Estimating the Gains from Liberalizing Services Trade: The Case of Passenger Aviation

Estimating the Gains from Liberalizing Services Trade: The Case of Passenger Aviation Anca Cristea University of Oregon David Hummels Purdue University, NBER Brian Roberson Purdue University May 2012

Liberalization and services trade We know relatively little about how services trade is affected by efforts at liberalization. Why? Measurement: Service trade data are highly aggregated; values, not P and Q Policy change is difficult to quantity Literature relies on cross country comparisons, and existing rules are complex. Service trade lib. occurs along with other domestic reforms, and technological change (e.g. finance, telecomm). E.g. Contrast cutting tariff on Mexican steel ball bearings from 15 to 5% with guaranteeing market access in business services 2

We focus on passenger aviation. Why? International passenger aviation is important Big (US + EU = $190bn/year) A key input into merchandise trade (Poole 2009, Cristea 2011) knowledge flows (Hovhannisyan and Keller 2011), Other services (GATS mode 2, mode 4) Not obvious that liberalization will generate benefits Liberalization may result in consolidation/collusion Distribution of gains may be uneven The data are a thing of beauty. 3

Data on aviation and policy change We have a nice policy experiment: From 1992 2007, the US signs 87 bilateral Open Skies Agreements that liberalize trade in passenger aviation. We have detailed firm level transactions data on US passenger aviation, 1993 2008. Prices, quantities, routes offered, carriers competing for precisely defined services e.g. coach ticket from IND > ORD > CPH 4

Air Passenger Traffic 5

Trends in Airfares Dotted line: US BTS price index (Fisher) exact match of ticket Solid line: all DB1B data; estimate period dummies including true origin dest FE 6

Past Regulatory Regime Chicago Convention (1944): failed attempt to set multilateral agreements on air services Countries negotiated air service agreements (ASA) on a bilateral basis. These are typically characterized by Market access restrictions pre defined points of origin and destination Limits on entry and capacity fixed number of designated airlines and limited flights Price control: advance double approval for all airfares 7

Eg: US China Aviation Treaty 1980 Only 2 carriers per country can offer service Flights allowed only between LA, SF, NY, Honolulu Beijing, Shanghai Tokyo is only 3 rd country city from which airlines can operate in serving market Carriers can offer two flights per week for a given route Price changes must be submitted to DC, Beijing for approval two months in advance. 8

Bilateral Open Skies Agreements Remove most existing restrictions No limit on carriers, routes, capacity Price setting at carriers discretion Grant new benefits Extensive beyond market rights Allow inter airline cooperation agreements Alliances, code shares 9

Chicago Gateway to gateway Copenhagen beyond Indianapolis Rome 10

Timing of open skies agreements Agreements are signed sequentially; order weakly correlated with GDP Europe spread throughout sample. NZ 1997, Australia 2008 See Table A1 Demark Sweden Germany Portugal Italy France UK, Spain 11

Model (quick sketch) Multiple carriers serving each country pair Each carrier has preferred low cost city, can offer service out of other cities, but at higher cost E.g. Delta prefers Atlanta but could fly Newark. Carriers differ in which cities are low cost Two stage game One: firms commit to aggregate capacity (# planes) Two: firms allocate capacity to city pairs and play Cournot. Extension of Anderson Fischer (1989) multi market oligopoly to case of multiple heterogeneous firms and markets. Examine outcomes (p, q, markups, # routes & carriers) in two cases Post OSA. Entry into all cities is permitted. Pre OSA. Entry allowed in subset of permitted cities. 12

Carrier entry under route restrictions cp, carriers prefer to allocate capacity to low cost locations With pre OSA route restrictions, two things can happen. Carriers with high cost on permitted routes devote low capacity or stay out of city market entirely, reducing competition; or These carriers enter permitted routes, and increase competition, but industry average cost of service goes up Eliminating route restrictions allows carriers to sort on cost Capacity constraint encourages reallocation of planes away from high cost routes permitted pre OSA Some routes see entry of carriers, others see exit Industry average cost drops. 13

