Hidden Baggage: Behavioral Responses to Changes in Airline Ticket Tax Disclosure

Finance and Economics Discussion Series Divisions of Research & Statistics and Monetary Affairs Federal Reserve Board, Washington, D.C. Hidden Baggage: Behavioral Responses to Changes in Airline Ticket Tax Disclosure Sebastien Bradley and Naomi E. Feldman 2018-057 Please cite this paper as: Bradley, Sebastian, and Naomi E. Feldman (2018). “Hidden Baggage: Behavioral Responses to Changes in Airline Ticket Tax Disclosure,” Finance and Economics Discussion Series 2018-057. Washington: Board of Governors of the Federal Reserve System, https://doi.org/10.17016/FEDS.2018.057. NOTE: Staff working papers in the Finance and Economics Discussion Series (FEDS) are preliminary materials circulated to stimulate discussion and critical comment. The analysis and conclusions set forth are those of the authors and do not indicate concurrence by other members of the research staff or the Board of Governors. References in publications to the Finance and Economics Discussion Series (other than acknowledgement) should be cleared with the author(s) to protect the tentative character of these papers.

Hidden Baggage: Behavioral Responses to Changes in Airline Ticket Tax Disclosure Sebastien Bradley and Naomi E. Feldman∗ July 2018 Abstract We examine the impact of a January 2012 enforcement action by the U.S. Department of Transportation that required domestic airlines to incorporate all mandatory ticket taxes in advertised fares. We show that the more prominent display of taxinclusive prices is associated with significant reductions in consumer tax incidence and demand along more heavily-taxed itineraries. Ticket revenues are commensurately reduced. These results suggest a pronounced degree of inattention to taxes not included in advertised fares and reinforce prior findings from the literature on tax salience in a quasi-experimental context characterized by economically-significant taxes and endogenous prices. JEL Codes: H31, H22, D90, D18 Keywords: tax salience, airlines, ticket taxes, tax incidence ∗Bradley: SchoolofEconomics,LeBowCollegeofBusiness,DrexelUniversity,1023GerriC.LeBowHall, 3220 Market Street, Philadelphia, PA 19104, sbradley@drexel.edu. Feldman: Research and Statistics, FederalReserveBoardofGovernors,20thandCStreets,NW,Washington,DC20551,naomi.e.feldman@frb.gov. We are grateful to Jan Brueckner, Mian Dai, Philipp D¨orrenberg, Matt Freedman, Jacob Goldin, Tatiana Homonoff,LauraKawano,RogerMcCain,NirupamaRao,NathanSeegert,KostasSerfes,JoelSlemrod,and Matthew Weinberg as well as discussants and other conference and seminar participants from the National TaxAssociation,DrexelUniversity,theInternationalInstituteofPublicFinance,theFederalReserveSystem Conference on Regional Analysis, the Southern Economic Association, the University of Michigan, Oxford University, Indiana University, Swarthmore College, and Cornell University for helpful comments. Jacob Adamcik, Christopher Collins, Aleksandra Kirilakha, Katherine Richard, and Lena Yemelyanov provided excellent research assistance. The statistical analysis of international airline ticket data compiled by the BureauofTransportationStatistics, U.S.DepartmentofTransportationwasconductedunderarrangements that maintain legal confidentiality. The views expressed are those of the authors and do not reflect concurrence by the U.S. Department of Transportation, the Board of Governors of the Federal Reserve System, or members of its research staff. Each author declares that s/he has no relevant or material financial interests that relate to the research described in this paper. 1

1 Introduction A growing body of literature has established that tax salience (i.e. visibility or transparency) can have a pronounced effect on behavioral responses to taxation for a variety of tax and tax-like instruments.1 One of the earliest and most robust findings from this literature is that consumers often fail to fully internalize total tax-inclusive prices when base prices and sales taxes are disclosed separately, as is the norm for U.S. retail sales (Chetty, Looney and Kroft (2009); Feldman and Ruffle (2015)). Due to the primarily experimental nature of the prior literature vis-`a-vis consumption tax salience, however, seller pricing behavior and possible exploitation of salience effects remain less studied. If consumers are inattentive to low-salience taxes, such that the elasticity of demand with respect to taxes is less than the elasticity of demand with respect to tax-exclusive (base) prices, producers will generally find it easier to pass taxes through to consumers and will bear a smaller share of the burden of the tax.2 Conversely, an increase in tax salience should lead to diminished tax incidence on consumers and a reduction in producer revenue (Chetty, Looney and Kroft, 2009). We evaluate the effects of increased tax salience on airline ticket pricing, demand, and revenues in the context of a regulatory change to the advertising of commercial airline tickets mandated by the U.S. Department of Transportation (DOT), whereby U.S. air carriers and online travel agents were required as of January 26, 2012 to incorporate all mandatory taxes and fees in their advertised fares. To our knowledge, this constitutes the only case in the U.S. of a regulatory switch from tax-exclusive to tax-inclusive pricing regimes, such that this paper represents the first quasi-experimental test of the basic experimental framework from the prior literature. Moreover, the market for air travel presents a unique setting in which 1See, for example, Chetty, Looney and Kroft (2009), Goldin and Homonoff (2013), Feldman and Ruffle (2015), and Feldman, Goldin and Homonoff (2018) (sales and excise taxes); Finkelstein (2009) (electronic tolls); Bradley (2017) (property taxes); or Chetty and Saez (2013) and Feldman, Katuˇsˇc´ak and Kawano (2016) (earned income and child tax credits). 2If consumers are wholly inattentive such that the tax elasticity of demand is zero, it is easy to see that the tax will fall entirely on consumers, at least in the short run. (See Chetty, Looney and Kroft (2009) or Reck (2016) for a discussion of the implications of longer-run budgetary adjustments.) This situation is indistinguishable in a static environment from complete pass-through resulting from infinitely elastic supply in a perfectly competitive market in long-run equilibrium, even where taxes are fully salient. 2

to examine tax salience due to the fact that airline ticket taxes are economically large and account for a non-trivial fraction of total airfares.3 Prior to 2012, U.S. domestic airlines had been allowed to advertise fares exclusive of specific (unit) tax amounts while publishing ticket taxes and fees at later stages of the onlineticket-buyingprocess,therebyrenderingvariationinroute-specificunittaxesrelatively invisible to consumers in their initial search stages. In that environment, learning about the variationinunittaxesamongrouteswouldrequireconsumerstoinitiatetheticket-purchasing process multiple times for different flight itineraries, thereby forcing (attentive) consumers to exert costly effort to compare tax-inclusive prices. Justifying the effort to initiate this process, however, would require prior knowledge of the existence of substantial potential variation in unit taxes—precisely the type of environment where inattention and failure to “learn by noticing” may be particularly of concern (DellaVigna (2009); Hanna, Mullainathan and Schwartzstein (2014)). Whereascognitivebiaseshaveservedtomotivatetheimplementationofvariousconsumer protections, primarily in the area of financial products,4 the DOT’s full-fare advertising rule represents the first instance of an application of tax salience considerations to U.S. federal regulations. These“fullfareadvertisingrules”(henceforthFFARinourterminology)provide a unique opportunity to study the importance of limited attention in modulating consumer responses to taxation and to quantify the magnitude of taxpayer optimization errors that arose under the prior low-salience ticket tax regime.5 3Average and median unit tax amounts in our sample of U.S. international flights amount to roughly $100, or 16 percent of the average total ticket price. 4See Barr, Mullainathan and Shafir (2009) for a broad discussion of arguments in favor of these types of regulations. Examples of such policies include the Pension Protection Act of 2006 (intended to promote automatic enrollment in retirement savings plans), elements of the Dodd-Frank Act (i.e. the mandatory provisionofmortgageescrowaccountstonewhomebuyers)ortheCreditCardAccountabilityResponsibility and Disclosure Act of 2009 (minimum payment disclosures). 5Equivalently, ticket taxes may be viewed through the lens of partitioned pricing as a type of “shrouded attribute”(GabaixandLaibson,2006). Tothispoint,FFARalsorequiredairlinestoprovidemoreprominent linkstoinformationregardingbaggagefees,whichrepresentaclearexampleofpartitionedpricingsimilarto cases considered elsewhere in the behavioral literature, such as printer ink cartridges (Gabaix and Laibson, 2006), shipping costs (Hossain and Morgan, 2007), or booking fees (Blake et al., 2017). Brueckner et al. (2013) examine the incentives for baggage fee unbundling and their resulting impacts on airline revenues, albeit without discussing the role of consumer inattention. Agarwal et al. (2014) provide a methodology 3

Using restricted-use (international) ticket data from the Bureau of Transportation Statistics’ Origin and Destination Survey (DB1B) over a period of 19 quarters surrounding the DOT rule change, we find that the more prominent presentation of tax-inclusive air fares following the implementation of FFAR is associated with a sharp decline in pass-through rates for unit ticket taxes. Prior to FFAR, airlines passed through nearly the entire tax onto consumers in the form of higher base and total fares, while in the post-FFAR period, only about 25 cents of every dollar of unit taxes is passed onto consumers. In addition, pass-through rates for other sources of airport- and route-specific costs that were not the subject of the new disclosure rules (e.g. runway fees, gate fees, navigation charges, noise and emissions fees, etc.) were not significantly affected. We also find that reductions in pass-through rates were generally largest in more highly concentrated markets, consistent with the elementary textbook theory of tax incidence under imperfect competition. Airlines thus appear to have partially insulated inattentive consumers from perceived fare increases due to the implementation of tax-inclusive pricing through large offsetting reductions in base fares. On balance, reduced ticket tax pass-through rates combined with the negative effects of unit taxes on ticket demand in the post-FFAR period together translate into significant reductions in airline ticket revenues along higher-tax routes, consistent with the predicted consequences of increasing tax salience and substitution away from high-tax routes. Controlling for all unobserved determinants of quarterly passenger demand by origin-destination city market and instrumenting for carriers’ endogenously-chosen base fares using a measure of competing carriers’ route availability, we find that a $5 increase in unit taxes (roughly equal to the average standard deviation in tax amounts within origin-destination markets) is associated with a 4.3 percent reduction in passenger volume in the post-FFAR period. for measuring the effects of fee disclosure on consumer welfare with an application to baggage fees. We are not able to assess empirically the effects of FFAR with respect to baggage fees due to lack of available data. However, we expect any potential effects to be heavily muted in our sample given our emphasis on international flights, where the major U.S. legacy carriers and their code-share partners have historically waived baggage fees on travelers’ first piece of checked luggage. 4

Allowing for attenuation of these demand effects due to reduced pass-through, price and quantity effects resulting from the same $5 tax increase contribute to a net reduction in airline ticket revenue of 2.4 percent.6 These effects reflect a relatively high elasticity of demand with respect to advertised fares within origin-destination city markets, and we cannot reject equal demand sensitivity to tax and non-tax fare components after the adoption of FFAR. Taken together, these findings provide strong quasi-experimental support for the main conclusions and predictions about the consequences of inattention to commodity taxes in the tax salience literature. However, relative to the field- and lab-generated experimental evidence presented in Chetty, Looney and Kroft (2009) and Feldman and Ruffle (2015), respectively, a key distinction in our setting is the ability of airlines to adjust pre-tax prices and—in the longer-term—route availability. Demand responses to more salient tax information are therefore attenuated through diminished tax incidence on consumers. Our results also help to inform the relatively narrow literature on commodity tax incidence, including Poterba (1996); Besley and Rosen (1999); or Carbonnier (2013), and we provide the first large-scale estimates of airline ticket tax pass-through rates.7 Given the nature of the market for air travel, our estimates serve as a test of the theoretical predictions on tax incidence in imperfectly competitive markets (Anderson, de Palma and Kreider (2001); Weyl and Fabinger (2013)) and complement recent estimates by Marion and Muehlegger (2011) and Conlon and Rao (2015) that emphasize the effects of market structure and supply conditions on tax incidence. Finally, our results also extend the literature devoted 6Perhaps not surprisingly, U.S. airlines have lobbied extensively to prevent and subsequently reverse the implementationofFFAR.Consistentwiththeseobjectives,theU.S.Housepassedthe“TransparentAirfares Act” in June 2014, which would have allowed airlines to revert to advertising tax-exclusive fares. The bill failed to reach the Senate before the conclusion of the 113th Congress. The FAA Reauthorization Bill of 2018—whichpassedtheU.S.HouseonApril27,2018byavoteof393/13andcurrentlyawaitsconsideration intheSenate—wouldlikewiseeliminatetax-inclusivepricingrequirements. Ourwithin-marketidentification strategydoesnotallowexaminationofaggregatedemandorrevenueeffectswhichcouldhaveresultedfroma perceptionofincreasedfaresfollowingtheadoptionofthefull-faredisclosureregimeor—correspondingly—its reversal, but the airline industry’s opposition to FFAR offers prima facie evidence of such concerns. 7Huang and Kanafani (2010) exploit variation in U.S. passenger facilities charges in order to obtain estimates of ticket tax incidence. Their results are limited to very modest variation in tax amounts across a sample of 50 U.S. airports. Karlsson, Odoni and Yamanaka (2004) provide descriptive evidence on effective ticket tax rates for domestic U.S. airfares. 5

to studying the impact of consumer disclosures, including Agarwal et al. (2014, 2015), and Keys and Wang (2018). The remainder of the paper is organized as follows: Section 2 describes the motivation for FFAR and its precise details in the context of the DOT’s ongoing regulatory action, Section 3 characterizes the data used in our analysis, Section 4 presents a general estimation framework, Section 5 presents and discusses our empirical results, and Section 6 concludes. 2 Full-Fare Advertising Rules The DOT’s full-fare disclosure rule was issued on April 20, 2011 and subsequently implemented on January 26, 2012 after a delay requested by U.S. air carriers to comply with technical deployment requirements. Strictly speaking, FFAR was not so much a regulatory change as an enforcement action. Under C.F.R. §399.84, airlines and travel agencies were already required to include all carrier-imposed charges (including fuel surcharges) as well as most government-imposed taxes or fees. However, the DOT had previously exercised discretion in terms of enforcement and exempted specific (unit) taxes and fees that were imposed on a per-passenger basis. Ad valorem taxes, including the U.S. domestic transportation tax, already appeared in airlines’ posted prices prior to 2012. Likewise, airport charges levied on per-movement rather than per-passenger basis, such as most runway fees, air navigation charges, noise and emissions fees, etc., could not be broken out as a separate passenger charge and would therefore have also been incorporated into airlines’ base fares before the imposition of FFAR. The DOT’s stated motivation for implementing FFAR in 2012 focused on consumer protection and concerns related to consumers being mislead as a result of tax- and fee-inclusive prices being less than fully transparent. Figure 1 highlights the nature of the potential challenge facingconsumers in selectingairline ticketsif tickettaxes are not immediately disclosed in advertised fares and consumers exhibit limited attention. The figure shows 18 possible 6

