VOLUME 5: ISSUE 2 SEPTEMBER 2012

Transcription

1 VOLUME 5: ISSUE 2 SEPTEMBER 2012 The Michigan Journal of Business (ISSN# ) is published semiannually by undergraduate business students on behalf of the Stephen M. Ross School of Business at the University of Michigan. Communication should be addressed to: Michigan Journal of Business 701 Tappan Street Ann Arbor, MI mjbquestions@umich.edu Website: EDITORIAL OBJECTIVE The Michigan Journal of Business intends to provide undergraduate students worldwide with a platform for exceptional work in the field of business. The Journal seeks to publish distinguished theses, empirical research, case studies, and theories in issues relating to areas of Accounting, Economics, Finance, Marketing, Management, Operations Management, Information Systems, Business Law, Corporate Ethics, and Public Policy. The Journal is distributed and cataloged in prestigious university libraries around the world, and is enlisted in the Directories of Open Access Journals (DOAJ), a scholarly journal database that enlists more than 3000 of the world s leading publications. The contemporary business environment is exceedingly complex. Analyzing this real world phenomenon through traditional applications of theories often yield a suboptimal understanding of the world. The Journal, accordingly, encourages work that takes an interdisciplinary approach to understanding a topic and emphasizes the importance of incorporating the knowledge of liberal arts into an area of interest. By providing a venue to recognize high quality work, the Journal gives an incentive for students to explore their area of interest, rewarding them with the experience to share the power of knowledge with others. The Journal s mission and philosophy parallel the mission of the University of Michigan, the premier research university in the United States.

2 2 THE MICHIGAN JOURNAL OF BUSINESS CONTRIBUTOR INFORMATION The Journal only accepts works from undergraduate students or works completed during undergraduate study. Each manuscript submitted should include a short abstract, author information, and any acknowledgements. Papers will be evaluated based upon sound analysis, originality of argument, and novelty of research. For more information on submitting article for publication, please visit REVIEW PROCESS The organization is entirely student-run, with an editorial staff of about 20 of the top students at the Department of Economics and the Stephen M. Ross School of Business at the University of Michigan. Each semester, the Michigan Journal of Business calls for papers from undergraduate students around the world. Throughout the semester, the editorial board carefully reviews, selects, and edits exceptional work for publication. Faculty willing to advise the Journal is formed from each department to give minor oversight for the project. Throughout the process, a blind review process is implemented to ensure an impartial review of all submissions. EDITORIAL BOARD The editorial board is composed of the top undergraduate students at the Stephen M. Ross School of Business and the Department of Economics at the University of Michigan. They are selected through a competitive application process based on academic merit and aptitude in writing. Past and present members of the MJB editorial board include a former Ross valedictorian who now attends Yale Law School, an editor who now attends Michigan Law School, and editors with professional experience at firms including Goldman Sachs, McKinsey & Co., Morgan Stanley, and JPMorgan.

3 3 WINTER 2012 EDITORIAL BOARD EDITOR-IN-CHIEF Scott Suh SENIOR EDITORS Victoria Greenstein Roger Zhong MANAGING EDITOR Stephanie Chueh Katy Beth Deschenes DIRECTOR OF FINANCE Parth Thakkar DIRECTOR OF PRODUCTION AND MEDIA Yahya Syed ASSOCIATE EDITORS Ajuj Arbol Ivy Chen Ho Yin Fong Jonathan He Michael Hu Habib Khan Gregory Kohler Michael Kopinsky AJ Malhotra Kush Patel Suchee Shah Derek Shan Xiangting Shi Andrew Simon Kabir Sodhi David Tao Srivaths Venkatachari Shaun Yu Qiao Zhang

4 4 THE MICHIGAN JOURNAL OF BUSINESS FACULTY SUPERVISION The Journal is supervised by Professor Tammy Feldman, Lecturer of Business Economics and Public Policy, and Professor Scott Moore, Arthur F. Thurnau Professor of Business Information Technology and BBA Program Director at the Ross School of Business. The following members of the Ross School of Business faculty also provide minor oversight to the editorial board during the review process: Anocha Aribarg Ph.D., University of Wisconsin Assistant Professor of Marketing George Cameron III J.D., Ph.D., University of Michigan Professor Emeritus of Business Law Tammy Feldman Ph.D, Harvard University Lecturer of Business Economics and Public Policy David Hess J.D., Iowa, Ph.D., University of Pennsylvania Assistant Professor of Business Law and Business Ethics Aneel G. Karnani Ph.D., Harvard Associate Professor of Strategy; Chair of Strategy Scott A. Moore Ph.D., University of Pennsylvania Arthur F. Thurnau Professor; BBA Program Director; Associate Professor of Business Information Technology Andrea Morrow M.A., University of Michigan Director of Writing Programs; Co-Coordinator BBA Communication Program Dana M. Muir J.D., University of Michigan Arthur F. Thurnau Professor of Business Law Anu Nagarajan Ph.D., University of Michigan Lecturer of Strategy Lisa A. Pawlik M.B.A., University of Michigan Adjunct Lecturer of Business Communication; Co-Coordinator of BBA Communication Program Josh Pierce Ph.D., Michigan State University Assistant Professor of Finance at University of South Carolina Lloyd Sandelands Ph.D., Northwestern Professor of Management and Organizations & Professor of Psychology Dennis G. Severance Ph.D., University of Michigan Accenture Professor of Business Info. Tech. Amitabh Sinha Ph.D., Carnegie Mellon University Assistant Professor of Operations and Management Science Valerie Y. Suslow Ph.D., Stanford Associate Dean for Degree Programs; Louis and Myrtle Moskowitz Research Professor of Business and Law; Associate Professor of Business Economics and Public Policy David B. Wooten Ph.D., University of Michigan Associate Professor of Marketing Lynn P. Wooten Ph.D., University of Michigan Clinical Assistant Professor of Strategy and Management & Organizations

5 5

6 6 THE MICHIGAN JOURNAL OF BUSINESS (ISSN# ) Copyright 2010 Stephen M. Ross School of Business at the University of Michigan. This work may be reproduced and redistributed, in whole or in part, without alteration and without prior written permission, solely by educational institutions for nonprofit administrative or educational purposes provided all copies contain the following statement: Copyright 2010 Stephen M. Ross School of Business at the University of Michigan. This work is reproduced and distributed with the permission of the Stephen M. Ross School of Business at the University of Michigan. No other use is permitted without the express prior written permission of the Stephen M. Ross School of Business at the University of Michigan. For permission, contact: The Michigan Journal of Business, 701 Tappan Street, Ann Arbor, MI 48109,

7 7 VOLUME 5: ISSUE 2 SEPTEMBER 2012 CONTENTS Editor s Note 8 The "Southwest Effect" Revisited: An Empirical Analysis of the Effects of Southwest Airlines and JetBlue Airways on Incumbent Airlines from 1993 to 2009 STEVEN M. WU Massachusetts Institute of Technology 11 The Effects of Laddering and Spinning in Underwriter Manipulation of IPOs LUDWIG CHINCARINI ALEX ETZKOWITZ JONATHAN KADISH University of San Francisco 43 To Host or Not to Host? A Comparison Study on the Long-Run Impact of the Olympic Games ANITA MEHROTRA University Of California, Berkeley 61

8 8 THE MICHIGAN JOURNAL OF BUSINESS EDITOR S NOTE The Michigan Journal of Business is proud to present its tenth edition during the 2012 winter semester at the University of Michigan. Established in 2008, the Journal remains strongly committed to its mission of recognizing exemplary undergraduate research. The editorial board would like to thank everyone that contributed to the success of this edition, which we believe is amongst our strongest yet. With the continued support from students and faculty at the Stephen M. Ross School of Business, we were able to distribute our printed publication to more than 200 university libraries across 5 continents. In addition to being catalogued in multiple academic databases, our Journal is open-access and available online free-of-charge. Following an extensive call-for-papers process involving authors from across the world, the editorial board was able to review 25 articles for publication. We would like to thank these submitting authors, whose papers continue to make our publication possible. Submissions varied significantly by topic and format, but were consistently impressive overall. After careful consideration, the editorial board selected three articles for publication in this issue. This issue begins with The Southwest Effect Revisited: An Empirical Analysis of the Effects of Southwest Airlines and JetBlue Airways on Incumbent Airlines from 1993 to 2009 by Steven Wu. In this paper, Wu examines pricing trends caused by low-cost carrier incursion onto legacy carrier routes using various regression models. He contributes additional insights into incumbent carrier pricing and overall price dispersion to the existing literature. The second article is The Effects of Laddering and Spinning in Underwriter Manipulation of IPOs submitted by Ludwig Chincarini, Alex Etzkowitz, and Jonathan Kadish. Submitted during an IPO market recovery, the authors examine an expanded dataset to explain IPO underpricing during the internet bubble. Their findings provide further insights into effects that are partially explained by theories presented in the previous literature. The final article is To Host or Not to Host? A Comparison Study on the Long-Run Impact of the Olympic Games by Anita Mehrotra. Mehrotra draws conclusions about Olympic host countries using a long-term outlook in her analysis to look beyond short-turn economic effects. Again, the editorial board would like to thank everyone who contributed to this edition of the Journal. In particular, we are grateful to the Thomas C. Jones Center for BBA Education and the Student Government Association at the Ross School of Business for contributing the funds to make this project possible. Happy reading! Scott Suh Editor-in-Chief

9 9

10 10 THE MICHIGAN JOURNAL OF BUSINESS

11 11 The "Southwest Effect" Revisited: An Empirical Analysis of the Effects of Southwest Airlines and JetBlue Airways on Incumbent Airlines from 1993 to 2009 Steven M. Wu 1 Massachusetts Institute of Technology Abstract The expansion of Southwest Airlines and JetBlue Airways has sparked new empirical interest in the effects of low-cost carriers (LCC) on existing airfares. Namely, empirical studies have attempted to capture the threat, or potential competition, of an entrant. This paper examines incumbent airline prices as a result of potential and actual competition from both Southwest Airlines and JetBlue Airways from 1993 to 2009 by analyzing mean airfares as well as price dispersion on incumbent routes. I incorporate a panel OLS with fixed effects model as well as GLS model with random effects. Consistent with recent literature, this paper finds that legacy incumbents cut fares significantly when threatened by Southwest Airlines. However, low-cost incumbents do not exhibit the same magnitude of pre-emptive price cutting. When threatened by JetBlue, neither legacy nor low-cost carriers cut fares significantly, suggesting that incumbents react differently when threatened by Southwest versus JetBlue. The evidence of increased price dispersion is mixed with price dispersion decreasing on legacy carrier routes as a result of Southwest threat and entry but increasing on legacy carrier routes as a result of JetBlue threat and entry. 1 The author graduated from the Massachusetts Institute of Technology in June 2011 with a BS in Economics, a BS in Mathematics, and a minor in Management Science with a concentration in Finance. This paper was written during the academic year as part of his senior thesis. He thanks Professors Sarah Ellison and Nancy Rose of MIT, Professor James Feyrer of Dartmouth College for their guidance and comments and Professors Chad Syverson of the University of Chicago, Severin Borenstein of UC Berkeley and Bogdan Daraban of Shenandoah University for providing sample data and codes. Brad Shapiro and Chris Palmer of MIT and Kerry Tan of The Ohio State University also gave helpful guidance.

12 12 THE MICHIGAN JOURNAL OF BUSINESS I. Introduction The Airline Deregulation Act of 1978 dramatically altered the competitive landscape of the US airline industry, ushering in an era of head-to-head competition between low-cost carriers (LCCs) and legacy carriers. In particular, the continued expansion of Southwest Airlines, the most profitable player in the LCC space today, has become a principal driving force behind the growth of LCCs and the ubiquity of low fares across routes in general. The impact of LCCs has attracted an increasing amount of empirical attention within the industrial organization literature in the past. Unlike legacy carriers which utilize a hub-and-spoke network and operate with a variety of different aircrafts, LCCs operate within a point-to-point network, allowing them to implement considerable flexibility in routes flown and operate with lower costs. For example, the number of passengers flying LCCs more than doubled from 1997 to 2007 and LCCs entered a total of 598 routes from 1997 to 2007 (Tan 2010). From 1993 to 2004, Southwest alone nearly tripled its revenues from $2.3 to $6.5 billion (Goolsbee and Syverson 2008). Figure 1 depicts the growth in airports serviced by Southwest and JetBlue from 1993 to The main objective of this paper is to extend upon previous empirical work on the impact of LCC entry threat by focusing on the particular expansionary nature of Southwest Airlines and JetBlue Airways, the two most dominant LCCs in the industry today. Within the airline industry literature, entry threat captures the scenario in which a particular LCC has begun operations in two endpoints of a route, but has not started flying the route itself. 2 Because an incumbent airline senses the increased probability that an entrant may potentially enter the route, existing fares would likely decrease well in advance of actual entry as the incumbent looks to deter entry or to generate brand loyalty among existing customers and "cushion" the impact of imminent competition. With Southwest and JetBlue's staggering expansion in the past decade which has witnessed the consolidation of several legacy carriers (i.e. United Airlines- Continental merger in May 2010) and the bankruptcy of others (Japan Airlines in January 2010), the airline industry is an excellent place to examine the strategic nature of entry deterrence. 2 Tan (2010) provides a succinct distinction between potential (or entry threat) and actual competition. Suppose Southwest operates out of both Boston Logan International Airport (BOS) and Philadelphia International Airport (PHL) and that it services the BOS-PHL route. Actual competition is said to occur if the incumbent on the route, say US Airways, services the route at the same time as Southwest. In this case, there is no entry threat because actual entry onto the route is guaranteed. Suppose Southwest also operates out of New York La Guardia (LGA) but not the BOS-LGA route. In this case US Airways potentially competes with Southwest on the BOS-LGA route at the time that Southwest operates out of both airports but not the actual route itself. Here, the BOS-LGA route is threatened by Southwest and US Airways, the incumbent on the route, may likely cut fares before any actual Southwest competition.

