mopdf.com

all contributors to the post-HSR reduction in arrival delay minutes. In addition, HSR entryleads to a reduction of the taxi-in time of 1.39 minutes on average at the destination airportsthat are impacted by the HSR.We consider and rule out a variety of alternative explanations for our findings. First,to address the possible contamination from the treatment to the control flights, we test theimpact of the HSR entry on air time and find that the HSR entry does not cause congestionin the air corridor for the control flights. Second, to address the concern that the reductionin arrival delays might be the result of a deliberately prolonged scheduled duration, ratherthan a genuine improvement in OTP, we test the impact of the HSR entry on scheduledduration and rule out this alternative explanation. Third, to examine whether a reductionin the number of passengers on the treatment flights, which leads to faster check-ins, coulddrive our findings, we use a subsample of flights during China’s holiday periods when we areensured that all airports and airlines operate at full capacity. We still find that the HSRentry leads to significant reductions in fight delays. Fourth, to address the possibility that ourresults are driven by some flights with more serious delays being either eliminated or being reassigned with new flight numbers, we focus on a subsample of flights that existed both beforeand after the HSR entry. We find that the improvement in the OTP of treated fights remainsin this subsample of flights. We also consider and rule out other alternative explanations,such as more favorable military/air traffic control and more favorable scheduling to less busytime slots, among others. Finally, we conduct placebo tests using a fictitious treatmentgroup, or a fictitious treatment date, and both placebo tests confirm that the competitioneffects we estimated are not caused by other spurious factors.The richness of our flight data also allows us to better understand the heterogeneity inthe service quality response to the competition from the HSR entry. We find that nonhub airlines and flights on short-to-medium routes (air distance within 1,200 km) are moreresponsive to HSR entry than their respective counterparts. We also extend our analysisto cover the sample period up to September 2015, by when 10 additional HSR lines wereintroduced.4 We find that our results are robust to the extension of the larger sample.Finally, we conduct a back-of-the-envelope calculation of a lower bound of the monetaryvalue of the time saving for air travelers on the routes affected by the HSR entries.This study adds to the literature that examines the causal effects of competition. Nickell (1996), Schmitz Jr (2005) and Matsa (2011) present evidence of the positive effects ofcompetition on productivity and product quality, while Gal-Or (Gal-Or) suggests a negativerelationship. Aghion et al. (2005) and Hashmi (2013) document an inverted-U relationship4Table A1 in the Appendix summarizes the opening dates of all these HSR lines. Many cities wereconnected to Beijing following the completion of some of these new HSR lines. For instance, the Beijing–Guangzhou HSR line was launched on December 26, 2012, and the Shanghai–Kunming HSR line was launchedon September 14, 2014.3

between import competition and innovation, whereas Cornaggia et al. (2015) reveal a negative impact of bank competition on innovation. Our findings also complement the literatureon airline competition. An extensive amount of empirical work shows that competitionwithin the airline industry improves the OTP of flights (Mazzeo, 2003; Rupp et al., 2006;Prince and Simon, 2009, 2015; Greenfield, 2014; Goolsbee and Syverson, 2008). To the bestof our knowledge, ours is the first study that provides causal empirical evidence for how airline OTP is affected by a plausibly exogenous competition shock from a different sub-sectorin the transportation industry.This study also contributes to the growing literature on the economic impacts of transportinfrastructure projects. Much of the literature explored the effects of urban transportationimprovements in roads and railways on urban growth, urban form, congestion, and tradecost (Baum-Snow, 2007; Duranton and Turner, 2011, 2012; Baum-Snow et al., 2017; Donaldson, 2018). In addition, the literature has shown that HSR has a positive influence onintercity mobility (Chen, 2012; Tierney, 2012), market integration (Zheng and Kahn, 2013),population density, and employment (Lin, 2017; Levinson, 2012). However, some studiesargue that HSR primarily benefits large cities, as opposed to small counties (Zheng andKahn, 2013; Qin, 2017). Moreover, recent studies that examined the impacts of HSR on theairline industry focus on the market share and price response (Behrens and Pels, 2012; Yangand Zhang, 2012; Fu et al., 2012). This study contributes to this strand of literature byexamining the causal impacts of China’s HSR on the non-price characteristics of the airlineindustry, which provides policy implications for other countries that may be contemplatingto build a HSR network.The remainder of the paper is structured as follows. In Section 2, we provide a briefbackground on the HSR networks and the airline industry in China. In Section 3, wedescribe our dataset and present summary statistics. In Section 4, we present our empiricalstrategies and the main results. In Section 5, we discuss various alternative explanations andpresent falsification tests. In Section 6, we provide a back-of-the-envelope calculation for alower bound estimate of the time value from the improvement in OTP. Finally, in Section 7,we conclude.2Background on the HSR and Flight Delays in ChinaAfter 20 years of development and expansion, China’s high-speed railways, which aredesigned for speeds of 250 to 350 Kilometers per hour (kph), have become the largest andmost extensively used HSR network in the world. China’s HSR network plan, which is oftendubbed “the Eight Vertical and Eight Horizontal plan,” is based on eight major HSR linesfrom the north to the south (the eight “verticals”), and another eight major HSR lines from4