round-trip itineraries between New YorkCity’s John F. Kennedy airport (JFK) and Tel Aviv (TLV) ranked by total tax-inclusive total fares versus tax-exclusive base fares (a rank of 1 designating the lowest fare).8 As shown, itineraries above the 45-degree line are relatively more expensive in ordinal terms than their base fare rank would suggest, whereas itineraries below the line ought to be more attractive to consumers than their base fare rank would suggest. Thus, for example, one of the two least expensive itineraries on a tax-inclusive basis, JFK DL TLV :: TLV DL JFK, would only be ranked 10th out of 18 flights in tax-exclusive terms (in a three-way tie). A consumer might consequently be more inclined to choose JFK LY CDG LY TLV :: TLV LY JFK (in a three-way tie for the second lowest base fare), despite this itinerary ranking eighth in tax-inclusive terms, and costing about $40 more than the lowest-cost ticket overall. More broadly, much of the differences across total fare amounts can be attributed to relatively wide variation in tax amounts, ranging from a low of $89 for a non-stop Delta flight to a high of $195 for an EL AL flight with a layover in Paris Charles de Gaulle (CDG) in both directions. Table 1 underscores the specific sources of underlying tax variation by presenting a breakdown of unit taxes for three sample itineraries linking JFK and TLV. The first row lists the base fare, or the fare that would have been advertised to consumers at the first stage of the ticket-buying process pre-FFAR, whereas the total fare inclusive of all taxes and fees (bottom row) would have only appeared at a later stage. As the table illustrates, there are numerous country-specific taxes and fees built into the final prices. As all flights in our data 8We define an itinerary as a sequence of flight segments and ticketing carriers, while a route represents a sequence of flight segments only (i.e., departing and arriving airports, including an origin, final destination, and all stopovers). An origin-destination airport pair encompasses all possible itineraries connecting the same origin and final destination airports. The latter are nested within origin-destination city pairs, which comprise all airports within a 100-mile radius of the largest population center in the area. For example, JFK DL TLV :: TLV AF CDG AF JFK represents a round-trip itinerary between New York City’s John F. Kennedy Airport and Tel Aviv’s Ben Gurion Airport with an outbound flight on Delta Airlines and a return trip (with a layover in Paris’s Roissy Charles de Gaulle Airport) operated by Air France. The corresponding route, offered by potentially multiple carriers, would simply be JFK TLV :: TLV CDG JFK and the origin-destination airport pair is simply JFK :: TLV. The origin-destination city pair consists of potentially multiple origin and destination airports located within a 100 mile radius of New York City. This includes 7 airports in the vicinity of JFK, including New York’s La Guardia (LGA); Newark, New Jersey (EWR); and Philadelphia, Pennsylvania (PHL); along with 3 secondary airports. 7

originate or end in the U.S., all incur U.S. taxes and fees. The remaining taxes and fees are determined by the set of foreign airports where the flight “touches down” for a layover or as a final destination, and, in some cases, by route or airline flown. If taxes and fees are not taken into account by the consumer at the time of initial fare selection, one may think that column (3) offers the lowest price. Once presented with the additional cost attributable to taxesonasubsequentscreenintheticket-buyingprocess, theconsumermightinfer—without comparable information from other itineraries to suggest otherwise—that taxes would apply uniformly across ticket choices.9 Instead, the itinerary in column (3) is clearly the most expensive of the three options once taxes and fees are included in the total price. The net effect of this type of route-specific tax variation within and across the 300 largest internationalorigin-destinationcitymarketsservedbyU.S.carrierscanbeseeninFigure2in terms of either unit tax amounts (2a) or effective tax rates (i.e. unit taxes as a percentage of average total fares; 2b). As shown, Western European and Caribbean destinations (purple circles and light blue squares, respectively) tend to exhibit among the highest unit tax amounts as well as the highest standard deviation thereof, reflecting a combination of high taxes at destination airports as well as increased taxes accruing at stopover points on longer routes. Relative to total fares, Caribbean destinations trigger by far the highest effective tax rates (Figure 2b), which exceed 30 percent in certain cases. Suggestive evidence of passengers substituting toward lower-taxed routes as a result of FFAR within this set of 300 origin-destination city markets is shown in Figure 3. The figure depicts average four-quarter changes in the high-tax share of passenger volume accruing to the set of routes in the top and bottom quartiles of the ticket tax distribution (based on a balanced panel of ever-available route offerings within origin-destination city market). 9Evenpost-FFAR,popularfareaggregatorwebsitesandairlines’ownwebsitesrarelyfeaturethecomplete breakdown of taxes and fees by levying country that appears in Table 1. This reduces the probability that a consumercouldlearn,forexample,thatlayoversinCDGcontributeroughly$90inadditionaltaxesandfees relative to a non-stop flight that avoids CDG. Informally, a large fraction of commenters on this paper have reportedbeingsurprisedtolearnthattickettaxesarenot constant. Thisviewislargelycorrectfordomestic flights,suchthatmore“experience”inpurchasingU.S.domesticairtravelmayactivelydeterlearningabout the true scale of variation that exists for international flights. 8

As shown, the share of passengers traveling via relatively high-tax routes was generally growing prior to the implementation of FFAR and remained roughly unchanged in 2012Q1 and 2012Q2 (during which time only a fraction of travelers would have been exposed to taxinclusive pricing at the time of ticket purchase). Beginning in 2012Q3, however, high-tax routes experienced persistent declines in volume share in favor of lower-taxed routes, with this effect gradually tapering off after eight quarters. 3 Data The primary data for this project are drawn from the restricted-use (international) portion of the DOT’s Origin and Destination (O&D) Survey (DB1B) for the period 2009Q4-2014Q2, used in conjunction with information on airport charges from RDC Aviation plus detailed fare compositions scraped via a flexible fare search platform. The DB1B data consist of a 10 percent sample of all complete ticketed itineraries sold by U.S. reporting carriers and are reported quarterly, based on date of travel. From this sample, we extract only the set of international itineraries that either originate or terminate at a U.S. airport (i.e. “outbound” and “inbound” itineraries, respectively). Crucially, these data include all route and carrier characteristics, as well as the number of passengers traveling, distance flown, fare class, and the total tax- and fee-inclusive fares paid per passenger. The DB1B ticket data do not, however, provide a breakdown of the fare composition. We consequently rely on data from RDC Aviation and fare scrapes to construct a historical database of itinerary-specific ticket taxes and non-tax charges, which we match to the DB1B data in order to back out tax-exclusive prices (i.e. base fares). This process involves a complex series of steps, which we describe in much greater detail in Appendix A.1. In essence, this procedure requires parsing information from RDC Aviation on all applicable airport charges for a sample of over 50000 unique quarterly airport-route-aircraft combinations in order to separate individual charge items into either government-imposed taxes and 9

fees (levied on a per-passenger unit basis and thus, affected by FFAR) or non-tax charges (levied on a per-movement basis, and thus, unaffected by FFAR). Performing this decomposition in turn relies on fare construction information that we gleaned from over 30000 online fare searches performed over the period December 30, 2014 - January 29, 2015. Each scraped itinerary yields an extract of all applicable ticket tax codes, descriptive names, and corresponding dollar amounts, thereby enabling us to flag matching charges from the RDC database at the airport-route level and assemble these across all relevant airport-route segments on a historical basis.10 This group of initial charge and fare queries represents all routes in the DB1B sample flown by more than 36 passengers (in either direction) over the 2012Q4-2013Q3 period (i.e. averaging at least one passenger per day in the full 100 percent sample).11 A sample concordance between the set of scraped French ticket taxes levied on a roundtrip flight PHL DL CDG and the corresponding set of per-passenger charges for arriving or departing flights in Paris (CDG) as reported by RDC Aviation are given in Table A1. We confirmthattheremainingsetof9chargesidentifiedbyRDCasbeingleviedatCDG—shown in Table A2—are indeed levied on per-movement basis and fall broadly into the general categories of air navigation, infrastructure, noise, parking, runway, or terminal charges.12 We divide the resulting total amount for these non-tax charges according to the seating 10The tax amounts recovered through our web-scraping procedure present only a static snapshot of applicable taxes from early 2015. We do not use these scraped tax amounts directly in our analysis due to the risk that this approach might introduce classical measurement error whose variance would grow the further we extrapolate post-FFAR tax amounts backward through time, thereby potentially biasing our results in favor of finding increased consumer sensitivity to ticket taxes in the more recent past. We do, however, use the scraped tax amounts to cross-validate our calculations based on RDC Aviation’s airport charges database and use this information to improve our ticket tax calculator. An earlier draft of this paper using only scraped tax amounts (adjusted historically for bilateral exchange rate movements and a complete history of applicable U.S. ticket taxes) presents qualitatively similar results to the ones presented here, and ultimatelylendslittlecredencetothesemeasurementerrorconcerns. Resultsinvolvingscrapedtaxamounts are available upon request. 11These routes account for approximately 60 percent of total passenger volume. We exclude lower-volume routesfromoursetofinitialqueriesoutofconcernthatchangesinpassengertrafficalongtheseroutesmight be subject to a high degree of unexplained variability. 12Other categories of charges, such as government charges, can encompass either taxes levied on a perpassenger or per-movement basis and require special care. 10

capacity of the corresponding aircraft in order to allocate these on a per-ticket basis.13 A similar set of tax and non-tax charges likewise apply for the arriving and departing flight segments at PHL, which we consequently combine with the set of applicable charges at CDG to construct complete tax and non-tax charge amounts for the full PHL DL CDG itinerary.14 Matching our resulting itinerary-specific unit tax and non-tax charge amounts to the full setofticketeditinerariesintheDB1Byieldsover45000uniquematcheditinerarieswithvalid tickettaxinformation,coveringmorethan4.5millionpassengertripsovertheperiod2009Q4- 2014Q2.15 After subtracting itinerary-specific tax and non-tax charge amounts from total ticketed fares to recover a measure of tax-exclusive base fares, we aggregate each matched observationinthequartertDB1Bsampletothecarrierc, routei-levelanddefinemeasuresof totalpassengervolumeandpassenger-weightedaveragebasefares. Collectively,ciconstitutes a unique itinerary whose endpoints define an origin-destination airport pair j, and origindestination city market pair k (i.e., the product category). Consistent with other applications of the DB1B data in the literature (see e.g., Brueckner (2003) for a careful description), we focus exclusively on round-trip, coach-class, non-award travel.16,17 WealsoexcludeticketsflaggedbyDOTasinvolvingunrealisticallyhighcosts-permile(conditionalonfareclass), aswellasallticketeditinerariesfeaturingmultipletripbreaks (i.e. extended stopovers) which may trigger the application of different taxes.18 Likewise, we 13See Appendix A.1 for a description of data sources used in making determinations of aircraft usage. 14U.S. ticket taxes on international flights consist of 6 distinct tax codes. We rely on multiple U.S. government sources, including the FAA, DHS, USDA, and CBP, to construct a complete historical record of airport-specific U.S. ticket taxes rather than use the RDC database for this purpose. 15For matching purposes, itineraries associated with Northwest or Continental Airlines as the reporting carrier in the earlier quarters of the DB1B data are re-coded as their merger partners, Delta and United Airlines, respectively. 16We apply multiple criteria based on cost-per-mile for defining award travel. See Appendix A.1.1 for details. Award travel thus defined appears to account for up to 10 percent of passenger volume. 17For tickets featuring different fare class segments, we define an itinerary as coach-class so long as the coach portion of the itinerary accounted for at least 90 percent of miles flown. Tests of differential FFAR reactions by class of service (not shown) suffer from low power. As a result, we cannot conclude whether first and business class travelers are any more or less sensitive than coach passengers to the implementation of tax-inclusive pricing. 18TheUKAirPassengerDuty, forexample, isonlypayableonflightsoriginating intheUK.Thetaxdoes not therefore generally apply to international flights with a layover in the UK, unless the layover exceeds 24 hours in duration. SimilarrulesapplytoflightsegmentswithintheU.S.aspartofaninternationalitinerary, with differing application of domestic transportation and segment taxes depending on the duration of these 11

omititinerariesinvolvingU.S.territories, Alaska, orHawaiiduetotheapplicationofdifferent U.S. ticket tax rules.19 Finally, we exclude all group tickets covering more than 9 passengers on the grounds that these are likely to involve negotiated fares whose purchasers (e.g. tour operators or the U.S. government) are unlikely to be subject to the same behavioral biases as individual consumers. We ultimately limit our analyses to the top 300 international origin-destination city markets(rankedbytotaloutboundandinboundpassengervolumein2011), eachofwhichare servicedbyanaverageof6.5availableitinerariesandaccountfor55percentoftotalpassenger volumeinourmatchedDB1B-taxsample. Thisrestrictionhasthevirtueofexcludingthinner markets where idiosyncratic variation in passenger demand may be especially prevalent and contribute to statistical imprecision. Unreported sensitivity analyses involving the complete sample of 498 city markets for which we have non-missing ticket tax and non-tax charge data (and non-zero within-market variation therein) account for 62 percent of matched passenger volume and yield qualitatively similar, yet less precisely-estimated results, consistent with this last concern. Table 2 reports basic summary statistics from our final estimation sample. As shown, total tax- and charge-inclusive fares (TotalFare) average $750, while mean and median specific taxes (UnitTaxes) are roughly $100, with a standard deviation of approximately $45.20 Non-tax charges (NonTaxCharges) account for a relatively smaller fraction of total fares and amount to roughly $20, albeit with a greater degree of dispersion around the mean than unit taxes. Mean and median passenger volume per quarter remain quite modest, with the spread between these and the large standard deviation giving a partial glimpse at the highly skewed nature of itinerary traffic. Owing in part to the difficulty of assembling ticket domestic layovers. 19With respect to U.S. territories, exceptionally high passenger volume moreover likely reflects the transportation of U.S. military personnel, the majority of whom presumably do not book their own air travel. 20For comparison, within origin-destination city markets, unit taxes exhibit a mean volume-weighted standard deviation of just under $5 across all 300 markets. In addition, the volume-unweighted average difference between the highest and lowest taxed itineraries within an O&D city market is roughly $26 (not shown). 12

tax and charges data for secondary airports, we see that the median itinerary in our sample involves zero layovers, and in practice, we are unable to match any itineraries featuring more than 4 layovers (i.e. 6 flight segments). 4 Model 4.1 Tax Incidence and Tax Salience Despite FFAR having had no effect on the true level of ticket taxes owed, heightened awareness of these tax amounts should yield a shift in the tax burden from formerly-inattentive consumers onto producers—in proportion to the extent of de-biasing induced by the switch to tax-inclusive pricing. Depending on the magnitude of the resulting reduction in base fares, consumers may have been more or less shielded from perceiving prices as varying by the full amount of unit ticket taxes in the post-FFAR period. Consequently, changes in tax incidence due to FFAR are not only informative with respect to the costs of consumer inattention but are also indicative of the remaining potential for consumer demand to show marked reactions to FFAR.21 We adapt Chetty (2009) and Chetty, Looney and Kroft (2009) to derive predictions regarding the effect of tax salience on the economic incidence of a unit tax in perfectly competitive versus monopoly markets. Under the standard neoclassical theory of tax incidence, net-of-tax producer prices (e.g., base fares), p, adjust to the imposition of a unit tax, t, according to the relative elasticities of supply and demand, where the latter elasticity is assumed to be the same regardless of whether changes in gross-of-tax consumer prices (q = p + t) are driven by changes in net-of-tax prices or taxes. However, if consumers are subject to limited attention and taxes are less than fully salient, this introduces the possi- 21This situation differs from the “sufficient statistics” approach advocated by Chetty, Looney and Kroft (2009), whereby estimates of tax incidence can be recovered as a function of the tax and price elasticities of demand(whichdifferonlyduetoinattention)andtheelasticityofsupply. Here,weinferinattentiondirectly fromestimatedchangesinelasticitiesofpassengerdemandconditional onfinalpricesadjustingendogenously to mitigate the consequences of increased tax salience. 13