13 The "Southwest Effect" Revisited 13 The remainder of this paper is organized as follows: Section 2 discusses existing literature on this topic. Section 3 discusses the data collection methodology. Section 4 suggests an empirical strategy to examine the effect of entry threats and the motive for preemptive price cutting. Section 5 presents the empirical findings and discussions. Lastly, Section 6 concludes the paper. Figure 1: Southwest and JetBlue aggressively expanded operations out of metropolitan airports from 1993 to II. Literature Review The airline transportation literature is dense with studies imparting the salutary impacts of LCC entry on airfares and air travel. Morrison (2001) found that the estimated savings (or decrease in competitor airfares), due to actual, adjacent and potential competition from Southwest, totaled $12.9 billion by 1998, of which $3.4 billion of these savings were directly attributable to Southwest's lower fares. Tan (2010) examined both legacy and LCC responses to several potential entrants and found that legacy incumbents cut fares more aggressively than LCC incumbents. Bennett and Craun (1993) studied Southwest's expansion into California in the early 1990s by examining discrete price drops on the Oakland-Burbank route and found that Southwest's operations resulted in a 55% decrease in prices as well as a sixfold increase in passenger traffic. On a similar note, Dresner, Lin and Windle (1996), examined the spillover effects of Southwest by examining fares on routes from nearby airports and suggested that the presence of a low-cost carrier on a particular route induced competitive price pressures in the form of spillovers onto other routes,

14 14 THE MICHIGAN JOURNAL OF BUSINESS resulting in higher passenger traffic and increased consumer welfare. 3 It is only until recently that economists have started to examine what happens to prices before low-cost entry. Goolsbee and Syverson (2008) examined how incumbents respond to the threat of Southwest entry as opposed to actual entry by analyzing average airfares from with the following question in mind: Are entry threats credible? They restricted their sample to Southwest airports only and focused on scenarios where Southwest started operating in both endpoints of a route but before it actually started flying the route. Controlling for airport-specific cost shocks, they found that incumbents cut fares significantly when threatened by imminent Southwest entry and that over half of Southwest's total impact on incumbent fares occurred from mere threat alone, with fares initially dropping 17% in the quarter 4 of threat (ie Southwest operates both endpoints) and ultimately dropping 29% three years after route entry. Goolsbee and Syverson's (2008) findings that airlines do in fact cut prices prior to entry run counter to the classic view of limit pricing, promulgated by the Chicago School, which implies that airlines are static players and should not cut prices before they have to. Other traditional arguments center around the notion that preemptive price-cutting is irrational because it entails decreased short-term profits and costly competitive actions with no material impact on future profits (Goolsbee and Syverson 2008). So then, why might an incumbent preemptively cut fares before actual competition has occurred? Theoretical articles by Klemperer and Roberts (1982) and Fudenberg and Tirole (1986) suggest that incumbents may resort to preemptive price cutting as a signaling mechanism to appear as if they too are low-cost and subsequently deter entry in hopes of reaping monopoly-like profits on passengerheavy routes. 5 Another theoretical construct in the entry deterrence literature is "predatory pricing," a scenario in which an incumbent engages in a war of attrition against the entrant by slashing fares well below costs and thereby sacrificing short-term profits in hopes of inducing exit. Once the entrant exits, the incumbent charges supracompetitive prices in a monopolistic setting. Because the incumbent firm is often better established and financially stronger, it may be able to sustain predatory practices in order to achieve greater profits when 3 The authors studied how airlines operating from Washington Reagan National (DCA) might change fares as Southwest enters Washington-Baltimore (BWI). DCA and BWI serve air travelers in the greater Washington DC area and are located within 50 miles of each other. 4 The term "quarter" is understood to be fiscal quarter. Quarter 1 is the 3-month period beginning with January 1 and ending with March Alternatively, incumbents may welcome entry if the entrant is a good candidate for an alliance, merger or buyout.

15 The "Southwest Effect" Revisited 15 exit is all but certain. Both arguments assume that entry threats are credible. Figure 2 delineates incumbent prices surrounding LCC entry and exit events in the event of predatory pricing and entry deterrence. Here, P 0, P 1, and P 2 represent the pre-entry, post-entry and post-exit equilibrium prices of the incumbent. Figure 2: Illustrative impact of Southwest entry and exit events on incumbent airfares. Source: Daraban and Fournier (2008) The predatory pricing argument has received mixed reception within empirical experiments. McGee (1958) was among the the first skeptics of predatory pricing. By examining the Standard Oil Company back in 1911, McGee (1958) found that predatory pricing did not drive out competing refiners and that Standard Oil achieved its monopoly through other means such as mergers and acquisitions. On the other hand, Milgrom and Roberts (1982) argued that predatory pricing can result from perfectly rational behavior in anti-competitive settings. Indeed, the airline industry is a perfect example. Several landmark antitrust suits brought forth by low-cost carriers against legacy incumbents have reignited interest in whether firms engage in predatory pricing in order to drive out competitors. In the 1999 case United States of America vs. AMR Corporation, the US government alleges that American Airlines priced its fares and products well below costs, attempting to exclude competition via "reputation for predation." Other recent cases have involved Air Canada, Quantas and Deutsche Lufthansa, each involving an injunction against the more established incumbent for manipulating fares to exclude competition.

16 16 THE MICHIGAN JOURNAL OF BUSINESS Other authors have examined the price distribution resulting from LCC entry. Borenstein and Rose (1994) and Gerardi and Shapiro (2009) both provide compelling but contrasting evidence on the effects of competition on directional movements in price dispersion. Borenstein and Rose (1994) used cross-sectional data to analyze price dispersion on routes and concluded that the absolute difference in fares between two passengers on a route is roughly 36 percent of the airline's average ticket price. In addition, this dispersion was more pronounced on routes with more competition or lower flight density. In contrast, Gerardi and Shapiro (2009), utilizing a panel data, found that price dispersion decreases with competition. Apart from the airline industry, entry deterrence has been closely researched in several other industries. Ellison and Ellison (2007) tested for strategic entry deterrence in the drug and pharmaceutical industry by examining the behavior of pharmaceutical incumbents just prior to losing patent protection. They found that in markets of intermediate size, incumbents reduced advertising immediately prior to patent expiration as evidence of strategic entry deterrence. Along the same vein, Dafny (2005) investigated whether hospitals and other medical establishments also engage in strategic entry deterrence by examining the growth in Medicare claims for electrophysiological studies (EP), a corrective heart procedure, after a 1990 Medicare policy that effectively lowered entry barriers for hospitals seeking to perform EPs. Dafny (2005) found that the growth in the volume of EPs was highest in markets with intermediate attractiveness, or markets where potential entrants are mostly on the fence about entering. In other words, incumbent hospitals in markets facing the most uncertainty around potential competition performed more EPs than hospitals in markets where entry was either extremely likely or unlikely. 2.1 Contribution to the Literature This paper incorporates recent findings on low-cost carrier competition into a comprehensive analysis and expands the empirical findings in the literature in several key ways. Namely, this paper: (1) uses panel data on airfares to examine the extent of incumbent price-cutting from Southwest and JetBlue potential and actual competition during a window of nearly 20 years on both the legacy and low-cost carrier cohorts, (2) examines pricing behavior on routes that Southwest eventually exits to determine patterns of predatory pricing, (3) documents fare-cutting across routes with varying market concentrations of legacy and LCC carriers and across other route characteristics, (4) and analyzes price dispersion on Southwest and JetBlue threatened routes using the Gini coefficient as a measure of price inequality.

17 The "Southwest Effect" Revisited 17 III. Methodology The data used in this paper originates from the Bureau of Transportation Statistics' Airline Origin and Destination Survey (DB1B) database from the first quarter of 1993 through the fourth quarter of The DB1B database contains quarterly data on airfares and provides a random 10% sample of all domestic tickets from reporting carriers. I pulled the following data from the DB1B database: the origin and the destination airports identified by the airport's three letter airport code (i.e. Boston Logan is "BOS"), the fare reported by the carrier on a specific route, the operating carrier, the reporting carrier, the ticketing carrier, the type of trip (i.e. one-way, roundtrip, etc), the coupon type (i.e. first-class), distance printed on the itinerary, the market distance flown, and the number of passengers who paid that particular fare. The obtained raw DB1B data are at the itinerary level meaning that each observation provides the carrier fare for a particular passenger itinerary in a particular quarter. Additional steps were taken to aggregate and clean the raw data. First, all observations were aggregated to the route-carrier-quarter level to resemble a panel dataset with each observation reflecting the average carrier fare, weighted by the number of passengers who paid that fare, on a route in a given quarter. Hence, the average market fare can be expressed below as: where pijt is the reported fare on route j serviced by carrier i in quarter t, nijt is the number of people who paid that particular market fare, and Nijt is the total number of passengers flying carrier i on route j in quarter t. Second, because the raw DB1B lists three separate carrier variables for each observation (reporting carrier, operating carrier and ticketing carrier), a simplifying assumption was made to identify each observation with only the ticketing carrier. 6 Because an air-traveler chooses a particular airline based on the price of the airline that issues the ticket and not based on the airline that actually operates or reports the fare, the ticketing carrier was used to simplify the analysis, similar to Tan (2010). Market share values were calculated by dividing the total number of tickets issued by that carrier in a given route and quarter by the total number of tickets issued by all the carriers on that route. 6 The reporting carrier refers to the carrier that reported the fare to the Bureau of Transportation Statistics. The operating carrier refers to the carrier that actually flies the route. The ticketing carrier refers to the carrier that issues the passenger the ticket for the flight. While they are the same in most instances, the three variables may differ under certain codeshare agreements whereby a regional airline operates the route under a legacy carrier name.

18 18 THE MICHIGAN JOURNAL OF BUSINESS Third, the following was used to clean the raw DB1B data and narrow the original data set. Any observation with fares below $20 or fares above $9998 was dropped from the dataset. Furthermore, the Standard Industry Fare Level (SIFL) dataset was obtained to rule out fares five times the SIFL for that particular route. Tickets with more than 2 coupons for a one-way trip or more than four coupons for a round-trip were dropped. Observations with an unidentified ticketing carrier or connecting flights were dropped. All observations involving a change of planes or to a non-us destination were dropped. Only one-way and roundtrip were included in the sample---open-jaw trips were excluded. 7 Moreover, consistent with previous literature, I focused only on routes flown by the following carriers: American, AirTran, Alaska, Continental, Delta, Frontier, JetBlue, Northwest, Southwest, Spirit, United, and US Airways. This effectively yields a sample of six legacy carriers and six low-cost carriers. Fourth, because fares are affected by a variety of route or airport characteristics (i.e. distance, size of air traffic hub), various types of routes were identified in the empirical exercise to examine the effect of LCC entry across these route characteristics. Flights that stop in either Florida or Las Vegas were identified as leisure routes, as those routes tend to have relatively higher concentration of tourists who travel for leisure. Routes with a LCC market share of over 90% were marked as LCC routes and routes with a legacy market share of over 90% were marked as legacy routes. Routes connecting large endpoint hubs were also marked as these routes tend to be congested and more passenger-heavy. 8 Four airports -- New York LaGuardia (LGA), New York JFK (JFK), Chicago O'Hare (ORD) and DC Reagan National (DCA) -- had quotas on landing slots at some point during my sample period. Routes connecting any of these airports were marked as "slot-controlled" routes. Lastly, the Southwest and JetBlue routes were marked. All routes in which Southwest or JetBlue had presence in both endpoints of a route from 1993 to 2009 were included in the regressions. I confirmed that the routes in my cleaned dataset were in fact Southwest (or JetBlue) threatened routes through press releases (via Lexis-Nexis) and through Southwest's and JetBlue's corporate websites. The variables t 0, t e, and t d will be used throughout the paper to denote the quarters of threat (entry into second endpoint), actual route entry and exit, respectively. 9 This effectively ensures that all routes included in the 7 Open-jaw trips are those from airport A to airport B but a return trip from airport B to airport C involving a change of planes. 8 The BTS classified an airport as a "large" airport if its market share was at least 1% of total enplaned passengers in Similar to Daraban and Fournier (2008), I mark the exit quarter t d as the first quarter that SW's market share drops below 3% preceded by at least four consecutive quarters with market share above 3% or greater. JetBlue did not exit any routes in my dataset.

19 The "Southwest Effect" Revisited 19 regressions are aggregated and "aligned" with respect to Southwest or JetBlue threat and entry events to resemble an event study. I then create Southwest and JetBlue binary variables that turn on if the current quarter of an observation happens to be the quarter during which Southwest or JetBlue threatened, entered or exited that route. 10 In order to capture the lagged effects of all quarters surrounding Southwest's and Jetblue's entry into both the endpoint and the route, dummy variables were also generated up to 4 quarters before the threat period to three quarters after the exit period. For instance, the time dummy variables Southwest_Threat t0-2 and Southwest_Threat t0-1 represent two quarters and one quarter prior to Southwest's endpoint entry and Southwest_Entry t0+1 and Southwest_Entry t0+2 designate one quarter and two quarters after Southwest's entry into the route, respectively. 10 As an illustrative example, Southwest entered PHL in quarter 2 of 2004 and BOS in quarter 2 of For all routes connecting PHL and BOS, the Southwest_Threat dummy equals one if that observation occurred in quarter 2 of 2004 and 0 otherwise. The Southwest_Threat dummy equals one if a PHL- BOS observation occured in quarter 2 of 2009 and is 0 otherwise.

20 20 THE MICHIGAN JOURNAL OF BUSINESS As a technical note, because each Southwest or JetBlue dummy variable essentially turns on for one quarter and remains off otherwise, they are mutually exclusive. In order to isolate the impact of Southwest quarter over quarter, I would take the difference of the reported coefficients. For instance, if the reported coefficient of the dummy in quarter t 0 (initial threat quarter) were ß 0 and the reported coefficient in quarter t e (actual entry quarter) were ß 1, the actual impact of Southwest entry alone vs threat is ß 1 - ß 0. Here ß 1 would be the cumulative impact of entry and threat. From 1993 to Southwest threatened and entered a total of 1,141 routes and JetBlue entered a total of 588 routes. In all, the dataset contains 31,388 unique carrier-route-quarter Southwest observations with a mean log-fares of 5.13 and a standard deviation of For the JetBlue regressions, there are 22,793 unique carrier-route-quarter observations with a mean log-fares of 5.12 and a standard deviation of These results are summarized in Tables 1, 2 & 3. IV. Methodology As mentioned in the previous sections, the empirical models in this paper attempt to capture incumbent pricing responses as a result of LCC threat and entry under the framework of an event study. Similar to Goolsbee and Syverson (2008), I run a probit regression to first determine the probability that Southwest flies a route in a given quarter, conditional on the number of endpoints that Southwest has already operated out of in the previous quarter. From Table 4, the probability that Southwest enters a route given that Southwest operated out of only one endpoint is a 0.26%. However, dual presence in endpoints raises this probability to 17.8%---an increase by a factor of 68. Furthermore, as suggested by Goolsbee and Syverson (2008), not only does Southwest presence in both endpoints raise the probability of entry into the route, the mere announcement, speculation of threat of entry into the endpoint should heighten incumbents' perception of Southwest's imminent entry.

