.743*** (0.063) 0.573*** (0.079) 0.917*** (0.074) 0.278** (0.117) 0.135 (0.125) 0.621*** (0.030) 0.256*** (0.029) 1356 0.622 1.157  -0.771*** (0.039) 7.90*** (1.753) 3.576*** (0.849) -1.372** (0.599) 0.115 (0.155) -1.246* (0.709) 1.351* (0.716) 1356 0.866 .705  -0.738*** (0.044) 0.633* (0.340) 1.082** (0.421) 3.367*** (0.323) 2.921*** (0.307)  Note: Statistical significance of 1%, 5%, and 10% are indicated respectively by ***, **, and *.  data gives me point-to-point information within Canada, whereas Canadian international trade data gives me cross-border information. Focusing on merchandise trade, McCallum (1995) and Helliwell and Verdier (2001) both use provincial data for within Canada trade, but they employ provincial trade with U.S. states in order to estimate the effect of the Canada-U.S. border. There is no comparable province-state service trade data set available. By using country-level international data, I am estimating the average effect of crossing the Canadian border to the fifty-nine trading partners in the sample. The results are therefore not directly comparable to those of the previous studies.  91  1372 0.915 .784  Distance Effects over Time 3.5  3  Distance Effect  2.5  2  Country Fixed effects GDP/GDPPC Fixed Effects and GDP/GDPPC Country-Year Fixed Effects  1.5  1  0.5  0 1992  1994  1996  1998  2000  2002  2004  2006  Year  Figure 4.6: Distance Effects for Canadian International Trade in Services  An important technical issue arises as a result of this approach. Since my aim is to estimate a national border effect, I include a dummy equal to one when trade is within Canada and zero otherwise. But since the set of within-Canada observations is equal to the set of observations involving provincial trade, I cannot also include importer and exporter fixed effects. This is because the border dummy would be a linear combination of the provincial dummies. I am therefore constrained to use the appended na¨ıve gravity equation for this exercise. The results of the previous section give me some comfort that the missing-variable bias may not be too strong for this sample. In order to include the ‘Culture’ variables in the regression, I code each of them equal to one for interprovincial trade.  92  Table 4.9: International and Provincial Stacked Data  Model : Depvar: intcpt  (1) (2) lcomserv lcomserv -0.730 -0.782 (0.528) (0.537) border 1.240*** 1.269*** (0.204) (0.244) distance -0.761*** -0.761*** (0.033) (0.033) origin GDP 0.845*** 0.846*** (0.038) (0.039) destination GDP 0.727*** 0.727*** (0.028) (0.028) origin GDP per capita 0.467*** 0.469*** (0.045) (0.045) destination GDP per capita 0.048 0.049 (0.036) (0.036) 1 for common official primary language 0.891*** 1.269*** (0.077) (0.244) 1 for pair ever in colonial relationship -0.060 -0.064 (0.086) (0.084) 1 for regional trade agreement in force -0.045 -0.044 (0.112) (0.112) 1 for common legal system 0.571*** 0.569*** (0.085) (0.085) border2 -0.056 (0.193) border3 0.001 (0.193) border4 0.003 (0.189) border5 -0.192 (0.189) border6 -0.155 (0.191) border7 0.004 (0.194) N 1500 1500 2 R 0.741 0.741 RMSE .969 .97 Note: Statistical significance of 1%, 5%, and 10% are indicated respectively by ***, **, and *.  93  The estimating equation is identical to that of the appended na¨ıve model of section 4.4, and Table 4.9 displays the results. Column 1 is the base-line regression. The coefficient implies a border effect of e1.24 = 3.5. In column 2 I have interacted the border dummy with each of the years to identify whether border frictions have diminished over time. The insignificant coefficients suggest there has been no change over the years 1997-2004. Border frictions can lead to a large ratio of internal trade to foreign trade for two reasons. First, the frictions might lead some cross-border commerce that might have taken place to simply vanish. Second, the cross-border trade may be re-directed to domestic business partners. The framework employed in this essay cannot distinguish which channel is more important in explaining why trade is larger at home, all else equal. Suppose that the entire border effect was due to lost foreign commercial opportunities, without compensating-alternative opportunities at home. Then we could say that a border effect of 3.5 implies international trade is 3.5 times smaller than it would be without the effects of the border. Multiplying the total value of exports (about $30 billion in 2004) by this factor gives us an upper bound for the dollar value of lost exports due to the border: 3.5 × $30 billion ≈ $100 billion. Of course, in reality Canadian firms are not $100 billion worse off because they are also protected in the home market by the same border. But the number gives us a benchmark. A recent CBC article reports that Canadian officials estimate border frictions to cost, “as much as $16 billion a year, or one percent of GDP.” This number does not seem unreasonable given the analysis of this essay.  