the east to the west (the eight “horizontals”). Beijing is regarded as the most crucial startingpoint of the vertical lines.The Beijing–Shanghai HSR line is the first medium- and long-haul Beijing-outbound HSRtrack that links Beijing to 26 other domestic destinations (see Table A1 in the Appendix).5 11of the 26 cities are linked with Beijing by non-stop commercial flights.6 The Beijing-ShanghaiHSR operates 45 trips in each direction on a daily basis. The opening of Beijing-ShanghaiHSR reduces the traveling time by train from around 13 hours to 4-5 hours for the 1,318 kmjourney. Given that a direct flight between the two cities takes about two hours of air time,so even with the longer travel time from the city to the airport than to the train station andthe longer boarding time for flights, train travel was clearly a much more time-consumingoption prior to the HSR entry. However, the introduction of the Beijing–Shanghai HSRline on June 30, 2011 changed the situation completely by reducing the HSR travel timesubstantially; moreover, the HSR is almost always punctual by the minute. In this sense, weinterpret the entry of HSR as a serious competition to the airline industry, particularly forshort-to-medium distance journeys.The Chinese airline industry has experienced tremendous growth in the past 30 years,with air passenger traffic growing from 18.2 billion in 1987 to 837.8 billion in 2016.7 Despitethis huge growth, China’s airline market is still in its nascent stage, and suffers from pooroperational efficiency and management. According to the 2018 world airport punctualityreport, none of China’s airports is ranked in the top 20 in terms of OTP.8In this study, we focus on the flights departing BCIA. BCIA has been the world’s second busiest airport in terms of passenger traffic since 2010, but it ranked only 44th out ofChina’s 76 international airports in punctuality as of 2017. Specifically, of the 286,602 flightsdeparting BCIA in 2017, only 53.7% departed on time, and departure delays averaged ataround 48.5 minutes.9 The chronic and often unpredictable delays in BCIA are one of themajor complaints from travelers through BCIA.5Spanning a distance of 117 km, the Beijing–Tianjin HSR line is the first Beijing-outbound HSR. However,owing to the short distance, there are no flights between Beijing and Tianjin.6These 11 cities are Changzhou, Hangzhou, Hefei, Jinan, Nanjing, Ningbo, Qingdao, Shanghai, Wenzhou,Wuxi, and Xuzhou, which are denoted by the red train signs in Figure 1.7Source: 0224 42760.html.8Source:https://www.oag.com/hubfs/Free Reports/Punctuality League/2018/PunctualityReport2018.pdf.9The number is calculated using the data collected from Feichangzhun. Source: Airport.5

3Data and Summary StatisticsThe flight data used in this analysis was obtained from a leading data company thatfocuses on commercial aviation. The dataset in the baseline analysis contains 865,967 nonstop flights, scheduled by 41 airlines, departing from Beijing to 113 domestic destinationsbetween January 1, 2009 and December 25, 2012.10 Figure 1 presents the 113 destinations(denoted by the red, green and black train signs, as well as the blue airport signs) thathave non-stop flights from Beijing. Focusing on the sample period between January 1, 2009and December 25, 2012 ensures that the Beijing–Shanghai HSR line, which opened on June30, 2011, is the only Beijing-outbound HSR in the analysis; it also guarantees a sufficientlylong pre- and post-HSR time window. To ensure that the results can be generalized to thefull sample, we repeat the main analysis using an expanded sample from January 2009 toSeptember 2015. In one of the robustness checks, we also use international flights as a controlgroup.[Figure 1 About Here]For each flight in our sample period, we have the flight number, flight date, scheduleddeparture and arrival times, actual gate departure and gate arrival time stamps, time spenttraveling from the gate to the runway (taxi-out time), time spent traveling to the gateafter landing (taxi-in time), and time spent in the air (air time). We illustrate the variouscomponents of flight duration in Figure 2. Following Prince and Simon (2015), we define aroute as a directional Beijing–destination pair for any carrier that provides non-stop services.For instance, for flight CA1515, the destination city (e.g., Shanghai) refers to a route, CA(China Air) stands for an airline company, and CA1515 represents a flight.[Figure 2 About Here]Following the existing literature, we construct two measures of OTP for both the arrivaland departure delays (Mayer and Sinai, 2003; Goolsbee and Syverson, 2008; Prince andSimon, 2009, 2015). Specifically, Arrival Delay in minutes (ADM) represents the differencebetween the scheduled and the actual arrival times. Arrival Delay 15 minutes (ADD15) isa dummy variable equal to 1 if a flight arrives at the gate at least 15 minutes late, and 0otherwise. We use the same approach to construct Departure Delay in minutes (DDM) andDeparture Delay 15 minutes (DDD15).To address the possibility that airlines could manipulate the OTP by artificially inflatingthe scheduled duration (Mayer and Sinai, 2003; Prince and Simon, 2015), we construct twoalternative measures of OTP: Actual Travel Time (ATT) and Excessive Travel Time (ETT).ATT is the time difference between the scheduled departure time and the actual arrivaltime, which measures the actual travel time because any passenger needs to be at the gate10Cities without direct flights from Beijing are excluded from the analysis.6