21 The "Southwest Effect" Revisited 21 Similarly, a probit regression was also performed to determine the probability of JetBlue's entry into a route conditioned on the number of endpoint airports with JetBlue presence in the preceding quarter. Single presence of JetBlue, like Southwest's single presence scenario, does not indicate anything meaningful. However, JetBlue's dual presence on the endpoints of a route gives only a 6.9% probability that JetBlue enters the route in the next quarter- -roughly a third of Southwest's 17.8%. This suggests that incumbents may react less drastically to JetBlue's threat given that JetBlue's dual presence is less indicative of actual entry. 4.1 Mean-fare Regression Model Prior studies such as Morrison (2001) utilized cross-section models in order to measure the effect of low-cost competition on incumbent airfares. However, because cross-section models do not necessarily account for unobserved characteristics inherent to a particular airline or route, I employ both fixed- and random-effects panel regressions to account for these unobserved effects. The logarithm of mean airfares were calculated to be the dependent variables in the specification and the coefficients on the Southwest or JetBlue dummies are the main covariates of interest. The baseline regression, borrowing from Goolsbee and Syverson's (2008) empirical methodologies, is given below: (5.1) where ln(p ijt ) is logged mean airfares for incumbent i flying route j in quarter t, γ ij and μ it measure carrier-route and carrier-quarter fixed effects re-

22 22 THE MICHIGAN JOURNAL OF BUSINESS spectively, θ ijt measures the effect of route concentration (using the Herfindahl index), X t accounts for controls such as seasonality and route characteristics (i.e. leisure route, LCC route, etc), and ε ijt is random noise. I included four dummies for four quarters prior to t 0, one dummy for the first quarter after t 0, one dummy for two quarters after t 0 and a single dummy for three or more quarters after t 0. I also included three dummies after the exit quarter t d. The coefficients on the time dummies Southwest_Threat measure the effect of Southwest's presence in both endpoints of a route on incumbent airfares while the coefficients on the Southwest_Entry time dummies measure the effect on incumbent airfares surrounding the period when Southwest actually flies the route relative to the excluded period (i.e. 5 quarters before the threat and 4 quarters after exit). Furthermore, the Southwest_Exit time dummies capture discernible shifts in prices after Southwest exits a route. The fixed-effects (FE) regressions are weighted by the number of identical itineraries per reported fare level, and standard errors are clustered at the carrier-route level. The Herfindahl index, a measure of market concentration on a route, is included to control for competitive effects. I report both FE and RE for the regression specification in (5.1) and perform a Hausman test to determine the suitability of each model. 11 While the FE is the more appropriate model to use (Hausman statistic 0) and the one that is predominantly used in the airline literature today, I include RE regression results for both Southwest and JetBlue main regressions as the results are similar to those derived from FE. 4.2 Gini Coefficient Regression The second model specification attempts to measure the discrete changes in the price dispersion rather than the average price and is modeled on Gerardi and Shapiro's methodology (2009). In order to measure price dispersion, I first calculated the Gini coefficient on a route as a measure of the dispersion in fares paid using the following from Borenstein and Rose (1994): (5.2) where N is the number of different fare levels reported by carrier i on a 11 Recall that the Hausman test checks the validity of the null hypothesis that the constant term is uncorrelated with the error term ε ijt. If the Hausman statistic is large (i.e. p<0.05), then we can reject the null hypothesis and conclude that FE is a more efficient estimator than RE.

23 The "Southwest Effect" Revisited 23 route, fare m is the reported fare for the m th ticket, and PAX m is the reported number of passengers traveling at that fare. A Gini coefficient of zero corresponds to perfect uniformity in prices: everyone pays the same price. On the flipside, a Gini coefficient of one means that everyone pays different prices. Then, in order to obtain an unbound statistic (recall that the Gini coefficient is strictly lodd between zero and one), the log-odds Gini ratio, G ijt, was calculated by ln. Similar to the mean-fare regression specification, I include Southwest dummies surrounding the quarter of SW threat, SW entry and SW exit: (5.3) where G ijt lodd is the log-odds Gini coefficient of carrier i on route j in quarter t. We may expect G ijt lodd to decrease due to Southwest threat and entry. A plausible explanation here is that an increase in competition, or merely the threat of competition in the future, may induce a decrease in an incumbent's market power (Gerardi and Shapiro 2009). Thus, as the power to price discriminate effectively diminishes, we would expect to observe less variation in the price distribution. On the flipside, price dispersion can increase as routes become more congested with LCCs, as empirically verified by Borenstein and Rose (1994). In this scenario, incumbents may charge higher prices for a price inelastic segment of the customer base, since those customers are likely to continue flying that carrier despite higher fares and lower fares for price elastic consumers. This effectively widens the tails of the price distribution and thus increases price dispersion. V. Discussion of Results 5.1 Threat of Entry from Southwest The regression results from specification 5.1 is shown in Column (1) of Table 5 in the Appendix and Figure 3 in the Appendix. Southwest's presence in both endpoints, but before flying the route, equates to a drop in prices of 10.6% (1 - e ) in quarter t 0, the threat quarter and is significant at the 5% level. The reported coefficients of the Southwest_Threat dummies in quarters t 0-4, t 0-3, t 0-2, and t 0-1 show that fares decrease slightly, but imprecision

24 24 THE MICHIGAN JOURNAL OF BUSINESS in my estimation strategy precludes these results from being statistically significant. On routes where Southwest threatens, but does not enter for at least three quarters after entry into second airport (i.e. t 0 + 3$ to t $), fares drop 15.5%, reflective of the entry deterrence behavior of incumbents. By the time Southwest enters these routes in quarter t e, fares have dropped slightly over 14% (1 - e ) relative to the excluded period. Ultimately, fares drop 18% one to two quarters after Southwest flies the route and 23% at least three quarters after Southwest route entry. Both results are significant at the 1% level. Column (2) of Table 5 reports the dummy coefficients of the LCC incumbents. Contrary to the aggressive price cutting seen from legacy incumbents, low-cost incumbent fares do not drastically decrease fares in quarter t 0. On routes where Southwest threatened for at least three quarters before entry, fares decreased by only 7.4% and the coefficients are largely insignificant. In fact, we don't observe a sizable drop in prices until at least the first quarter after Southwest actually flies the route (t e +1) where fares drop 9.9% and ultimately drop 12.2% 3 quarters after route entry. The disparity between the results in Columns (1) and (2) can be partially explained by a low-cost entrant's strategy, which is to match, or undercut, the prices of a low-cost carrier incumbent. Because legacy fares tend to be much higher than LCC fares on a given route, legacy carriers are more prone to decrease their prices in response to the entrant's lower fares. This downward price pressure is only transient in low-cost carrier incumbents case as low-cost carriers are less engaged in direct price competition with one another. To determine whether the trend in prices is suggestive of actual predatory pricing or simply a direct product of Southwest competitive effects, we examine what happens to fares after Southwest exits a route. During the sample period from 1993 to 2009, Southwest exited three airports: Detroit Metro (DET) in 1993, San Francisco (SFO) in 2001 and Houston George Bush International (IAH) in Columns (1) and (2) of Table 5 reports the coefficients on the Southwest exit dummy variables. Overall, the results show an increase in fares in the legacy incumbents cohort particularly between the quarter of exit t d and one quarter after t d. In fact, fares were 23% lower from t e + 3 to t e + 12 but only 11% lower during t d, suggesting a 12% spike in fares during those two periods. While subsequent post-exit dummy coefficients are statistically insignificant, due to the small sample of routes that Southwest exited, we observe that prices were in general higher during Southwest's post-exit quarters than before exit. The post-exit dummy coefficients for the low-cost incumbent in Column (2) did not increase to the same extent as the legacy fares. One possible explanation for the discrepancy in pricing behavior between legacy incumbents and 12 Southwest would later reenter SFO in 2007.

25 The "Southwest Effect" Revisited 25 low-cost incumbents is that legacy carriers, due to their size and considerable market share, can afford to raise prices knowing that customers on those routes have few low-cost alternatives after Southwest exits. For instance, selecting for routes with endpoints at either SFO or IAH, the two airports Southwest exited from 2001 onward, legacy carriers had a mean total market share of 73.1% on these routes with the largest single LCC average market share at 5.1%. If legacy carriers were indeed hit by the Southwest Effect of low fares, it would seem rational for them to increase fares once Southwest leaves to recoup for losses as a result of Southwest competition. 5.2 Sensitivity Analysis Table 6 reports the coefficients of the dummy variables across several specification criteria in order to see whether incumbent pricing behavior alters significantly across different route characteristics. Note that the sensitivity analysis restricted incumbents to legacy carriers only since the LCC cohort was much smaller and harder to break down by category. Furthermore, this portion of the analysis only looked at Southwest as a potential entrant since there was more data to base the analysis on. Column (1) of Table 6 reports the regression results for the baseline specification. Columns (2), (3) and (4) report the coefficients for leisure routes, LCC routes and legacy routes. 13 While incumbents on all three route types ultimately reduced fares, price cuts on the LCC routes and the legacy routes were most prominent. In Column (2), one can see that legacy carriers flying to a tourist destination were highly sensitive to potential competition from Southwest. Because neither Las Vegas nor the major airports in Florida (except Miami) are large legacy carrier hubs, flyers who travel to these destinations are most likely flying to their final destinations rather than using these airports as a connecting hub. Moreover, flyers traveling to these destinations are likely tourists with lower demand elasticity, meaning that they are most likely choosing the lowest fare they can find. Hence, legacy carriers are more likely to respond by cutting fares in order to stay competitive with Southwest and other LCCs, even under mere threat alone. Columns (3), (4), and (5) report the coefficients on LCC routes, legacy routes and routes involving large endpoint hubs. Compared to the pooled results in Column (1), the results in Columns (3) and (5) suggest that fares decreased more drastically on routes with heavy legacy carrier concentration and those involving dominated hubs. The results in Column (4) are consistent with 13 Recall that leisure routes were routes connecting airports in Florida or Las Vegas, LCC routes were identified as routes where the mean market share of LCCs was at least 90%, and legacy routes were those with mean market share of legacy carriers at least 90%

26 26 THE MICHIGAN JOURNAL OF BUSINESS Morrison (1997) who concluded for instance that, despite Delta Air Lines's dominance in Salt Lake City (SLC), fares in SLC were 16% lower as a result of Southwest presence in 1996 and 39% lower when compared to non-southwest airports. Because legacy concentrated routes and dense traffic hubs have higher fares to begin with -22% higher than other airports according to Morrison (1997) - it appears logical that fares would decrease more from potential entry vis-a-vis routes with that are less heavily traveled. Column (6) of the sensitivity reports the coefficients for those airports with slot controls (LGA, JFK, ORD and DCA). Upon inspection, one can see that Southwest has only marginal impact during the quarters surrounding the threat period and the results are largely inconclusive except three quarters after entry with fares dropping 13.3%. In theory, slot-controlled airports should in fact have lower fares than airports that are not slot-controlled since the slot system is meant to decrease congestion and encourage LCC entry and regulating the number of takeoffs certain airlines are permitted to have. There are two main reasons why fares on these routes are immune to the Southwest Effect. First, the slot-system is inefficient. Morrison (1997), in a testimony before the US House of Representatives in 1997, concluded that the fare premia on routes involving ORD, LGA and DCA in 1996 were 11 to 15% higher than flights involving other airports. Second, all of these four airports are hubs of major legacy carriers: American Airlines' hubs include ORD and JFK, US Airways is DCA's largest carrier, and Delta Air Lines has major operations out of LGA. Hence, many passengers flying to other destinations must fly through one of these hub airports, suggesting that these passengers have higher demand inelasticity, which translates into higher market power for incumbent airlines. It would appear logical that incumbents on these routes can afford to keep prices high until Southwest actually enters. 5.3 Evidence of Entry Deterrence Table 7 displays the effect on airfares for routes that Southwest entered immediately upon entry, routes that Southwest still threatens but has not entered for over 12 quarters after the threat quarter and routes connecting airports within 50 miles of a Southwest airport in Columns (2), (3) and (4) respectively. The behavior of incumbents on pre-announced routes is significantly different - none of the coefficients on the quarters before actual entry is significant nor substantially negative, suggesting that incumbents do not preemptively slash fares when Southwest entry is certain. Only in the actual entry quarter do fares drop. In Column (3), we see that incumbents do not significantly slash fares until at least two quarters after entry and that fares decreased by 16.2%

27 The "Southwest Effect" Revisited 27 between three quarters and 12 quarters after the threat period, suggesting that even on routes on which Southwest does not fly, the mere possibility of Southwest flying in the future drives fares lower. Lastly, Column (4) examines routes in nearby airports and suggest that there is no discernible pattern in airfares on routes connecting hubs within close proximity to a Southwest airport. The results in Column (4) run counter to the findings of Dresner, Lin and Windle (1996) who concluded spillover effects onto other routes existed, although the authors focused only on actual entry. 5.4 Threat of Entry by JetBlue The mean fare regressions for JetBlue are reported in Table 8. In contrast to the Southwest results, it appears that neither legacy nor low-cost carriers respond to JetBlue threat. The reported coefficients in the threat periods are neither sizable nor significant, suggesting that JetBlue has minimal effects on incumbent fares. The greatest drop in fares occurs around one to two quarters after JetBlue entry, with fares dropping 7.6% and 8.8% in quarters t e + 1 and t e + 2 respectively for legacy incumbents and dropping 11% during the same periods for LCC incumbents. Upon further inspection, it does not appear that fares drop further afterwards with the reported coefficients on t e + 3 to t e +12 appearing small and insignificant for both incumbent cohorts. This suggests that incumbents react less aggressively to JetBlue's threat and entry than to Southwest's. These results are graphically depicted in Figure 4 in the Appendix. Several reasons can be hypothesized as to why fares are lower on the Southwest routes. First, because of its larger network as well as a longer-standing reputation as an low-cost airline, Southwest may elicit a stronger response from incumbents. Tables 3 and 4 show that Southwest not only threatened and entered into twice as many routes as did JetBlue during our sample period but entry is almost three times as likely to occur if Southwest establishes dual presence in the endpoints than if JetBlue establishes dual presence (17.8% vs. 6.9%). In 2003, Southwest served roughly seven times more passengers (65.7 million vs 9 million) and roughly ten times (2,800 daily flights vs 252 flights) more daily flights than JetBlue did. 14 Yet, a second hypothesis is that Southwest engages in more direct price competition with an incumbent by more frequently undercutting existing fares rather than by simply matching the fares. Table 10 reports the frequency and percentage that either Southwest or JetBlue enters a route with an average price higher than, equal to, or below an incumbent's average price weighted 14 Chris Woodyard,``Pitting Southwest vs. JetBlue," \emph{usa TODAY}.

28 28 THE MICHIGAN JOURNAL OF BUSINESS by the number of identical itineraries. 15 From the data, it is rare for either Southwest or JetBlue to set a price higher than an incumbent's average fare, which helps explain why incumbent prices would fall after entry. Moreover, upon closer inspection, one can see that Southwest tends to undercut its legacy incumbents more so than JetBlue. For instance, on routes with American Airlines as the incumbent, Southwest undercut AA 57.3% upon entry, while Jet- Blue undercut AA 31.5% upon entry. If Southwest sets initial fares lower than the incumbent's existing fares more often than does JetBlue, then incumbents will be more likely to react in the form of larger cuts. Third, differences between Southwest's and JetBlue's operating models may shed additional insight into Southwest's slight advantage in the low-fare battle. Southwest's competitive advantage germinates from its dense operations out of smaller and less-conveniently located airports, helping the airline save money on landing fees. With lower operating costs, Southwest has the financial flexibility to lower costs without compromising profitability. On the other hand, JetBlue has attempted to develop a business model that resembles both the legacy carrier and LCC models. 16 In particular, JetBlue has aimed to "modernize" the customer traveling experience by affording free DirectTV, XM Satellite Radio, leather seats, in-flight entertainment and other amenities - an aspect of consumer travel that Southwest had not paid as much attention to. Because legacy incumbents may realize that they are not simply competing with JetBlue on a pure price basis, drops in existing fares may not be as sensitive to JetBlue competition as to Southwest competition. 5.5 Price Dispersion The regression results of specification (5.3) is reported in Table 9 in the Appendix and shown in Figure 5 of the Appendix. Column (1) shows that the log-odds Gini ratio decreased in general on routes threatened by Southwest relative to the excluded period. The greatest drop in the Gini coefficient occurs in quarter t 0 with the Gini dropping 9.3%. This suggests that fares on Southwest threatened routes became more uniform, especially during the quarters surrounding the threat and entry events. The regression results in Column (1) are consistent with Gerardi and Shapiro's (2009) findings that price dispersion decreases as competition increases on a route. The intuition here is that Southwest competition diminishes incumbents' market power and thus their ability 15 Following Tan (2010), I constructed a $20 window such that price matching occurs if the entrant's average fares falls within the $20 window. In order for the entrant to have set a price higher (lower) than the incumbent's price, the entrant's average fare must be at least $20 greater (lower) than the incumbent's average fare. Each observation is weighted by the number of identical itineraries. 16 Stephen Ellis, "The Decline of Southwest and the Rise of JetBlue," The Motley Fool.