Effects on price and quantity Prices Depends on combination of cost and markup effects. Industry average costs drop everywhere due to sorting, but markup effect varies by city. Prices can rise if there is enough exit from a city. Quantities Price change due to cost, markup variation affects quantities. Unconstrained carriers will choose higher aggregate capacity. Outside of model: quality Consumers may value flight frequency, better connections. Firms capacity choice could include airplane quality. 14

Other Restrictions (future work) Advanced notification of price changes difficult for carriers to manage yields (capacity utilization) if demand is uncertain. planes fly with empty seats despite MC > 0, and the passengers who would fly if price dropped. Restrictions on alliances and code shares A core rationale for national service providers is to prevent monopolization by foreigners Relaxing restrictions may allow carriers to realize comparative advantage on routes, and increase capacity utilization or, allow them to pursue anti competitive market share agreements. 15

Does OSA liberalization lead to welfare gains? Its not obvious. Provisions in OSA could raise or lower prices by affecting average cost of entrants and/or markups raise or lower quantities sold by Affecting prices Changing the set of routes on offer Affecting service quality (quantity net of prices) affect the distribution of gains Redistribute them between carriers and consumers Redistribute them away from consumers on permitted routes and toward consumers on not permitted routes 16

Empirics How do OSA s affect prices, quantities, entry/exit of carriers, routes? Use diff in diff strategy compare traffic growth rates pre and postliberalization Compare OSA to non OSA countries in same period control for year specific shocks to technology, input prices, demand. Combine estimates into a consumer welfare calculation: changes in a quality and variety adjusted price index. 17

Traffic data by route (city pair) x carrier T100 International Segment data Firm level: all air traffic for domestic &foreign air carriers All non stop flight segments crossing the US border Number of passengers, departures, available seats No price data Doesn t track connecting flights When I fly Indy to Chicago to Copenhagen only Chicago Copenhagen in data. 19

Price and Quantity Data Origin Destination Passenger Survey Transaction data: 10% sample of int l airline tickets air fare paid service characteristics (dist, # segments, transit airports, class) all segments of the itinerary and carrier(s) Many tickets involve joint production of several carriers Does not cover non immunized carriers 20

Air Passenger Traffic Nonstop Routes: grow from 870 to 1444 True OD Routes: grow from 28k to 40k 21

Estimate the impact of OSA on traffic Difference in difference estimation method for the number of U.S. passengers abroad: 93 ln Z OSA ln Y / L ln L X 93 93 jt, 1 jt, 2 jt, 3 jt, jq t jt, Z is growth (relative to 1993) in a measure of passenger traffic Index j = country, q = qtr t = year 22

Estimate the impact of OSA on traffic Difference in difference estimation method for the number of U.S. passengers abroad: 93 ln Z OSA ln Y / L ln L X 93 93 jt, 1 jt, 2 jt, 3 jt, jq t jt, Income, population growth: absorb change in traffic demand for country j Year effects: absorb common cost shocks, trend growth in air travel; Country x quarter FE: allow differences in traffic for country j season q 23

Estimate the impact of OSA on traffic Difference in difference estimation method for the number of U.S. passengers abroad: 93 ln Z OSA ln Y / L ln L X 93 93 jt, 1 jt, 2 jt, 3 jt, jq t jt, To pick up effect of OSA Dummy: OSA = 1 for any year that agreement is in effect Interact OSA dummy with vector D( 3) to D(+5) for the age of the OSA agreement allows us to identify pre existing trends, lagged effects of signing 24

Total traffic 25

Open Skies and Traffic Growth Traffic Share covered by Open Skies (%) 1993 2000 2008 Cumulative Growth (%) Total Passengers Non-Stop Routes True O&D Routes Nafta 0 0.0 53.2 102.5 122.6 27.6 Latin America & Caribbean 0 28.5 41.0 93.9 85.5 40.8 OECD Europe 7.7 43.3 100.0 76.0 12.4 36.5 Europe & Central Asia 0 37.0 60.4 245.2 53.8 205.4 Southeast Asia & Pacific 0 22.2 32.6 38.8 8.2 49.4 Middle East & N. Africa 0 8.9 7.1 102.2 16.7-1.4 TOTAL 79.4 66.0 109.1 26

