Competition and price dispersion: evidence from airline and high-speed rail competition in China

Abstract

The introduction of the Beijing–Shanghai high-speed rail (Launch) and the Wenzhou bullet-train collision (Crash) in China provided a unique occasion to study the effect of competition on price dispersion. Results show that price dispersion increased by about 48% after the Launch, as airlines offered greater discounts to price-elastic consumers than to price-inelastic consumers. By contrast, price dispersion decreased by about 55% after the Crash, as airlines raised prices more for price-elastic consumers than for price-inelastic consumers.

1. Introduction

Economic predictions for the effect of market competition on price dispersion are often mixed.¹ The standard microeconomic theory argues that more competition will lead to lower price dispersion. This is because a competitive firm cannot price discriminate, as it is a price-taker, but a firm with some monopoly power can (e.g., see Hayes and Ross, 1998). Thus, as more competitors enter a market, incumbent firms find it harder to charge a price above the marginal cost, which leads to smaller price dispersion. However, some authors emphasize the dominance of brand effect in differentiated product markets (Borenstein, 1985; Holmes, 1989; Gale, 1993; Gale and Holmes, 1993).² They argue that consumers in imperfectly competitive markets often have different demand elasticities due to brand preference. Therefore, as competition intensifies, if a firm offers greater discounts for price-elastic consumers than for price-inelastic consumers, the market price may become more dispersed. Moreover, the price dispersion might be sensitive to both the quantity and the quality of competition.

This paper adds to the literature by examining the effect of high-speed rail competition on the airline industry. In particular, the introduction of high-speed rail on the heavily traveled Beijing–Shanghai route enables us to consider the impacts of intensified competition on airline price dispersion. On July 1, 2011, the Beijing–Shanghai high-speed rail system (the Jinghu HSR or the HSR) was officially put into operation (the Launch). The HSR poses a real competition to air travel and thus generates a negative demand shock to the airline industry in the passenger markets between Beijing and Shanghai. HSR also adds a qualitative dimension in the sense that now there is a viable (timewise) substitute for folks with a fear of flying. The data we collected before and after the Launch enabled us to formally evaluate the impact of intensified competition on airfares.

In less than a month, the Wenzhou bullet-train collision (the Crash) occurred on July 23, 2011. Two high-speed trains collided in the suburb of Wenzhou, Zhejiang province, China; 40 people were killed and more than 100 injured. The train involved in the accident—a first-generation bullet train—was traveling at about 153 km (95 miles) per hour—much slower than an HSR train—which averages roughly 350 km (217 miles) per hour.³ This tragedy immediately raised public concern about the safety of China’s HSR and generated a positive demand shock to the airline industry (“China Tries to Calm Fears after Train Crash,” Wall Street Journal, July 26, 2011). The data, therefore, also allow us to study the effects of reduced competition on price dispersion.

Using 26,860 airfare observations for flights along and off the Beijing–Shanghai HSR from June 20 to August 8, 2011, we estimate the effects of competition on price dispersion with a difference-in-difference (DID) approach. For passengers traveling along the HSR, we find that after the Launch airline companies offer greater discounts to price-elastic consumers than to price-inelastic consumers. As a result, price dispersion in airline markets along the HSR increased as the market became more competitive. Our analyses also show that after the Crash airlines raised prices for elastic consumers but barely change prices for inelastic consumers. Consequently, price dispersion decreased as the airline market became less competitive. Meanwhile, we find that the two events barely affected price dispersion for flights not connected by HSR at that time. Taken together, our results indicate that lower-end fares are more responsive to the intensity of competition than higher-end fares and that there exists a positive relationship between competition and dispersion.

Our study is related to an extensive literature on the effects of competition on price dispersion. The empirical studies in this area date at least back to Mathewson (1983) and Dahlby and West (1986), and both find a negative relationship between competition and price dispersion in the insurance market. Due to some special features of the airline industry (customers with different demand elasticities, data availability, etc.), there is a huge literature on airfare dispersion, including the seminal work of Borenstein and Rose (1994) and Stavins (2001), who find a positive relationship between market competition and price dispersion; Gerardi and Shapiro (2009) and Gaggero and Piga (2011), who report a negative effect of competition on price dispersion; and the recent work of Dai et al. (2014) and Kim et al. (2022), among many others.

