Incentives and Competition in the Airline Industry

Transcription

1 Incentives and Competition in the Airline Industry Rajesh K. Aggarwal D Amore-McKim School of Business Northeastern University Hayden Hall 413 Boston, MA r.aggarwal@neu.edu Carola Schenone McIntire School of Commerce University of Virginia Rouss and Robertson Halls Charlottesville, VA schenone@virginia.edu May 3,

2 Incentives and Competition in the Airline Industry Abstract We examine how performance changes at airlines in response to a change in incentives for their executives. When airline executives are given bonuses based on on-time arrival performance, on-time arrival does indeed improve. Interestingly, competitors on the same routes also improve their on-time performance, even when the executives of the competitors did not receive bonuses based on their own on-time performance. As a result, there is no improvement in airline financial performance. Our results suggest that incentives simply heighten competition in on-time performance, which then erodes any possible financial gains from improved service. Keywords: Incentives, Competition, Airlines. JEL classification : G30, G34, G32

3 1 Introduction While incentives for executives have been the subject of an enormous amount of academic research, very little is known about what managers do in response to incentives. For example, much of the managerial incentives literature focuses on grants of stock and stock options, and then examines performance measures such as stock or accounting returns. What is unexplored is what actions managers take in response to incentives to a ect stock or accounting returns. Do managers reduce labor costs, or build new factories, or enter new markets, or try to increase customer satisfaction, or restructure their span of control, etc.? In short, there are an enormous number of actions managers could take in response to changes in incentives, and even isolating which actions are feasible for managers is a daunting challenge. As a result, research has focused on end outputs (stock returns or accounting measures) and ignored managerial inputs (actions). However, in principal-agent models (e.g., Holmstrom and Milgrom (1987)), it is managerial actions that are directly influenced by incentives, while outputs are measured with noise. Thus, understanding how managerial actions respond to changes in incentives is a crucial step in understanding how incentives influence firm performance. Further, relatively little is known about how competitors respond to incentives. Specifically, when a firm changes managerial incentives, this presumably induces managers to take certain actions, and may induce or provoke a strategic response from industry competitors. Indeed, firms may alter incentives strategically for precisely this reason (see Aggarwal and Samwick (1999)). The empirical challenge is again in isolating the actions taken by managers in response to incentives, and the actions taken by competitors in response to another firm s changes in incentives. We address these issues by looking at a specific industry, the airline industry, and at specifically delineated incentives that have measurable outcomes very specifically tied to actions. Notice that we do not have the actual actions themselves. Instead, we argue that the outcomes we measure are a relatively noise free transformation of actions, and thus represent a valid measure of actions. We focus on bonuses provided by airlines to executives for on-time arrival performance. On-time arrival is a key metric for customer satisfaction in the airline industry. Over the past 20 years, virtually every airline has adopted on-time 1

4 arrival as a measure used in the provision of executive bonuses. Importantly for us, the introduction of bonuses tied to on-time arrival has been staggered in time across airlines. We are able to identify when each airline adopted on-time arrival bonuses. We examine the following questions. Do airlines respond to the introduction of on-time arrival incentives and how? Do they adjust the frequency or capacity utilization of flights? Do they adjust their schedules or game the incentive by increasing scheduled flight times? How do competitors respond to the introduction of incentives? Who benefits from the introduction of incentives the airlines or consumers? Taken together, the answers to these questions provide a fairly complete picture of how the introduction of incentives a ected airlines, as we now discuss. The Bureau of Transportation Statistics provides data for on-time performance for all su ciently large airlines in the United States. Thus, we can directly see how an airline s on-time performance has varied in response to changes in incentives based on on-time arrival. Moreover, we are able to disaggregate on-time performance into departure times, scheduled flight times, actual flight times, and arrival times, thus more precisely seeing how the airline responds to the introduction of the incentive for executives. Further, since on-time performance is a key metric of customer satisfaction, it is also an area in which airlines can and do compete with each other. We are able to see how competitors respond to the introduction of an on-time arrival incentive by rivals. For example, competition in on-time performance may be either competition in strategic complements or strategic substitutes. If competition is in strategic complements, then instituting an on-time arrival incentive may cause both the initiating firm and its rivals to toughen competition. This would increase consumer surplus but not result in better financial performance for the firm that initiates an on-time arrival incentive. Executives receiving an on-time performance bonus might strategically change the number of scheduled and performed flights in an attempt to turn around the planes faster and meet departure and arrival times. Executives could also introduce changes in the number of passengers and seats available as a way to expedite the boarding and deplaning of passengers, making it easier for the carrier to depart on time. Thus, we examine changes in capacity, both the number of flights as well as passengers and seats. 2