before the scheduled departure and will not leave the gate at the destination until the actualarrival time. ETT is the difference between ATT and the minimum feasible travel time. Theminimum feasible travel time refers to the minimum travel time of the same flight observedeach month, which serves as a benchmark for determining the travel time when a flightis free of any external influences such as air congestion, weather shocks, and air corridormilitary controls. Therefore, ETT controls for any unobserved or observed time-varyingexternal influences and is immune to airline scheduling manipulations. In Figure 3, we plotthe distributions of ATT for the Beijing-outbound flights to the 11 destination cities alongthe Beijing-Shanghai HSR, as well as that of the HSR trains.11 Figure 3 shows that the ATTfor flights (denoted by the black lines) exhibits large variations.[Figure 3 About Here]In Table 1, we provide the summary statistics of the OTP measures and other variablesat the individual flight level. In the post-HSR period, the mean values of ADM, DDM,ATT, and ETT increase for both the treatment and control flights, which is a reflection ofthe rapid growth of the passenger travel industry in China, but it is interesting to note thatthe increases in the treatment group are smaller. The summary statistics at the aggregatedairline-route-month level (12,499 airline-route-month observations) are reported in Table A2in the Appendix.[Table 1 About Here]4Empirical Strategies and Main ResultsIn this section, we first present evidence that the HSR entry poses real competition tothe airline industry on the impacted routes. We then describe our empirical strategies, themain empirical results, and various robustness checks.4.1HSR Entry as a Competition Shock: Evidence from SupplySide ResponseWe have argued that HSR is a disruptive transportation technology that poses competition to air travel, particularly for short-to-medium distance journeys. In this subsection,we provide direct evidence that the HSR entry indeed is a competition shock to the airlineindustry by examining the supply-side response of the airlines. Specifically, we examine theimpact of HSR entry on the number of flights operated in a given route at the flight-month,11According to thepercent for departureshandle/10986/31801.the red vertical line inlatest World Bank report, the punctuality rate of HSR service in China is over 98and over 95 percent for arrivals. Source: https://openknowledge.worldbank.org/Therefore, we consider the travel time invariant for HSR travel, which is denoted byFigure 3.7

the airline-route-month, and the route-month levels. To study the average monthly flightsupply responses to the Beijing-Shanghai HSR entry, we run the following regressions:Yi,m α β · Treatmenti · Afterm µi γm i,m ,Yj,d,m α β · Treatmentj,d · Afterm θj ηd γm j,d,m ,Yd,m α β · Treatmentd · Afterm ηd γm d,m ,(1a)(1b)(1c)where i, j, d, and m index the flight number, the airline, the route (or destination), andyear-month, respectively. Yi,m , Yj,d,m , and Yd,m represent the number of flights by flightmonth, by airline-route-month, and by route-month, respectively. Treatmenti , Treatmentj,d ,and Treatmentd are dummy variables that takes value 1 if the flight i, the airline-route (j, d),and the route d, respectively, belongs to the 11 HSR destinations connected to Beijing by theBeijing-Shanghai HSR. Aftert is a dummy variable that takes the value 1 after June 30, 2011,and 0 otherwise. Flight fixed effect µi is included in Eq. (1a); airline fixed effects θj androute (or destination) fixed effects ηd are included in Eq. (1b); and the route (or destination)fixed effects ηd are included in Eq. (1c). Year-month fixed effects γm are included in all threeequations. The standard errors are clustered at the flight-, airline-route-, and route-level inEqs. (1a), (1b) and (1c), respectively.Table 2 reports the results. It shows that the coefficients for the interaction terms arenegative and statistically significant, suggesting that the number of flights to the treateddestination cities decreases by 8.42% ( 1 exp( 0.088)) to 17.47% ( 1 exp( 0.192))more than that of control destination cities in the post-HSR period. The results are consistentwith both the anecdotal evidence and the findings in Fu et al. (2012).12 We consider therelative reduction in flight supply as direct evidence that the HSR entry poses a seriouscompetition shock to the airlines.13[Table 2 About Here]4.2HSR Entry and Flight Delays: Baseline ResultsOur basic specification to examine the causal effects of the Beijing-Shanghai HSR entryon the OTP of the treated flights is a DID regression at individual-flight level:Delayi,j,d,t α β · Treatmenti,j,d · Aftert µi δhour ζt i,j,d,t ,12(2)Source: utes.13However, as we will show in Figure A2 , the overall number of flights departing BCIA has been goingup in this period because the Chinese air travel industry is proliferating.8