29 The "Southwest Effect" Revisited 29 to price discriminate. As a result, prices become less dispersed on routes that are more competitive. Column (2) reports the coefficients for the low-cost incumbents. Similar to the mean-fare regressions, the price dispersion results are inconclusive with the coefficients negative during quarters before Southwest entry, and positive after entry. Perhaps because of the smaller distribution in fares set by low-cost carriers to begin with, there is little subsequent effect and movement in price dispersion that can be picked up with the methodologies presented. Columns (3) and (4) contain the analogous results for routes threatened by JetBlue. Interestingly, the results in Column (3) suggests that price dispersion rose on JetBlue-threatened routes. The general increase in the Gini coefficient quarter-to-quarter is consistent with Borenstein and Rose's (1994) findings that price dispersion increases as routes become more competitive. Since those who frequently fly accumulate credit through frequent flyer and other reward programs and would subsequently lose out on those rewards by switching to another airline, brand-loyal consumers are willing to pay a higher fare rather than incur a switching cost. Hence, incumbents may keep higher-end fares the same knowing brand loyal consumers will likely not switch airlines while decreasing discount fares to lock in the price-sensitive consumers. Under the aforementioned scenario where higher-end fares are approximately the same level while the discount fares are cut, the price distribution has become flatter which would likely result in the increase in price dispersion we see in Column (3). VI. Conclusion The impact and threat of low-cost carriers on incumbent fares have been well documented in the field of applied economics and industrial organization. The purpose of this study is to add additional insight into why incumbent airlines reduce fares as a result of LCC entry and whether this behavior changes across various incumbent and route types. In addition, this paper also briefly examined discrete changes in the Gini coefficient of both Southwest and Jet- Blue routes as a result of the threat and entry events and offered suggestive evidence in line with previous research. Using airline data on US domestic flights from 1993 to 2009, I find evidence that legacy incumbents engaged in preemptive and aggressive price-cutting to Southwest threat and entry but in much milder fashion to JetBlue threat and entry. Furthermore, low-cost incumbents did not react significantly to either Southwest or JetBlue threat, although fares ultimately declined upon actual entry. Using sensitivity analysis across route types to extend upon the main

30 30 THE MICHIGAN JOURNAL OF BUSINESS regression, I concluded that fares on leisure routes, legacy-dominated routes and routes connecting endpoints with high traffic volume all witnessed severe drops in fares from Southwest threat and entry. Routes involving airports that were slot-controlled surprisingly were not all that much affected from Southwest threat. Furthermore, by examining fares on routes that Southwest eventually exited, this paper provided some suggestive evidence of predatory pricing as fares in the post-exit periods gradually rose. Lastly, I also examined changes in price dispersion and concluded that legacy carrier prices became more uniform as a result of Southwest threat and entry and less uniform as a result of JetBlue entry. The application of empirical models to the airline industry and the findings presented in this paper have powerful policy implications. If the impact of Southwest is as beneficial to the average traveler as the research suggests, policies should be designed to promote low-cost carrier presence in markets with high congestion and fares. There is significant value in understanding the dynamic nature of strategic entry deterrence as it applies across industries - future empirical work on this topic should garner considerable attention from both academia and industry.

31 The "Southwest Effect" Revisited 31 References Bennett, Randall and James Craun, "The Airline Deregulation Revolution Continues: The Southwest Effect," Washington, DC: Office of Aviation Analysis, US Department of Transportation, Berry, Steven and Panle Jia, "Tracing the Woes, an Empirical Analysis of the Airline Industry," American Economic Journal: Microeconomics, v. 2(3), pp. 1-43, Borenstein, Severin, "The Evolution of U.S. Airline Competition," The Journal of Economic Perspectives 45-73, Borenstein, Severin, "Hubs and High Fares: Dominance and Market Power in the U.S. Airline Industry," The RAND Journal of Economics, p , Borenstein, Severin and Nancy Rose, "Competition and Price Dispersion in the U.S. Airline Industry," The Journal of Political Economy, p , Cairns, Robert D. and John W. Galbraith, "Artificial Compatibility, Barriers to Entry, and Frequent-Flyer Programs," The Canadian Journal of Economics, Vol. 23, No. 4, Dafny, Leemore, "Games Hospitals Play: Entry Deterrence in Hospital Procedure Markets," Journal of Economics and Management Strategy Vol. 14, p , Daraban, Bogdan and Gary Fournier, "Incumbent Responses to Low-cost Airline Entry and Exit: A Spatial Autoregressive Panel Data Analysis," Research in Transportation Economics Vol. 24, p , Dixit, Avinash, "A Model of Duopoly Suggesting a Theory of Entry Barriers," The Bell Journal of Economics, Vol. 10, No. 1, p , Dixit, Avinash, ``The Role of Investment in Entry-Deterrence," The Economic Journal, Vol. 90, No. 357, p , Dixit, Avinash, ``The Role of Investment in Entry-Deterrence," The Economic Journal, Vol. 90, No. 357, p , Dresner Martin, Chris Lin and Robert Windle, "The Impact of Low-Cost Carriers on Airport and Route Competition," Journal of Transport Economics and Policy, Vol. 30(3), p , Ellis, Stephen, "The Decline of Southwest and the Rise of JetBlue," The Motley Fool Ellison, Glenn and Sara Ellison, "Strategic Entry Deterrence and the Behavior of Pharmaceutical Incumbents Prior to Patent Expiration," MIT Working Paper, Fudenberg, Drew and Jean Tirole, "A Signal Jamming Theory of Predation," Rand Journal of Economics, Vol. 17, p , 1986.

32 32 THE MICHIGAN JOURNAL OF BUSINESS Gerardi, Kristopher and Adam Shapiro, "Does Competition Reduce Price Dispersion? New Evidence from the Airline Industry," The Journal of Political Economy, p , Goolsbee, Austan and Chad Syverson, "How Do Incumbents Respond to the Threat of Entry? Evidence from the Major Airlines," The Quarterly Journal of Economics, p. 1-37, Klemperer, Paul, "Markets with Consumer Switching Costs," The Quarterly Journal of Economics, p , Klemperer, Paul, "Competition when Consumers have Switching Costs: An Overview with Applications to Industrial Organization, Macroeconomics, and International Trade," The Review of Economic Studies, Vol. 62, No. 4 p , Klemperer, Paul, "Collusion via Switching Costs: How ``Frequent-Flyer" Programs, Trading Stamps, and Technology Choices Aid Collusion," Working Paper 786, Graduate School of Business, Stanford University, May McGee, John, "Predatory Price Cutting: The Standard Oil (N.J.) Case," Journal of Law and Economics Vol. 1, p , Milgrom, Paul and John Roberts, "Limit Pricing and Entry Under Incomplete Information: An Equilibrium Analysis," Econometrica Vol. 50, p , Morrison, Steven A., "Actual, Adjacent, and Potential Competition: Estimating the Full Effect of Southwest Airlines," Journal of Transport Economics and Policy, p , Rosenthal, Robert, "A Model in which an Increase in the Number of Sellers Leads to a Higher Price," Econometrica, p , Tan, Kerry, "Incumbent Response to Entry by Low-Cost Carriers in the US Airline Industry," Ohio State University Working Paper, p. 1-33, Vowles, Timothy, "The Effect of Low Fare Air Carriers on Airfares in the US," Journal of Transport Geography, p , Vowles, Timothy, "The Southwest Effect In Multi-Airport Regions," Journal of Air Transport Management, p , Winston, Clifford and Steven A. Morrison, "The Fare Skies: Air Transportation and Middle America," The Brookings Institution, Fall Woodyard, Christopher, "Pitting Southwest vs. JetBlue," USA TODAY, July 5, 2004.

33 The "Southwest Effect" Revisited 33 Appendix 1. Tables

34 34 THE MICHIGAN JOURNAL OF BUSINESS

35 The "Southwest Effect" Revisited 35

36 36 THE MICHIGAN JOURNAL OF BUSINESS

37 The "Southwest Effect" Revisited 37

38 38 THE MICHIGAN JOURNAL OF BUSINESS

39 The "Southwest Effect" Revisited Charts

40 40 THE MICHIGAN JOURNAL OF BUSINESS

41 41

42 42 THE MICHIGAN JOURNAL OF BUSINESS

43 43 The Effects of Laddering and Spinning in Underwriter Manipulation of IPOs Ludwig Chincarini Alex Etzkowitz Jonathan Kadish 1 University of San Francisco Abstract Using a unique and larger sample of companies from going public from classified by both spinning and laddering manipulation, we are able to disentangle the effects of laddering and spinning and find that much of the underpricing during the internet bubble was due to laddering rather than spinning. In fact, we find that a laddered IPO is expected to be underpriced by 110% compared to a non-laddered IPO. The effects of spinning are insignificant once laddering is simultaneously considered. Our study lends further support to the laddering theory of Aggarwal et al. (2006) as a reason for IPO underpricing during the internet bubble. 1 The authors would like to thank Gary Smith for comments. Contact: Ludwig Chincarini, CFA, Ph.D., is an Associate Professor at the University of San Francisco, School of Management, 2130 Fulton Street, San Francisco, CA chincarinil@hotmail.com. Phone: Alex Etzkowitz's contact is ace02006@mymail.pomona.edu, and Johnathan Kadish's contact is jonathan. kadish@gmail.com.

44 44 THE MICHIGAN JOURNAL OF BUSINESS I. Introduction During the period first-day initial public offering (IPO) returns averaged 65%, compared to the long-run average for first day returns of 10% (Loughgran and Ritter (2004)). First day IPO returns during the internet bubble period were so large that the traditional explanations of IPO underpricing were no longer sufficient. Before 1998, explanations for IPO underpricing focused on the riskiness associated with the future earnings of issuer firms and informational frictions between the various parties involved in an IPO. 2 New theories which attempted to reconcile the seemingly inexplicable returns of the period with economic theory point to the importance of market manipulation and the realignment of issuer firm incentives. In this paper we will examine two forms of market manipulation, spinning and laddering, and test the hypothesis that these illegal activities are primarily responsible for the abnormally large first-day IPO returns from Ritter and Beatty (1986) provide a theoretical model of IPO underpricing consistent with portfolio theory, positing that underpricing is positively related to the uncertainty associated with a firm s prospects. Additional risk must be offset by higher expected returns so underwriters may intentionally underprice to attract investors to the IPOs of especially risky firms. Ritter and Beatty (1986) hypothesize that more speculative ventures tend to have smaller IPO offerings and thus are more underpriced. Notably, the SEC requires more speculative ventures to provide detailed disclosures for the uses of funds raised by equity sales in the Use of Proceeds section of the filing. As predicted, they find that offer size is negatively associated with underpricing while the number of uses for IPO proceeds is positively associated with underpricing. Ritter (1984) uses similar logic in an attempt to explain the hot issue market of Although he finds a correlation between risk and average initial returns in some sectors, he finds evidence that underwriters were simply exploiting issuers by underpricing in the natural resource sector. Barry et al. (1990) provide further evidence for this hypothesis by showing that the IPOs of older firms with respect to their date of formation tend to be less underpriced than those of newer firms. They argue that older companies have more established reputations and longer earnings histories and thus there is less uncertainty associated with future earnings. Unfortunately, simple models of issuer firm risk 2 In perhaps the first attempt to theoretically explain IPO underpricing, Baron (1982) argues that underwriters have superior information about capital markets and are compensated for it by issuers through underpricing. Rock (1984) suggests that the presence of informed investors forces underwriters to underprice new issues relative to their true value in order to ensure that uninformed investors purchase the issues. Welch (1989) argues that underpricing is utilized by high-quality firms in an attempt to distinguish themselves from low-quality firms for whom underpricing is relatively more costly.

45 The Effects of Laddering and Spinning in Underwriter Manipulation of IPOs 45 fail to explain much of the jump in underpricing during the internet bubble period. Loughran and Ritter (2004) propose the changing risk hypothesis that firms going public in the internet bubble period were unusually speculative ventures but find that it explains only a small part of the increase in underpricing. Other explanations focus on the incentives of issuer firms, underwriters, venture capitalists, auditors and lawyers involved in IPOs. The grandstanding hypothesis proposed by Gompers (1996) suggests that young venture capital firms have strong incentives to rush IPOs to market in order to establish their reputations and increase the likelihood of raising money for follow-on funds. Therefore these firms are more willing to transfer wealth from themselves (pre-ipo investors) to the purchasers of newly issued securities through the practice of IPO underpricing. Lee and Wahal (2004) empirically test the Gompers' hypothesis over the period from and report results in favor of the existence of grandstanding. Additionally, they find that IPOs backed by venture capital firms (or VC-backed IPOs ) in general had significantly larger underpricing than non-vc-backed IPOs during this period. Demers and Lewellen (2003) provide evidence that similar incentives may have been present for issuer firms during the internet boom due to substantial advertising and marketing benefits. They find that IPO underpricing is positively associated with both subsequent newspaper article mentions in the Lexis-Nexis database and with issuer firm website traffic. Beatty and Welch (1996) find that high-quality auditors and well-compensated lawyers are both associated with less underpricing. Moreover, they find that the negative relationship between underwriter prestige and underpricing in the 1980s was actually reversed in the 1990s. Although Loughran and Ritter (2004) find that VC-backing and prestigious underwriters are both positively associated with underpricing during the bubble period, these explanations fail to explain the increase in underpricing in general from For example, they find that non-vc backed IPOs and IPOs with low-prestige underwriters were still underpriced by over 30% during the bubble period, more than double the average underpricing from In response to the failure of traditional models of underpricing to explain the period, Ljunqvist and Wilhelm (2003) argue that underpricing during the internet bubble is a function of increased ownership fragmentation, and smaller proportional CEO ownership stakes. In this formulation, CEOs and other issuer firm executives had weaker incentives to bargain with underwriters for higher prices because they held smaller proportional stakes in their firms. However, Loughran and Ritter (2004) argue that there is scant empirical evidence for this explanation and further note that although

46 46 THE MICHIGAN JOURNAL OF BUSINESS their proportional shares were smaller, in terms of the total dollar value of holdings CEOs held more equity in their IPOs during the bubble period than they had in the previous three years. More recent attempts have focused on aspects of what Loughran and Ritter (2004) call the changing issuer objective function hypothesis. They specifically refer to two separate changes in issuer objectives: spinning and analyst lust. Spinning refers to the practice of allocating intentionally underpriced IPO shares as a payoff to corporate clients, while analyst lust refers to the desire of issuer firms to receive favorable analyst coverage. An additional factor that likely contributed to underpricing during the late 1990s was tie-in agreements whereby purchasers agree to buy additional shares of an issue in the aftermarket in exchange for access to the IPO. Aggarwal et al. (2005) investigate this phenomenon, commonly referred to as laddering, and find that laddered IPOs between 1998 and 2000 had seven-times higher first day returns relative to IPOs without tie-in agreements. Although the spinning and laddering hypotheses have each been tested in isolation, no paper has attempted to determine the relative effects of these two actions on underpricing. We determine the impact of each action using a new data set that incorporates a variety of legal complaints. We hypothesize that laddering and spinning will both be positively associated with underpricing, and that the inclusion of these two variables will explain much of the extreme underpricing during the period. The paper is organized as follows: section II describes the basic types of IPO market manipulation examined in this paper; section III discusses the data and methodology used in this paper; section IV discusses the empirical results; and section V concludes. II. Types of IPO Market Manipulation 2.1: The Spinning Hypothesis The spinning hypothesis suggests that issuer firm executives and venture capital investors will hire underwriters that underprice in exchange for separate side payments made directly to the individual decision makers. Spinning is a process through which these side payments can be made. According to Liu and Ritter (2010), ``Spinning is the allocation by underwriters of the shares of hot initial public offerings (IPOs) to company executives in order to influence their decisions in the hiring of investment bankers and/or the pricing of their own company s initial public offering."