5 Another possibility is that airlines might try to game the on-time arrival incentive. Bennett et al. (2016) show that managers in general do try to strategically achieve specific compensation targets. In our context, a specific compensation target would be an on-time arrival bonus. When faced with this new incentive, executives might strategically increase the scheduled amount of time for a flight, thus making it easier for the flight to arrive on-time and the executives to achieve their bonus. We find that when airline executives are given bonuses based on on-time arrival performance, on-time arrival does indeed improve. Delays decrease by about a minute, or 8%, of the average non-negative arrival delay (negative delays are flights arriving early) relative to a carrier without an incentive after saturating our specifications with fixed e ects. The fraction of flights delayed 15 minutes or more (an industry standard definition for a flight being late) also decreases by about 8%. Because we see both arrival delays and departure delays, we show that most of the decrease in arrival delays is due to airlines decreasing departure delays getting departing flights out of the gate closer to on time. When we restrict our sample to problem markets those markets where more than 30% of flights in a quarter are late by 15 minutes or more we find that arrival delays shrink by over 5.5 minutes. Interestingly, competitors on the same routes also improve their on-time performance, even when the executives of the competitors did not receive bonuses based on their own on-time performance. The magnitudes are similar, with the minutes delayed decreasing by about 10% and the fraction of flights delayed 15 minutes or more decreasing by about 8%. When both executives of the own and rival airline receive bonuses based on on-time performance, we find that on-time performance on the routes on which they compete is worse. Much of this result when both competitors have an incentive is driven by United Airlines, which initiated its on-time arrival incentive while it was in bankruptcy. For capacity, passengers, seats, and number of flights all increase for an airline with an on-time arrival incentive, while the e ect on rival carriers without an incentive is negligible. For airlines with incentives, passengers increase more than seats, so that load factors or capacity utilization increase. The key point here is that airlines with an on-time arrival incentive do not strategically decrease capacity or utilization to make it easier to achieve bonuses. 3

6 For other types of strategic gaming, such as increasing the scheduled time of the flights, we again find little evidence of such gaming. On average, carriers with an on-time arrival incentive decrease their scheduled minus actual flight times, but by a tiny amount (7 seconds). Rival carriers do increase their scheduled minus actual flight times, but again by a tiny amount (20 seconds). We do find that one carrier, Alaska Airlines, did substantially increase its scheduled minus actual time by almost 4 minutes after introducing an incentive, possibly indicating strategic gaming at this carrier. As for financial performance, there is no improvement in airline profitability, as both revenues and expenses increase after the incentive is initiated. Our results suggest that incentives simply heighten competition in on-time performance, creating consumer surplus and eroding any possible financial gains from improved service. We are able to exploit the staggered introduction of incentives across carriers to more precisely estimated each airline s response to the introduction of an incentive. We use a di erence-in-di erence specification where we consider an airline before and after introducing an incentive (treated) versus other airlines without an incentive (control) on the same routes. Here we find e ects that are 50% larger after introducing an incentive relative to our baseline specifications. This di erence is due to examining the two years around the treatment, rather than the entire sample. In our di erence-in-di erence specifications, we are also able to examine the e ect of the introduction of incentives by one carrier on its competitors that do not have an incentive. We find similar results to our full sample competitors respond to the introduction of incentives by reducing their delays on the routes on which they compete. Finally, to address endogeneity concerns, we consider fuel price shocks as exogenous changes in the environment. When fuel prices increase, airlines will tend to fly more slowly in order to conserve fuel. Put di erently, the costs to trying to be on time increase. For airlines with an on-time incentive, the incentive may (partially) counteract the costs to being on-time. For airlines without an incentive, the costs to being on-time will predominate. We consider three fuel price shocks and find some evidence that airlines without an incentive decrease their on-time performance relative to those airlines with an incentive. This paper is organized as follows. We describe our data and variable construction in Section 2. Section 3 presents our econometric specifications and details our identification 4

7 strategy. Section 4 presents our results and Section 5 presents a di erence in di erence specification and results. Section 6 presents a natural experiment. Finally Section 7 concludes. 2 Data We select the airline industry as we have measurable outcomes very specifically tied to actions, that respond to the particular incentive given to a manager. Furthermore, the airline industry yields a vast amount data for studying competitive reactions among industry rivals. Specifically, airlines compete over many di erent markets, and over di erent time periods. For instance, at any given point in time, a carrier s faces di erent competitors in di erent markets; and for any given route, a carrier is likely to face di erent competitors at di erent times as carriers enter and exit routes. For example, United Airline s competitors in Denver- San Francisco in the first quarter of 2001 are not the same set of competitors United faces in the route between Dallas Forth Worth and Newark in the first quarter of Thus, a carrier s business strategy at a point in time and in one market might di er from its strategy on a di erent market at the same time; and the strategy at one market might di er across time depending on which carriers entered exited that market. In fact, the literature on the airline industry (see for example, Berry (1990), Brueckner et al. (1992), and Gerardi and Shapiro (2009)) claims that carriers make decisions that are route specific as opposed to making nation wide decisions. Therefore, even if we focus on a few airlines, the strategic decision each one makes in any one market during a specific time period, is likely to di er from the strategic response chosen at the same time but in a di erent market where the set of product market competitors is di erent. The airline data in our work is a compilation of di erent databases from the O ce of Airline Information at the Bureau of Transportation Statistics (BTS). The sample ranges from January 1 st, 1993 to December 31 st, During this time window, most of the large US carriers instituted an incentive, and by the end of the sample, the only carrier without an incentive, or that had not merged with another carrier that had an incentive, is Jet Blue. Data on executive contracts and bonuses were gathered manually using firm proxy statements. 5