bility that consumers may respond differently to changes in prices that arise from changes in base prices compared to changes that arise from tax changes. We model this possibility by allowing consumers to perceive a fraction θ ≥ 0 of the true tax amount, q = p + θt, θ such that observed consumer demand can be expressed as D(q ) = D(p + θt). θ = 1 in θ the full-attention, full-salience case (as in the neoclassical model), and D(q ) = D(q). By θ assumption, taxes that are included in posted prices are fully salient: θ = 1. At the Qtr>2012Q1 other extreme, θ = 0 corresponds to complete inattention or zero salience (i.e. consumers completely ignore the tax when making purchasing decisions). More generally, θ represents the degree of tax salience (consumer inattention) and can be measured as the ratio of the price elasticities of demand with respect to the tax price versus the base price (evaluated at the perceived tax-inclusive price): ε = θ∂D q θ = θε . D,q|t ∂q D(q ) D,q|p θ Starting from the assumption of perfect competition, total differentiation of the market clearing condition D(p+θt) = S(p) yields dp ∂D/∂q θ·ε |t D,q = − ≡ − (1) dt ∂D/∂q −∂S/∂p ε − q θε |p D,q p S,p dp dq t·ε D,q = = − (2) dθ dθ ε − q θε D,q p S,p where ε = ∂S p represents the elasticity of supply at the net-of-tax price. S,p ∂p S(p) When θ = 1, equation (1) produces the standard full-optimization result, whereby the incidence of a unit tax on producer prices is proportional to the magnitude of the elasticity of demand relative to the magnitude of the combined elasticities of demand and supply. The tax burden borne by producers—all else equal—is hence increasing in θ, conditional upon a nonzero demand elasticity. Unsurprisingly, producers bear none of the tax burden (dp = 0) dt when the tax is fully obfuscated from inattentive consumers and θ = 0. This situation is empiricallyindistinguishablefrommorestandardresultsinvolvingperfectlyinelasticdemand (ε = 0) or perfectly elastic supply (ε = ∞), as in a perfectly competitive market in D,q S,p long-run equilibrium. Independent variation in θ (induced by FFAR), t, and the degree 14

of market competition are therefore key to separately identifying tax salience effects from demand and supply elasticity effects in our analysis. Equation (2) characterizes the impact of full de-biasing resulting from a shift in saliency regime (e.g. from tax-exclusive to tax-inclusive pricing in a world where consumers are fully inattentive to taxes that are not advertised in posted prices) on both net-of-tax and grossof-tax prices in the presence of pre-existing taxes. As equations (1) and (2) suggest, small changes in pass-through rates of ticket taxes to total fares resulting from the adoption of FFAR could result either from θ ≈ 1, ε ≈ 0, or ε ≈ ∞ (or some combination PreFFAR D,q S,p thereof). Regardless of salience effects, consumers might consequently be unaffected by FFAR if the market for international air travel were perfectly competitive and subject to constant marginal costs. Of course, the airline industry is not generally considered to be perfectly competitive; nevertheless, we exploit the fact that individual markets may differ widely in their degree of market concentration. In the case of imperfect competition, the monopolist confronted by inattentive consumers must solve the modified profit maximization problem maxp·D(p+θt)−C(D(p+θt)) p which yields the conventional Lerner Formula, with the modification that the marginal cost of production, C(cid:48)(·), and elasticity of demand, ε , are implicit functions of θ: D,q (cid:20) D(p∗ +θt) 1 (cid:21) p∗ 1+ ≡ C(cid:48)(D(p∗ +θt)) (3) ∂D/∂q p∗ |p C(cid:48)(D(p∗ +θt)) ⇔ p∗ = (4) 1+ 1 q θ ∗ ε p∗ D,q|p where p∗ is the profit-maximizing net-of-tax price for the monopolist. 15

By the Implicit Function Theorem, (cid:104) (cid:105) −θ 1− D(·)D(cid:48)(cid:48)(·) −C(cid:48)(cid:48)(·)D(cid:48)(·) dp∗ (D(cid:48)(·))2 = (5) (cid:104) (cid:105) dt 1+ 1− D(·)D(cid:48)(cid:48)(·) −C(cid:48)(cid:48)(·)D(cid:48)(·) (D(cid:48)(·))2 A zero salience tax (θ = 0) again delivers full pass-through onto consumers, but θ otherwise plays a more nuanced role depending on the underlying nature of demand. For illustration, weconsidertwosimplifyingcasesinvolvingconstantmarginalcosts,C(cid:48)(·) = κ: lineardemand or constant demand elasticity. Assuming linear demand of the form D(p+θt) = a−b(p+θt), 1 (cid:104) a(cid:105) 1 p∗ = κ+ − θt (6) 2 b 2 dp∗ 1 = − θ (7) dt 2 dp∗ 1 = − t (8) dθ 2 Following standard principles of tax incidence, a fully-salient tax (θ = 1) hence falls equally on both consumers and the monopolist. Correspondingly, full de-biasing leads to the net-oftax producer price falling by exactly half of the unit tax amount, or $0.50 per dollar. This suggests a large potential impact of FFAR on ticket-tax pass-through rates in imperfectlycompetitive markets (assuming approximately linear demand), even if demand is otherwise relatively inelastic or airlines face near-constant marginal costs. If demand instead exhibits constant elasticity of the form D(p+θt) = A(p+θt)−b, such that ε = −b and ε = −θb, then D,q|p D,q|t κ p∗ = (9) 1+ 1 q θ ∗ εD,q p∗ dp∗ 1 dq∗ (1−θ)+ε D,q = −θ· ⇒ = (10) dt 1+ε dt 1+ε D,q D,q dp∗ 1 dq∗ = −t· = (11) dθ 1+ε dθ D,q 16

A fully-salient tax in this (admittedly special) context will be overshifted onto consumers whenever ε < −1. Contrary to the perfectly-competitive case or the linear demand D,q monopoly case, θ thus amplifies rather than attenuates tax incidence on consumers, and de-biasing due to the adoption of FFAR could conceivably raise profit-maximizing net-oftax prices. The effect of an increase in tax salience on tax incidence in any given market therefore depends upon market structure and the curvature of marginal costs and demand. 4.2 Empirical Specifications We estimate the average dq empirically following Weyl and Fabinger (2013) and Conlon dt and Rao (2015) as the share of each dollar in ticket taxes that is passed through to total fares according to the following general specification in order to measure consumer ticket tax incidence pre- and post-FFAR: TotalFare = α+β UnitTaxes +β UnitTaxes ×I[Qtr > 2012Q1] cit 1 cit 2 cit t +˜γX˜ +η +ν +(cid:15) (12) ij ct kt cit TotalFare represents the average total fare paid by consumers for a flight operated by cit carrier c on route i in quarter t. Unit taxes (UnitTaxes ) are defined at the corresponding cit itinerary level, and the post-FFAR period indicator, I[Qtr > 2012Q1] , is set to 1 in all perit ods falling after the first quarter of 2012 and is zero otherwise.22 As a placebo test, we extend (12) with controls for pre- and post-FFAR effects of non-tax charges (NonTaxCharges ) cit because these were always required to be included in advertised fares. Beyond these main variables of interest, X˜ represents a vector of route and origin-destination airport pair charij acteristics, including categorical indicators for the number of connecting flight segments as 22Given that the DOT disclosure rules only went into effect on January 26, 2012 and that our ticket data are dated only by the quarter flown, it is uncertain what fraction of 2012Q1 travelers would have been exposed to FFAR at the time of ticket purchasing. We consequently omit 2012Q1 from our analysis altogether and acknowledge the possibility that a shrinking fraction of passengers traveling in the following threequartersmighthavestillpurchasedtheirticketsunderthepre-FFARregime. Airlinesdonotgenerally allow ticket purchases more than 10 months prior to the date of travel. 17

well as cubic polynomials in distance flown, market concentration, capacity utilization, and the log of carrier passenger volume at the (U.S.) airport of origin (for both domestic and international flights). η accounts for unobserved time-varying carrier-specific attributes ct that might be correlated with the tax salience effects of FFAR, such as pre-existing variation in the transparency of tax information on carriers’ own websites, or differences in the existence of baggage fees and their associated disclosure. Seasonality effects and secular trends influencing origin-destination city-pair pricing are captured in ν . Remaining unobserved kt sources of variation in total fares are attributed to (cid:15) . An implicit assumption is that cit unobserved (non-tax) determinants of route i ticket prices are uncorrelated (conditional on distance, carrier, etc.) with the timing of FFAR or ticket tax amounts, such that these do not represent a source of omitted variable bias.23 In our preferred specification involving a full set of origin-destination city × quarter (ν ) and carrier × quarter (η ) fixed effects, identification rests on within-quarter variation kt ct in ticket taxes and total fares across itineraries serving the same city pairs, allowing for the relationship between taxes and total fares to vary pre- and post-FFAR. β is thus the 2 difference-in-differences estimator of the change in ticket tax pass-through rates associated with FFAR and reflects the impact of de-biasing (i.e. bringing the tax elasticity of demand into alignment with the price elasticity of demand), conditional on market supply conditions. Weallowthisde-biasingeffecttovarymoregenerallywithmarketconcentrationandcapacity utilization in later specifications to test for heterogenous effects related to these supply conditions. In all but our basic specifications, estimation of pre- and post-FFAR passthrough rates for non-tax charges alongside unit taxes offers a valuable comparison given 23A potential concern in this context is that if taxing authorities are responsive to changes in passenger demand (e.g. such as if airports compete actively for volume), unit tax amounts may respond endogenously to tax salience effects. Given the asynchronous timing between our measurement of ticket tax amounts and the DB1B’s reporting of passenger volume on the basis of the date of travel as opposed to the date of purchase,thiswouldtendtobiasourestimatesoftheeffectofthefull-fareadvertisingrulestowardzero(i.e. because an endogenous reduction in ticket taxes due to a reduction in ticket purchases in the prior quarter, for instance, would be partially matched with a continued decline in passenger traffic in the quarter(s) after the rate cut). A qualitative assessment suggests that unit taxes change infrequently and are generally committed to long-term budgetary outlays, such as funding for infrastructure improvement projects. 18

thatonlythelatterweresubjecttonewdisclosurerulesunderFFAR.Accountingfornon-tax charges in this manner helps corroborate the validity of our general difference-in-differences identification strategy, despite our inability to exploit more precise timing variation as a result of the manner in which the DB1B data are recorded. Our empirical strategy with respect to estimating the effects of FFAR on additional demand outcomes involves a similar difference-in-differences or triple-differencing (accounting for non-tax charges) approach. Adding controls for average base fares to the empirical model yields a simple adaptation of (12): ln(Y ) = α+[β BaseFare +β BaseFare ×I[Qtr > 2012Q1] ] (13) cit 1 cit 2 cit t +β UnitTaxes +β UnitTaxes ×I[Qtr > 2012Q1] +˜γX˜ +η +ν +(cid:15) 3 cit 4 cit t ij ct kt cit where Y alternately represents itinerary-level passenger volume or tax-exclusive ticket revcit enue. In the latter case, we exclude β BaseFare + β BaseFare × I[Qtr > 2012Q1] 1 cit 2 cit t from our model in order to measure the combined impact on ticket revenue coming from both endogenous price responses (i.e. changes in pass-through rates) as well as changes in passenger demand. Otherwise, base fares are measured as the difference between total fares and unit taxes and—where included—non-tax charges, such that base fares are defined differently depending on whether we break out non-tax charges alongside unit taxes, and we allow consumers to exhibit differing price elasticities of demand pre- and post-FFAR. Naturally, the simultaneous determination of prices and quantities in our demand specification will yield biased ordinary least squares estimates of the semi-elasticity of demand with respect to base fares, and this issue is further compounded by the possibility of endogenous variation in pass-through rates resulting from FFAR. We consequently use a measure of exogenouspricecompetitionasaninstrumentforbasefares(aloneandinteractedwiththesame I[Qtr > 2012Q1] indicator) and estimate (13) via both OLS and two stage least squares t (IV) for comparison. Our preferred instrument for this purpose is measured as the number of 19

itineraries offered by competing carriers (excluding code-share or alliance partners) servicing the same O&D airport pair in a given quarter (based on the full DB1B sample).24 If ticket taxes were fully salient prior to FFAR, we should expect demand for airline tickets to be equally sensitive to changes in appropriately-instrumented base fares, β , as 1 to variation in unit taxes in the pre-period, β , or non-tax charges. Correspondingly, β 3 4 ought to equal β (assumed to be zero) in this case. In the alternative, θ ≡ β3 2 Qtr<2012Q1 β1 measures consumer inattention in the pre-FFAR period, whereas θ ≡ β3+β4 mea- Qtr>2012Q1 β1+β2 sures consumer inattention post-FFAR. By assumption, consumers are expected to optimize fully with respect to taxes when these are included in posted prices. θ −θ Qtr>2012Q1 Qtr<2012Q1 hence reflects the extent of de-biasing associated with the more salient presentation of unit taxes under full-fare advertising. It is important to note that changes in passenger volume in response to FFAR may have arisen either through shifts in aggregate demand (such as if inattentive consumers perceived airfares to have risen across the board as a result of FFAR) or through cross-itinerary substitution. Increased tax salience might for instance induce consumers to substitute towards itineraries with fewer layovers, or layovers at more lightly taxed airports, to avoid the accumulation of unit taxes at each departing and arriving airport along their route. By exploiting within origin-destination city market × quarter variation in unit tax amounts, our identification strategy addresses only the latter channel. As such, our estimates cannot readily be translated into aggregate demand or aggregate ticket revenue effects. 24SeeLederman(2007)orBerryandJia(2010)forsimilarIVstrategiesandtheuseofcomparableinstrument(s) in the literature. We also consider the use of cost-shifter instruments constructed as an interaction of trip distance and quarterly jet fuel or oil (West Texas Intermediate) prices, or 6-month NYMEX futures thereof. Given heterogeneity in airline fuel and exchange rate hedging strategies coupled with unobserved airport-specific variation in delivered dollar-denominated fuel prices, these instruments suffer from instrument weakness in most tests. Results are available from the authors upon request. 20