People have studied the relationship between competition and price dispersion in other markets as well, including the traditional retailing market (especially gasoline market, e.g., Barron et al. (2004), Lewis (2008), and Chandra and Tappata (2011); car market, e.g., Goldberg and Verboven (2001); supermarket industry, e.g., Eden (2018); online retailing market, e.g., Sorensen (2000), Clay et al. (2001), Clemons et al. (2002), and Baye et al. (2004); and agricultural market, e.g., Aker (2010) and Vukina and Zheng (2010), just to name some.

Despite the huge literature on this subject, we have not seen any study that considers both intra- and inter-industry competition, with its inherent quantitative and qualitative dimensions, as we do here with the airlines and HSR industries, which is our first contribution to the literature.

Second, previous studies of price dispersion often rely on intra-industry data and must address simultaneity bias concerns. The simultaneity question arises here because, on the one hand, price dispersion depends on the intensity of competition; on the other hand, the extent of price dispersion determines the market structure through firms’ entry or exit decisions. To solve this issue, both Borenstein and Rose (1994) and Gerardi and Shapiro (2009) employ the instrumental estimation method. However, they obtain conflicting results. In contrast, we avoid the inherent simultaneous endogeneity bias in previous studies by taking advantage of the exogeneity of the two events (i.e., Launch and Crash).⁴ These two inter-industry events generate exogenous demand interventions for the airline market, as the occurrence of the two events does not depend on price dispersion; they do, however, affect competition in airline markets along the HSR. Consequently, our approach enables us to establish the causal effects of competition on airfare without instrumenting for competitive intensity, as do previous studies (cf. Borenstein and Rose, 1994; Gerardi and Shapiro, 2009). Notice that there are several previous studies using deregulations (e.g., Esplin et al. (2020)) or technology innovations (e.g., Aker (2010)) in some industries to study the effect of consequent competition change on dispersion, but our inter-industry events are generally more likely to be exogenous than the intra-industry events they considered.

Finally, our study also contributes to the understanding of the economic outcomes of China’s HSR. China started its HSR project in 1999 and launched its first bullet train in 2008. As of 2020, China’s HSR operating mileage has reached 37,900 km. Despite the rapid development of China’s HSR, empirical studies of its economic outcomes are rare.⁵ In a recent study, Fang et al. (2023) find that the competition from the entry of HSR leads to significant reductions in travel delays on the affected airline routes. Lin (2017) shows that an HSR connection increases city-wide passenger flows by 10% and employment by 7%. Zheng and Kahn (2013) find that China’s network of bullet trains reduces the cost of megacity growth and facilitates market integration, which further stimulates the development of second- and third-tier cities. Our study is among the first to study the economic effects of HSR on China’s airline industry.

The rest of the paper is organized as follows: Section 2 describes institutional features and data. Section 3 constructs several measures for price dispersion. Section 4 examines in detail how airline price dispersion changes with the intensity of competition induced by the two inter-industry events. Section 5 further examines how airlines price discriminate across consumers in different price quantiles following the two events, and Section 6 concludes.