47 The Effects of Laddering and Spinning in Underwriter Manipulation of IPOs 47 Ljungqvist and Wilhelm (2003) briefly mention the rise of directed share programs (DSPs), whereby issuers allocate a predetermined proportion of shares in an IPO to ``friends and family" including employees of the underwriter, customers and strategic corporate partners. This practice is distinct from spinning in that it is the issuer and not the underwriter who is ``directing" shares. Griffith (2003) points out that issuers control the allocation of a miniscule proportion of shares in an IPO through DSPs, as opposed to underwriters who control the vast majority of shares. Further, this practice, unlike spinning, is generally legal as long as it is in compliance with SEC and/or NASD rules. Ljungqvist and Wilhelm show that from DSPs became increasingly popular, with the fraction of DSPs rising from 19% in 1993 to 91% in Additionally, DSPs consistently enter Ljungqvist and Wilhelm s regressions on underpricing with positive coefficients significant at the 5% level. Although this evidence does not directly confirm the spinning hypothesis, it is highly suggestive as issuers and underwriters likely face similar incentives to please potential clients, business partners, and others by providing access to severely underpriced IPOs. Loughran and Ritter (2004) argue that the spinning hypothesis is broadly consistent with the evidence because top-tier underwriters were more associated with underpricing during the internet bubble period. They use toptier underwriters as a proxy for ``the ability and willingness to spin." However, without directly observing which decision makers had access to spun IPOs, the links between spinning and underpricing remain tenuous. Liu and Ritter (2010) aim to verify the relationship between spinning and underpricing by using a sample of 56 IPOs in which executives were allegedly spun by the infamous investment banker Frank Quattrone. They find that the first day returns of these IPOs were 23% higher than non-spun IPOs with similar characteristics. Although they find that both spinning and all-star analyst coverage are positively and significantly correlated with underpricing, they argue that these two explanations account for only ``9% of the 65% average underpricing during the bubble." In our work, due to the use of a larger dataset on spinning, and since we examine both spinning and laddering simultaneously, we believe these hypotheses may have more explanatory power than these authors have found. 2.2: Laddering Hypothesis Aggarwal et al. (2005) examine the practice of laddering, in which IPO investors agree to purchase shares in the aftermarket of an IPO in exchange for being granted access by the underwriter to invest in the IPO at the offer price. Underwriters use laddering to create excess demand which serves to artifi-

48 48 THE MICHIGAN JOURNAL OF BUSINESS cially inflate the price of the issuer firms stock in the very short run. Logically, if investors are willing to agree to pay a higher price than the offer price in the aftermarket, they must believe that the true value of the shares lies above the offer price as determined by the underwriter. In other words, laddering can only occur when an underwriter intentionally underprices an IPO relative to its knowledge of the available demand for the issuer firm s equity. Aggarwal et al. show that during the period from , IPOs that were laddered exhibited significantly different returns than those that were not. Interestingly, they also find that laddered IPOs subsequently underperform compared to non-laddered IPOs in the long run. This corroborates the findings from Purnanandam and Swaminathan (2004) who suggest that first day underpricing is negatively associated with the long-run performance of IPOs. Finally, Hao (2007) proposes that more expected underpricing unrelated to laddering will result in increased laddering and thus even more underpricing. This positive feedback loop between underpricing and laddering may explain both the size and persistence over time of underpricing from The class action lawsuits filed in late 2000 from which we draw much of our data potentially imposed new costs on laddering and broke the positive feedback loop Hao refers to even before the SEC began investigating IPO malfeasance in and regulations were tightened in : Recurring Trends Within the Czech Republic, three commodities nuclear reactors (84), electrical machinery (85), and vehicles (87) account for the largest increase in exports as well as the imports. These findings corroborate a strong intraindustry trade that allows the Czech Republic to support a growth in exports with a growth in imports of the same commodities. For vehicles (87), however, the intra-industry trade masks Outward Processing Trade that can be explained with a deeper level of disaggregation. While, aircraft and spacecrafts (88) are exiting Czech Republic s trade structure as a whole, this commodity is one of the few drastic changes within the import trade structure. III. Data and Methodology In order to test our hypotheses, we use data from 4,717 IPOs from IPO data is primarily from Thomson ONE Banker (T1B). We define 3 See 4 See

49 The Effects of Laddering and Spinning in Underwriter Manipulation of IPOs 49 underpricing as IPO Underpricing = where P C is the price at the end of the first day of trading and P O is the price designated by the underwriter for the IPO. We test the impact of spinning and laddering 5 on IPO underpricing controlling for underwriter prestige 6, previous venture capital investment (from T1B), issuer firm age (from T1B), issue size (from T1B), industry earnings volatility and industry price earnings (PE) ratio (from COMPUSTAT). Table 1. Summary Statistics for Regression Variables In order to create the dummy variables for spinning and laddering, data were pooled from a variety of sources including the U.S. Securities and Exchange Commission (SEC) and the Financial Industry Regulatory Authority (FINRA) complaints, class action lawsuits and previous published work. The majority of the data for laddering was obtained from 309 coordinated class action lawsuits filed in the United States District Court for the Southern District of New York beginning in These cases were collectively known as IPO Securities Litigation, 21 MC 92 (SAS), and were eventually settled for a total of $586 Million in The IPO s for each of the 309 issuer firms listed in the Consolidated 5 Source: 6 Source: 7 Source:

50 50 THE MICHIGAN JOURNAL OF BUSINESS Amended Complaints are considered to have been laddered in our data. 8 Additional data on laddering was gathered from SEC complaints against J.P. Morgan Securities Inc. 9, Morgan Stanley & Co Inc. 10, and Goldman, Sachs & Co. 11 Each of these three firms eventually reached settlements with the SEC numbering in the tens of millions of dollars. 12 As with the class action lawsuit, any IPOs mentioned in these complaints receive a 1 for our laddering dummy variable. To create the dummy variable for spinning, we use a methodology similar to Liu and Ritter (2010) to identify IPOs for which executives either at the issuer firm or other corporate clients received side payments in the form of IPO shares in their personal brokerage accounts. Liu and Ritter (2010) assemble data from on IPOs spun by a prominent investment banker, Frank Quattrone, while he worked at Deutsche Morgan Grenfell (DMG), Credit Suisse First Boston (CSFB), and Solomon Smith Barney (SSB). IPOs that were spun at DMG and CSFB are taken from Government Exhibit 2051 dated March 21, 2000 from the case against Quattrone. IPOs spun at SSB were taken from information supplied to the Office of the Attorney General of New York through a Freedom of Information Act request. We classify all 56 IPOs listed in Liu and Ritter s Appendix Table IA-1 as having been spun. 13 Further, we include the IPOs allegedly spun in a class action suit filed in the Northern District of California against CSFB which are not contained in Ritter and Liu s data. 14 It should be noted that all IPOs listed in the Northern District of California class action suit are classified as having been laddered and spun whether or not they are contained in Ritter and Liu s data as the case alleges that CSFB engaged in both practices. Finally, we include all IPOs allegedly spun in SEC complaints against Citigroup Global Markets Inc. 15, Robertson Stephens Inc. 16, and NASD s allegations against Piper Jaffray & Co. 17 To our knowledge, the data we have collected on spinning and laddering represent the most exhaustive dataset currently available (see Table 2) and

51 The Effects of Laddering and Spinning in Underwriter Manipulation of IPOs 51 Table 2. Summary Statistics for Spinning and Laddering Dummy Variables The industry PE and industry earnings volatility variables are created using data from Compustat for the years For industry PE (variable Industry_PE), we use the average of the price to earnings ratios of all of the firms with the same SIC code in the same year. We use the industry PE ratio from the previous year to ensure that all the information captured by our variable would have been available to the market at the time of the IPO. Theoretically, the industry PE ratio should not enter significantly in our regressions. Underwriters use a variety of valuation methods when setting the offer price for an IPO including most commonly the dividend discount model (DDM), discounted free cash flow models, and valuations based on price/earnings and/or price/cash flow of comparable firms. Deloof et al. (2002) show that compared to the DDM, valuation methods such as the PE ratio tend to arrive at higher values. However, unless investors rely more on PE models than underwriters there should be no wedge between IPO offer prices and the closing price on the first day of trading based on industry PE. Kim and Ritter (1999) provide further evidence for why a simple measure such as industry PE should not affect underpricing: The difficulty of using comparable firm multiples for valuing IPOs, without further adjustments, leaves a large role for investment bankers in valuing IPOs. Because using the midpoint of the offer price range results in smaller prediction errors than using comparables, investment bankers apparently are able to do superior fundamental analysis. In addition, investment bankers are able to achieve additional valuation accuracy by canvassing market demand before setting a final offer price. While much attention has been focused on the wide variation between the offer price and subsequent market prices that occurs in practice, our results suggest that the pricing precision would be much worse if a mechanical algorithm was used instead.

52 52 THE MICHIGAN JOURNAL OF BUSINESS Therefore, a significant positive relationship between industry PE and underpricing would suggest that naive investors may be overpaying for issues in highly valued or hot sectors compared to the more complex analyses of underwriters. This result would confirm Ljungqvist and Wilhelm s proposition that investors were simply optimistic in the extreme, and the incentives of both issuer firms and underwriters were to fan the flames of excessive optimism. Alternatively, a significant negative relationship would suggest that underwriters are relying more heavily on PE valuation models compared to investors who may be taking a more conservative DDM approach. For industry earnings volatility (variable Industry_EarningsVol) we take the average of the standard deviations of quarterly earnings from the eight quarters in the previous two years from all companies with the same SIC code. We expect that industry earnings volatility will enter our regressions with a positive coefficient as investors should demand a premium for holding the equity of riskier ventures. If more IPOs occurred in industries with high earnings volatility from and the coefficient is positive as expected, this would provide evidence in favor of Loughran and Ritter s changing risk hypothesis that IPO firms during this period were unusually risky. Underwriter Reputation and Age are both taken from Jay Ritter s website. 18 The methodology for the Underwriter Reputation data amassed by Ritter is largely taken from Carter and Manaster (1990). For Age we simply take the difference in years between the founding date and the offer date. The variable Size refers to the dollar value of the IPO, which corresponds to the offer price times the number of shares issued. VC_backed is a dummy variable representing whether or not an issuer firm received funding from a venture capital firm prior to its IPO. Data for Size and VC_backed are both taken from T1B. IV. Empirical Results We perform regressions over two time periods: and Results are shown in Table 3 with the standard errors reported in parentheses. 18

53 The Effects of Laddering and Spinning in Underwriter Manipulation of IPOs 53 Table 3. OLS Underpricing Regressions Results: Dependent Variable IPO_ Underpricing Regression 1 is the standard result for considering spinning in isolation of laddering. It shows that spinning is significant in explaining IPO underpricing. However, it leaves out laddering. Regression 2 indicates that laddering, underwriter reputation, industry PE and age all significantly impact underpricing. As expected, laddering has a positive effect on underpricing. The laddering coefficient of indicates that a laddered IPO is expected to be underpriced by an additional 110 percentage points compared to non-laddered IPOs, controlling for other factors.

54 54 THE MICHIGAN JOURNAL OF BUSINESS This is a striking result especially considering that our variable for spinning is is no longer statistically insignificant at any confidence level and exhibits a coefficient of only All of the laddered IPOs in our data occurred during the period from ; therefore, it appears likely that laddering was responsible for a large portion of the unusually high underpricing during this period. 19 Contrary to Lee and Wahal (2004) we do not find a statistically significant relationship between venture capital backing and IPO underpricing, which may be driven by the fact that we use a different dataset to construct our VC dummy variable. As expected, we find a statistically significant positive relationship between underwriter prestige and underpricing. The coefficient of in regression 1 is somewhat difficult to interpret due to the construction of the Carter-Manaster underwriter ranking. Therefore in regressions not reported in the table, we rerun regression 1 with a dummy variable with a value of one for ``top-tier" underwriters with a Carter-Manaster rank of 8.0 or greater and zero otherwise in place of the original underwriter reputation variable. The top tier underwriter dummy variable enters with a coefficient and is statistically significant at the 5% level. Therefore top tier underwriters underprice by only approximately 2.56 percentage points more than their competitors. We further examine only the period with this variable, although do not report it in this paper. The coefficient for the top tier underwriter dummy increases to and remains statistically significant at the 5% level. Further, this is the only subperiod for which underwriter rank is statistically significant. This is consistent with the results from Loughran and Ritter (2004) who only find a statistically significant positive relationship between top tier underwriters and underpricing from The results of Regression 1 also indicate a statistically significant relationship between issuer firm age and underpricing, although the economic significance of the relationship is quite modest. In our sample the median age of a firm at the time of the IPO is just under 8 years. The 75th percentile age is years. In terms of expected underpricing, our coefficient on firm age predicts that the 75th percentile firm in terms of age will be underpriced by 1.04% ( x ( )) less than the median age firm. This relationship is somewhat stronger during the period in which the same increase in age is associated with a 2.72% ( x ( )) decrease in underpricing. The coefficient for issue size is positive but not sta- 19 One might argue that large underpricing was likely to lead to lawsuits, which is the majority of our sample. This would cause a simultaneity problem, however we would argue that the causation is primarily driven from laddering to underpricing. Also, the law suits could have been on other issues and not related to laddering per se.