8 2.1 Data Sources Airline Stylized Facts and Performance Based Incentives Data on executive bonuses were collected from firm proxy statements (Form DEF 14a). All proxy statements were read to see if bonuses depended on on-time arrival or performance. Firms generally disclose key measures for firm bonuses. Firms are not required to disclose how much of the bonus depends upon each measure, and they generally do not do so. Table 1, Panel A shows when each airline in our sample initiated an on-time arrival incentive. These incentives are quite sticky, and once instituted are not rescinded. It is important to highlight how the introduction of incentives has been staggered across time and carriers. This allows us to isolate the e ect of incentives on individual carriers. The first carrier to introduce an ontime performance incentive was Continental in 1994; by the end of our sample period, 2010, the only surviving carriers without an incentive are Northwest (acquired by Delta) and Jet Blue (America West did not introduce an incentive but was absorbed by US Airways in 2005; in turn, US Airways was absorbed by American Airlines in 2013). Six out of then of the airlines in our sample have operated under Chapter 11 Bankruptcy protection during our sample period: Continental, America West, United, Delta, Northwest, and US Airways (twice). Panel B shows when individual airlines enter and leave bankruptcy. To avoid any confounding e ect that operating under Bankruptcy protection might have on arrival and departure delays, as well as capacity changes, we will control for the year-quarters where each of the bankrupt carriers operated under Court protection. See, for example, Borenstein and Rose (1995), Borenstein and Rose (2003), Ciliberto and Schenone (2012a), and Benmelech and Bergman (2008) for the e ect of bankruptcy on airline competition. 1 The airline industry has recently seen a wave of mergers that further consolidated the number of carriers. At the onset of our sample period there are ten large US carriers, by the end, the number has been reduced to three legacy carriers and Southwest. Such degree of consolidation can further impact competition and airline behavior, and therefore we control for mergers in our empirical work (see Kim and Singal (1993), and Oliver (2003))). Panel C 1 Other papers look at the relationship between airline delays and the following variables: competition: Rupp et al. (2006), market concentration: Mazzeo (2003), bankruptcy: Ciliberto and Schenone (2012b). 6

9 shows when mergers are announced, completed, and when the merged airlines operated as a unified airline, as well as the surviving airline name Airline Performance Data On-time Arrival and Departure Outcomes The Bureau of Transportation Statistics (BTS) collects on-time performance data reported by US certified air carriers that account for at least one percent of domestic scheduled passenger revenues, and reports it in the database named: Airline On-Time Performance Data. Carriers report daily flight information for the routes they serve. 2 Specifically, it contains scheduled, and actual, departure and arrival times, and scheduled and actual flight times. 3 As will be explained below, we collapse this daily flight data to quarterly data for our empirical analysis. Capacity and Utilization Data Executives receiving an on-time performance bonus might strategically change the frequency of flights served, or the number of transported passengers, to aid in turning planes around in amoree cientway,thusimprovingarrivalanddeparturetimes.toaccountforthispossibility, we study changes in capacity around the time the carriers introduced the performance incentive. Capacity data is from the T100 Domestic Segment of Form 41-Tra c data provided by the BTS and includes items such as passengers enplaned, seats available, load factors, and 2 A route is a non-stop segment of a market. And a market is defined by the origin and end destination of a trip. This distinction is important. A route is a segment in the passenger s itinerary, this means, it is a nonstop trip. A market is created by a trip break, which are points in the itinerary at which a passenger is assumed to have stopped for a reason other than changing planes. For example: an itinerary BOS-SFO- LAX-SFO-BOS would have two routes BOS-SFO and SFO-LAX, and the market is BOS-LAX. The trip break occurred at LAX. Data on on-time performance and capacity is at the route level. 3 This data, as well as the capacity data and financial performance data is available through the BTS maintained website: 7

10 number of departures performed and scheduled. This database is not a sample of flights, but a record of all monthly flight segments between an origin and destination airport located within the US boundaries or its territories for US certified air carriers that account for at least one percent of domestic scheduled passenger revenue. This data is at the aircraft type level, i.e., a carrier, on a route, on a given day, flying a specific aircraft (a route can be served with di erent aircrafts depending on the time of day the flight occurs). As with the ontime performance data, we collapse this data to a quarterly basis and merge it with the ontime performance measure. Financial Outcomes It is reasonable to expect that an on-time bonus incentive results in operational e ciency gains and reduced delays. This, in turn, might result in increased demand, and consequently, improve the carrier s financial performance. To analyze the financial impact of the performance based incentive, we collect data from Schedules B1 and P1.2 of the Domestic Segment Form 41-Financial Data provided by the BTS. For large certified U.S. carriers with annual operating revenues of 20M or more, Schedule B1 contains quarterly operating balance sheet statements, and includes items such as current and total assets, cash, accounts receivables, short and long term debt, etc. Schedule P1.2 provides quarterly profit and loss statements, and includes operating revenues, expenses and profits, depreciation and amortization, etc. The unit of observation for this data set is a carrier j, year-quartert. 2.2 Variable definitions Operations Performance Measures Arrival delays are calculated as the di erence between scheduled and actual arrival times. The BTS defines actual arrival time as, The time the aircraft touches down upon arrival, and scheduled arrival time as The scheduled time that an aircraft should cross a certain point (landing or metering fix). For each reporting carrier j, flyingrouter, onanygivenday,we 8