4.7  Conclusion  This chapter analyses Canada’s interprovincial and international service trade. Identification of a provincial border effect is complicated by the nature of service transactions, and disentangling border and distance effects may not be possible without city-to-city trade data. I employ alternative specifications suggested by the recent literature to test whether the apparently large provincial border effects in Canada are instead due to extreme short-distance attenuation of trade, and find some evidence that this is the case. Looking to the future, an event study of the TILMA may also shed light on its merit and effectiveness. Canada’s international trade in services is subject to substantial distance effects,  94  in line with earlier estimates of the distance effect for goods trade. The effect of distance on service trade has remained constant since the early Nineties, in spite of technological advances in electronic communication. A possible explanation is that companies may have been increasingly substituting foreign direct investment for cross-border service provision for more distant countries. Such behaviour would exert an upward pressure on the effect of distance over time. This hypothesis remains to be tested. If the estimated distance effect continues to hold into the future, the threat from outsourcing may remain minimal. I estimate the national border effect for service trade by combining the interprovincial and international data sets. Interprovincial trade (excluding own-province observations) provides point-to-point transaction data within the country. International trade provides cross-border transaction data. I control for various cultural/institutional differences between countries and estimate a national border effect of 3.5.  95  Chapter 5 Conclusions Barriers to trade have implications for the pattern of international economic activity as well as the strategies of the firms engaged in international trade. In this thesis I explore some of these implications, with a particular focus on trade in cultural (motion picture) and professional services. In Chapter Two I develop a simple model of international trade in a heterogeneous unit-demand product. Foreign revenues are rising in variety quality and destinationcountry size and wealth. Fixed costs of entry imply that only varieties that are appealing enough will be exported. Using a direct measure of movie quality, I test how well U.S. movie exports adhere to this hierarchy. Graphical techniques suggest that selection is important, but also that destination-movie-specific demand shocks play a large role in foreign entry decisions. I use direct and revealed measures of movie quality to look for a systematic role for movie quality in export decisions. Estimates suggest that a one-standard-deviation increase in quality from the average leads to a three-to-eight percentage-point increase in the probability of entry. This compares with an overall probability of 17% in the sample. I exploit data on movie genre to estimate a measure of cultural distance (“Hollywood distance”) between destination countries and the U.S. This measure of cultural distance is correlated with geographic distance, but uncorrelated with prominant other indices of cultural distance. Hollywood distance is associated with lower bilateral trade volumes between the U.S. and its trading partners, but higher trade volumes for cultural goods. The result might be due to the combination of two opposing forces. On the one hand, cultural distance increases transaction costs as it is associated with lower trust. On the other hand, people are interested in consuming 96  cultural goods from exotic places. A growing body of work has suggested that manufacturing firms learn about their export profitability through exporting. Chapter Three adds to that literature by considering a new type of product. Motion picture distributors can only release movies once in any destination country, and have no scope for intra-market learning. On the other hand, potentially large idiosyncratic differences in