55 The Effects of Laddering and Spinning in Underwriter Manipulation of IPOs 55 tistically significant. We also find a positive relationship between industry PE and underpricing which is statistically significant at the 1% level. A one standard deviation increase in industry PE results in a 3.92% ( x ) increase in underpricing. During the period a one standard deviation increase in industry PE results in a 5.25% (83.57 x ) increase in underpricing and the relationship remains significant at the 5% level. Industry earnings volatility, however, does not have a statistically significant coefficient in regression 1. Contrary to theory, the coefficient on industry earnings volatility is negative in our main regression and many subsample regressions we conducted but not reported in the paper. 20 V. Conclusion Our results provide considerable evidence in support of the hypothesis put forth by Aggarwal et al. (2006) that laddering explains most of the unusual IPO underpricing from the late 1990s. In total, we found that 316 IPOs were laddered between 1998 and 2000 compared to 819 that were not. For the 819 IPOs which were not laddered, mean underpricing was only 25.3% and median underpricing was only 10%. These figures compare to 17.7% mean underpricing and 6.9% median underpricing for the entire period. In short, outside of laddering activity, IPOs during the internet bubble period were only partly different from IPOs in the broader period. This is not to say that the shares bought directly as a result of laddering agreements fully account for the increase in underpricing during the period. On the contrary, we believe there are a number of behavioral explanations which help to explain the predictive strength of laddering in our regressions. First, as explained by Hao (2006) there is likely a positive feedback loop between expected underpricing and laddering. If investors believe that an offer price is too low, laddering will be more likely to occur, thus driving up the aftermarket price and creating an even wider gap between the offer price and the aftermarket price. The informational cascades discussed by Welch (1992) also provide insight on this matter. For example, if investors are misinterpreting laddering agreements as strong buy signals from supposedly informed investors, they may drive the price of an issue even higher than the value stipulated in laddering agreements In a regression using only the period this relationship becomes significant at the 5% level. Increasing earnings volatility by one standard deviation is associated with a one percentage point decrease in underpricing. 21 Of course, it is not clear that investors would have been aware of these agreements.

56 56 THE MICHIGAN JOURNAL OF BUSINESS Further, we believe that the relationship discussed by Loughran and Ritter (2004) and Liu and Ritter (2010) between spinning and underpricing may actually be capturing the effects of laddering, which is omitted from their studies. If we perform a similar exercise and remove the laddering dummy variable from our main regression, the coefficient on our spinning dummy jumps from to and becomes significant at the 1% level. When our laddering dummy is included, the spinning dummy is statistically insignificant at any confidence level. This suggests that underpricing during the internet bubble period was driven much more by manipulation in the aftermarket than the changing issuer objective function hypothesis of Loughran and Ritter (2004). Perhaps our most interesting result is the finding that the PE ratio of firms within an issuer firm s industry has a statistically significant impact on IPO underpricing. This suggests that IPO investors were willing to pay relatively more for issues in industries that are highly valued by the market compared to what underwriters believe they are worth. This affect strengthened during the internet bubble period, and disappeared after the bubble burst in In analysis not reported in this paper, we find that industry PE is no longer statistically significant during the period. These findings correspond with Ljunqvist and Wilhelm s (2003) suggestion that investors were simply optimistic in the extreme. However, they apply not only to the internet bubble period, but to the broader period from During this period, the S&P 500 rose from to , meaning that a simple buy-and-hold strategy would have yielded over 15% per year not including dividends. The incredible bull market experience of may have irrationally driven investors toward issues in hot sectors, similar to the phenomenon that Ritter (1984) discusses in the natural resource sector. Our result is especially interesting when considering the long-run fair market value of IPO issues. Purnanandam and Swaminathan (2004) show that IPOs are overvalued at the offer price, tend to run up in the after market and revert to fair value in the long run. This finding questions the very notion of IPO underpricing. It appears that underwriters tend to fundamentally overestimate the value of an IPO even if the market believes this valuation to be too low. Purnanandam and Swaminathan (2004) suggest that investors overconfidence about private knowledge of IPO firms may be responsible for initial aftermarket overvaluation. However, our results suggest that individuals may not even have any firm-specific information on which to trade. On the contrary, for over a decade investors systematically overvalued IPO issues of firms simply because they operated in industries to which

57 The Effects of Laddering and Spinning in Underwriter Manipulation of IPOs 57 the market attached high valuations. Underwriters may not have completely adjusted their models to account for buy-side irrationalities but in terms of long-run value they still overpriced somewhat. Underwriters threw fuel on the fire of investor overconfidence when they engaged in laddering agreements, but there is little evidence to suggest that they underpriced IPOs relative to fair value in the long-run during the period examined. This paper suggests that future studies should not ignore the incentives and behavior of IPO buyers, as first-day IPO performance is as much a function of their actions as underwriters and issuer firms.

58 58 THE MICHIGAN JOURNAL OF BUSINESS References Aggarwal, R. K., A. K. Purnanandam, et al. (2005). "Underwriter manipulation in initial public offerings." Unpublished working paper, University of Minnesota. Baron, D. P. (1982). "A model of the demand for investment banking advising and distribution services for new issues." Journal of Finance 37(4): Barry, C. B., C. J. Muscarella, et al. (1990). "The role of venture capital in the creation of public companies* 1: Evidence from the going-public process." Journal of Financial Economics 27(2): Beatty, R. P. and J. Ritter (1986). "Investment Banking, Reputation and the Underpricing of Initial Public Offers." Journal of Financial Economics 15: Beatty, R. P. and I. Welch (1996). "Issuer expenses and legal liability in initial public offerings." JL & Econ. 39: 545. Carter, R. and S. Manaster (1990). "Initial public offerings and underwriter reputation." Journal of Finance 45(4): Deloof, M., W. De Maeseneire, et al. (2002). "The valuation of IPOs by investment banks and the stock market: empirical evidence." Working Papers. Demers, E. and K. Lewellen (2003). "The marketing role of IPOs: evidence from internet stocks." Journal of Financial Economics 68(3): Gompers, P. A. (1996). "Grandstanding in the venture capital industry." Journal of Financial Economics 42(1): Griffith, S. J. (2003). "Spinning and Underpricing-A Legal and Economic Analysis of the Preferential Allocation of Shares in Initial Public Offerings." Brook. L. Rev. 69: 583. Hao, Q. (2007). "Laddering in initial public offerings." Journal of Financial Economics 85(1): Kim, M. and J. R. Ritter (1999). "Valuing IPOs." Journal of Financial Economics 53(3): Lee, P. M. and S. Wahal (2004). "Grandstanding, certification and the underpricing of venture capital backed IPOs." Journal of Financial Economics 73(2): Liu, X. and J. R. Ritter (2010). "The Economic Consequences of IPO Spinning." Review of Financial Studies. Ljungqvist, A. and W. J. Wilhelm Jr (2003). "IPO pricing in the dot-com bubble." The Journal of Finance 58(2): Loughran, T. and J. Ritter (2004). "Why has IPO underpricing changed over time?" Financial management: Purnanandam, A. K. and B. Swaminathan (2004). "Are IPOs really underpriced?" Review of Financial Studies 17(3): 811.

59 The Effects of Laddering and Spinning in Underwriter Manipulation of IPOs 59 Ritter, J. R. (1984). "The" hot issue" market of 1980." Journal of Business 57(2): Rock, K. (1986). "Why new issues are underpriced." Journal of Financial Economics 15(1-2): Welch, I. (1989). "Seasoned offerings, imitation costs, and the underpricing of initial public offerings." The Journal of Finance 44(2): Welch, I. (1992). "Sequential sales, learning, and cascades." The Journal of Finance 47(2):

60 60 THE MICHIGAN JOURNAL OF BUSINESS

61 61 To Host or Not to Host? A Comparison Study on the Long-Run Impact of the Olympic Games Anita Mehrotra 1 University of California, Berkeley Abstract Macroeconomic research on sporting mega-events, like the Olympic Games, often focuses on the short-run economic impact of individual countries. In this paper, I explore the long-run effect of the Olympics on host countries, in general. I analyze hosts in relation to runner-up countries, i.e. countries that come in second to the hosts in the bidding process. Upon rectifying anomalies in the data set and including control variables for the initial state of the economy and population, I find that hosts long-run GDP per capita (GDPpc) is negative in comparison to runner-up countries at a statistically significant level. This suggests that a one-time spike in government expenditure may lead to long-run detrimental effects: a reverse multiplier effect perpetuates the fall in investment demand and consumption levels back to pre-olympic levels. My results extend the view that individual countries experience a negative economic impact, to the group of host countries in general. 1 Anita Mehrotra graduated with a double major in Pure Mathematics and Economics from the University of California, Berkeley in This paper was completed as an honors thesis under the guidance of Professor Emmanuel Saez. The author extends a special thanks to Slavic Languages librarian Alan Urbanic, Vladimir Asyrian, Professor Roger Craine, and especially to her family for their love and support.

62 62 THE MICHIGAN JOURNAL OF BUSINESS I. Introduction When the first Olympic Games were hosted in Olympia, Greece in 776 BC, perhaps few knew that they would go on to become one of the most widely watched mega-events in the world. Today, the Games have morphed into a display of the host country s ability to sponsor an international, mega-sporting event. In this paper, I attempt to determine the long-run economic impact of hosting the Olympic Games. I define long-run as the ten-year period after the year the Games are held, consistent with current literature in which [t]he longer-term impact of the Olympics is identified as last[ing] for at least a decade after the Games (PricewaterhouseCoopers 2004, p. 18). Much of the existing research in this area focuses on the immediate, short-run effects of hosting the Games. Few studies have assessed the longrun economic impact of hosting the Games on a large group of countries. In addition, most of the existing research provides analysis and insight on only one or two countries at a time (Hotchkiss et al 2003; Madden and Gisecke 2007; Matheson 2006; Veraros et al 2004). As a result, it is difficult to draw general conclusions about whether there is economic justification in bidding for and hosting the Olympics (Brunet 1995; Owen 2005; Whitson and Horne 2006). In this paper, I extend the current research by analyzing long-run effects on hosts gross domestic product per capita (GDPpc). A positive effect will support some existing literature that suggests an economic benefit to hosting the Games (Hotchkiss et al 2003; Veraros et al 2004). A negative economic impact, on the other hand, will reinforce the remaining majority of existing literature (Jones 2001; Matheson 2006; PricewaterhouseCoopers 2004). Ultimately, this paper aims to provide future host countries with a better understanding of the long-run, macro-economic impact of hosting the Olympics. I compare the group of countries that hosted the Olympics from 1972 to 1998 (inclusive) with the corresponding group of first runner-up countries (the group of countries that bid to host the Olympics but do not receive the right to do so, and are second only to the host country). I analyze relative GDPpc, adjusted for inflation and in current USD. The year that the Games are hosted is considered year 0, and data for ten years before and ten years after the Games are averaged per year, across each group. Since the host country is announced seven years prior to the year that the Games are hosted (Olympic. org), this data is normalized to year 7, or seven years before the Olympics are held. Upon graphing the resulting data points, I notice a relatively negative impact on the average, normalized GDPpc for host countries starting at approximately year 3, or three years before the Games are held, and continuing till year +10, or ten years after the Games are held. I test this null

63 To Host or Not to Host? A Comparison Study on the Long-Run Impact of the Olympic Games 63 hypothesis using a simple ordinary least squares regression and find that there is a statistically significant, negative impact on hosts GDPpc. This suggests a long-run, negative economic impact on hosts output per capita. II. Literature Review For a brief two weeks after the Opening Ceremony flame is lit, an international spotlight is focused on the host country. Similarly, research in sports economics often highlights the short-term macroeconomic impact of hosting the Olympics. Such research, however, focuses on only one or two countries at a time. This analysis therefore fails to provide substantive evidence for future host countries, and especially for those that are concerned about the long-run impacts on their economies. 2.1: Short-Run Economic Impact on Individual Host Countries Some macroeconomists suggest that hosting the Games increases international awareness and therefore results in a positive, albeit short-lived, economic effect on real output (GDPpc). For example, Hotchkiss et al (2003) used standard and modified differences-in-differences techniques to determine whether regions near Atlanta saw a change in employment levels after the 1996 USA Olympics. They concluded that the Games boosted employment by 17% in counties affiliated with and close to Olympic activity. Hotchkiss et al additionally used a random-growth model test across multiple metropolitan statistical areas (MSA s) and found that employment rose by an additional 11% in these MSA s post-olympics, relative to other similar MSA s that were not near Atlanta. Then in 2004, Veraros et al extended such short-term analysis to the financial sector, studying the impact of the 2004 Olympic host announcement (which occurred in 1997) on investment levels. They observed an increase in foreign direct investment and concluded that there was a statistically significant, positive effect on the Athens Stock Exchange. On the other hand, the Milan Stock Exchange, located in the runner-up country, was barely impacted. Such research indicates that in the short-run, the macroeconomic impact of hosting the Olympics may be beneficial for the host. On the other hand, a large realm of existing literature indicates the exact opposite: that the Olympics actually have a negative short-run impact on host GDPpc. Jones (2001) notes that countries are often disillusioned by biased pre-games estimates about the positive economic rewards of hosting. As a result, they are economically unprepared to deal with the after-effects of a more modest (or negative) profit. This can, in turn, affect short-run expenditure allocations and the government s budget. For instance, although the 1988

64 64 THE MICHIGAN JOURNAL OF BUSINESS Calgary Games in Canada successfully generated a recorded profit of more than $130 million, the few scholarly studies that examined the economic impacts suggest that the direct impacts [were] not as a great as official rhetoric implie[d] (Whitson and Horne 2006). Such embellishments are not an anomaly to sporting mega-events in general: in his paper on the regional economic effects of hosting the Rugby World Cup, Jones (2001) concludes that the commonly-used input-output (IO) tables 2 over-represent [ex ante tourism] activity resulting from special events. These complex IO tables are rooted in inter-industry relationships that are, in turn, based on a region s normal production patterns. However, these normal patterns clearly do not hold during a sporting mega-event. As Matheson (2006) notes, ex ante estimates also exaggerate the net economic benefits because they are biased by committees that need an infusion of taxpayers money. Even without this bias, however, ex ante studies can suffer from any one of three primary theoretical deficiencies: 1. The substitution effect. This occurs when local spending, rather than additional foreign spending, is poured into the sporting megaevent. The result is not new economic activity [but] rather a reshuffling of local spending. Matheson goes on to suggest that a more accurate estimate would entirely exclude local resident spending from economic impact estimates. 2. Crowding out. An influx of tourists supplant, rather than supplement, the regular tourist economy, minimizing additional profits that are generated from hosting the Olympics. Since host cities tend to be popular tourist destinations, Olympic tourists substitute regular tourists and lead to nearly no additional profit for host countries. 3. Leakages. These occur if tourist spending does not wind up in the pockets of local residents... [even if] the taxes used to subsidize these events are paid for by local taxpayers. Profits may leak to outof-town companies who set up temporary booths during the Games. Since many do not employ in-town residents, nor do they lose a large percentage of their profit to the city or national government, a majority of the returns are pocketed by the companies rather than utilized for 2 Input-output tables are used to model the dynamic time path of the economy that tracks towards continually shifting equilibria. They are used as a way to present a realistic picture of the impacts of dynamic characteristics of economic structure and change (West and Jackson, 1998). Recently, computable general equilibrium models (CGE models), which have improved upon IO models with their incorporation of fixed factors and substitution effects, are slowly displacing IO tables. However, they, too, are at risk of making over-optimistic estimates of the net benefits (Giesecke and Madden, 2007).