5 Results 5.1 Tax Incidence Table 3 presents the results from the estimation of Equation (12). All specifications include the full set of controls in X˜. These are suppressed from Table 3 for brevity but can be found in Appendix Table A4. Additionally, Column 1 controls for carrier × quarter fixed effects, while Columns 2 and 3 further incorporate origin-destination city-pair × quarter fixed effects and represent our preferred specifications. Large differences between Columns 1 and the others in estimated pass-through rates in both the pre- and post-FFAR periods highlight the importance of controlling for unobserved time-varying product characteristics which might otherwise yield a spurious association between ticket taxes and total fares. Based on the results in Column 2, ticket-tax pass-through in the pre-FFAR period is approximately 0.99, consistent with consumers having borne essentially all of the tax burden prior to 2012, either because of relatively low “true” elasticity of demand (high elasticity of supply) or because of a high degree of consumer inattention. Only this last possibility, however, can explain the sharp reduction in average pass-through rates following the adoption of tax-inclusive pricing. In the post-FFAR period, the ticket tax pass-through rate falls by 0.74 (β ), so 4 that, on net, every dollar increase in unit taxes is associated with a 25 cent increase in total fares. Three-quarters of every dollar in ticket taxes are thus borne by the airlines in the post-FFAR period, in marked contrast to the pre-FFAR period when consumers bore the entire tax. Column 3 of Table 3 introduces our measure of itinerary-specific non-tax charges. Due to the manner in which non-tax charges are levied (i.e. on a per-movement basis instead of per-passenger), these constitute a cost of airline operations much like any other, and their inclusion in advertised fares was consequently unaffected by FFAR. Thus, non-tax charges serve as a type of placebo control in that pass-through rates for these charges should have remained unchanged in the post-FFAR period and, furthermore, should be similar to unit 21

tax pass-through rates once both are treated equally: namely, once both are required to be presentedupfrontaspartofasingletax-andfee-inclusivepricepost-FFAR.Theresultsshow that pass-through in the pre-FFAR period for unit taxes is little changed from column (2) at 0.958 cents for every dollar of unit taxes. Non-tax charges, however, show a significantly lower pass-through rate in the pre-period (p-value = 0.065), consistent with their inclusion in posted prices precluding airlines from shifting these itinerary-specific costs fully onto consumers. Moreover, pass-through rates for non-tax charges are virtually unchanged post- FFAR—as expected, given that the rule change did not impact the presentation of these charges to consumers. Pass-through rates in the post-FFAR period of 0.247 and 0.373 for unit taxes and non-tax charges, respectively, are not statistically-distinguishable (p-value = 0.708), consistent with both sets of costs being presented in an equally-salient manner. 5.2 Passenger Demand and Tax-Exclusive Total Revenue Table4hasanidenticalstructuretoTable3butfocusesonthepost-FFAReffectofunittaxes on itinerary-level passenger volume (columns 1-2) and total revenue (column 3). Ordinary leastsquares(OLS)resultsarepresentedincolumn1; however, asdiscussedinsection4, OLS estimates in this specification are likely to suffer from endogeneity bias. As always, failing to account for the simultaneous determination of equilibrium prices and quantities should yield positively-biased OLS estimates of the price semi-elasticity of demand. Indeed, as shown in the previous section, ticket prices were themselves endogenously impacted by FFAR, with larger reductions in base and total fares arising along higher-taxed routes. We consequently implement an instrumental variables (IV) strategy using the number of itineraries offered by competing carriers servicing the same O&D airport pair as our preferred instrument in order to focus on exogenous variation in base fares. IV estimates are reported in column 2 of Table 4 and reveal multiple important results. First, although the point estimates suggest the possibility of some heightened demand sensitivity with respect to base fares in the post-FFAR period, we cannot reject that changes 22

in base fares affect demand similarly in both the pre- and post-FFAR periods. The same is true for non-tax charges. Unit taxes, however, show no statistically significant impact on demand in the pre-period but show a large negative impact in the post-period. Moreover, t-tests of the equality of estimated coefficients show that we cannot reject equality of the impact of base fares and non-tax charges on demand either pre- or post-FFAR (p-values of 0.311 and 0.483, respectively), whereas we can reject equality of each with unit taxes in the pre-FFAR period (p-values of 0.000 and 0.005, respectively). Post-FFAR, we cannot reject a test of equality of demand effects due to base fares, unit taxes, or non-tax charges (p-value = 0.126). Consumers’ under-reaction to components of the total price that are not fully salient (i.e. unit taxes in the pre-FFAR period) serves as further evidence of the pronounced effects of limited attention. Adoption of FFAR, however, is associated with significant de-biasing, such that when base fares, unit taxes, and non-tax charges are all included in total fares in an equally salient manner, consumers respond to each equally—consistent with the standard theory of (attentive) consumer behavior. The results from column 2 can also be interpreted as estimated elasticities, presented in the bottom half of the table. In order to convert our semi-elasticity estimates into directlycomparable price elasticities of demand, we evaluate each of our point estimates in relation to the average value of total fares (about $750). The bottom panel of Table 4 reports these calculations. An increase in base fares equal to 1 percent of total fares (roughly $7.50) in the pre-FFAR period thus implies a 3.23 percent reduction in demand. This elasticity increases slightly in absolute terms in the post-period but is statistically-indistinguishable from the pre-period. The elasticity of demand with respect to non-tax charges at -2.25 and -3.44 is statistically similar to that of base fares in both the pre- and post-FFAR periods, respectively. In contrast, the demand elasticity with respect to unit taxes is positive and statistically-insignificant in the pre-period but negative and significant in the post-period. Demand hence falls by 6.36 percent in response to an increase in unit taxes of an amount equal to one percent of total fares in the post-period. While larger than the elasticity of 23

demand with respect to base fares, a 95 percent confidence interval around our estimate of the unit tax elasticity of demand spans a range of approximately -3.45 to -9.26, and we cannot reject that the post-FFAR base fare and unit tax elasticities are equal.25 Our estimated elasticities fall at the high end of the range of elasticity estimates for air travel reviewed in Gillen, Morrison and Stewart (2003) or InterVistas (2007), which combine studies based on domestic and international travel, the latter markets tending toward higher elasticities given the relative importance of leisure travel. Berry and Jia (2010) document a trend toward increasing elasticities between 1999 and 2006 and report a main estimate of 1.05 for the latter period based on U.S. domestic flights only. It is worth noting, however, thatelasticityestimatesbasedonDOTticketdatafromthepre-FFARerawillsystematically understate consumer sensitivity to advertised (base) fares as a result of inattention to the unit tax portion of total fares reported in the DB1B.26 Furthermore, it is also important to emphasize that the source of identifying variation in our analyses arise within O&D city market, such that our estimates of demand responses depend fundamentally upon patterns of consumer substitution across itineraries serving the same origin and destination. This is a much narrower source of identifying variation than in most studies of airline demand, andconsumersmayreasonablyviewitinerarieswithinsuchnarrowly-definedmarketsasmore highly substitutable than itineraries serving the same general regions, origins, or destinations (separately). A key parameter of interest with respect to tax salience is the degree of taxpayer inat- 25Taken seriously, consumers could exhibit hyper-sensitivity to unit taxes for several reasons, especially in the short-run aftermath of the adoption of FFAR. While ticket taxes are now included in advertised prices, they are still enumerated before final purchase, thereby calling special attention to their magnitude. Moreover,consumersmighthaveexperiencedinitialshockattheshiftinpricingnorms(i.e. the“Hawthorne effect”), an effect to which particular carriers might have advertently or inadvertently drawn attention in their roll-out of FFAR pricing rules. Spirit Airlines, for instance, made an explicit point of alleging on their website that the new DOT rule was requiring airlines to “hide” taxes from consumers (i.e. by rolling these into a single total fare). A newly-attentive—or surprised—consumer might plausibly have exhibited tax aversion as a result, at least temporarily. 26Interestingly, the Wall Street Journal reported a claim by Delta Airlines in December 2017 that for every dollar increase in ticket taxes (specifically, U.S. passenger facility charges), demand falls by one percent. Based on the typical average domestic fare of $300 quoted in the same article, this implies an elasticity of -3, precisely in line with our calculations. (https://www.wsj.com/article_email/ airports-want-to-raise-ticket-fees-airlines-say-no-fight-ensues-1512729000-lMyQjAxMTI3NDAwODgwMjg5Wj/) 24

tention measured as the ratio of the estimated elasticity of demand with respect to taxes relative to the elasticity of demand with respect to tax-exclusive prices. As discussed in Section 4, the post-FFAR change in this ratio provides a direct measure of the change in consumer inattention resulting from the implementation of full-fare advertising. Somewhat trivially, using our IV estimates from column 2 and taking into account the degree of statistical imprecision surrounding our point estimates, we cannot refute full inattention in the pre-periodandthusassuming—consistentwithourestimates—thatconsumersoptimizefully when taxes are included in advertised fares, full de-biasing as a consequence of FFAR.27 By way of comparison, Chetty, Looney and Kroft (2009) document a degree of inattention of approximately 0.35 under sales tax-exclusive pricing, such that their experimental introduction of tax-inclusive pricing on grocery store shelves is associated with a change in inattention of 0.65. It is a priori ambiguous whether to expect more or less severe inattention to ticket taxes under tax-exclusive pricing given the combination of larger financial stakes (i.e. more costly optimization errors) and fewer learning opportunities or experience to eradicate biases in the context of ticket taxes on international airfare, but our evidence suggests that the latter mechanism dominates. As shown in Table 3, unit tax pass-through rates fell from approximately 1 to 0.25. For the average ticket sold post-FFAR along higher-taxed itineraries, this should constitute a significant loss in ticket revenue through reduced base fares. Moreover, the results from column 2 of Table 4 establish that increased ticket tax salience could lead to further possible revenue losses through reductions in passenger demand. Column 3 of Table 4 presents estimates of these combined price and quantity effects on itinerary-level ticket revenues exclusive of unit taxes and non-tax charges (measured in logs). Consistent with the prior results,unittaxesinthepre-FFARperiodhavenostatistically-significantimpactonrevenues while non-tax charges have a significant negative impact (reflecting both incomplete pass- 27More precisely, θ = ∂ln(Passengers)/∂UnitTaxes = 0.304 = −0.69, with a 95% confidence Qtrt<2012Q1 ∂ln(Passengers)/∂BaseFare −0.438 interval spanning the range [−2.06,0.67], and θ = 0.304−1.179 =1.59, with a confidence interval Qtrt>2012Q1 −0.438−0.113 spanning the range [0.90,2.28]. 25

through ofthe latter charges, as well as their negative demand effects). Post-FFAR, however, a $5 increase in unit taxes (approximately equal to the average within-market standard deviation of unit taxes in our estimation sample) is associated with a 2.4 percent reduction in ticket revenue.28 For comparison, a simple back-of-the-envelope calculation of tax-exclusive revenue losses attributable to the product of price and quantity effects identified in Table 3 and Table 4, column 2, would instead imply a 2.8 percent reduction in ticket revenue for a tax increase of the same magnitude. These represent large potential losses in ticket revenue (from comparably large variation in unit tax amounts relative to average within-O&D city market variation therein) and lend strong justification for the U.S. airline industry’s intense and persistent efforts to reverse FFAR through lobbying and public relations campaigns. Moreover, it is also possible that carriers may have compensated for lower base fares and ticket revenue through increased reliance on product unbundling and the use of less heavily regulated add-on fees, such as baggage and check-in fees, seat upgrades, in-flight meals and service, etc., whose costs to consumers we do not observe in the ticket data.29 If airline markets were perfectly competitive, this is precisely the expected response due to carriers adjusting to restore equality between (fee-inclusive) prices and marginal cost (Agarwal et al., 2015). Though this last characterization surely does not apply to airlines, we are unable to assert that our estimates of ticket revenue losses or reduction in unit tax pass-through rates represent a pure transfer of surplus from airlines to consumers.30 28i.e. e(0.05∗(0.195−0.682))−1)=−0.024. 29The airline industry has likewise lobbied heavily—and thus far successfully—to prevent the DOT from requiring more prominent disclosure of add-on fees. 30Appendix A.3 characterizes the evolution of the largest U.S. carriers’ sources of revenue from international and domestic operations on the basis of quarterly financial statement information compiled by the DOT. With the possible exception of United/Continental, it does not appear that the implementation of FFAR coincided with a sharp break in carriers’ reliance on add-on fees. In the United/Continental case, the shift in reliance on add-on fees as a source of revenues more likely reflects the coincident timing of merger-related restructuring and opportunities afforded by the alignment of business practices at that time. 26

5.3 Heterogeneity in Pass-through: Market Concentration and Capacity Utilization In this section we consider the possibility of heterogeneous effects of FFAR on tax incidence as a function of market supply conditions, including market concentration and capacity utilization. According to the basic theory of tax incidence—based on linear demand and fully-salient taxes—taxes should fall relatively more heavily on firms in less competitive markets.31 This prediction has not been tested for less than fully-salient taxes, let alone in an environment where the degree of salience (and changes therein) may depend in part on the availability of competing product offerings in order for consumers to make informative comparisons. We compute a Herfindahl-Hirschman Index (HHI) of market concentration based on carrierrevenueshareswithinorigin-destinationairport-pairsinthefullDB1Bsample—regardless oftheavailabilityofmatchingtaxinformation,classofservice,andoutboundversusinbound, round-trip versus one-way status—and we divide this number by 10000 to obtain HHI values ranging from 0 (perfect competition) to 1 (monopoly). Mean and median HHI levels in our estimation sample are 5600 and 5000, respectively, such that what are typically considered “competitive” markets based on an adaptation of the classification introduced by Borenstein and Rose (1994) account for just over half of all observations, and monopolistic markets account for only approximately 5 percent of observations.32,33 We allow for market concen- 31In the simple case of constant marginal cost production, taxes will be born entirely by consumers in a perfectly competitive market, but shared equally with a monopolistic firm. The tax incidence effects of complete inattention versus perfect competition with constant marginal costs are thus isomorphic. More broadly, pass-through rates in imperfectly-competitive markets depend not only on the relative elasticities of supply and demand, but also the curvature of demand, with the result that full or over-shifting of taxes onto consumers are in principle possible. See Weyl and Fabinger (2013) for an extended discussion. 32Translation of Borenstein and Rose’s 1994 definition of monopoly, duopoly, and competitive markets (originallybasedoncarriersharesofthenumberofdailyflights)intominimumthresholdHHIvaluesimplies that markets with an HHI of less than 4050 are considered “competitive.” Values of HHI falling between 4050and8100(i.e. correspondingtotherangeofHHIvaluesinamarketinwhichtwofirmscollectivelyhold a 90 percent market share yet where no single firm holds 90 percent individually: 2∗452 =4050≤HHI < 8100=902) constitutes a “duopoly”, and a “monopoly” is defined as having an HHI of at least 8100. 33Independent of the usual caveats regarding the use of HHI as a measure of market competitiveness, we areunabletomeasureHHIpreciselyduetothefactthattheDB1Bdataonlyincludeinformationonforeign carriersthroughtheircode-sharingagreementswithU.S.reportingcarriers. Wemayconsequentlyunder-or 27

tration to affect pass-through rates by extending our basic empirical specification with an interaction of unit taxes and a cubic polynomial in HHI (pre- and post-). We depict the resulting partial effect estimates evaluated over the distribution of HHI deciles in Figure 4. There are a number of notable observations to make here. First, we find that we cannot reject complete pass-through at all levels of market concentration in the pre-FFAR period. Logically, if consumers do not react to changes in unit taxes due to their inattention, then market concentration is irrelevant to pass-through. Second, at higher levels of competition (lowerlevelsofHHI),thepost-FFARinteractionwithHHIshowsnearcompletepass-through of unit taxes, consistent with standard theoretical predictions with respect to marginal cost pricing in competitive markets. Following the adoption of tax-inclusive pricing, however, pass-through rates for unit taxes are shown to drop most sharply in more highly concentrated (i.e. “duopoly”) markets, consistent with a combination of substantial de-biasing and standard tax incidence results under imperfect competition and linear demand. In these less competitive markets, pass-through is strictly less than one in the post-FFAR period, and even negative over part of the range (albeit not statistically different from zero) before rising slightly in the top HHI decile. One possibility in this context is that tax salience is lower and remains lower—despite the implementation of tax-inclusive pricing—in markets where fare comparisons are largely impossible due to the presence of a single dominant carrier in the market, thereby offsetting otherwise lower pass-through rates due to monopolists’ pricesetting behavior. As such, market concentration may play a dual role with respect to FFAR, in terms of both conventional cost pass-through effects as well as in terms of modulating intrinsic consumer attentiveness and tax saliency. Figure 5 provides comparable evidence of heterogeneous pass-through rates as a function of (standardized) capacity utilization across O&D airport pairs within city markets. As overstate the true degree of market concentration depending on the importance of direct competition from foreign carriers versus the treatment of code-share or alliance partners. Measured market concentration is predictably somewhat higher when we treat all members of the SkyTeam, Star, and OneWorld alliances as belonging to a single firm. We nevertheless obtain qualitatively similar results using a measure of HHI defined on the basis of airline alliances. See Brueckner (2003) for a discussion of airline competition with respect to alliances and code-sharing agreements. 28

airlines and airports bump into capacity constraints at high levels of capacity utilization (e.g. because of an inability to readily deploy larger aircraft types or acquire new landing slotsintheshortterm), weexpectthistobereflectedinalowerelasticityofsupplyandlower rates of ticket tax pass-through. Consistent with this conjecture, pass-through rates in both thepre-andpost-FFARperiodsareindeeddecreasingmodestlyincapacityutilization, albeit not significantly so statistically speaking. Moreover, the spread between pre- and post-FFAR pass-through rates at comparable rates of capacity utilization remains virtually unchanged over the capacity utilization distribution—despite estimation of complete interaction effects between unit taxes and a cubic polynomial in capacity utilization (pre- and post-). This suggestsarelativelyinsignificanteffectoftheelasticityofsupplyonpass-throughratesordebiasing, once other market characteristics—including market concentration—are accounted for among our general set of controls. 6 Conclusion We find that the switch from tax-exclusive to tax-inclusive pricing of airfares mandated by the DOT in 2012 had a significant impact on ticket tax incidence and consumer demand. Contrary to the standard presumptions of well-informed rational consumer behavior, this confirms that tax salience plays a prominent role in affecting market outcomes when consumers suffer from limited attention, even in cases involving relatively large purchases and high effective commodity tax rates. The implementation of FFAR is associated with a significant decline in unit tax pass-through from near-complete pass-through under the previous tax-exclusive pricing regime to a rate of roughly 25 cents on the dollar in the post-FFAR period—comparable to the rate of pass-through on the set of non-tax charges which were always subject to disclosure in advertised fares. Moreover, the decrease in pass-through is relatively larger in less competitive markets, consistent with the basic textbook theory of tax incidence under imperfect competition. 29