where Delayi,j,d,t is one of the six OTP measures for flight i of airline company j departingfrom Beijing to destination d on date t. Treatment is a dummy variable equal to 1 for treatedflights. β is the parameter of interest to be estimated, which captures the difference in theaverage post-HSR delays of a treated flight relative to the post-HSR delays of a control flight.µi refers to the flight fixed effect (flight number), capturing the unobserved factors that mayaffect flight delays at the flight level. The term δhour represents the hour fixed effects for theflight’s scheduled departure time, which account for any hourly variations that may affectflight delays, such as the airport congestion and weather conditions.14 We also include thedate fixed effects ζt to eliminate any seasonal and national trends. The standard errors areclustered at the destination city level to capture the potential heteroskedasticity of the errorterms across the destination cities.[Table 3 About Here]Table 3 presents the estimation results for Eq. (2). The estimated coefficients onTreatment · After are consistently significantly negative in all columns, which suggests thatflights facing the new competition from the HSR entry improve their OTP in the post-HSRperiod relative to control flights. Specifically, at the intensive margin (as shown in Columns1 and 3), on average, the HSR entry reduces the arrival and departure delays for the treatment flights by 2.54 minutes (about 14.51%) and 5.28 minutes (about 14.47%) more than forthe control flights. At the extensive margin (as shown in Columns 2 and 4), treated flightsin the post-HSR entry period are less likely than the control flights to experience arrival(departure, respectively) delays longer than 15 minutes, by 2.5 (3.4, respectively) percentagepoints. Using the alternative measures of OTP in Columns 5 and 6, we find very robustresults indicating that the HSR entry reduces ATT and ETT by 4.73 and 3.92 minutes,respectively.Table A3 in the Appendix reports the estimation results when we include route fixedeffects interacted with the year-month fixed effects and airline fixed effects interacted withthe year-month fixed effects to address any omitted factors at the route (or destination) andairline level. The results are consistent with our baseline results in Table 3.4.3Parallel Pre-Trends and Dynamic Effects of the HSR EntryIn this subsection, we verify the parallel pre-trend assumption that is necessary for thevalidity of the DID approach we used in estimating Eq. (2). We estimate the followingequation to verify the parallel pre-trends between the treatment and control flights, and to14Note that we do observe substantial changes in the scheduled departure time for the same flight becauseof the rapid growth of the commercial airline industry in China in this period. Thus we can include bothflight fixed effects and departure hour fixed effects.9

capture the dynamics of the improvement of the OTP to the entry of the HSR:Delayi,j,d,t α s 5Xβs · Treatmenti,j,d · 1{t Quarters } µi δhour ζt i,j,d,t(3)s 4where t Quarters is a binary indicator which takes value 1 if the date t is in quarters { 4, 3, 2, ., 0, ., 3, 4, 5} before/after June 30, 2011. The coefficient βs measuresthe difference in the response of OTP compared with the first 12 months (benchmark periodfrom January 1, 2009 to December 31, 2009) in our sample period between the treatment andcontrol flights. More specifically, the coefficient β0 measures the immediate response in OTPduring the quarter of the HSR entry. The coefficients β1 , . . . , β5 measure the responses inthe first to the fifth quarter following the entry o

transportation.1 China is a perfect testing ground to analyze the competition between HSR and airlines for several reasons. First, China has the largest and most extensively used HSR network in the world; second, the airline industry in China is rapidly growing both in the number of scheduled