taste across markets mean distributors can not infer too much from any given success or failure. Nonetheless, I show that distributors do appear to adjust their entry strategies based on prior-market performance. The correlation between past “surprises” and entry decisions could be due to omitted factors in the initial forecasts, but results show that while this is likely a factor, unrealized surprises have a much smaller effect on entry decisions than past surprises, leading to the conclusion that learning is indeed taking place. As firms move toward the “day and date” simultaneous release strategy to fight international piracy, they lose the ability to use the information from prior markets, adding to the cost of piracy. In Chapter Four I analyse Canada’s interprovincial and international service trade. Identification of a provincial border effect is complicated by the nature of service transactions, and disentangling border and distance effects may not be possible without city-to-city trade data. I employ alternative specifications suggested by the recent literature to test whether the apparently large provincial border effects in Canada are instead due to extreme short-distance attenuation of trade, and find some evidence that this is the case. Looking to the future, an event study of the TILMA may also shed light on its merit and effectiveness. Canada’s international trade in services is subject to substantial distance effects, in line with earlier estimates of the distance effect for goods trade. The effect of distance on service trade has remained constant since the early Nineties, in spite of technological advances. A possible explanation is that companies have been increasingly substituting foreign direct investment for cross-border service provision for more distant countries. Such behaviour would exert an upward pressure on the effect of distance over time. This hypothesis remains to be tested. If the estimated distance effect continues to hold into the future, the threat from outsourcing may remain minimal. I estimate the national border effect for service trade by combining the interprovincial and international data sets. Interprovincial trade (excluding own-province  97  observations) provides point-to-point transaction data within the country. International trade provides cross-border transaction data. I control for various cultural/institutional differences between countries and estimate a national border effect of 3.5.  98  Bibliography Abramitzky, R. and I. Sin, 2010, “Book translations as idea flows: The effects of the collapse of Communism on the diffusion of knowledge,” mimeo Stanford University. Akhmetova, Z., 2010, “Firm Experimentation in New Markets,” mimeo. Albornoz, F, H.F. Calvo Pardo, G. Corcos, and E. Ornelas, 2010, “Sequential Exporting,” mimeo. Anderson, James E., 1979, “A Theoretical Foundation for the Gravity Equation,” American Economic Review, 69, 106-16. Anderson, S., A. de Palma, and J.F. Thisse, 1992, Discrete Choice Theory of Product Differentiation, MIT Press. Anderson, James E. & Eric van Wincoop (2003): “Gravity with Gravitas: A Solution to the Border Puzzle,” American Economic Review, 93(1), 170-92. Arkolakis, 2010, “Market Penetration Costs and the New Consumers Margin in International Trade,” Journal of Political Economy, 118(6), 1151-1199. Baldwin, R.E. and J. Harrigan, 2007, “Zeros, Quality and Space: Trade Theory and Trade Evidence,” NBER Working Paper No. 13214. Bergstrand, J., 1985, “The Gravity Equation in International Trade: Some Microeconomic Foundations and Empirical Evidence,” Review of Economics and Statistics, 67 (August), 474-81. ———, 1989, “The Generalized Gravity Equation, Monopolistic Competition, and Factor-Proportions Theory in International Trade,” Review of Economics and Statistics, 71(1), 143-53. 99  Bernard, A., J. Eaton, B. Jenson and S. Kortum, 2003, “Plants and Productivity in International Trade,” American Economic Review, 93(4), 1268-1290. Bernard, A. and B. Jenson, 1999, “Exceptional exporter performance: cause, effect, or both?” Journal of International Economics, 47, 1-25. Bernard, A. and B. Jenson, 2004, “Why Some Firms Export,” Review of Economics and Statistics, 86, 561-569. Bernard, A., S. J. Redding and P. K. Schott, 2009, “Multi-Product Firms and Trade Liberalization,” Working Paper 09-21, Center for Economic Studies, U.S. Census Department. Blum, B. & A. Goldfarb, 2006, “Does the internet defy the law of gravity?” Journal of International Economics 70(2), 