11 calculate the minutes of arrival delays for each one of the flight in that route, as the di erence between actual and scheduled arrival times ArrivalDelay j,dmy,r = ArrivalT ime j,dmy,r SchedT ime j,dmy,r. If a flight arrives ahead of time, ArrivalDelay j,dmy,r < 0, which masks the e ective measure of delays for carrier j in route r during that quarter. Since we are interested in actual flight delays, and don t want to confound this measure with early arrivals, we replace negative arrival delays with 0 delayed minutes. We then aggregate the data to a year-quarter observation for each carrier-route pair, as follows: For each carrier j, in route r, during year-quarter t, weaddtheminutesofarrivaldelay,anddividethesumbythetotal number of flights carrier j completed P in route r, duringintheday-month-yeardmy of yearquarter t: MeanArrDelay j,r,t =.Departuredelays,MeanDepDelay j,r,t are defined similarly. 4 dm2t ArrDelay j,dmy,r NuFlights j,r,t The BTS collects data on actual flight times, and also on scheduled flight times. The latter are reported to the Computer Reservation System (CRS) to allows consumers and travel agencies organize and compare trips. Scheduled and actual flight times are recorded for each carrier, for every flight in a give route on a daily basis. Just as explained above, we collapse this carrier-route-day-flight data to a carrier-route- year-quarter unit of observation: CRS Scheduled j,r,t is the average scheduled flight time carrier j reports on route r during year-quarter t. Similarly for the actual elapsed flight time Actual Elapsed j,r,t. To capture whether managers receiving an on-time performance bonus pad the scheduled times in an attempt to show more on-time arrivals, we calculate the the di erence between the scheduled and actual flight time, P ad j,r,t = CRS Scheduled j,r,t Actual Elapsed j,r,t. 5 Table 2, Panel A, reports summary statistics on scheduled minus actual flight time, as 4 There are several issues we deal with using the raw data from the BTS. First, we eliminate observations for which arrival, departure, as well as actual and scheduled flight times is missing. Second, we drop observations for which arrival delays are more than 600 minutes (10 hours), early arrivals more than (180 minutes (3 hours). We also drop observations for which departure delays are more than 600 minutes, and for which early departures exceed 30 minutes. 5 As before, several adjustments need to be done to the raw BTS Data. First, in calculating actual and elapsed times we account for any time zone di erences between origin and destination airports. Specifically, we adjust the origin departure time to be the equivalent of the destination time to compute actual flight times as the di erence between the actual arrival times and the time-zone-adjusted departure time. Second, we drop flights for which the actual elapsed time is less than 15 minutes, or more than 720 minutes (12 hours). 9

12 well as arrival and departure delays (where early arrivals/departures are coded as negative (Panels B and C report results where early arrivals/delays are coded as no delays), for the entire sample of airlines used in our estimations. We require that a carrier must fly a route for the entire time a carrier-route is in our sample. We split our observations into four mutually exclusive categories: 1) None no carrier on a specific route in that quarter has an on-time incentive; 2) Own the specific carrier on a specific route in that quarter has an on-time incentive; 3) Rivals at least one competitor carrier on a specific route in that quarter has an on-time incentive; 4) Own Rivals both a specific carrier and at least one competitor on a specific route in that quarter have on-time incentives. Panel A shows that if there are no on-time incentives for any carrier on a route, then arrival delays are on average 8.44 minutes. If a carrier has an on-time incentive, then its arrival delays are 5.84 minutes. Interestingly, if a carrier s rival on a route has an incentive, then the carrier s delays are 5.40 minutes, possibly suggesting that there is a competitive response to a rival carrier having an incentive. Finally, if both carriers on a route have an incentive, then arrival delays are minutes. This result seems anomalous, and we return to it later. Departure delays are presented in the second row of Panel A. The di erence between the CRS reported scheduled flight time and the actual elapsed flight time, CRS Scheduled j,r,t Actual Elapsed j,r,t is presented in the third row. If carriers wanted to pad the flight time as a response to on-time arrival incentives, then we should see a significant increase in CRS Scheduled j,r,t, and we do see some evidence of that in Columns (2) and (3). As should be true, MeanArrDelay j,r,t = MeanDepDelay j,r,t (CRS Scheduled j,r,t Actual Elapsed j,r,t ), up to rounding error. The total number of route-carrier-quarter observations in our sample is given in the last row of Panel A. The first four columns of Panel B of this table report arrival and departure delays where early arrivals and departures have been coded as no delays (negative minutes delayed replaced by 0 delayed minutes). Results are consistent with those in Panel A and, not surprisingly, larger in magnitude. The industry standard definition of a flight being delayed is if the flight arrives at least 15 minutes late relative to its scheduled time. Reducing the frequency of flights that are 10