Accountingfortheseendogenouspricingresponses—anovelfeatureofourquasi-experimental framework relative to the prior experimental literature on tax salience—we also show that a $5increaseinunittaxes(approximatelyequaltotheaveragestandarddeviationofunittaxes within O&D city market) is associated with a 4.3% reduction in itinerary-level passenger volume along higher-taxed routes. In sharp contrast to evidence from the pre-FFAR period, consumers in the post-FFAR period are thus equally sensitive to tax-driven changes in total fares as they are to changes in total fares resulting from changes in underlying base fares. Given the within-market nature of our identification strategy, we attribute this reduction in demand to cross-itinerary substitution as consumers sought out lower-taxed routes. The combined impact of reduced ticket tax pass-through and reduced passenger demand (in relation to the portion of the tax still born by consumers) together imply that a $5 increase in unit taxes is furthermore associated with a 2.4% reduction in airline ticket revenue. While our within market-by-quarter identification strategy and data limitations do not allow us to calculate the impact of FFAR on aggregate ticket revenues—let alone airline profits—these results point to a substantial transfer of surplus from airlines to consumers whose precise magnitude is subject to the aforementioned caveats about possible compensating adjustments in reliance on add-on fees. The airline industry’s persistent and ongoing attempts to reverse FFAR serve as prima facie evidence of its negative effects on producer welfare due to increased tax incidence on airlines, as well as possible reductions in aggregate demand due to the perception of higher prices. These findings emphasize the profound influence which disclosure rules may have in light of the prevalence of cognitive biases. This represents a potentially-fruitful avenue for promoting consumer welfare through regulation and tax policy design. However, this should be tempered by the possibility of fostering unintended consequences. Consideration of possible such consequences in the context of FFAR—such as through the increased use of add-on fees as a source of revenues or through extensive-margin itinerary entry and exit supply decisions—is left for future work. 30

References Agarwal, Sumit, Souphala Chomsisengphet, Neale Mahoney, and Johannes Stroebel. 2014. “A Simple Framework for Estimating Consumer Benefits from Regulating Hidden Fees.” Journal of Legal Studies, 43(S2): S239–S252. Agarwal, Sumit, Souphala Chomsisengphet, Neale Mahoney, and Johannes Stroebel.2015.“RegulatingConsumerFinancialProducts: EvidencefromCreditCards.” The Quarterly Journal of Economics, 130(1): 111–164. Anderson, Simon P., Andr´e de Palma, and Brent Kreider. 2001. “Tax Incidence in Differentiated Product Oligopoly.” Journal of Public Economics, 81: 173–192. Barr, Michael S, Sendhil Mullainathan, and Eldar Shafir. 2009. “The Case for Behaviorally Informed Regulation.” In New Perspectives on Regulation. , ed. D. Moss and J. Cisternino, 25–61. Cambridge, MA:The Tobin Project. Berry, Steven, and Panle Jia. 2010. “Tracing the Woes: An Empirical Analysis of the Airline Industry.” American Economic Journal: Microeconomics, 2(3): 1–43. Besley, Timothy J., and Harvey S. Rosen.1999.“SalesTaxesandPrices: AnEmpirical Analysis.” National Tax Journal, 52(2): 157–178. Blake, Tom, Sarah Moshary, Kane Sweeney, and Steve Tadelis.2017.“PriceSalience and Product Choice.” Borenstein, Severin, and Nancy L. Rose. 1994. “Competition and Price Dispersion in the {U.S.} Airline Industry.” Journal of Political Economy, 102(4): 653–683. Bradley, Sebastien J. 2017. “Inattention to Deferred Increases in Tax Bases: How Michigan Homebuyers are Paying for Assessment Limits.” Review of Economics and Statistics, 99(1): 53–66. Brueckner, Jan K. 2003. “International Airfares in the Age of Alliances: The Effects of Codesharing and Antitrust Immunity.” Review of Economics and Statistics, 85(1): 105– 118. Brueckner, Jan K., Darin N. Lee, Pierre M. Picard, and Ethan Singer. 2013. “Product Unbundling in the Travel Industry: The Economics of Airline Bag Fees.” CESifo Working Paper, , (4397). Carbonnier, Cl´ement. 2013. “Pass-through of Per Unit and ad Valorem Consumption Taxes: Evidence from Alcoholic Beverages in France.” The B.E. Journal of Economic Analysis & Policy, 13(2): 837–863. Chetty, Raj. 2009. “The Simple Economics of Salience and Taxation.” NBER Working Paper, , (15246). Chetty, Raj, Adam Looney, and Kory Kroft. 2009. “Salience and Taxation: Theory and Evidence.” American Economic Review, 99(4): 1145–1177. 31

Chetty, Raj, and Emmanuel Saez. 2013. “Teaching the Tax Code: Earnings Responses to an Experiment with EITC Recipients.” American Economic Journal: Applied Economics, 5(1): 1–31. Conlon, Christopher T., and Nirupama S. Rao. 2015. “Wholesale Prices, Retail Prices and the Lumpy Pass-Through of Alcohol Taxes.” DellaVigna, Stefano. 2009. “Psychology and Economics: Evidence from the Field.” Journal of Economic Literature, 47(2): 315–372. Feldman, NaomiE., and BradleyJ. Ruffle.2015.“TheImpactofIncluding,Addingand Subtracting a Tax on Demand.” American Economic Journal: Economic Policy, 7(1): 95– 118. Feldman, Naomi E., Jacob Goldin, and Tatiana Homonoff.2018.“RaisingtheStakes: Experimental Evidence on the Endogeneity of Taxpayer Mistakes.” National Tax Journal, 71(2): 201–230. Feldman, Naomi E., Peter Katuˇsˇc´ak, and Laura Kawano. 2016. “Taxpayer Confusion: Evidence from the Child Tax Credit.” American Economic Review, 106(3): 807–835. Finkelstein, Amy. 2009. “EZ-Tax: Tax Salience and Tax Rates.” Quarterly Journal of Economics, 124(3): 969–1010. Gabaix, Xavier, and David Laibson. 2006. “Shrouded Attributes, Consumer Myopia, and Information Suppression in Competitive Markets.” Quarterly Journal of Economics, 121(2): 505–540. Gillen, David W., William G. Morrison, and Christopher Stewart. 2003. “Air Travel Demand Elasticities: Concepts, Issues and Measurement.” Department of Finance Canada, Ottawa, Canada http://www.fin.gc.ca/consultresp/airtravel/ airtravstdy_1-eng.asp. Goldin, Jacob, and Tatiana Homonoff. 2013. “Smoke Gets in Your Eyes: Cigarette Tax Salience and Regressivity.” American Economic Journal: Economic Policy, 5(1): 302–336. Hanna, Rema, Sendhil Mullainathan, and Joshua Schwartzstein. 2014. “Learning Through Noticing: Theory and Evidence from a Field Experiment *.” The Quarterly Journal of Economics, 129(3): 1311–1353. Hossain, Tanjim, and John Morgan. 2007. “...Plus Shipping and Handling: Revenue (Non) Equivalence in Field Experiments on eBay.” B.E. Journal of Economic Analysis and Policy: Advances in Economic Analysis and Policy, 6(2): 1–27. Huang, Edward, and Adib Kanafani. 2010. “Taxing for Takeoff: Estimating Airport Tax Incidence Through Natural Experiments.” InterVistas. 2007. “Estimating Air Travel Demand Elasticities.” International Air Transport Association, Montreal, Canada https://www.iata.org/whatwedo/Documents/ economics/Intervistas_Elasticity_Study_2007.pdf. 32

Karlsson, Joakim, Amedeo Odoni, and Shiro Yamanaka. 2004. “The Impact of Infrastructure-Related Taxes and Fees on Domestic Airline Fares in the {US}.” Journal of Air Transport Management, 10(4): 283–291. Keys, Benjamin J., and Jialan Wang. 2018. “Minimum Payments and Debt Paydown in Consumer Credit Cards.” Journal of Financial Economics, (forthcoming). Lederman, Mara. 2007. “Do Enhancements to Loyalty Programs Affect Demand? The ImpactofInternationalFrequentFlyerPartnershipsonDomesticAirlineDemand.”RAND Journal of Economics, 38(4): 1134–1158. Marion, Justin, and Erich Muehlegger. 2011. “Fuel Tax Incidence and Supply Conditions.” Journal of Public Economics, 95: 1202–1212. Poterba, James M. 1996. “Retail Price Reactions to Changes in State and Local Sales Taxes.” National Tax Journal, 49(2): 165–176. Reck, Daniel H. 2016. “Taxes and Mistakes: What’s in a Sufficient Statistic?” Weyl, E. Glen, and Michal Fabinger. 2013. “Pass-Through as an Economic Tool: Principles of Incidence under Imperfect Competition.” Journal of Political Economy, 121(3): 528–584. 33

Figure 1. Tax-Inclusive Versus Tax-Exclusive Fare Rankings: New York City (JFK) to Tel Aviv (TLV) JFK LX ZRH LX TLV|TLV BA LHR BA JFK $(4301+156.06) JFK DL TLV|TLV BA LHR KL JFK $(2372.5+140.96) JFK AF LHR BA TLV|TLV DL JFK $(2372.5+137.46) JFK LY LHR LY TLV|TLV LY LHR LY JFK $(869+203.36) JFK LY CDG LY TLV|TLV LY CDG LY JFK $(873+195.36) JFK AF CDG AF TLV|TLV AF CDG AF JFK $(889+164.16) JFK LY TLV|TLV LY CDG LY JFK $(873+157.56) JFK LY TLV|TLV LY LHR LY JFK $(873+153.66) JFK LY LHR LY TLV|TLV LY JFK $(873+153.46) JFK DL TLV|TLV AF CDG AF JFK $(889+126.46) JFK AF CDG AF TLV|TLV DL JFK $(885+127.46) JFK LY CDG LY TLV|TLV LY JFK $(869+143.46) JFK DL TLV|TLV DL AMS DL JFK $(901+108.16) JFK AZ FCO AZ TLV|TLV DL JFK $(885+112.66) JFK LY ZRH LY TLV|TLV LY JFK $(873+118.86) JFK BA LHR BA TLV|TLV BA LHR BA JFK $(795+187.36) JFK LY TLV|TLV LY JFK $(869+104.96) JFK DL TLV|TLV DL JFK $(885+88.96) )knaR( eraF latoT 02 51 01 5 0 0 5 10 15 20 Base Fare (Rank) Dollaramountsinparenthesesalongsideeachitineraryrepresentbasefares+unittaxes. Source: BasefareandtaxamountsaredrawnexclusivelyfromonlinefaresearchesperformedbetweenDecember30,2014and January25,2015(ITASoftware,non-DB1B). 34