384-405. Cavalli-Sforza,L. L., P. Menozzi and A. Piazza, 1994, The History and Geography of Human Genes, Princeton, NJ: Princeton University Press. Chaney, T., 2005, “Liquidity Constrained Exporters,” University of Chicago mimeo. Chen, M. X. and M. O. Moore, 2010, “Location decision of heterogeneous multinational firms,” Journal of International Economics, 80(2), 188-199. Cowen, T., 2002, Creative Destruction: How Globalization Is Changing the World’s Cultures, Princeton, NJ: Princeton University Press. Crozet, M., K. Head and T. Mayer, 2009, “Quality Sorting and Trade: Firm-level Evidence for French Wine,” CEPR Discussion Paper 7295. Deardoff, A. V., 1984, “Testing Trade Theories and Predicting Trade Flows,” in Handbook of International Economics, Volume I, ed. by R. Jones and P. Kenen. Amsterdam: North-Holland. ———, 1995, “Does Gravity Work in a Neoclassical World?” in J.A. Frankel (ed.), The Regionalization of the World Economy, Chicago: Chicago University Press. De Vany, Arthur, 2004, Hollywood Economics, London: Routledge.  100  Disdier, A.-C., L. Fontagn´e, T. Mayer and S. Tai, 2010, “Bilateral Trade in Cultural Goods,” Review of World Economics, 145(4), 575-595. Disdier, A.-C. and K. Head, 2008, “The Puzzling Persistence of the Distance Effect on Bilateral Trade”, Review of Economics and Statistics, 90(1), 37-48. Disdier, A.-C., K. Head and T. Mayer, 2010, “Exposure to foreign media and changes in cultural traits: Evidence from naming patterns in France,” Journal of International Economics, 80(2), 226-238. Dyen, I., J.G. Kruskal and P. Black, 1992, “An Indo-European classification: a lexicostatistical experiment,” Transactions of the Amer. Philosph. Society, 82: Part 5. Eaton, J., M. Eslava, C.J. Krizan, M. Kugler, and J. Tybout, 2011, “A Search and Learning Model of Export Dynamics,” mimeo. Eaton, J. & S. Kortum, 2002, “Technology, Geography, and Trade,” Econometrica, 70(5), 1741-79. Eaton, J., S. Kortum and F. Kramarz, 2008, “An Anatomy of International Trade: Evidence from French Firms,” NBER Working Paper No. 14610. Elberse, A. and J. Eliashberg, 2003, “Demand and Supply Dynamics for Sequentially Released Products in International Markets: The Case of Motion Pictures,” Marketing Science, 22(3), 329-354. Eliashberg, J., A. Elberse and M. Leenders, 2006, “The Motion Picture Industry: Critical Issues in Practice, Current Research, and New Directions,” Marketing Science, 25(6), 638-661. Epstein, E.J., 2005, “Send in the Aliens: They’re the last hope for the foreign box office,” Slate, http://www.slate.com/id/2125153. Feenstra, R., 2002, “Border Effects and the Gravity Equation: Consistent Methods for Estimation,” Scottish Journal of Political Economy, 49(5), 491-506. Ferreira, F. and J. Waldfogel, 2010, “Has A Half Century of World Music Trade Displaced Local Culture?”, NBER Working Paper No. 15964. 101  Finney, A., 2010, The International Film Business: A Market Guide Beyond Hollywood, Routledge, New York. Francois P. and T. Van Ypersele, 2002, “On the protection of cultural goods,” Journal of International Economics, 56(2), 359-369. Ganesh, J., V. Kumar and V. Subramaniam, 1997, “Learning effect in multinational diffusion of consumer durables: An exploratory investigation,” J. Acad. Marketing Sci., 25(3) 214-228. Gatignon, H., J. Eliashberg and T. S. Robertson, 1989, “Modeling multinational diffusion patterns: An efficient methodology,” Marketing Science, 8(3) 231247. G´en´ereux, P. and B. Langen, 2002, “The Derivation of Provincial (Inter-Regional) Trade Flows: The Canadian Experience,” presented at the 14th International Input-Output Techniques Conference at the Universit´e du Qu´ebec a` Montr´eal, Canada. Goettler, R. and and P. Leslie, 2005, “Cofinancing to Manage Risk in the Motion Picture Industry,” Journal of Economics & Management Strategy, 14(2), 231261. Grossman, G. and E. Helpman, 1991, Innovation and Growth in the World Economy, Cambridge, MA: MIT Press. Hanson, G. and Chong Xiang, 2008, “Testing the Melitz Model of Trade: An Application to U.S. Motion Picture Exports,” NBER Working Paper No. 14461. Harrigan, J., 1996, “Openness to Trade in Manufactures in the OECD,” Journal of International Economics, 40, 23-39. Head, K. and T. Mayer, 2010, “Illusory Border Effects: Distance Mismeasurement Inflates Estimates of Home Bias in Trade,” in van Bergeijk, P.A.G. and S. Brakman (eds), The Gravity Model in International Trade: Advances and Applications, Cambridge University Press, 165-192. Head, K., T. Mayer and J. Ries, 2009, “How Remote is the Offshoring Threat,” European Economic Review. 