13 delayed by this definition (15 minutes late or more) is the typical goal of an on-time incentive in an executive bonus. The right four columns of Panel B show the frequency with which flights are delayed 15 minutes or more. If no carrier on a route has an incentive, flights are delayed 15 minutes or more 24% of the time upon arrivals. If a carrier has an incentive, this drops to 21%, and drops to 18% if a rival carrier has an incentive. If both a carrier and its rival have an incentive, the frequency of flights arriving 15 minutes late or more is 26%, which again seems anomalous. We define problem routes for each carrier, as those in which a carrier s flights are delayed more than 15 minutes more than 30% of the time, prior to the carrier s introduction of the incentive, and look at di erences in arrival delays before and after the implementation of the incentive. Panel C reports these statistics. The left four columns show the average minutes delayed for problem markets, and the right four columns show the frequency with which flights are delayed more than 15 minutes in a problem market. Note that these average delays can still be less than 15 minutes and the frequency of delays less than 30%, since the definition of a problem market is that at least 30% of the flights on that route were delayed 15 minutes or more in the prior quarter. Nonetheless, delays in problem markets seem to be quite persistent. We will later focus on these problem markets, as these are presumably the markets most in need of reductions in delays. Note that the number of observations in problem markets is much smaller than the full sample. Executives receiving an on-time performance bonus might strategically reduce the number of scheduled and performed flights in an attempt to turn around the planes faster and meet departure and arrival times. Executives could also introduce changes in the number of passengers and seats available as a way to expedite the boarding and deplaning of passengers, making it easier for the carrier to depart and arrive on time. We use the capacity data reported in the T100 Domestic Segment. The T-100 Domestic Segment database is not a sample or survey of flights; it is a census that records all monthly nonstop flights between an origin and a destination airport located within the US boundaries or its territories. It reports detailed information at the route-carrier-year-quarter-aircraft level. This means for each carrier, flying a route on any given month, we have specific data on the type of aircraft used to serve each flight on that month, the total number of available seats that month and the actual number of passengers 11

14 transported. 6 We collapse the data to a quarterly unit of observation taking the sum of the relevant variables across all the aircrafts a flown on month m in route r by carrier j : Q j,r,t = P ma2t Q j,ma,r where Q j,ma,r is one of the following capacity measures: record of seats available, actual number of enplaned passengers, departures scheduled, and departures performed, for each aircraft a flown in the month corresponding to year quarter t. Wedefine quarterly load factor as the ratio of revenue passenger per mile to the available seats per mile. 7 Table 3 reports summary statistics for capacity and capacity utilization. To normalize the number of seats and passengers for a carrier in a year-quarter, we divide seats available (passengers) by the total number of scheduled (performed) flights by that carrier, in Seats the year-quarter and in that route: Seats j,r,t = j,r,t Scheduled Flights j,r,t and P assengers j,r,t = Passengers j,r,t Performed Flights j,r,t. Panel A includes all markets while Panel B reduces the sample to those defined as problem markets. The carriers that introduce the incentive see an increase in the number of seats per scheduled flight available as well as the number of enplaned passengers per performed flight. Similarly for competitors of the carrier that introduced the incentive. There is a slight in capacity utilization for the carrier that introduces the incentive, as well as the competitors of such carriers, and a slightly more significant increase when a carrier and at least one rival have an incentive in the same market at the same time: Load factor increases from 0.58 when no carrier has an incentive 0.73 when a specific carrier and a rival in a route at the same time have an incentive. 6 The T100 is a complete census of non-stop flight segments. We use this data to identify non-stop routes, and routes that are not served in a consistent manner (e.g., sporadic service between an origin and destination). We drop route-carrier-year-quarter observations for which the quarterly sum of scheduled or performed, departures is less than 24 (routes served less than 8 times a month by a carrier), and also drop observations for which the quarterly sum of passengers, or seats, is less than 100. We also drop observations for which the route includes either an origin or destination not ranked, by the total number of passengers transported, in the top 100 origin and destination airports. These are standard filters in the literature to avoid small, unrepresentative routes that are sporadically served. 7 The BTS defines a passenger as: Any person on board a flight who is not a member of the flight or cabin crew, and available seats as Installed seats in an aircraft (including seats in lounges) exclusive of any seats not o ered for sale to the public by the carrier; provided that in no instance shall any seat sold be excluded from the count of available seats. 12