Figure 2. Variation in Unit Taxes Across and Within Origin-Destination City Markets (2011Q4) CLE LHR EWR PVG DFW HKG EWR FRA MIA EZE LAX PVG SFO OMRNDL ICNEWR ICN LAX ICN SEA ICN MIA GND DCA TLV LAX MNL EWR POS MISAF AOU HAKG IAH ICN EWR SIN DTDWC AFR PAOS EWR PEK DTLWAX EP WNVRGRT HKG BUF AUA IAH GND SFO ICN EWR CDG SFO PVG JFK BCN EWR AMS MSP MBJ DFW CDGEWR POP IAD MUC SFO ZRH EWR LHR C E F M LL W T CB I L AO R S OO S HR J S S J D S U DS U T ASJ F T T SUTJ WT UJ L U A S S U D J L J CO U A U S C D SL S RX A C TE S F D L A S O T S S O L J T SP T P J A P A U RD JEU T E T E X U D K F K L K W G PG U VGU A G E UA W AEE ORO WW DBA RP C R OK R DA A S G G B PE G GG U U O W J S A U D U F A UA M F T RA F K A O A L W D E M B B W U Z A B O B E R D Z M A E I A M T DS E HD X L LL H OE W F E P E AC AM C M O WRA E WN R XX D L L E E U D W S T B L EP R C EMX M B X B T A F DA Z R C B H W A C L ZZ MA EB O E T B CXE L Z U X LV E U E CM XR L Z X MAM E W AE A O M N SO N D E YC S XXE H S ESX S I ARSM EC E U X A F X N M I PC C A C MLADA B TX I A A AS E SAA N O T B E A S D G U L U P K U L L DSN C C D B BB EI W A UFC L SMM J N W PB H T C NCNCDO NV E N SD N S MY N S A U O EWT DBQCA E F PASAANL SOSMTGBM L IUSA U R E U C RV A DDU OX F FA R T UEE G N R E DL K N L L W S C S D HABU O Y L M R MIKLS FCMXHG T V C M M N O T D M O W S N K CL B A W D L T B YL I MCNQC U RXSMSTCM MN R C U F MMO R G E L C E F I U PM F E C E E E EU C R C LC I CCU UM N T C CC C R E A D CE CXEC E X U CC A CC W CEE CC F X N XC W XUX G VCN A C U U L UUUNNX B U S UUC O XE UX X UU H N X C UUXUUUUU G A G UN V U I UN R NNN D NNNUX G EN G NNN O NNNNNN T N R N R P NN LW N A R CC AL BP U TM U B MCR S A I U V U ML R D Y L W P IND D MT PKDE U G A A S T FHDL PW IB I CE HU A LNR X D T Y WPN O NN T LW NALDN ADCN ANNA A NA Y UA ANEXAFANSIUAAS U U AAX A SS A C SWASAAAG S I SS A A S AAS A L GUUUS I E UAU T H GARAA W S UAA K U B F A R I O F O ND L AG F S L N W Y M U S E CP D B M E B D W PHV B S NIM A BJ CS IE A GX U R TT GA J LFLK N H L OML M F S MM AE M K OK A E B R B MB M M SO M X I B I XB B W M N NM D M DJ Y B D B B J B JJ KM B B L I BB J J J C J B K I O K JBNJJ KA O IN I J G IN G MNKA M ICTN MS LO PIK AK I C NICN D E DG W G R PUJ M M H I P A K C DIDOBBDB E TL LL CFDUWUE A ER P P AWLFN X IP M U U B P P P P A PUPPJP J D U M UUU T UUJUU L J C J L JJJJ JJ P OP B O O U P P O P F O R P PDO LPRM IAHE WEZRE EZE L IA A DH OBDAD X SFMS E OLT FFRC L L E SFLNLOW A HSDA Z OP X L L R E E LE LL LHL H L ZH L HHZH HHRR H ERE RRR RR R AU D S E N LH L R HR )sralloD( noitaiveD dradnatS 04 03 02 01 0 50 100 150 200 Average Unit Tax per Ticket (Dollars) (a) MIA GND EWR POS ORD LRM ORD BOG FLL MBJ BDL PUJ LAX PVG FLL SJU BWI NAS CHS CUN CVG CUNDCA CUNGSO CBUDNL KIN MCO KIN MKE MBJ ATL POP BDL POP D O C D RS A A IS O AF T D T F P R HO W L O E E E P D G H K W W PEB P S P I N EKGO V R R E V D K S M G E A IG B P S A S F W S O O E NF H S O F F I L RR O A K R F O M O A E E I D L H O T C E T S O XSD Z W A N T L W N FG F E I Y R P F X R T C L VR L O R DW V D R L V L T N L N A U G I GV C D H A I HR T G X C ERL L C K DL KT RWT N A E U G GAV I G B A I U X W SRAR A TR H L S A J H G E R T ULGAU U E T L W V B I SX I G I E L A C GOF C R T W J D L I N S C I GC L T F CA F M J R E E O K N E X J N WO I UB W XA F W M B E F R E G O PH K S R RA W C E R W D U S A DJ L S UT P M O X D S R T U A A R LA CS L T T SF A T L A M O X D JA L S Y J G W P T D X U C U B A E I S A O S S V M U N A Z L K NEM G D U J J G R A E E E T U UW D E U S N G D B X L S F S N J A LRZ U S B M E W F D JE E ED A T E U D Z K AB W F E W C T Z E H U W O XB B R A E R K B S S C Z A L B G S P F M E B N G T O A M E Z I D T A O Z A S L X E U R W T E S P F E D P E M U A D W Y W B I R M W D A M C V N S ZE R M G A H C I E C R F D B FL E A U S W T U E A Z G D O D N G M C F C L E D X A E R S G T U A G O E D DZ Z S F P B T W E L AM M U X E R B M F G F W A T E W L A W A I D WE O HY B E Z F AX L M I M R XL M M Q E R X A M M CS A E P E M A C X I S H MAD U A E E X E H L X O A D E OI B L X X G A A E W X A A C L S X E O U X S X H U A L R Y N C E G P J A T C A A E R W U G U C P D S M U U S Y Z C E N N V T C A R X H S E E E A U L C B Y G D N U O X A C P F M O N D U M E N U R N C U C E AL N L C D E C W O A A U N A A C U X K W U CM AO R S R N L X U S N U N A C D M R G B E D M E N A A C U W S W S S M AN C E U C C M U N L A O P L R A X U S A O A A V M N EL A S S N P D N M D S U X G S H B AX B D D XL A S A S C F F A U E O L N JN C B E C C M B F S O P M W L D HX F P C U UN N A O A U O Q B EB U R C E R P O D I S L LSA C C N TA M M D N P J J D M A U N H M E B A C H F A K U U L T UF A K U A A B U N X CU W B U U N R C B A N L L N N W JI U J E A JD L C F A N N J C T H I X A M A MA L C W P A T D EU U L U R U N B A S H H B K U U E A C NL RK N N A W U A R J I B N X J X N I A M I A C S A N I P NP D E I A M U B L C U L U D A U W L HR S H P H J B S P P M N M U J J C I N R D Y I R D R D L N I T A C H C K T P U A HU L OX D C L M E U P B I S N C C D R T H R N U N M N U U C H M N U U F R D A N N A J F U E M S K A W D N N B L N S S A X N L C T E S A M C W F A S H LC AL X W I S B C SI E R E U B U C M MJ C M E U W L U P N S U L B A S U N E H I F R H M M N J T P L N X R M R C H K PS K M C M E L E U R Y I G SU N D B E X S J P K T J B NC S P P U C D I L W O A N U D M J L C BC I PJ B M C E W B A U K F U O A B J M E I L NI K J N A S J L P NW B I T N U L A L J R L H H J S B K P R R U I O N F P POP DCA POS MIA M LR C M O POP )tnecreP( noitaiveD dradnatS 6 4 2 0 OMA GUA EWR POS MIA GND BOS GND MSP GUA ORD GND SEA LHR ORD LRM ORD BOG STL NAS FLL MBJ LAXS FGON GDND BDL PUJ LAX PVG FLL SJU CVG T U C S U C N C DHU CSN A C CUUNNGSO CBUDNL KIN BWI NAS MCO KIN 0 D O CD RS A A IS O AF T D T F P R HO W L O E E E P D G H K W W PEB P S P I N EKGO IV R R E V A D K S D M G E A IG H I B P S A S A F W E S O O E NF H S O F D N N F I L R S RD O A K R F O M R OA E E I IE D L G H O D C C T C E T S O T X L SD Z W A N A T F A R L WN N FG AF E I Y R P F X R T C LW V RS U L H OX RG DW P DV D R L VL T N S L S N F K B AU H G IGV CRC H T H A A O F I HR T GO G S X C E X A L URL K L K N D O L KT F S RVWT N G A E G U G GV N T I G O G B G AI U X W RS H RALR A R R HR L I S U V A J S H K G E R T U G T GAUU U E T I L G W B I N SX I G I E L A C GOSF C R T W J D L I N S FC I GC L T F SCA F MO J R E E O K N D E D EX J N WO I I UB W XA A F W FM B D A E E F R E G O PH H K W SR R N A W UC I E C R W D G N U S A D D S J LS UT P A M BO X D L S R A S U D I T U A A R LA C E S L L T C T M S T S F A TB L M E L AA DMC O X D N JA L S Y J G N S G WP F SA T D O G A X U E SC U B A E I S C L A O S F O S G U V M SG U N A S UA N Z L K N A E EM G V O D U JJ G A R U A U U E E AS E BT B U X DU Z W D B G E P U S N BG D B AX A S L F AZ Z S E S T C F S P N H O J A C L A RZ U C E S D M E N S B F B M EW W F BD JE D N A XE M E L G M A T O E O C F W U D Z E K A Z B 1 W F E X BT W T S O Z B C G EA H X E U W O X I Z B B B R S E O B D Z I R K Uv G B E S S S C O C Z 0 A E B Z A C R L B G P A U F M J ED B D N G EL Z G T O A M A D E Z I D U A e O A Z A C G S EL X E U R W E S P F E E E D P A E M U A D W Y W S B I Z R Z Mr S W D A M C B V N AS E Z E E B R M F GA H C I E C aZ R F D B NFL E A U S WO O S TU E A Z E G D O D N G M C F C F L E E S D X P A E RP S G T U A G O g E O DD DZ ZZ S F I M PB B T T W S E L T AM M U X E ER B M EF G F W A T E O W Y L A S A S W A I M D WE OH N e Y B E Z F AX S A L P L M I M R X B L M M Q E ER D X A L P M M CS A CE M O P E X M A F C X I S TH MAD U A E EXD B E A H E L S X W O Y A DE U OI B L X X G A N A E G W L MX X A A C L S S XE A I O U B X S X P NA H U A SL R Y L N C M E G P J T A D T C A A P A C E N R n W U G U CS P DY S M U U U S Y Z C A D M E A N N M V T C A RX S S E E E A A SH U E L U C B I Y E G D i N U A E A X A P C P FM M O O N N D A U M X t L N E X N H H R N C U C E A E A U L N L C D E C W S N X A A U M N NA A M U X X K W U B CM A O F O TA C R S S N C LA X E N S NN O U N A P C R S D M RU G CB FE M S H X E A A A A C TA U DW O N S U W S SN M aS ANC S E U C Y C U U N LA A O P N A M O L R A X U S P A D O A A N V M N S U L S A S A S NR P U D E N M x D SF U X G S A H BA A B A I D D X P L U A S N X AF W J S C F F A S UU E O L N I JN C B E C C U M S C B F S PSD O P M W L D HX N FS P C U UN B N A O A U OJ T Q S T B E I BU R C N E R P E O D T I A L A L U SA C L C FN TA A M M DN P JJ D M A U C N HM E A B A C S H F U K RF F U U L T L P B UFA K U AA B U DN U S X CUW B U L U R C B A A R N L L AN N W U N N JI U J T E A I F N JD L C F A N N J C T H I X TA A M A A L J CA A W P W A T D EU U S L C U R U N B A S H C H B K U S U A C N P E L RK N MN RA W U A R J I P U B N J X N U I A M A I A C S A N HI P NOP D N E I A M J U B L C U L U P D A U W L S HSA OR S H P H J B S J P P M U N M U J J YU C I N S A R MDR Y I R D R D L N I ( JT A C H C K T RA P U A Y X HUF M C LO 2 X D C L M E M U P P P B I S R N C C D R M T H I C S R P N U N M H N U U C H M S N U U M F R Y U P U B DA N 0 N A X J F U P E M S K A W D N N G B L N B U J J N e S S A P X N LC T E CS AM CP W FP J S A J SO H LC AL XW L UI U O S B C SI E r RE S E U B U C M M J J J C M E U W M L U P N S U c L S B CAU N E P H I F R H MB M N P J T U L N X H R M J J R C H e K PS NK C M A XM E L E U R Y I X U S N D B M E X LS J n P KNT J B H C K K P R P U C D I L W R A I N E U D N D M J L C B t C I PJ U B M C E W B A ) U K F U O A B J P ME I L NI K J N A S J LU P NWB I T N U J L A L JR L H H J S B K P R R U I O A N F P T P L O P P OPBDL D P C O A P POS 3 M 0 IA M LR C M O POP (b) noitaiveD dradnatS 8 6 4 2 0 0 10 20 30 Average Unit Tax ETR Caribbean Central America South America Western Europe Middle East South Asia Southeast Asia East Asia Oceania Airportlabelsrefertothelargestsingleorigin-destinationairport-pairbypassengervolumewithineachorigin-destinationcity market. Effectivetaxrates(ETR)arecomputedastheratioofunittaxespaidtototalfares. Eachpointinthefigureis computedfromatleasttwouniquerouteswithdifferentunittaxes. AveragetaxamountsandETRsareallmeasuredona passenger-weightedbasiswithinorigin-destinationcitymarketpairs. Source: DB1BandRDC. 35

Figure 3. Four-Quarter Changes in High-Tax Route Volume Shares )pp( erahS xaT−hgiH ni egnahC Y/Y 6 4 2 0 2− 4− 2011q1 2011q4 2012q3 2013q2 2014q1 Quarter Routesarecategorizedashigh-andlow-taxrelativetothetopandbottomquartiletaxamountswithinorigin-destinationcity marketpair,respectively,andarebasedonabalancedpanelofever-availablerouteofferings. Onlyorigin-destinationcity marketsfeaturingatleastonehigh-taxandonelow-taxrouteareincluded. Source: DB1BandRDC. 36

Figure 4 stceffE lanigraM egarevA 5.1 1 5. 0 5.− 1− Unit Tax Pass−Through by HHI Decile .28 .33 .37 .43 .5 .61 .71 .82 .91 Market Concentration (HHI) Pre−FFAR Post−FFAR Averagemarginaleffectestimatesarebasedonamodelfittedwithpre-andpost-FFARunittaxesandairportcharges interactedwithacubicpolynomialinHHI.Whiskerbarsrepresent90percentconfidenceintervals. Source: DB1BandRDC. 37

Figure 5 stceffE lanigraM egarevA 5.1 1 5. 0 5.− Unit Tax Pass−Through by Capacity Utilization Decile −1.08 −.66 −.31 −.01 .19 .35 .53 .72 .99 Standardized Capacity Utilization Pre−FFAR Post−FFAR Averagemarginaleffectestimatesarebasedonamodelfittedwithpre-andpost-FFARunittaxesandairportcharges interactedwithacubicpolynomialinHHI.Whiskerbarsrepresent90percentconfidenceintervals. Source: DB1BandRDC. 38

Table 1. Sample Tax and Fare Decomposition: New York City (JFK) to Tel Aviv (TLV) JFK-TLV JFK-FCO-TLV JFK-CDG-TLV TLV-JFK TLV-JFK TLV-CDG-JFK (1) (2) (3) Base Fare $885.00 $885.00 $873.00 Fare 279.00 279.00 873.00 (of which non-tax charges)a 16.86 22.86 39.67 Fuel surcharge (YQ or YR) 606.00 606.00 0.00 Total Taxes and Fees $88.96 $112.26 $195.36 US Intl Departure and Arrival Tax (US) 35.00 35.00 35.00 US Sep. 11 Security Fee (AY) 5.60 5.60 5.60 US Passenger Facility Charge (XF) 4.50 4.50 4.50 USDA APHIS Fee (XA) 5.00 5.00 5.00 US Immigration Fee (XY) 7.00 7.00 7.00 US Customs Fee (YC) 5.50 5.50 5.50 Israel Departure Tax (IL) 26.36 26.36 26.36 Israeli Security and Insurance Surcharge (AP)b 16.00 Italy Passenger Service Charge Departure (MJ) 1.10 Italy Council City Tax (HB) 9.10 Italy Security Charge (VT) 3.10 Italy Embarkation (IT) 10.00 French Intl Passenger Service Charge (QX) 52.20 French Airport Tax (FR) 38.20 TOTAL FARE $ 973.96 $ 997.26 $ 1,068.36 a Aircraft-specificnon-taxchargesarebasedon2014Q2levels(incurrentU.S.dollars)andincludevariousairportfees, includingtake-offandlandingcharges;parkingandterminalfees;noiseandenvironmentalcharges;navigationcharges;etc. Chargesareallocatedonaper-passengerbasisassuming100percentseatingcapacityutilization. b APappliesonlytoflightsoperatedbytheIsraelinationalairline,ELAL. Source: ITASoftwareandRDC. 39

Table 2. Quarterly Ticket and Itinerary Characteristics: 2009Q4-2014Q2 Mean Median Std. Dev. Average Ticket Characteristics ($00s): TotalFare 7.50 6.07 3.67 cit BaseFare 6.19 4.91 3.48 cit Itinerary Characteristics: UnitTaxes ($00s) 1.08 0.98 0.45 cit NonTaxCharges ($00s) 0.23 0.18 0.19 cit Passengers 44.03 16.00 95.63 cit Distance 5.10 3.74 3.28 i Layovers 0.79 0.00 0.96 i HHI 0.56 0.50 0.24 jt Load 85.84 87.33 7.09 cit LnOriginVolume 11.79 12.17 1.42 cjOt Itineraries 45.27 35.00 42.13 jt Itineraries 29.94 21.00 30.54 c−jt Observations (Itinerary-Quarters) 24,712 Observationsincludeonlyround-tripflightswithaU.S.originandexcludeallbusiness,first-class,andawardtravel. Data from2012Q1areomitted. Itinerary-levelstatistics(excludingpassengervolume)arepresentedonapassenger-weightedbasis. Distanceismeasuredinthousandsofmiles. SeeTableA3forvariabledefinitions. Source: DB1BandRDC. 40

Table 3. Ticket Tax Pass-Through Y = TotalFare (1) (2) (3) cit UnitTaxes 0.768*** 0.992*** 0.958*** cit (0.100) (0.312) (0.307) UnitTaxes ×I[Qtr > 2012Q1] 0.700*** -0.743* -0.711* cit t (0.134) (0.418) (0.413) NonTaxCharges - - 0.351*** cit - - (0.127) NonTaxCharges ×I[Qtr > 2012Q1] - - 0.022 cit t - - (0.186) Controls: Layovers x x x i Distance (cubic) x x x i HHI (cubic) x x x jt LnOriginVolume (cubic) x x x cjOkt Load (cubic) x x x cit Fixed Effects: Carrier × Qtr (η ) x x x ct O&D City × Qtr (ν ) x x kt Observations 25,175 24,712 24,712 R-squared 0.854 0.964 0.964 ***p<0.01,**p<0.05,and*p<0.1. Standarderrorsclusteredbyorigin-destinationairport-pairappearinparentheses. Observationsareweightedbypassengervolume. Source: DB1BandRDC. 41