2. Institutional background and data

It is useful to offer some institutional background so the reader understands the qualitative dimension of the airline–HSR competition and appreciates the extent to which the two may be substitutes along the Beijing–Shanghai market. Officially put into operation on July 1, 2011, the Beijing–Shanghai HSR connects China’s two largest cities. The rail line spans 1318 km (820 miles) and covers a region that accounts for 25% of China’s population and 30% of its gross domestic product (GDP). The route has 22 stations, 10 pairs of which are connected by flights (Figure B1 in  Appdendix B).⁶ After the launch, travel by HSR became a substantial competitor for the airline industry. During the sample period, a direct flight from Beijing to Shanghai normally takes two and a half hours and costs, on average, about US$170 for an economy fare.⁷ In comparison, the nonstop HSR from Beijing to Shanghai takes 5 hours—but costs about US$89 for a second-class seat. Therefore, compared with air travel, the HSR provides a more time-consuming (not counting the airport security and check-in formalities) but a less expensive alternative.⁸ One naturally supposes that the HSR provides a more attractive alternative to passengers with more elastic demands than those with less elastic demands. The main body of the analysis is devoted to confirming this conjecture and studies its implications for airfare dispersion.

Although the launch of the HSR was a largely expected event, it posed an exogenous demand shock to the Chinese airline industry for the following reasons.⁹ First, although building the HSR was a strategic decision made by China’s state government, the Ministry of Railways (MOR) is responsible for the development of new rail infrastructure.¹⁰ State planning for the HSR can be traced back to December 1990 when the MOR submitted a proposal to the National Congress to build an HSR between Beijing and Shanghai. The construction of the HSR began in April 2008. In contrast, the Chinese airline industry is overseen by the Civil Aviation Administration of China (CAAC), which is under the administration of the Ministry of Transport. As it took more than 20 years for HSR to go from proposal to launch, it is hard to connect its launch with market fundamentals in the airline industry.

Second, although airlines can adopt some preemptive strategies in response to HSR’s entry, their choices are limited by CAAC regulations. In particular, capacity adjustments are almost impossible. According to the regulations, Chinese airlines can only adjust their major flight schedules twice a year, in the spring and fall.¹¹ The 2011 spring schedule is effective from the end of March to the end of October. In the meantime, airlines can only make temporary adjustments under specific circumstances, such as extreme weather conditions and air-traffic-control matters. Consistent with these regulations, our data show that the total number and capacity of flights on the routes we examine remain unchanged after the launch of the HSR. In the sample, before the Launch, there were 106 flights per day for routes along the HSR and 105 after the Launch. The total available seats were about 37,626 per day before the Launch and 37,499 seats after the Launch.¹²

Our sample includes 16 Chinese domestic airline routes, six of which are not along the HSR and serve as the control group. All ticket information was obtained from Ctrip.com, which is a company listed on the National Association of Securities Dealers Automated Quotations that provides full-scale travel services including transportation tickets, lodging, packaged tours, car rentals, and so on.¹³ In Q3 of 2018, about 38% of the airline tickets were sold through Ctrip.com, which was ranked No. 1 in China, in terms of market share. Like their counterparts in the United States (e.g., Expedia.com), Chinese travel agencies, traditional and online alike, obtain ticket information from TravelSky, the nation’s monopoly state-owned global distribution system (GDS). For all airline inquiries made on Ctrip.com, the bottom of the confirmation page reads “All flight information is provided by China Civil Aviation Information Network Co. Ltd (TravelSky Technology Limited).”¹⁴ The CAAC regulations allow all travel agencies to negotiate airline commissions for all ticket sales after October 2008. As a major online travel agency, Ctrip.com receives a commission of up to 10% of the airfare.¹⁵ These facts warrant that the airfare information on Ctrip.com pertains to price movement made by the airlines rather than Ctrip.com.¹⁶

We collected our data using a web spider on a daily basis. For each flight, we collected information such as airfare for economy seats with the number of days booked in advance, departure time, and airplane type. We obtain other information, such as the average population and per capita GDP of cities, from China’s Sixth National Population Census, which was conducted in 2010. Tables C1 and C2 in Appendix C define the main variables used in the study and present their summary statistics, respectively.