Decompose changes in traffic Write: EM IM Qjt EM jt * IM jt 1 1 1 OSA effect OSA effect on the EM on the IM Intensive margin = air traffic on existing city pair routes (continuing service) Extensive margin = flight service on routes never offered before 1. simple counts of routes 2. Passenger weighted counts of routes (in manner of Feenstra 1994) based on t 3 weights. 3. Could also count carriers as distinct varieties (United and Delta flights from Indy > CPH are different services) Replace Z in estimating equation with components above Recall: pre existing bilateral ASAs specifically restrict entry to particular routes, carriers 27

Total traffic Growth in New Routes Traffic on Existing Routes Extensive margin is much larger when using simple counts, much smaller if we use route x carriers 28

Carrier entry and exit Use T100 data to examine the distribution of entry/exit across routes. Carriers enter routes with sparse competition, exit routes with many firms competing 29

Understanding the channels OSA could raise or lower prices Reduced unit costs from rationalized operations, economies of route density; and lower markups generated by net entry Consolidation creates collusion, higher markups Conditional on prices, OSA could raise or lower quantities by changing service quality Flight frequency, connectivity, use of preferred carriers Reduced incentive for firms to compete by overinvesting in quality 30

Estimating price equation Use O D ticket data to estimate changes in prices for a given true origin destination route r. Starting from about 40 million tickets: Aggregate all tickets within a given route r at time t We might have 10 different ways to get from Indy to CPH, on many different carriers We create a (passenger weighted) average price for route r, from country j, time t. Use diff in diff: how do average prices on OSA routes change relative to non OSA routes? 31

All possible routings to get from IND to MEL are aggregated for a given year, but we keep track of average characteristics (distance, number of segments) Sydney SF Melbourne Indianapolis LAX 32

Controls in price equation Cost shocks Control for route FE, ticket characteristics (distance, number of segments) rjtvarying Economies of route density (population & number of possible destinations reached by each airport) Include time FE (costs common to all routes in a time period) Route time varying cost shocks (fuel*dist, insurance*geographic region) Include OSA, and OSA connect dummy OSA: direct effect on traffic originating/terminating in OSA ctry OSA connect: indirect effects for traffic connecting through an OSA country but originating or terminating elsewhere E.g. fly through Denmark to get to Italy. 33

Cost shifters: ATA data Fuel ranges from 9 27% of total cost in this period Route x Time varying Time varying 34

Price Regressions: (DB1B) Dependent variable: Economy Class Airfare (log) (1) (2) OSA 0.004-0.015*** [0.005] [0.005] OSA Connect * Distance Share -0.105*** [0.009] OSA * Share US Origin (outbound) No. Connections (log) 0.238*** 0.243*** [0.008] [0.008] Observations 599,533 599,533 R-squared 0.203 0.204 Control Variables: Cost shifters: Ticket Distance Fuel*Distance Aircraft Insurance*World Region Trip characteristics: One way Avg. Number of Connections Outbound Traffic Density: US state Population Foreign Country Population Total Departures at Origin Total Departures at Destination Total Direct Routes (country) Other: Partial Liberalization 35

Entry and exit Use T100 data to examine the distribution of entry/exit across routes. Carriers enter routes with sparse competition, exit routes with many firms competing 36

Price effects by entry/exit (outbound) Dependent variable Economy Class Airfare (log) All Routes Net Exit Net Entry OLS OLS OLS OSA -0.001 0.043*** -0.023*** [0.007] [0.008] [0.007] Economy Class Fare Observations 545,345 433,592 480,365 R-squared 0.056 0.057 0.053 Sample: only tickets that match gateway gateway routes; outbound flows only 37

Estimating Quantity equation Use O D ticket data to estimate changes in demand on a given true origin destination pair r. More general than T 100, has all segments and all destinations (not just gateways); can control for prices Include all tickets with same origin destination Prices instrumented with fuel*distance, insurance costs*region interactions Demand shifters: Population, income; bilateral trade; number of segments OSA variable measures increase in traffic conditional on prices, other demand shifters. 38