Table 4. Itinerary-Level Passenger Volume and Tax-Exclusive Ticket Revenue Y = ln(Passengers) ln(Revenue) cit cit (1) (2) (3) (a) BaseFare -0.008 -0.438** cit (0.011) (0.180) - (b) BaseFare ×I[Qtr > 2012Q1] -0.067*** -0.113 cit t (0.015) (0.202) - (c) UnitTaxes 0.289** 0.304 0.195 cit (0.121) (0.218) (0.119) (d) UnitTaxes ×I[Qtr > 2012Q1] -0.791*** -1.179*** -0.682*** cit t (0.198) (0.350) (0.195) (e) NonTaxCharges -0.017 -0.303** -0.169* cit (0.092) (0.134) (0.087) (f) NonTaxCharges ×I[Qtr > 2012Q1] -0.154 -0.163 -0.135 cit t (0.128) (0.161) (0.122) Elasticity of Demand w.r.t.: Base fares: Pre-FFAR -3.23** (1.31) Post-FFAR -4.05*** (0.88) Unit taxes: Pre-FFAR 2.31 (1.67) Post-FFAR -6.35*** (1.48) Non-tax charges: Pre-FFAR -2.25** (0.98) Post-FFAR -3.44*** (0.86) Controls: Layovers x x x i Distance (cubic) x x x i HHI (cubic) x x x jt LnOriginVolume (cubic) x x x cjOt Load (cubic) x x x cit Fixed Effects: Carrier × Qtr (η ) x x x ct O&D City × Qtr (ν ) x x x kt Observations 24,712 24,712 24,712 R-squared 0.836 0.780 0.866 Kleibergen-Paap F-Stat 15.27 ***p<0.01,**p<0.05,and*p<0.1. Standarderrorsclusteredbyorigin-destinationairport-pairappearinparentheses. Observationsareweightedbypassengervolume. p-valuesoftestsofequalityofestimatedcoefficientsfromcolumn(2): (a)=(c): 0.000;(a)=(e): 0.311;(c)=(e): 0.005; (a)+(b)=(c)+(d)=(e)+(f): 0.126. Source: DBIBandRDC. 42

A Appendix - For Online Publication A.1 Data A.1.1 Data Construction The data for our analysis consist of two main components: (1) the restricted-use (international) portion of the DOT’s Origin and Destination Survey (DB1B) data, featuring ticketlevel itinerary characteristics, total fares paid, and passenger counts for a 10 percent sample of all tickets redeemed by U.S. reporting carriers on a quarterly basis, and (2) detailed historical data on tax and non-tax airport charges from RDC Aviation, measured on an airport-route-aircraft-specific basis. The combined data span the period 2009Q4 (the earliest quarter of broad data availability from RDC Aviation) through 2014Q2. We define international itineraries as those that originate in the U.S. en route to a foreign destination (“outbound” itineraries) or the reverse (“inbound” itineraries). In line with other applications of the DB1B data (see, e.g., Brueckner (2003) for a careful description), we apply multiple sample restrictions to ensure a relatively homogeneous product sample. We exclude at the outset all one-way or multi-leg itineraries (i.e. itineraries involving a sequence of destinations which the DOT distinguishes from layovers using flags for directional breaks); itineraries involving at least one first-class or business-class segment (accounting for at least 10 percent of miles flown); group tickets featuring 9 or more passengers, and all tickets flagged by the DOT as involving implausibly high prices per mile flown. This latter restriction appears targeted at fares in excess of $0.90 per mile (not inflation adjusted), albeit with unspecified exceptions, and covers approximately 1-1.5% of ticketed itineraries. We extend this restriction to exclude all fares—without exception—that exceed $1 per mile (0.4% of the remaining sample). At the opposite end of the distribution, we exclude all tickets with fares of less than $0.02 per mile or base fares of less than $0.01 per mile (or $50) as likely award travel.34 Depending on quarter, this exclusion eliminates 34Carriersarenotrequiredtodistinguishawardversusnon-awardticketsforpurposesoftheDB1B.Dollar- 43

approximately 7-10% of ticketed itineraries from our sample. Among the remaining set of round-trip coach-class non-award tickets, we further exclude ticketed itineraries that fall outside of the 5th-95th percentile of the fare distribution within itinerary-quarter to limit the potential influence of promotional offers or last-minute purchases. Multiple steps are required to match the resulting DB1B sample with data on airportlevel tax and non-tax charges from RDC Aviation in order to decompose total fare amounts in the DB1B into base fares, ticket taxes, and non-tax charges. Airport-specific tax amounts are commonly dependent on flight distance or route (with differing levels of tax for transatlantic versus intra-EU versus domestic flight segments, for instance), or whether passengers are exiting the airport versus catching a connecting flight. Taxes may also differ on rare occasion according to operating airline. Airport-specific non-tax charges—such as runway fees, emissions and noise charges, etc.—likewise vary along multiple dimensions, but—with the exception of navigation and terminal (international versus domestic) charges—do not depend on route. Instead, non-tax charges vary primarily according to aircraft maximum take-off weight (MTOW), seating capacity and class configuration, and engine type of arriving and departing aircraft. Where possible (i.e. for all flight segments involving a U.S. airport), we utilize data from the DOT’s Form 41 T-100 Segment database (domestic and international) to identify model types, seating capacity, and load factors for aircraft operated by U.S. and foreign air carriers along specific flight segments in a given quarter. A passenger traveling on the outbound leg of a round-trip flight Philadelphia to Paris on Delta Airlines (i.e. PHL DL CDG) in 2014Q2 would be identified, for example, as having flown on a 171-seat Boeing 757-200. In the case where airlines use multiple aircraft types in a given quarter to service the same segment, we use information for the most-heavily used valuedtotalfarethresholdsarecommonlyusedelsewhereintheliteraturetomakethisdistinction,butthese ignore the fact that consumers remain responsible for paying certain ticket taxes on award travel. Ideally, we wouldpreferto exploitexogenously-flaggedawardtickets forpurposesofconducting sensitivityanalyses, but we remain concerned that many low dollars-per-mile fares may represent erroneous entries, and we do not have the ability—outside of U.S. ticket taxes—to identify which taxes apply to award travel and which do not. A large number of exact $0 fares is indicative of misreporting as all international award tickets remainsubjecttonon-zerotaxes. Thedistributionoftotalfarespermilereachesalocalnear-zerominimum density at $0.02 per mile, hence our choice of threshold. 44

aircraft based on passenger volume. The combined quarterly airport, route, airline, aircraft model, and seating capacity information is then fed into RDC Aviation’s query system to extract the per-passenger tax and non-tax charges applied to the departing aircraft at PHL and the arriving aircraft at CDG. The corresponding inbound flight segment—which could in principle involve a different aircraft—triggers additional departure charges at CDG and arrival charges at PHL. In cases where we cannot use the T-100 database to match flight segments to aircraft, such as for flight segments between foreign airports, we scraped this information through FlightAware’s Flight Finder API in November 2016. This procedure yields aircraft tail numbers and model types for recent and upcoming flights, which can in turn be matched to the DOT’s Form 41, Schedule B-43 annual aircraft inventory for aircraft owned by U.S. carriers to determine the relevant seating capacity of the aircraft. We assume for this purpose that airlines’ selection of aircraft to service particular flight segments is largely fixed over time. Absent valid tail number information (U.S. carriers only), we defer to RDC Aviation’s airline-specific fleet information to determine seating capacity. Where neither the T-100 database nor FlightAware’s Flight Finder yield any specific matching aircraft, we use information either from adjacent quarters or for the most common model of aircraft utilized over the same or similar routes in the same quarter.35 Each round-trip itinerary in the DB1B requires data for at least two sets of arriving and departing charges, one at each endpoint of the passenger’s journey. The addition of a single layover in either direction adds two additional flight segments and thus two further sets of charges. In a typical case, a single query for airport charges for a particular airport-route-aircraft-airline combination yields multiple applicable charges and corresponding charge amounts (converted to nominal U.S. dollars at prevailing quarter-average bilateral 35Astheforegoingdescriptionimplies,ourcalculationofnon-taxchargesisinevitablysubjecttomeasurement error from multiple sources. To the extent that such measurement error is uncorrelated with the level of true charge amounts, the resulting attenuation bias may be offset by our allocation of non-tax charges to ticket prices assuming 100% capacity utilization, which will tend to understate passengers’ true share of total non-tax charges when flights operate at less than full capacity. 45

exchange rates). Charge amounts are classified by RDC Aviation as being either “permovement” or “per-passenger” and fall broadly into ten categories: air navigation, aircraft security, government, infrastructure, noise, parking, passenger, passengersecurity, runway, or terminal charges. Charges are further distinguished by their applicability to arriving versus departing aircraft and terminal versus connecting passengers. Altogether, we utilize data for nearly 320000 charge amounts which we have to classify as passenger-specific tax amounts or non-tax charges before stringing these together for a sequence of flight segments into total itinerary-level taxes and charges for itineraries appearing in the DB1B. Regrettably, the RDC data are not reliably coded in this manner, much less linked to specific international airport tax codes as defined by the International Air Transport Association (IATA). In order to make use of the RDC data, we therefore construct a concordance of ticket and airport tax codes along with their corresponding names for a representative sample of flight segments by pulling these details from fare searches using ITA Software, which we link based on description and dollar amount to named per-passenger charges in the data from RDC Aviation. Where applicable, we also consult the underlying government source documents to confirm our name- and amount-matching procedure. As a representative example, Table A1 lists the set of taxes levied by the French airport authority for a round-trip flight PHL DL CDGusingtheprecisenamesandIATAtaxcodesgivenbyITASoftware, alongsidethesetof matching charge elements from RDC which forms the basis for our ticket tax concordance. We are thereby able to pass the list of charges from the RDC database (e.g., 16 charge elements for transatlantic flights arriving in and departing from CDG as of 2014Q2) through our ticket-tax concordance to come up with a complete historical record of tax amounts by IATA tax code, and we confirm that the remaining charges in the RDC database for which we do not have a matching tax represent per-movement non-tax charges (such as parking and landing fees, etc.). Table A2 provides an illustration of the latter types of fees, as applied to the same PHL DL CDG flight. Otherwise, we treat charges that are levied on a 46

per-passenger basis without matching our tax concordance as miscellaneous ticket taxes.36 Out of the 253 unique tax codes represented in our original pull of over 30000 scraped fare searches, we are thus able to use our ticket tax concordance to construct complete historical records for 94 foreign tax codes from the RDC data with a high degree of precision.37 We are furthermore able to construct complete histories of the six applicable U.S. ticket taxes (International Departure and Arrival Taxes (US), September 11th Security Fees (AY), Passenger Facility Charges (XF), APHIS Fees (XA), Immigration Fees (XY), and Customs Fees (YC)) from various sources, including the Federal Aviation Administration, Department of Homeland Security, Department of Agriculture, and Customs and Border Patrol.38 This allows us to directly identify applicable tax amounts for all itineraries involving these 94 foreign and 6 domestic ticket tax codes based on our scraped fare search results (subject to caveats about variation in amounts owed for specific airport or ticket taxes based on route or airline), whereas without known tax amounts by tax code, we instead compute total tax amounts from miscellaneous airport-level taxes by adding these up across itinerary flight segments. We apply a similar procedure to sum non-tax charges (by airport-route-aircraft-airline) into a single itinerary-specific total amount and allocate these to passenger fares assuming 100% capacity utilization. Duetothecomplexityoftheaforementionedprocedureforconstructingahistoricalrecord of itinerary-specific ticket tax amounts, this inevitably requires us to impose one further important restriction on our sample. Namely, we limit our ticket tax queries to the set of routesflownbynofewerthan9passengersinasinglequarterovertheperiod2012Q4-2013Q3 36We strive to avoid failed matches due to minor string mismatches in the naming of charges over time. Nevertheless, some such mismatches are largely inevitable. Failed matches can also reflect more substantive changes in applicable taxes over time, such that current IATA tax codes and descriptions may not capture ticket taxes that have been replaced or eliminated. 37Oftheremainingtickettaxcodes, 30representpresumptiveadvaloremtaxes—typicallyonlyapplicable to inbound flights—which were already required to be included in advertised fares prior to FFAR and thus do not figure in the calculation of applicable unit taxes. We distinguish ad valorem taxes from unit taxes by running separate regressions of each tax code on scraped base fares plus a scrape date indicator. All taxes featuring a statistically significant effect of base fares in excess of 0.5 percent and a regression R-squared of at least 0.5 are treated as as ad valorem. 38We are also grateful to Joakim Karlsson and the MIT Airline Ticket Tax Project for sharing data on airport-specific passenger facility charges at an earlier stage in this project. 47

(or 36 passengers over the full four quarters) in the DB1B (i.e. implying an average of at least 1 passenger per day in the full 100% ticket sample). This is intended to mitigate undue influence from large idiosyncratic changes in passenger volume (measured in logs) along very low-volume itineraries. Likewise, we exclude all observations from itineraries involving relatively low-volume ticketing carriers (i.e. below the 1st percentile of the distribution of carrier volume in the year 2011) out of concern that many of these carriers—including many foreign and charter operators—may not have been subject to FFAR. As described in the next section, we nevertheless compute certain key control variables prior to applying these last sample restrictions using data from the full DB1B without regard to our ability to calculate matching ticket tax information. A.1.2 Variable Definitions BriefdescriptionsofthevariablesusedinouranalysesarepresentedinTableA3. HHI(HHI) and the number of total or competing itineraries within O&D airport pair (Itineraries) are each calculated using the full DB1B sample, whereas the remaining variables are all calculated exclusively within our sample of tax- and charge-matched itineraries. A.2 Extended Results Tables A4 and A5 display the full set of coefficient estimates from estimation of our main empirical specifications, and replicate the results shown in Tables 3 and 4, respectively. First stage IV results corresponding to the specification shown in column 2 of Tables 4 and A5 are reported in Table A6. Figure A1 plots quarterly pass-through rates from equation (12) estimated as an event study, whereby quarterly unit tax and non-tax charge pass-through rates are estimated via incorporation of a full set of quarterly interaction terms. Data from 2012Q1 are omitted for consistency with the main specifications in the body of the paper. The main message of this figure may be that our quarterly estimates suffer from low power and are subject to 48

large standard errors. Only in rare instances can we reject pass-through rates of either 1 or 0, especially for unit taxes. Pass-through rates for non-tax charges are generally more preciselyestimated. Thispatternreflectsournarrowidentificationstrategy,whichisbasedon relatively modest within-market × quarter variation in unit taxes and the crude mapping of FFARtopassengers’quarteroftravel. Thoughsmallerinmagnitude,non-taxchargesexhibit greater within-market variation due to the fact that they are aircraft- and route-specific, whereas unit taxes are generally only route-specific (and with very rare exceptions, operating carrier). Statistical imprecision notwithstanding, the remaining key message from Figure A1 is that our point estimates of unit tax pass-through are close to unity for most quarters prior to 2012, and these lie above the pass-through rate for non-tax charges for the corresponding period in all but one quarter pre-FFAR. Post-FFAR, unit tax pass-through rates appear to shift downward, as evidenced by point estimates near or below 0.5 in virtually every quarter. Moreover, these estimates are in many cases very similar to the point estimates for nontax charges, especially over the period 2012Q4-2013Q4, 2012Q4 being the first quarter that FFAR would have affected virtually all passengers identified in the DB1B sample of redeemed flight coupons. Subsequent divergence between pass-through rates for non-tax charges and unit taxes—with rates on the former exceeding the latter in the last 2 or 3 quarters of our sample period—are harder to explain except as a possible short-run over-correction. A.3 Add-On Fees Detailed data on airline charges and fees are relatively scarce. Nevertheless, U.S. air carriers’ quarterly financial statements provide a rough breakdown of sources of revenue from domestic and international operations. Of particular relevance to understanding the proliferation of carrier-imposed fees are baggage fees, cancellation and change-of-ticket fees, ticketing and check-infees, feesforseatassignmentsandupgrades, andchargesforin-flightfoodandbeverages, entertainment, Wi-Fi, pillows and blankets, etc. Only the first two longest-established of these fees are reported separately on Form 41, Schedule P-1.2. More broadly, Schedule 49