65 To Host or Not to Host? A Comparison Study on the Long-Run Impact of the Olympic Games 65 the city and its residents. Smith (2009) also suggests that countries are often unprepared to deal with ephemeral spikes in employment in the construction and tourism sectors. Workers who are temporarily hired and become unemployed after the Olympics can affect the economy adversely, although indirectly. Unemployment rates that return to higher, pre-olympic levels can dampen buyer optimism. This macroeconomic uncertainty, statistically defined as increased variance of individual income, can negatively impact national consumption levels and lower domestic output. Such uncertainty has historically impacted countries adversely, as shown in Romer s analysis on the US during the Great Depression (Romer 1990) and in Baker et al s recent report on the Great Recession (Baker et al 2011). Others, like Owen (2005), note that host countries may also endure sunk costs in the form of unused athletic housing and derelict stadiums that were expensive to build. This, coupled with biased pre-game estimates and dampened consumer optimism, can thus have short-run, negative impacts on hosts economies. 2.2: Long-Run Economic Impact on Individual Host Countries A small portion of the existing literature discusses the long-run impact on host countries. Of those that do (PricewaterhouseCoopers (PWC) 2004; Giesecke and Madden 2007), most document a negative impact on hosts economies. In an analysis on the five to seven years after Spain and Australia hosted the Games (1992 Barcelona and 2000 Sydney, respectively), PWC s 2004 European Economic Outlook reported a sharp slowdown in investment expenditure as preparations were finalized in [both] the run-up to the Olympics, as well as after the Games were over (p. 21). This, in turn, lowered GDPpc levels. Gieseke and Madden (2007) compared estimates from ex ante computable general equilibrium (CGE) models to ex post analyses, and found that in the long-run, the Sydney Olympics generated a net consumption loss of approximately $2.1 billion. This therefore provides the motivation for the first component of my two-pronged analysis: I attempt to extend such research and determine if there is a long-run economic effect of hosting the Games. I conclude that, consistent with a large portion of the existing literature, there is a relatively negative macroeconomic impact on hosts real GDPpc. 2.3: Economic Impact on the Group of Host Countries In both short- and long-run analyses, most researchers like Brunet (2005) and Shoval (2002) focus on individual host countries. Rose and Spiegel s 2010 study, however, stands out as an exception. In discussing their re-

66 66 THE MICHIGAN JOURNAL OF BUSINESS sults (generated from a comparison study of hosts with all other countries, including non-bidders ), the authors reveal an apparent Olympic effect on host and bidding countries. On average, countries that bid to host the Games experience a positive impact on net exports. Rose and Spiegel suggest that bidding actually signals a country s desire to be more open, or globally integrated. They use historical data to substantiate their theory: Our explanation seems to accord well with the facts, at least superficially. In July 2001, Beijing was awarded the right to host the Games of the XXIX Olympiad. Just two months later, China successfully concluded negotiations with the World Trade Organization, thus formalizing its commitment to trade liberalization. Nor is this a one-off coincidence. Rome was awarded the 1960 Games in 1955, the same year Italy started to move towards currency convertibility, joined the UN, and, most importantly, began the Messina negotiations that led two years later to the Treaty of Rome and the creation of the European Economic Community (EEC). The Tokyo Games of 1964 coincided with Japanese entry into the IMF and the OECD. Barcelona was awarded the 1992 Games in 1986, the same year Spain joined the EEC; the decision to award Korea the 1988 Games coincided with Korea s political liberalization. [Furthermore, this] correlation extends beyond the Olympics; the 1986 World Cup was held in Mexico coincident with its trade liberalization and entry into the General Agreement on Tariffs and Trade, the predecessor to the World Trade Organization. Such research thus begs the following question: what has been the GDPpc impact of hosting the Olympics, in general? This motivates the second component of my analysis, in which I focus my tests on the group of host countries in general. This paper therefore attempts to draw a conclusion about whether hosting the Olympics, in general, has a negative long-run macroeconomic impact. I define the group of host countries as my treatment group, while the first runners-up comprise a naturally formed control group. As a result, the implicit assumption is that the group of runner-up countries is nearly identical to the group of host countries with the main difference being that countries in the control group are not awarded the right to host the Games. I find that, on average, host countries experience a statistically significant, negative impact on real GDP per capita beginning approximately three years before the Games are hosted, and continuing in a negative direction for the next thirteen years.

67 To Host or Not to Host? A Comparison Study on the Long-Run Impact of the Olympic Games 67 III. Data and Descriptive Statistics 3.1: General Information To determine if there is an impact on host countries GDPpc, I use data from the World Bank Databank and the US Bureau of Labor Statistics. I chose this data based on the lists of bidding and host countries compiled from the International Olympic Committee s Voting Results ( ). However, since there was no runner-up country for the 1984 USA Games in Los Angeles, the list of host countries was slightly longer than the list of bidding countries. The majority of my data comes from the World Bank Databank. Other databases were sparser, missing information for a number of the countries on my lists. One viable alternative, Global Financial Data (GFD), provided GDPpc for nearly all countries beginning from However, I noticed a wide disparity in trends exhibited by GFD data compared with trends of data from other sources, like Penn World Tables, International Monetary Fund, and the US Bureau of Labor Statistics. Figure 1 provides an example of the large discrepancy in the GFD trend for one country s GDPpc data (Germany, 1960 to 2008). Though this incongruity was true for only some countries, World Bank data provided consistent data trends across all countries of interest. It was thus a more reliable data source. Figure 1. Gross domestic product per capita for Germany,

68 68 THE MICHIGAN JOURNAL OF BUSINESS Although more precise, the World Bank Databank only contained data from 1960 to Since I was interested in GDPpc for the ten years before each of the Olympic Games, I refined my lists to begin with the country that hosted the first Olympics after 1970 (ten years after 1960). Similarly, I needed GDPpc for the ten years after each of the Games, and thus further narrowed my list to end with the country that hosted the 1998 Olympics (ten years before 2008). My resulting panel data set included observations across twenty-nine countries (fifteen host countries and fourteen runner-up countries). 3.2: Constructing the Data Set My data set was missing GDPpc for three countries: Yugoslavia, for the years before 1994 (Yugoslavia hosted the 1984 Winter Olympics, so I needed data from 1974 to 1994); Germany, for the years before 1970 (Germany hosted the 1972 Summer Olympics, so I needed data from 1962 onwards); Russia, for the years before 1998 (Russia hosted the 1980 Summer Olympics and was a runner-up for the 1976 Olympics, so I needed data from 1966 onwards). During the process of researching missing data points, I determined that in some cases, it would be more accurate to eliminate certain subsets altogether. Since my set of macroeconomic data is small, I believe it is important to provide brief explanations of any decisions I made in this regard : Data on Yugoslavia Present-day Yugoslavia is comprised of Slovenia, Macedonia, Croatia, Serbia, Montenegro, Kosovo, and Bosnia and Herzegovina. However, very little documentation of GDPpc exists before 1994 for Kosovo and Montenegro. This was further complicated by the fact that some regions were parts of others (i.e. significant geographic overlap existed among various provinces). For example, parts of Croatia and Kosovo used to be a part of Serbia before they gained independence in 1991 and 2008, respectively. Since so much data was missing, and given the inevitable accounting difficulties that would arise even if such data was found, it was more precise to drop Yugoslavia altogether from my data set. Yugoslavia hosted the Winter Games in 1984, and because there was no runner-up country to the 1984 Summer Games, the lists (and thus the data) were now made up of an equal number of host and runner-up countries.

69 To Host or Not to Host? A Comparison Study on the Long-Run Impact of the Olympic Games : Data on Germany Data included for Germany is entirely of West Germany because only West Germany s GDPpc was documented before the fall of the Berlin Wall in The 1972 Olympic Games were hosted in Munich, located in West Germany. Since I was only interested in data from 1962 to 1982, all of the data I used is from West Germany. Inflation-adjusted, current USD GDPpc data for Germany is available from the US Bureau of Labor Statistics (BLS). Although this data exhibits similar trends to data from the World Bank post-1970, I needed to match the BLS data to the World Bank data. I scaled the BLS GDPpc from the first overlapping year (1970) so that it was equal to the 1970 World Bank GDPpc level. I then applied this scaling factor to previous years worth of data from the BLS. Figure 2 provides a visual representation of this process. Figure 2. Method for filling holes in Germany data set 3.2.3: Data on Russia Data on Russia s GDPpc was the most difficult to obtain for multiple reasons. First, I needed to gather data from before 1988, when Russia was still a part of the USSR. Thus, such data would have to include information on only the (geographical) portion of the USSR that comprises present-day Russia. This isolated data set was not available in English, so I turned to Russian Statistical Yearbooks from 1962 to However, translation from Rus-

70 70 THE MICHIGAN JOURNAL OF BUSINESS sian to English proved difficult, in part because of differing macroeconomic terminology. After extensive research, historical documentation revealed that manufacturing national income and national social product referred to what is now known as GDP and GNP (Gross National Product), respectively (Khomenko 2006). The currency s standard units proved to be the final hurdle, as yearly data during the 1960 s, 1970 s and 1980 s were listed in terms of a baseline of 100 from the closest five-year mark before the year of interest (for example, 1967 data was listed with 1965=100, 1971 data was listed with 1970=100, etc). The data was inflation-adjusted, but only to the most recent five-year mark. Then, in the 1990 s, recorders switched to tabulating data in terms of billions of Russian rubles. Given the above complexities in gathering GDPpc data for Russia, I concluded that my data set would be more accurate, albeit less complete, if I removed Russia entirely. Since Russia was a bidder in 1976 and then hosted the Summer Games in 1980, I dropped one country from both the host and runner-up lists. The number of countries in each list therefore remained equal. My current data set consists of twenty-six countries (thirteen in each of the host and runner-up groups), is in current USD, and is adjusted for inflation.

71 To Host or Not to Host? A Comparison Study on the Long-Run Impact of the Olympic Games 71 Table 1. Summary statistics of panel data (number of observations = 540) IV. Methodology I first tabulate the average, normalized GDPpc for host and runner-up countries from ten years before the Olympics are hosted to ten years after they are hosted in two matrices (Appendix, Tables B and C). Each column corresponds to one of thirteen countries, and each row is labeled with a year n, where n ϵ [ 10, +10]. Second, I average the GDPpc across all countries for each year, in order to obtain an average GDPpc per year for both the host and runner-up groups. I normalize these numbers to the year 7, i.e. the year in which the host country is announced (Olympics.org). Finally, I plot the normalized numbers in a line graph, and note a relatively negative impact on the host group s average GDPpc beginning at approximately year 2.5*.

72 72 THE MICHIGAN JOURNAL OF BUSINESS Figure 3. Graph of difference in average, normalized GDPpc Figure 4. Graph of average, normalized GDPpc

73 To Host or Not to Host? A Comparison Study on the Long-Run Impact of the Olympic Games 73 I run two regressions to test if this gap is statistically significant. Regression (1) uses the difference in GDPpc data (as depicted in Figure 3): Average normalized GDPpc = α+β 1 Host+β 2 Time+β 3 (Host Time)+ε Regression (2a) tests whether the gap is statistically significant using the original numbers from Figure 4: GDPpc = α+β 1 Host+β 2 Time+β 3 (Host Time)+ε In both regressions, Host = a binary variable; 1 if the country hosted the Olympics, 0 otherwise Time = a binary variable; 1 if year 3 (from the discussion* on page 16), 0 otherwise The variable of interest is the interaction variable Host Time. It measures the impact on GDPpc in a particular year depending on whether or not the country hosted the Games. Upon gathering additional data on initial GDPpc and population (at year 10), I run regression (2b): GDPpc = α+β 1 Host+β 2 Time+β 3 InitialGDPpc+β 4 InitialPop+β 5 (Host Time)+ε Regression (2b) improves upon (2a) greatly because it controls for the initial state of the country s economy and population. The interaction variable s coefficient now has a lower standard deviation and p-value, suggesting an improvement in the fit of the regression. I initially conclude that the coefficient on the interaction variable is negative, though not statistically significant. Finally, I graph and study the GDPpc of individual pairs of countries (Appendix, Figures D1 to D12). I find that in 1992, there is a large positive impact on the host s GDPpc relative to the runner-up country, and hypothesize that this pair is an anomaly. I check historical narrative evidence to ensure that I am not cherry-picking from my data set and discover an interesting fact: of the compared pairs, the 1992 Olympics is the only one in which a non-communist host country and a post-communist, war-torn regime are compared. As a result, the host country sees a relatively (huge) positive effect of hosting the Games, even if this economic benefit is partially because the runner-up coun-

74 74 THE MICHIGAN JOURNAL OF BUSINESS try was actually transitioning to a democratic country. This substantiates my hypothesis. Removing the pair of 1992 countries leads to a better fit (the R2 term moves up to 0.801). In addition, the negative coefficient on the interaction variable becomes statistically significant at the 95% level. I conclude that the effect of hosting the Olympics results in a statistically significant negative, long-run impact on host countries GDPpc in general. V. Results 5.1: Evaluation of Findings Table 2 lists results from regression (1) in which I test average, normalized GDPpc. On average, an individual could lose percent of their annual GDPpc as a result of increased spending in for the Olympic Games. However, this relationship is not statistically significant (p-value=0.250). Table 2. Regression (1) Results: Dependent variable: Average, normalized GDPpc per country Table 3 (page 21) has two components: the first portion (3(i)) lists regression (2a) results, while 3(ii) lists regression (2b) results. The results indicate a negative long-run impact of hosting the Olympics on GDPpc relative to runner-up countries as shown in Figure 4. In Table 3(i), there is an estimated negative impact of approximately USD $1500 per person and a standard deviation of approximately USD $1800. To improve regression results, I include

75 To Host or Not to Host? A Comparison Study on the Long-Run Impact of the Olympic Games 75 control variables for the initial state of the economy (InitialGDPpc) and initial population (InitialPop). I also experiment with the time value, i.e. the year in which Time switches from 1 to 0. Recall that from my graphical analysis, the negative trend begins somewhere between year 3 and year 2. In (2a), Time=1 if the year is greater than or equal to 3, and 0 otherwise. In (2b), I test what happens when Time=1 if the year is greater than or equal to 2, and 0 otherwise. Table 3(ii) captures these results. Both the standard deviation and p-value for the coefficient on Host Time fall (standard deviation= , p-value=0.130). Although this relationship is not statistically significant, it is close to the 0.1 threshold required for statistical significance at the 90% level. Interpreted in terms of real GDPpc I conclude, though not confidently, that in host countries, an average individual loses anywhere from $450 and $2150 (Table 3). Table 3. Regression (2) Results Dependent variable: Real GDP per capita per country

76 76 THE MICHIGAN JOURNAL OF BUSINESS Adjusting Time (1 if year 2), Including independent variables: initial GDPpc, population Because the p-value in Table 3(ii) is close to the required 0.10 measure in order to be statistically significant at the 90% level, I hypothesize that one pair of countries may be acting in the opposite direction. In particular, perhaps there is an anomalous pair of countries in which the host country experiences a large positive impact. To test this theory, I graph each pair of host and runner-up countries (Appendix, Figures D1 to D12). I notice a large positive gap between France (host country) and Bulgaria (runner-up country) in France s GDPpc increases dramatically compared to Bulgaria s beginning at year 5, or five years before the Olympics are actually held. This is the only pair in which the host is not impacted negatively, and is depicted in Figure 5 below.