We use the two reverse events, i.e., the Launch and the Crash, to identify the effects of competition on price dispersion. Specifically, we use data for the Launch, which are contained in the subsample from June 20 to July 22, 2011, to study how the intensified competition affects price dispersion, and for the Crash, which is contained in the subsample from July 1 to August 8, 2011,¹⁷ to study how airline companies respond to weakened competition from the HSR.¹⁸  Figure B2 in  Appdendix B shows the timeline for the two events and the subsamples used in the study. Each subsample spans about 2 weeks before and 2 weeks after the relevant event.¹⁹ One thus should interpret our results as primarily short-run effects; by focusing on short-run effects, we are in a better position to isolate the effects of demand interventions on price dispersion (because cost-side factors are unlikely to have changed much during the sample period).

To get a rough idea of whether the price dispersion changes around the two events, we first calculate the coefficients of variation (CV in short, equal to the standard deviation divided by the mean), ranges, and Gini coefficients of airfares for a given carrier and route on a given day, and we then plot the CVs, ranges, and Gini averaged over carrier–routes around the two events in Figures B3 and B4. From these plots, one can roughly see that the price dispersion measures increase after the Launch, while they decrease after the Crash. However, we want to point out that these dispersion measures do not separate out dispersion coming from the cost of serving different passengers, and we are going to get some better measures in the next section.

We end this section by demonstrating that Chinese airline pricing is a market-driven process. The CAAC began to deregulate airline pricing in 1992, and since 2004, Chinese airlines have been allowed to set ticket prices ranging from 45% below to 25% above the regulated benchmark. However, this rule is frequently violated due to ineffective enforcement. In a study of Chinese airline deregulation, Zhang and Round (2008) conclude that “the pricing of airfares in China’ s domestic market has, de facto, been deregulated, without a formal Deregulation Act such as in the US.” More recently, facing increasing competition from HSR, the CAAC took further steps to relax airfare restrictions—for example, in October 2013, the CAAC removed the lower bound restriction of 45% below the benchmark price on 31 domestic routes.²⁰ Appendix A contains a detailed study of how Chinese airlines price according to market conditions. The next section details our measurement of price dispersion.

3. Measuring price dispersion

Price dispersion normally refers to the variation in prices caused by price discrimination. One would only consider price dispersion to be present if price variation remains even after controlling for differences in the cost of serving different passengers. In the airline industry, both “systematic” peak-load pricing and “stochastic” peak-load pricing are considered to be cost related and can induce a distribution of prices rather than a single price. Borenstein and Rose (1994) make a clear distinction between the two types of peak-load pricing. Systematic peak-load pricing reflects variations in the expected shadow costs of capacity at the time the flight is scheduled and before any ticket has been sold, whereas stochastic peak-load pricing reflects demand uncertainty about individual flights that is resolved as the departure date approaches and tickets are sold. See Lott and Roberts (1991), Hayes and Ross (1998), and Dana (1999) for a more detailed discussion of the two pricing strategies.

Following Lach (2002) and Lewis (2008), we take advantage of the panel structure of the data and use ticket fixed effects to obtain measures of “pure” price dispersion. Specifically, we use the information on the exact time of departure and number of days booked before departure to capture some aspects of peak-load pricing and |$Flight{s_{jkt}}$|—the remaining flights at the time the ticket is booked—to account for parts of the stochastic demand pricing. Later, we will construct measures of price dispersion using residuals from estimating the following panel fixed-effects model:

where |${P_{ijkst}}$| is the log price of the flight that departs at clock time s on date t by carrier i on route j and booked k days in advance (k = 0, 1, 2, 3, 5, 7, 15, and 30).²¹  |$Even{t_t}$| refers to the two key events, i.e., |$Launc{h_t}$|and|$Cras{h_t}$|⁠. |$Launc{h_t}$| is set to 1 if the flight is scheduled to depart after July 1—when the HSR is officially launched—and zero otherwise. In contrast, |$Cras{h_t}$| equals 1 if the ticket is scheduled to depart after July 23—when the Crash occurred—and zero otherwise. |$Trea{t_j}$| is equal to 1 if the route is on the HSR and 0 otherwise. Notice that |$Trea{t_j}$| should have appeared separately in equation (1), but it is absorbed into the fixed-effect term |${u_{ijks}}$|⁠.