15 Regarding departures scheduled and performed, we see a large increase when a carrier introduces and incentive and when a carrier without an incentive competes with one that has an incentive. Results are even larger in magnitude when both a given carrier and at least one competitor on a specific route in that quarter have on-time incentives. Panel B reports similar results, but larger in magnitude, when considering the sample restricted to problem markets. If the on-time bonus incentive works to reduce delays and improve a carrier s e ciency, then the carrier s financial performance should improve. Lower delays can increase demand for the carrier s flights and allow carriers to increase revenues. However, if a carrier s ontime incentive translates into more competition across airlines serving the routes that the incentive-initiating carrier serves, the financial performance e ect of the incentive might be eroded. To consider this we use the financial data in Schedules B1 and P1.2 of the Domestic Segment Form 41-Financial Data, and use a carrier s quarterly operating revenue and expenses as a fraction of total assets, as well as operating profits, to test this. In Table 4, we present summary financial statistics on revenues, expenses, net income, and operating profit for the airlines sorted by whether they had an incentive or not prior to the measurement of financial performance. In general, we see reductions in revenue and expenses from before to after the introduction of an incentive. The reductions in revenues is greater, resulting in a reduction in operating profit and net income Market presence, Mergers, and Bankruptcy In the econometric specification we need to control for whether a carrier s presence in a market generates a particular competitive dynamic for all carriers serving that market, regardless of the existence of a performance based incentive. To control for carrier j s presence in route r at time t let: jinroute rt be a categorical variable equal to one if carrier j is active at time t in route r regardless of whether the observation is for carrier j or a competitor of j, andzerootherwise. During the last ten years the airline industry has seen a large number of mergers. US Airways merged with America West and began joint operations in 2006 under the merged 13

16 entity US Airways. US Airways subsequently merged with American Airlines and began joint operations in Delta and Northwest merged and began joint operations in 2010, under the surviving carrier Delta Airlines. United and Continental Airlines merged and reported jointly beginning in 2012 as United Continental Holdings. Some of these mergers are outside of our sample period. To avoid confounding the e ect that mergers could have on ontime performance, we include in the regression equations the following categorical variables: ijmerge rt =1forobservationsofmergingcarriersi and j during the year quarters t between merger completion and the start of joint reporting to the BTS, on routes r where i and j operate. M is the set of carriers that merge during the sample period. Of the ten airlines in our sample, six have operated under Bankruptcy Chapter 11 protection during our estimation period. Continental and America West enter our sample operating under Bankruptcy; the first to emerge is Continental in the second quarter of 1993 (just 2 quarters after the beginning of our sample), and America West emerged in the third quarter of United operated under Chapter 11 between December 2002 and February Interestingly, United introduced the on-time incentive in 2003 during this time window. Delta and Northwest filed for bankruptcy on the same day in 2005 and emerged on the second quarter of US Airways also operated under Bankruptcy Court protection at two different times during our sample: between 2002 and 2003, and then again between 2004 and 2005 when it emerged by merging with American Airlines. Previous work has shown that carriers operating under Court protection introduce changes that alter their performance and the way they compete in the product market. Thus, to control for any confounding e ect that can arise from operating in bankruptcy, we include bankruptcy control categorical variables: jbkt jrt =1ifcarrierj operates under Chapter 11 Bankruptcy at time t, andb is the set of carriers that at some point in our sample period operated under Chapter 11. Note that within our time period, the only carriers not operating under protection are: American, Jet Blue, Southwest, and Alaska Time Trends at Origin and Destination Airports There could be persistent correlation between negative unobserved current and future demand shifts in origins or destinations served by carriers that have introduced an incentive 14

17 relative to those origins and destination served by carriers that have not introduced an incentive. That is, over time, there could be origin- and/or destination- specific shocks that can alter the ontime performance of flights involving such origins and destinations. For example, during the financial crisis of 2007 some cities su ered more (e.g. Detroit), so flights into or out of airports in such cities could see a change in on time performance or capacity and utilization measures associated with the decline of the city s economy over time, but unrelated to whether a carrier serving that route has an incentive. To deal with this route specific unobservable correlation across time, we include linear origin and destination time trends. 8 Origin r,t =1fortheorigincorrespondingtorouter; similarlyfordest r,t. T imetrend t is a year-quarter time trend variable, taking the values 1 (first quarter 1993) to 72 (fourth quarter 2010). T imetrend t Origin r,t captures specific changes in an origin over time. Similarly for T imetrend t Dest r,t. 9 3 Econometric Specification We estimate a regression of the type: P jrt = Own Own jrt + Rivals Rival jrt + Own Rival Own jrt Rival jrt (1) + X r Origin(r) T imetrend t Origin r,t + X r Dest(r) T imetrend t Dest r,t + X jbkt jbkt jrt + X ij ijmerge rt + X j2b ij2m j jrt jinroute rt + " jrt P jrt is one of the performance measure for carrier j, inrouter, attimet: Arrival/departure delays, and scheduled minus actual flight times, available seats, enplaned passengers, scheduled and actual performed flights. To control for a carrier s presence in a market we include jinroute rt = 1 if carrier j serves route r at time t. destination-time trends are Origin(r) and Dest(r) respectively. The parameters for the origin and 8 See Friedberg (1998). 9 Note that this means that we estimate a parameter for each origin (and each destination) included in the estimation. 15