77 To Host or Not to Host? A Comparison Study on the Long-Run Impact of the Olympic Games 77 Figure 5. Graph of normalized GDPpc, 1992 Pair I use narrative historical evidence to ensure I am justified in eliminating this pair from my data set. (A more detailed explanation follows in section 5.2.) I then re-run regression (2b), sans the 1992 pair, and obtain a statistically significant p-value after year 3 of 0.019, as well as an increase in the R2 term to (Table 4). This suggests an improvement in the fit of the regression. It also allows me to conclude with 95% confidence that the average longrun impact on hosts GDPpc is negative.

78 78 THE MICHIGAN JOURNAL OF BUSINESS Table 4. Regression (2b) Results without France/Bulgaria 1992 Dependent variable: Real GDP per capita per country The negative impact on the two to three years before the Games are hosted is intuitively plausible: host countries prepare for the Games well in advance, with expenditures ramping up two to three years before they are held. Hosts spend on stadiums and athletic housing, restoration (or building) of mass transit networks, and improvements of expansive sewage systems leading to an increase in hosts expenditures. What is most interesting, however, is that there appears to be a negative impact on host countries for several years after the Games are held. Though an increase in government spending contributes to expansionary fiscal policy, spending on the Olympic Games may have a smaller multiplier effect than current research suggests (Ball, 1999). For example, host countries see an increase in foreign awareness and domestic investment. This, in turn, can lead to a rise in domestic and international demand for host countries products. It is subsequently propagated via the consumption channel of the classic macroeconomic equation for output: Y=C(Y-T)+ G+I(r)+ NX(r) However, this spike in consumption would be relatively short-lived. Coupled with Olympic infrastructure that often goes unused post-games, negative capacity effects can result in a decline in long-run output, mitigating the effect of increased consumption.

79 To Host or Not to Host? A Comparison Study on the Long-Run Impact of the Olympic Games 79 After the Olympics, host countries face a sort of reverse multiplier effect, where the multiplier works in the opposite direction: the non-recurring boost to expenditure results in a longer-run fall in demand as the economy returns to its pre-olympic equilibrium income (PricewaterhouseCoopers 2004, Giesecke and Madden 2007). It further provides a potential explanation for why the impact is negative for a long period of time after the Games are hosted. 5.2: The 1992 Summer Olympic Games I analyze historical evidence about France and Bulgaria during the early 1990 s in order to determine whether my hypothesis that they are an anomalous pair in my data set is valid. I find that out of all the pairs of countries on my list, France and Bulgaria are the only post-cold War coupling that compares a non-communist country (France) with a post-communist regime (Bulgaria). Like other countries transitioning to capitalism, Bulgaria experienced a painful shift as it emerged from the throes of Communism in the early 1990 s, when the first fully democratic parliamentary elections were held (the Union of Democratic Forces won). Characterized by massive unemployment, failure of uncompetitive industries, and national infrastructure backwardness, this period of severe social and economic turmoil eventually culminated in an economic and financial crisis (late 1996 early 1997). France, on the other hand, remained a capitalist state, entering the 1990 s shortly after trente glorieuses ( thirty glorious years ) of steady national economic development. As a result, the Olympics appear to have had a positive economic impact on France relative to Bulgaria. However, narrative evidence indicates that this wasn t the sole economic factor: France did not face the formidable economic challenges of transitioning from a Communist to non-communist government during the early 1990 s as Bulgaria did ( Bulgaria Historical Highlights ). Although my regression accounts for the initial state of the economy, it does not capture the economic impact during the middle years. As a result, my results are skewed because they do not take into account the economic disadvantage that Bulgaria faced. This is because I treat all years as equal, renumbering 1982 to year 10, 1983 as year 9, and so on. This method, though helpful in determining the overall effect on host countries in general, is susceptible to missing historical events as exemplified by the 1992 Olympic Games. In using narrative historical evidence, I confirm that unlike other pairs, France and Bulgaria are an exception. The increase in French GDPpc relative to Bulgaria during the ten years following the Games is not attributable solely to hosting the Olympics. Excluding this pair from my analysis results

80 80 THE MICHIGAN JOURNAL OF BUSINESS in a more accurate measure of the effect of hosting the Olympics in general. I ultimately conclude that there is a statistically significant, negative impact of the Games on host countries in general. 5.3: Difficulties and Possible Improvements 5.3.1: Expanding the Data The impact on host countries may be smaller than what it would have been if I had included additional countries in my tests. In particular, all of the countries analyzed in this paper are advanced economies, according to the International Monetary Fund (Appendix, Table A). I hypothesize that in the past, the process of bidding for the Olympics created a natural self-selection process in which developed countries engaged in the bidding,and eventual hosting, process. This makes sense intuitively: the cost associated with bidding for the Games can be very burdensome financially. However, this trend has recently begun to change, as less-developed countries are hosting megasporting events. Examples include China (hosted the 2008 Olympic Games), India (hosted the 2010 Commonwealth Games), South Africa (hosted the 2010 World Cup), and Brazil (will host the 2016 Summer Olympics). In order to accurately quantify the long-run economic impact of hosting the Games in general, an ideal data set would include information and analysis on less-developed host countries. Other improvements include finding a consistent method of handling GDPpc data from Yugoslavia and Russia (sections and respectively). Though economists must wait till such data is collected and becomes available, future research will provide an even more comprehensive picture of the general economic impact of hosting the Games. If researchers obtain accurate early data of the Olympics in the future, they will need to control for multiple important variables. I include some here: The advent of television and the beginning of televised Games in Games held after 1960 may have seen an increase in international viewership, awareness, and thus a potential increase in profits. As television-broadcasting rights became more valuable to both host countries and TV networks during the 1970 s and 1980 s, government spending on the bidding process increased dramatically ( Olympics and Television ). Inclusion of a binary variable e.g. Televised (1 if the Olympics were hosted before 1960 and 0 otherwise) may help mitigate such bias.

81 To Host or Not to Host? A Comparison Study on the Long-Run Impact of the Olympic Games 81 Table 5. Television broadcasting rights and its role in the Olympics, Revenue, inflation-adjusted Source: (Moreland) Olympics and Television The influence of international conflict. One of the biggest difficulties I faced in attempting to isolate the macroeconomic effect of the Olympic Games was the occurrence and thus the impact of other political events that occurred at or around the same time as the Games. The data set I used was from 1962 to 2008, so my countries of interest were limited to having hosted/been a runner-up for the Games from 1972 to Using control variables and historical narrative evidence to eliminate data points as necessary minimized such variance to a large degree. If, on the other hand, future research includes countries from earlier years, there must be a method to account for the economic after-effects of World War I and II, the Balkan Wars, the First and Second Sino-Japanese Wars, and the Vietnam War, among others. These wars had far-reaching effects that were often global in

82 82 THE MICHIGAN JOURNAL OF BUSINESS scale, and can adulterate results : Regressions: Areas of Improvement As is common with macroeconomic research in general, an econometric issue that I faced in this paper was omitted variable bias. A post-regression analysis situated in context of narrative evidence reveals that substitution effects and leakages (section 2) can exaggerate the impact: reshuffling of spending and returns to taxpayers may be inaccurately measured. Although my analysis attempts to account for other forms of bias, future research may include variables to better capture the substitution effect, like year-to-year changes in the average number of tickets purchased for sporting events before and after the Games are held. A spike in sporting-event ticket sales during or immediately before the Olympic year may account for the sudden increase in spending, and control for the substitution effect. Additional methods include using CGE models (described in footnote 2), as is demonstrated by Giesecke and Madden in their 2007 report on the 2000 Sydney Olympics. An analysis on profits for domestic versus foreign country-companies can help capture leakages, although this effect may prove more difficult to account for. VI. Conclusion The analysis in this paper provides some insight into the economic impact of hosting the Olympics. My results go beyond this in an effort to provide a better understanding of the long-run macroeconomic effect of hosting the Olympics in general. So why do host countries experience long-run negative economic effects, on average? This may be due in part to the fact that host countries experience a one-time increase in domestic investment. Hosts may also experience a spike in consumption of domestic assets. Instead of remaining at higher output levels, host countries GDPpc falls. Current literature suggests that this return to normal[cy] is then subject to a reverse multiplier-effect, decreasing long-run GDP per capita. However, because my data does not include GDPpc levels of less-developed economies, and my analysis is limited to the Games before the 1998 Olympics, it is challenging to determine the specific drivers of this negative difference. Other challenges common to sports macroeconomic studies persist, like the difficulty in measuring Matheson s (2006) substitution effect and leakages. These may contribute to omitted variable bias. Ultimately, my results provide a launching pad for additional research on the general long-run impact of hosting the Olympics. In this paper, I attempt

83 To Host or Not to Host? A Comparison Study on the Long-Run Impact of the Olympic Games 83 to provide researchers and governments with a better idea about the effect of hosting the Games in general, rather than a case-by-case analysis. This study aims to enable potential bidding countries and researchers to better assess the economic impact of hosting the Olympics on their citizens, and to focus investment in areas that will contribute to the growth of the country. The Olympic Games can undoubtedly be wielded as a tool to increase normal output above current levels, but only if policy-makers strategically align their expenditures with infrastructure that will have a long-run economic benefit (like building stadiums that can later be used by universities). Thus, my research opens the door to further analysis on why countries that host the Olympics experience long-run, negative effects on output in general.

84 84 THE MICHIGAN JOURNAL OF BUSINESS References International Olympic Committee. Factsheet: Host City Election Facts and Figures. Last modified July Accessed February 25, Documents/Reference_documents_Factsheets/Host_city_election.pdf. Season Workers Job posting site. Temporary Jobs at the Olympics. Accessed April 20, Smith, Alex. Building.co.uk, 2012 Olympics Creates 30,000 Construction Jobs. Last modified January Accessed April 20, co.uk/news/2012-olympics-creates construction-jobs/ article. Moskva: Statistika Volume 33 ed., s.v. Narodnoe Khoziaistvo RSFSR Annual gross domestic product per capita, inflation-adjusted via Bureau of Labor Statistics Database, htm (accessed April 20, 2011). Annual gross domestic product per capita, inflation-adjusted via Global Financial Database, html (accessed April 20, 2011). Annual gross domestic product per capita, inflation-adjusted via International Monetary Fund Database, (accessed April 20, 2011). Annual gross domestic product per capita, inflation-adjusted via International Monetary Fund Database, weo/2008/02/weodata/groups.htm#ae (accessed April 20, 2011). Annual gross domestic product per capita, inflation-adjusted via World Databank: World Development Indicators and Global Development Finance, (accessed 25 February 2011). Annual gross domestic product per capita, inflation-adjusted via Penn World Tables Database, (accessed April 20, 2011). Romer, Christina. The Great Crash and the Onset of the Great Depression, Quarterly Journal of Economics (1990): stable/ (accessed February 27, 2011). Baker, Scott, Nicholas Bloom, and Steven Davis Measuring Economic Policy Uncertainty, unpublished paper (2011). Ball, Laurence. Aggregate Demand and Long-Run Unemployment. Brookings Papers on Economic Activity. (1999): , sholden/e4325/ball-1999-aggregate-demand.pdf (accessed October 8, 2010). Brunet, Ferran. An Economic Analysis of the Barcelona 92 Olympic Games: Resources, Financing, and Impacts. Centre d Estudis Olímpics-Universitata Autònoma de Barcelona.(1995).

85 To Host or Not to Host? A Comparison Study on the Long-Run Impact of the Olympic Games 85 pdf (accessed February 27, 2011). Hotchkiss, Julie, Robert Moore, and Stephanie Zobay. The Impact of the 1996 Summer Olympic Games on Employment and Wages in Georgia. Southern Economic Journal. 3. no. 69 (2003): report53.pdf (accessed February 27, 2011). Jones, Calvin. Mega-events and Host-region Impacts: Determining the True Worth of the 1999 Rugby World Cup. International Journal of Tourism Research. (2001): onlinelibrary.wiley.com/doi/ /jtr.326/pdf (accessed March 4, 2011). Khomenko, Tatiana. Estimation of Gross Social Product and Net Material Product in the USSR Hitotshubashi University Research Unit for Statistical Analysis in Social Sciences (2006): 2. D pdf (accessed February 20, 2011). Madden, John and James Giesecke. The Sydney Olympics, Seven Years On: An ex-post Dynamic CGE Assessment. The Centre of Policy Studies and the Impact Project (2007): 11, (accessed January 21, 2011). Matheson, Victor. Mega-Events: The Effect of the World s Biggest Sporting Events on Local, Regional and National Economies. (2006). pages/docs/2744/ _matheson_events.pdf (accessed March 22, 2011). Moreland, Jennifer. Olympics and Television. Encyclopedia of Television. 2 (1997): Owen, Jeffrey. Estimating the Cost and Benefit of Hosting Olympic Games: What Can Beijing Expect from Its 2008 Games? The Industrial Geographer. 3. no. 1 (2005). (accessed February 27, 2011). PricewaterhouseCoopers. III The economic impact of the Olympic Games. PricewaterhouseCoopers European Economic Outlook. (2004): of%20olympics%20pwc.pdf (accessed January 21, 2011). Romer, David. Short-Run Fluctuations. Unpublished (2006). Rose, Andrew and Mark Spiegel. "Mega Sporting Events and International Trade." (accessed March 22, 2011). Rose, Andrew and Mark Spiegel. The Olympic Effect, National Bureau of Economic Research Working Paper No (2010). w14854 (accessed October 1, 2010). Shoval, Noam. A new phase in the competition for the Olympic gold: The London and New York Bids for the 2012 Games. Journal of Urban Affairs. 24, no. 5

86 86 THE MICHIGAN JOURNAL OF BUSINESS (2002): onlinelibrary.wiley.com/doi/ / /pdf (accessed March 4, 2011). U.S. Department of State: Diplomacy in Action, Bulgaria - Historical Highlights. Accessed February 24, (accessed April 3, 2011). Veraros, Nikolaos, Evangelia Kasimati, and Peter Dawson. The 2004 Olympic Games Announcement and its Effect on the Athens and Milan Stock Exchanges. Applied Economics Letters. 11 (2004): deas.repec.org/a/ taf/apeclt/v11y2004i12p html (accessed January 21, 2012). West, Guy and Randall Jackson. Input-output+econometric and econometric+inputoutput: Model differences or different models? Journal of Regional Analysis and Policy. 28, no. 1(1998): pastvolumes/1990/v28/ pdf (accessed February 27, 2011). Whitson, David and John Horne. The Glocal Politics of Sports Mega-events, Part 2: Underestimated Costs and Overestimated Benefits? Comparing the Outcomes of Sports Mega-events in Canada and Japan. The Editorial Board of the Sociological Review. 54. (2006): 2. Vernik, Aleksandr. IOC VOTE HISTORY. Last modified July 7, Accessed January 20,

87 To Host or Not to Host? A Comparison Study on the Long-Run Impact of the Olympic Games 87 Appendix A A.1 Table A Table A. IMF Classification of Advanced Economies

88 88 THE MICHIGAN JOURNAL OF BUSINESS A.2 Table B Table B. Data for host countries

89 To Host or Not to Host? A Comparison Study on the Long-Run Impact of the Olympic Games 89 A.3 Table C Table C. Data for runner-up countries