It should be noted that, unlike the Launch, the Crash is an unexpected event and airlines were unable to adjust their prices beforehand. Therefore, we should use the booking day instead of the departure day to define the Crash. However, as we will see toward the end of the section, the price dispersion is for an airline route on a particular departure day. To avoid overestimating the price dispersion after the Crash, we omitted observations that were booked before but scheduled to fly after the Crash in calculating price dispersion and thus define the Crash by the departure date.

Previous studies have established a weekend effect on airline pricing (see, e.g., Mantin and Koo (2010)), and thus, we include the dummy variable |$Weeken{d_t}$| to capture the difference between weekend and weekday pricing. Sometimes, invariant aspects of congestion pricing will be picked up by the ticket fixed effect |${u_{ijks}}$|⁠. In addition, we adopt the DID method to control for possible trends in airfare during the sample period.

Equation (1) was estimated for each event separately. The Launch sample is based on a subsample from June 20 to July 22, right before the Crash, whereas the Crash sample is from July 1 to August 8. Results are presented in Table C3. We find that both events have a significant impact on fares for flights along the HSR. In particular, ticket prices on average dropped by 11.8%, or about Renminbi (RMB) 94, after the Launch and increased by 13.3%, or about RMB 106, after the Crash.²²

Our interest here is in residual variation, which will be used to construct measures of price dispersion. Each residual, |${\hat \varepsilon _{ijkst}}$|⁠, simply reveals whether a ticket price was above or below its expected level relative to the average price of the reference group. The residual prices can be compared across tickets because scale (output) effects are absorbed in the overall constant. Furthermore, we interpret the residuals as the discriminatory airfares charged by different airlines for the same service after controlling for heterogeneities.

Following Lach (2002) and Lewis (2008), we use the residuals to construct three alternative measures of price dispersion: standard deviation, range, and interquartile range.²³ To get results comparable with the existing literature (see, e.g., Borenstein and Rose, 1994; Gerardi and Shapiro, 2009), we calculate a price dispersion measure |$P{D_{It}}$| for each carrier–route I on each day t. Take standard deviation, for example,|$P{D_{It}}$| is just the standard deviation of all those residuals |${\hat \varepsilon _{ijkst}}$| with departing date t and carrier–route combination |$ij$| = I. To be more specific, all the fare residuals (booked k days in advance, for any k) of Air China departing on July 1, 2011 (departing on any time of that day) for the Beijing–Shanghai route were used to calculate |$P{D_{It}}$| for I = Air-China–Beijing–Shanghai and t = July 1, 2011. In the end, we obtained 1541 observations of dispersion for the Launch and 1608 for the Crash.

It should be noted that the existing studies use transaction prices to construct measures of price dispersion (e.g., from Databank 1B (DB1B) of the Department of Transportation’s Origin and Destination Survey data, see, e.g., Borenstein and Rose, 1994; Gerardi and Shapiro, 2009). Price dispersion thus comes from variations across tickets for different flights (e.g., day of the week and time of the day), bought at different dates (in-advance purchase and weekday vs. weekend) and time (business vs. off-business hours), and more recently add-ons (e.g., upgrade from Economy to Economy Plus). By contrast, our measures of price dispersion are based on the price residuals, which are generally “cleaner” because price dispersion normally refers to the variation in prices caused by price discrimination but not the differences in the cost of serving different passengers, and we have taken out the cost variations measured by terms like |$Weeken{d_t}$| and |${u_{ijks}}$| from the transaction prices. In the robustness check of the next section, we also redo the analyses using price dispersion based on prices directly.

Table C4 presents the time average for price-dispersion measures divided by the average price for each sample. The mean airfare is RMB 777 for the Launch sample and RMB 786 for the Crash sample. Two points are in order. First, alternative price-dispersion measures for the two events are comparable to each other. Second, price dispersion continues to represent a sizable portion of airfares after controlling for heterogeneity: roughly speaking, dispersion of the middle 50% of airfares is about 14–15% of the average price. The difference between the 95th percentile and the 5th percentile is about 30% of average airfare. For comparison, Chinese airfare dispersion is lower than that found in US and European studies (e.g., in terms of Gini coefficient of airfares: 36% in Borenstein and Rose (1994), 44% in Gerardi and Shapiro (2009), and 70% in Gaggero and Piga (2011)).