102  Head, K. and J. Ries, 2008, “FDI as an outcome of the market for corporate control: theory and evidence” Journal of International Economics, 74(1), 220. Helliwell, J., 1996, “Do National Boundaries Matter for Quebecs Trade?” Canadian Journal of Economics, 29, 50722. ———, 1998, How Much Do National Borders Matter? Washington, DC: Brookings Institution. Helliwell, J. and G. Verdier, 2001, “Measuring Internal Trade Distances: A New Method to Estimate Provincial Border Effects in Canada,” Canadian Journal of Economics, 34(4), 1024-41. Helpman, E. and P. Krugman, 1985, Market Structure and Foreign Trade: Increasing Returns, Imperfect Competition, and the International Economy, The Massachusetts Institute of Technology Press. Helpman E., M.J. Melitz and Y. Rubinstein, 2008, “Estimating Trade Flows: Trading Partners and Trading Volumes,” Quarterly Journal of Economics, 12(3), 441-487. Hillberry, R. and D. Hummels, 2008, “Trade responses to geographic frictions: A decomposition using micro-data,” European Economic Review, 52(3), 527550. Hofstede, G., 1980, Culture’s consequences: International differences in workrelated values. Beverly Hills, CA: Sage. Hofstede, G., 2001, Culture’s consequences: Comparing values, behaviors, institutions and organizations across nations (2nd edn). Thousand Oaks, CA: Sage. Hofstede, G., 2002, “Dimensions do not exit: A reply to Brendan McSweeney,” Human Relations, 55(2). Holloway, I.R., 2011, “Foreign Entry, Quality, and Cultural Distance: ProductLevel Evidence from U.S. Movie Exports,” mimeo. Iacovone L. and B. Javorcik, 2008, “Shipping Good Tequila Out: Investment, Domestic Unit Values and Entry of Multi-product Plants into Export Markets,” mimeo University of Oxford. 103  Johnson R. C., 2009, “Trade and Prices with Heterogeneous Firms,” mimeo. Khandelwal, A., forthcoming, “The Long and Short (of) Quality Ladders,” The Review of Economic Studies. Kugler, M. and E. Verhoogen, 2008, “The Quality-Complementarity Hypothesis: Theory and Evidence from Colombia,” NBER Working Paper No. 14418. Lambert, R.A., 1986, “Executive Effort and Selection of Risky Projects,” RAND Journal of Economics, 17(1), 7788. Linneman, H., 1966, An Econometric Study of International Trade Flows, North Holland. Manova, K., 2010, “Credit Constraints, Heterogeneous Firms, and International Trade,” NBER Working Paper 14531. Manova, K. and Z. Zhang, 2009, “Export Prices and Heterogeneous Firm Models,” mimeo Stanford University. Maystre, N., J. Olivier, M. Thoenig and T. Verdier, 2009, “Product-Based Cultural Change: Is the Village Global?” CEPR DP7438. McCallum, J., 1995, “National Borders Matter: Canada-U.S. Regional Trade Patterns,” American Economic Review, 85(3), 615-23. McCalman, P., 2005, “International Diffusion and Intellectual Property Rights: An Empirical Analysis,” Journal of International Economics, 67, 353-372. Melitz, M. J., 2003, “The Impact of Trade on Intra-Industry Reallocations and Aggregate Industry Productivity,” Econometrica, 71(6), 1695-1725. Minetti, R. and S. Zhu, 2011, “Credit Constraints and Firm Export: Microeconomic Evidence from Italy,” Journal of International Economics, 83(2), 109-125. Neelamegham, R. and P. Chintagunta, 1999, “A Bayesian Model to Forecast New Product Performance in Domestic and International Markets,” Marketing Science, 18(2), 115-136.  104  Olivier J., M. Thoenig and T. Verdier, 2008, “Globalization and the Dynamics of Cultural Identity,” Journal of International Economics, 76(2), 356-370. Random House, 2010, “culture,” Dictionary.com Unabridged. http://dictionary.reference.com/browse/culture Rauch J. E. and V. Trindade, 2009, “Neckties in the Tropics: A Model of International Trade and Cultural Diversity,” Canadian Journal of Economics, 42(3), 809-843. Santos Silva, J.M.C. and S. Tenreyro, 2006. “The Log of Gravity,” Review of Economics and Statistics, 88(4), 641-658. Tinbergen, J., 1962, Shaping the World Economy, New York: The Twentieth Century Fund. Trebilcock, M. and D. Schwanen, 1995, Getting There: An Assessment of the Agreement on Internal Trade, Toronto: C.D. Howe Institute. Vanek, J., 1968, “The Factor Proportions Theory: The N-Factor Case,” Kyklos, 21, 749-56. Zaidi, H., 2010, “Going Bionic: Distributing Independent Films Internationally That Sell – Genres That Sell!,” Film Threat, http://www.filmthreat.com/features/22250.  105  

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
http://iiif.library.ubc.ca/presentation/dsp.24.1-0072656/manifest

Comment

Related Items