18 There are several sources of unobservable heterogeneity we need to control for. Thus " jrt comprises several terms: " jrt = u jr + u t + u jrt. First, u jrt is an idiosyncratic carrier-routeyear-quarter unobservable. Second, to control for the e ects that a particular carrier flying a particular route can have on the performance of that carrier in that route, but not in others routes, we include route-carrier fixed e ects u jr. For example, a carrier in a route might fly more modern planes that could impact its ontime performance, and use older planes in other routes. The routecarrier fixed e ect absorbs this route-carrier specific unobservable. Furthermore, if there are sample selection issues that arise when the unobservables that a ects a carrier s choice of entering route r are the same unobservables that impact a carrier s ontime performance in that route, but not in others, then route-carrier fixed e ects control for this carrier-route selection. 10. Finally, u t are year-quarter fixed e ects that control for performance changes that stem from seasonal and other exogenous shocks that can a ect performance outcomes and can therefore confound the e ect of incentives on performance. Year-quarter fixed e ects also control for any serially correlated industry-specific shock to demand that can impact performance measures. To study the e ect on financial outcomes stemming from performance based incentives, we modify specification (1) as these data are only available at the carrier j and year-quarter t level. To control for the potential that financial performance occurs with a lag relative to the implementation of the performance based incentive and consequent outcomes, we include the lagged incentive measures. The econometric specification is, ( F inp erformance ) jt = Own Own j(t 2) + Rivals Rivals j(t 2) T otalassets (2) + Own Rival Own j(t 2) Rival j(t 2) 10 See, for example, Nijman and Verbeek (1992) + X jbkt jbkt jrt + X ij ijmerge rt + " jt j2b ij2m 16

19 Where F inp erformance jt is one of the following measure of carrier j s performance in year quarter t: Revenues and expenses from operations, net income, and operating profit/loss. As in previous specifications, we include controls for the years a carrier operates under Chapter 11 bankruptcy protection, as well as controls for the quarters between the completion of a merger and the time the carriers begin joint reporting to the BTS. The unobservable is " jt = u jt + u t + u j,whereu jt is a carrier-year fixed e ects that captures changes within a carrier over time; u t is a year-quarter fixed e ect capturing seasonal changes that can a ect profitability and financial outcomes; and u j capture unobservable carrier specific e ects (such as managerial expertise in financial policy, etc.). 4 Results Table 5, Panel A, provides results from our baseline specification, equation (1). The left two columns examine the reduction in the minutes of arrival and departure delays. The omitted category is observations with no incentive at all (None), so coe cients should be interpreted as reductions or increases in delays in minutes relative to routes with no airlines having an incentive in that quarter. Airlines that institute an on-time arrival incentive significantly decrease their arrival delays by 1 minute. To gauge the magnitude of this estimate consider Table 2, Panel B, showing unconditional means on delays: An airline with no incentive has a non-negative arrival delay of minutes, while an airline with an incentive has an arrival delay of 9.95 minutes. The unconditional e ect of the incentive is the di erence 2.12 minutes. Table 5, Panel A, Column 1, shows that after controlling for route-carrier fixed e ects, year-quarter fixed e ects, origin and destination time trends, bankruptcy, mergers, and other carriers flying the same route, instituting an on-time arrival incentive reduces arrival delays by 1.00 minutes, or about 45 percent of the unconditional reduction in arrival delays of 2.12 minutes. For competitors, if a rival airline in the same route initiates an on-time arrival incentive, then the competitor airline that did not introduce the incentive still significantly reduces its arrival delays by 1.21 minutes after controlling for fixed e ects, etc. The unconditional reduction in arrival delays for competitors without an incentive is 2.98 minutes (

20 minutes from Table 2, Panel B). Thus, a rival airline introducing an incentive is associated with a reduction in delays by competitors in the same markets of 41 percent ,ofthe unconditional reduction in delays. If both airline competitors have an on-time arrival incentive, then there is no incremental reduction in delays. In this case, both having an incentive is associated with an increase in delays of 1.19 minutes. The competitive e ect on rival airlines is interesting in principle, if one airline introduces an incentive, the rival airline would be better o not introducing an incentive, while still getting the benefit of a reduction in delays. While the increase in delays when both have an incentive seems anomalous, as we demonstrate later, this e ect is primarily due to United and American Airlines. These two airlines compete directly at O Hare Airport in Chicago, and United instituted its incentive while in bankruptcy following the September 11, 2001 attacks. Though we control for origin and destination time trends, and we control for bankruptcies, there still appears to be a residual e ect that we cannot explain. We return to this issue later. The second column of Panel A repeats the analysis for departure delays and finds similar results. One conclusion from this is that arrival delays are mostly caused by departure delays the reduction in departure delays feeds through to reductions in arrival delays. The third and fourth columns of Table 5, Panel A, consider the fraction of flights delayed more than 15 minutes. Instituting an on-time arrival incentive is associated with a reduction in the fraction of flights delayed more than 15 minutes by a significant 2 percentage points. From Table 2, Panel B, the unconditional reduction in flights delayed more than 15 minutes is 3 percentage points, so again after controlling for fixed e ects, etc., we find that 2/3 of the reduction in delays is attributable to the on-time arrival incentive. For rivals, the fraction of flights delayed more than 15 minutes is also reduced by a significant 2 percentage points. If both competitors on a route have on-time arrival incentives, then the frequency of arrival delays increases by 2 percentage points for both airlines. Essentially, airlines that compete on the same routes and that both have an on-time arrival incentive are more likely to be competing on routes with greater delays (e.g., American and United on flights into or out of O Hare airport in Chicago). 18