P-1.2 classifies revenues into: transport revenues from scheduled passengers, mail, freight, baggage fees, revenue from charter operations, change/cancellation fees, miscellaneous operating revenues, and transport-related revenues. Fees for seat assignments and upgrades or ticketing fees are included in general transport revenues along with base fares, while inflight sales are included in transport-related revenues, which also incorporate revenue from code-share operations (flown by the partner airline), fuel sales, rental revenue, and revenue from maintenance performed for other carriers. Miscellaneous operating revenues includes pet transport fees (in the hold) and compensation for collection of passenger facility charges. Figure A2 plots the evolution of six sources of revenue as a share of total revenue from international operations for the eight largest carriers by international revenue. The only notable break in reliance on baggage and cancellation fees around the implementation of FFAR occurs in 2012Q1 for the newly-combined United/Continental in their first period of joint financial reporting. Previously, neither constituent carrier levied baggage fees for international travel to any significant degree. Baggage fees for domestic travel on the “new United,” meanwhile, decreased significantly as a share of total revenue (Figure A3) in the post-FFAR period. These apparent trend breaks appear to coincide with merger consummation, but it cannot be ruled out that these changes were implemented in reaction to reduced revenues from ticket sales on high-tax international routes. 50

Figure A1 etaR hguorhT−ssaP 4 2 0 2− 4− Unit Tax and Charges Pass−Through by Quarter 2010q1 2011q1 2012q1 2013q1 2014q1 Tax Non−Tax Charges Whiskerbarsrepresent95percentconfidenceintervals. Source: DB1BamdRDC. 51

Figure A2. Carrier Revenue Sources from International Operations 001 08 06 04 02 0 001 08 06 04 02 0 United/Continental Delta/Northwest 2010q1 2012q1 2014q1 2010q1 2012q1 2014q1 American US 2010q1 2012q1 2014q1 2010q1 2012q1 2014q1 )%( erahS euneveR Quarter 001 08 06 04 02 0 001 08 06 04 02 0 JetBlue Hawaiian 2010q1 2012q1 2014q1 2010q1 2012q1 2014q1 Alaska Spirit 2010q1 2012q1 2014q1 2010q1 2012q1 2014q1 )%( erahS euneveR Quarter 00108060402 0 00108060402 0 United/Continental Delta/Northwest American/US Airways JetBlue 2010q1 2012q1 2014q1 2010q1 2012q1 2014q1 Quarter Ticket Revenue Miscellaneous Revenue Baggage Fees Mail, Freight, and Charter Revenue Cancellation Fees Other Transport−Related Revenue Delta/NorthwestandUnited/Continentalbeganreportingcombinedfinancialstatementsin2010Q1and2012Q1,respectively. Revenuesfrommergedcarriersarecombinedfortheentiresample. Source: DOTForm41AirCarrierFinancialStatements,Sch. P-1.2. 52

Figure A3. Carrier Revenue Sources from Domestic Operations 001 08 06 04 02 0 001 08 06 04 02 0 United/Continental Delta/Northwest 2010q1 2012q1 2014q1 2010q1 2012q1 2014q1 American US 2010q1 2012q1 2014q1 2010q1 2012q1 2014q1 Quarter 001 08 06 04 02 0 001 08 06 04 02 0 JetBlue Hawaiian 2010q1 2012q1 2014q1 2010q1 2012q1 2014q1 Alaska Spirit 2010q1 2012q1 2014q1 2010q1 2012q1 2014q1 Quarter 00108060402 0 00108060402 0 United/Continental Delta/Northwest American/US Airways JetBlue 2010q1 2012q1 2014q1 2010q1 2012q1 2014q1 Quarter Ticket Revenue Miscellaneous Revenue Baggage Fees Mail, Freight, and Charter Revenue Cancellation Fees Other Transport−Related Revenue Delta/NorthwestandUnited/Continentalbeganreportingcombinedfinancialstatementsin2010Q1and2012Q1,respectively. Revenuesfrommergedcarriersarecombinedfortheentiresample. Source: DOTForm41AirCarrierFinancialStatements,Sch. P-1.2. 53

Table A1. Sample Tax and Charge Concordance: French Portion of PHL DL CDG Tax Description (IATA Code) RDC Charge Element RDC Item Detail Airport Tax Direct Passengers French Airport Tax (FR) National Surcharge Per Passenger France Civil Aviation Tax Per departing passenger Civil Aviation Tax Domestic And International (FR) to other states Per departing passenger, Passenger Fee international France Passenger Service PRM Fee Per departing passenger Charge International (QX) Check-In Counters Per passenger, (Supplemental Rate) other international Computer Check-in Per departing passenger and boarding (Crews) French Air Passenger Economy passengers Solidarity Tax Solidarity Tax (IZ) to other states Source: ITASoftwareandRDC. Table A2. Sample Non-Tax Charges: French Portion of PHL DL CDG (Aircraft-Specific) RDC Charge Element RDC Item Detail Terminal Charges Departing flights Tax on Air Transport Noise Pollution Acoustic Group 5a, Departure 06:00-18:00 Noise Level Coefficient Acoustic Group 5a, Departure 06:00-22:00 Fixed Power Supply (Landing) Category 1, other international, 400 Hz Fixed Power Supply (Take-Off) Category 1, other international, 400 Hz Aircraft Parking Fee Base charge, pier-side stands Aircraft Parking Fee Supplemental charge, pier-side stands 07:00-23:00 Aircraft Landing Fees MTOW over 40 tonnes De-icing Fees - Base Fee Per Landing Class 4 aircraft Source: RDC. 54

Table A3. Variable Definitions Variable Name Description Unit of Observation Source TotalFare Average total fare per ticket Itinerary-Qtr DB1B cit BaseFare Total fare net of ad valorem and unit taxes Itinerary-Qtr DB1B, RDC cit (and, depending on specification, non-tax charges) UnitTaxes Sum of all specific (unit) taxes levied on a Itinerary-Qtr RDC cit per-passenger basis, aggregated over all arriving and departing flight segments. NonTaxCharges Sum of all airport charges levied on a per-movement Itinerary-Qtr RDC cit basis, aggregated over all arriving and departing flight segments. ln(Passengers) Natural log of passenger volume Itinerary-Qtr DB1B cit ln(Revenue) Natural log of ticket revenue net of unit taxes Itinerary-Qtr DB1B, RDC cit (and, depending on specification, non-tax charges) Distance Total flight distance (in thousands of miles) Route DB1B i Layovers Number of connecting flights Route DB1B i HHI Herfindahl-Hirschman Index of market concentration O&D-Qtr DB1B jt based on ticketing carrier revenue shares in full DB1B sample (scaled to [0,1] interval) Load Percent of available seats sold on the U.S.-foreign Itinerary-Qtr T100 cit flight segment of the ticketed itinerary (normalized by O&D city pair mean and std. dev.) LnOriginVolume Natural log of number of passengers transported by Carrier-Origin-Qtr T100 cjOt ticketing carrier at U.S. airport of origin for all domestic and international flights Itineraries Number of available itineraries in full DB1B sample O&D-Qtr DB1B jt Itineraries Number of available itineraries excluding ticketing Carrier-O&D-Qtr DB1B c−jt carrier c’s own route offerings (including alliance or code-share operations) in full DB1B sample Dollar-denominatedfiguresaremeasuredinhundredsofcurrentdollars. O&Dreferstoorigin-destinationairportpairs. 55

Table A4. Ticket Tax Pass-Through Y =TotalFare (1) (2) (3) cit UnitTaxes 0.768*** 0.992*** 0.958*** cit (0.100) (0.312) (0.307) UnitTaxes ×I[Qtr >2012Q1] 0.700*** -0.743* -0.711* cit t (0.134) (0.418) (0.413) NonTaxCharges - - 0.351*** cit - - (0.127) NonTaxCharges ×I[Qtr >2012Q1] - - 0.022 cit t - - (0.186) Distance 29.647*** 33.994* 31.512* i (7.990) (17.998) (17.865) Distance2 10.782*** -6.785** -6.417** i (1.115) (2.810) (2.790) Distance3 -0.517*** 0.216** 0.205* i (0.045) (0.108) (0.107) I[Layovers=1] -71.686*** -21.894*** -21.943*** i (5.500) (4.107) (4.099) I[Layovers=2] -109.258*** -33.280*** -33.943*** i (7.220) (5.991) (5.929) I[Layovers=4] -107.537* -115.872*** -115.678*** i (61.591) (17.567) (17.575) HHI 51.533 661.766*** 687.970*** jt (170.581) (210.405) (208.671) HHI2 424.729 -1,062.927*** -1,102.260*** jt (319.866) (411.793) (408.013) HHI3 -448.029** 542.313** 557.449** jt (179.686) (235.436) (233.122) LnOriginVolume 339.867*** 185.596*** 182.505*** cjOt (36.657) (23.988) (23.771) LnOriginVolume2 -39.894*** -21.971*** -21.687*** cjOt (3.961) (2.617) (2.591) LnOriginVolume3 1.437*** 0.824*** 0.813*** cjOt (0.136) (0.091) (0.090) Load -5.142*** 4.082*** 3.895*** cit (1.760) (1.287) (1.274) Load2 -1.528*** 0.085 0.058 cit (0.454) (0.364) (0.364) Load3 0.039 -0.086*** -0.084*** cit (0.034) (0.028) (0.028) Fixed Effects: Carrier × Qtr (η ) x x x ct O&D City × Qtr (ν ) x x kt Observations 25,175 24,712 24,712 R-squared 0.854 0.964 0.964 ***p<0.01,**p<0.05,and*p<0.1. Standarderrorsclusteredbyorigin-destinationairport-pairappearinparentheses. Observationsareweightedbypassengervolume. Source: DB1BandRDC. 56

Table A5. Itinerary-Level Passenger Volume and Tax-Exclusive Ticket Revenue Y = ln(Passengers) ln(Revenue) cit cit (1) (2) (3) BaseFare -0.008 -0.438** cit (0.011) (0.180) - BaseFare ×I[Qtr >2012Q1] -0.067*** -0.113 cit t (0.015) (0.202) - UnitTaxes 0.289** 0.304 0.195 cit (0.121) (0.218) (0.119) UnitTaxes ×I[Qtr >2012Q1] -0.791*** -1.179*** -0.682*** cit t (0.198) (0.350) (0.195) NonTaxCharges -0.017 -0.303** -0.169* cit (0.092) (0.134) (0.087) NonTaxCharges ×I[Qtr >2012Q1] -0.154 -0.163 -0.135 cit t (0.128) (0.161) (0.122) Distance -0.348*** -0.206 -0.387*** i (0.097) (0.133) (0.096) Distance2 -0.031** -0.059*** -0.030** i (0.014) (0.021) (0.014) Distance3 0.001*** 0.002*** 0.001*** i (0.001) (0.001) (0.001) I[Layovers=1] -2.483*** -2.586*** -2.526*** i (0.033) (0.045) (0.033) I[Layovers=2] -1.927*** -2.084*** -1.978*** i (0.043) (0.065) (0.044) I[Layovers=4] -1.510*** -2.009*** -1.701*** i (0.092) (0.236) (0.092) HHI -1.579 1.580 -0.621 jt (1.152) (1.638) (1.198) HHI2 2.037 -3.046 0.556 jt (2.060) (3.013) (2.173) HHI3 -0.957 1.621 -0.201 jt (1.155) (1.700) (1.234) LnOriginVolume -1.794*** -0.961*** -1.525*** cjOt (0.204) (0.323) (0.205) LnOriginVolume2 0.211*** 0.112*** 0.178*** cjOt (0.022) (0.037) (0.022) LnOriginVolume3 -0.007*** -0.004*** -0.006*** cjOt (0.001) (0.001) (0.001) Load 0.013 0.031*** 0.020** cit (0.008) (0.011) (0.008) Load2 -0.054*** -0.054*** -0.053*** cit (0.006) (0.006) (0.006) Load3 -0.000 -0.001 -0.000 cit (0.001) (0.001) (0.001) Fixed Effects: Carrier × Qtr (η ) x x x ct O&D City × Qtr (ν ) x x x kt Observations 24,712 24,712 24,712 R-squared 0.836 0.780 0.866 Kleibergen-Paap F-Stat 15.27 ***p<0.01,**p<0.05,and*p<0.1. Standarderrorsclusteredbyorigin-destinationairport-pairappearinparentheses. Observationsareweightedbypassengervolume. 57 Source: DB1BandRDC.

Table A6. Itinerary-Level Passenger Volume - IV First Stages Y = BaseFare BaseFare ×I[Qtr >2012Q1] cit cit t (1) (2) Itineraries -0.006*** -0.000 c−jt (0.001) (0.000) Itineraries ×I[Qtr >2012Q1] -0.002 -0.009*** c−jt t (0.002) (0.001) UnitTaxes -0.079 0.793*** cit (0.306) (0.115) UnitTaxes ×I[Qtr >2012Q1] -0.761* -1.970*** cit t (0.418) (0.312) NonTaxCharges -0.549*** -0.127*** cit (0.127) (0.035) NonTaxCharges ×I[Qtr >2012Q1] -0.018 -0.394*** cit t (0.187) (0.141) Distance 0.317* 0.161 i (0.179) (0.165) Distance2 -0.062** -0.027 i (0.028) (0.026) Distance3 0.002* 0.000 i (0.001) (0.001) I[Layovers=1] -0.219*** -0.195*** i (0.041) (0.034) I[Layovers=2] -0.393*** -0.308*** i (0.058) (0.051) I[Layovers=4] -1.181*** -0.068 i (0.176) (0.152) HHI 4.388** 3.027* jt (2.003) (1.675) HHI2 -7.408* -5.804* jt (3.939) (3.302) HHI3 3.756* 3.180* jt (2.263) (1.901) LnOriginVolume 1.328*** 0.662*** cjOt (0.246) (0.190) LnOriginVolume2 -0.159*** -0.081*** cjOt (0.027) (0.021) LnOriginVolume3 0.006*** 0.003*** cjOt (0.001) (0.001) Load 0.042*** 0.030*** cit (0.012) (0.010) Load2 0.001 0.003 cit (0.004) (0.003) Load3 -0.001*** -0.001*** cit (0.000) (0.000) Fixed Effects: Carrier × Qtr (η ) x x ct O&D City × Qtr (ν ) x x kt Observations 24,712 24,712 R-squared 0.960 0.984 ***p<0.01,**p<0.05,and*p<0.1. Standarderrorsclusteredbyorigin-destinationairport-pairappearinparentheses. Observationsareweightedbypassengervolume. Source: DB1BandRDC. 58