Quantity Regressions (DB1B) Dependent variable: Number of Air Passengers OLS 2SLS (1) (2) (3) (4) Economy Class Airfare (log) -0.068*** -0.067*** -1.412*** -1.412*** [0.007] [0.007] [0.112] [0.112] OSA 0.048*** 0.088*** 0.062*** 0.002 [0.011] [0.012] [0.011] [0.015] OSA Connect 0.099*** 0.077*** 0.083*** [0.008] [0.009] [0.009] OSA*US origin share (outbound) 0.107*** [0.015] No. Segments (log) -1.255*** -1.269*** -1.269*** -0.930*** [0.033] [0.033] [0.033] [.049] Observations 599,619 599,619 599,606 599,520 R-squared 0.228 0.228 -- First Stage Statistics: F-Test of iv 112.2 108.2 Hansen's j stat 146.3 126.6 Instruments for Airfare: Ticket Distance Fuel*Distance Insurance*World Region Control Variables: Trip characteristics: Direct (non stop) Avg. Number of Connections Outbound Market size: US State Population US State Income Foreign Country Population Foreign Country Income Total Exports Total Direct Routes (country) Other: Partial Liberalization Caribbean Trend 39

D(qty) as a quality effect OSA s improve flight frequency, connectivity, use of preferred carriers; may also induce competition through better amenities on planes. To extract this ln q E ln p where X OSA rjt rjt rjt rjt rjt rjt jt rjt OSA quality effect (price equivalent) = /.107 /1.4 7.6% This attributes none of other sources of quality change (e.g. reducing number of flight segments) to the OSA 41

Welfare calculation To measure OSA relative to non OSA, we capture 1. Relative price movements from OSA price regression 2. Construct quality adjusted prices by netting off the effects of OSA on quality (measured as OSA effect on quantity net of prices from quantity regressions.) 2. Use quality adjusted prices to form relative price series; then apply Feenstra 1994 to get variety adjusted price index t P t t t 1 Variety adjusted price index 1/( 1) P t i p it p it1 Price index for common set w it r ii ii r p x ir p x ir ir ir Variety adjustment 42

Applying this to the policy experiment OSA direct OSA connect Sigma == 1.25 outbound inbound outbound inbound D Airfare (price effect) 0.000-0.026-0.060-0.053 D Quality (quantity effect net of prices) -0.098-0.038-0.102-0.102 D Quality Adjusted Price Index 0.902 0.936 0.839 0.846 D Lambda-ratio Variety Index 0.755 0.755 0.755 0.755 D Variety Adjusted Price Index 0.681 0.706 0.633 0.638 Drop in Price Index due to OSA (%) 31.91% 29.41% 36.70% 36.18% Sigma estimated using either variation across routes 43

Summary We use firm level transactions data to examine the effects of sequential bilateral liberalization of aviation markets. Diff in diff strategy compares changes pre/post OSA for signers relative to non signers. We find that OSA s Lower prices, raise (implicit) quality, expand route offerings Net effect: a (quality & variety adjusted) price index drops by 31 % relative to non signers. Additional findings Third party effects: non signers can connect through OSA countries. Benefits are not uniform Pre OSA service is concentrated on a few routes; OSA => exit, rise in prices 44

Supporting slides 45

Understanding the quantity channels Conditional on prices, OSA raise quantities. Relax capacity constraints Raise service quality To examine first channel, measure the extent of capacity constraints Load factor = passengers / seats How high was load factor pre OSA? Did load factor change as a result of OSA? 46

Is D(qty) due to relaxed capacity constraint? Pre OSA Load factor never exceeds 85%. Median 63.6% Load factors rise post OSA. Elasticity of load factor wrt OSA = 0.026. Organize routes into 8 bins by pre OSA load factors. Number is max load factor in that bin Height of bar is log change in passengers post OSA 47

Aside: load factors and prices Suppose carriers enter a route until P = AC conditional on making a flight, marginal cost of a passenger is very small. Then: AC = cost of a flight/number of passengers This is the inverse of the load factor ln P ln OSA ln load factor ln OSA.026 48