21 Another possibility is that airlines might try to game the on-time arrival incentive. When faced with a new incentive, executives might strategically increase the scheduled amount of time for a flight. To examine this, we look at the di erence between the CRS scheduled flight time and the actual elapsed time for flights, in the fifth column of Panel A, Table 5. An airline instituting an on-time arrival incentive decreases its scheduled flight time relative to the actual flight times by 0.13 minutes, or about 7 seconds on average. This result is statistically significant, but small in magnitude. More importantly, when an airline institutes an incentive, it does not appear to be the case that the di erence between scheduled and actual times are increasing, after controlling for fixed e ects, etc. In other words, introducing an incentive does not appear to result in strategic gaming. For rival airlines, however, there is a statistically significant increase in the amount of time aflightisscheduledforandtheactualflighttimeof0.34minutes,or20seconds.therefore, there is some evidence of strategic gaming by rivals in response to an airline introducing an incentive. If both have an on-time arrival incentive, then both will incrementally increase their scheduled flight times by 0.13 minutes, which is again small in magnitude That both airlines increase their scheduled times in a market in response to both having an incentive could suggest some degree of collusive behavior, but again, the e ect is small. In Panel B of Table 5, we restrict attention to problem markets. Here we find e ects that are large in magnitude. Having an on-time arrival incentive is associated with a reduction in arrival delays of 5.58 minutes. The e ect for rivals is much smaller, with a reduction of only 0.66 minutes. When both have an incentive, there is an increase in delays of 0.66 minutes, suggesting that problem markets with multiple carriers with an incentive remain problem markets despite the implementation of the ontime incentive. All of these e ects are similar for departure delays. In the third and fourth columns, we find that having an incentive is associated with a decrease in the frequency of delayed flights of ten percentage points for arrivals and nine percentage points for departures. There are no significant e ects on rivals or when both carriers have incentives. Finally, for the di erence between scheduled and actual flight times, there is an increase associated with having an incentive of 0.66 minutes, so there is some evidence of strategic 19

22 gaming in problem markets. However, the reduction in departure delays of 5.37 minutes is eight times larger, suggesting that the incentive is working as intended in these markets. For rivals, the di erence between scheduled and actual flight times times decrease. One concern with our results is a specific form of endogeneity. In particular, carriers may introduce an incentive when on-time performance is especially poor. If performance mean-reverts, then we may incorrectly ascribe the performance improvement to the incentive rather than simple mean reversion. Also recall that on-time incentives, once introduced, are sticky and remain in force. To address this issue, we first verify that poor performance does predict the introduction of an incentive. Table 6, Panel A, shows that lagged one-year arrival delays predict the introduction of an incentive in both logit and linear probability models. The same is true for individual carriers in Panels B and C. The two exceptions are Alaska Airlines, where two-year lagged arrival delays predict the introduction of an incentive, and United Airlines, which introduced its incentive in bankruptcy following the 9/11 attacks, when it had better on-time performance (possibly due to the reduction in air travel post 9/11). Second, to account for the mean reversion explanation, we use an Ashenfelter dip strategy, where we estimate specifications similar to those in Table 5, but drop observations from both the year prior to and the year of the introduction of an incentive. The idea is that we compare on-time performance post the introduction of an incentive to at least two years prior to the introduction of an incentive, thus removing the e ect of the particularly poor year that may have induced the introduction of the incentive. The results are in Table 7. In Panel A, where we focus on all markets, the results are similar to those in Table 5, Panel A, although slighly smaller in magnitude. In Panel B, where we focus on problem markets, the results are virtually unchanged relative to Table 5, Panel B. Thus, we conclude that mean reversion is unlikely to be driving our results. Another possibility is that carriers might reduce capacity in response to on-time arrival incentives. The idea here is that having more flights and passengers increases the likelihood of delays, either because it takes longer to board more passengers, longer to fuel flights that are more fully loaded, or because with more flights scheduled, any delays will compound throughout the airline system. Therefore, in order to make it easier to achieve an ontime bonus, carriers might reduce capacity. Table 8 considers the response of capacity and 20