Tables C5 and C6 further decompose price dispersion by treatment and control groups and before and after each event. As we observe from the two tables, while alternative measures of price dispersion for the treatment group (i.e., airlines along the HSR) increase after the Launch and decrease after the Crash, those for the control group decrease after the Launch and increase after the Crash.

In the next section, we perform more detailed econometric analyses to examine the effects of demand interventions on airfare dispersion.

4. Demand interventions and price dispersion

Given the presence and significance of the price dispersion (after controlling for ticket heterogeneities) in China’s airline market reported earlier, we now examine how price dispersion changes with more or less intensified market competition.

The key feature of our study is that we identify the effects of competition on price dispersion by the two exogenous intra-industry events, which leads to the following DID panel regression model:

where |$\ln P{D_{It}}$| represents log price dispersion for the airline–route combination I on departure date t.²⁴ Independent variables include the event dummy and the interaction term between the event dummy and the treatment group indicator. Similar to ), we control for the fixed effects of flight, route, and days prebooked through the fixed-effect term (⁠|${v_I}$|⁠).|$Mflight{s_{It}}$| is the average number of residual flights across different days prior to departure on departure date t. Finally, |${\omega _{It}}$| is the disturbance term.

As we argue earlier, the Crash made bullet trains a less competitive transportation mode than air travel, at least in the short period following the accident. Therefore, the Crash poses an exogenous demand shock that reduces competition among airlines along the HSR. Table C7 presents results for the effect of the Crash (i.e., less competition) on price dispersion. The first four columns are estimated for the full sample (i.e., the 39-day Crash sample). The most striking result is that price dispersion for airfare along the Jinghu HSR decreased significantly after the Crash and ensuing (decreased) competition from the HSR. The result is robust with alternative measures of price dispersion. For example, column (4) shows that, roughly speaking, price dispersion along the HSR dropped by about 55% (⁠|${\alpha _1} + {\alpha _2}$|⁠) despite the fact that price dispersion for routes that were not connected by HSR rose by about 33% (⁠|${\alpha _1}$|⁠) during the same period.

4.1. Robustness checks

As always with event studies, one needs to be careful about defining the study window (MacKinlay, 1997). Therefore, we first do a robustness check with different windows. We then redo the analyses to see the effects of the events on price dispersion measures based on airfares directly.

As the first robustness check, we estimate the same set of models using two subsamples with balanced observations before and after the Crash. The first subsample consists of observations that span 1 week before and 1 week after the Crash, and the second subsample of observations that span 3 days before and 3 days after the Crash. The results are presented in the second and last four columns of Table C7. Estimates for the interaction term between Crash and Treat are consistent and significantly negative, which indicates a significant and positive effect of competition on price dispersion.

Turning to the impact of the Launch on price dispersion, in contrast to the Crash, the Launch increased competition among airlines along the HSR simply because those who previously traveled by air between major cities on the Beijing–Shanghai route now had an alternative and competitive mode of transportation.

Table C8 presents results for the effect of the Launch on price dispersion. The first four columns of the table are estimated for the full sample (i.e., the 32-day Launch sample), and the rest are estimated with different study windows as robustness checks. Estimates for the cross-product term (Launch*Treat) are all significantly positive except for the last three columns, which indicates that airfare is more dispersed as a result of increased competition among airlines. Focusing on columns (4), (8), and (12) of Table C8, the extent of increased price dispersion is quite remarkable: a roughly 48% (⁠|${\alpha _1} + {\alpha _2}$|⁠) increase with the full 39-day sample, 24% with the 15-day sample, and 16% with the 7-day sample under alternative measures of price dispersion. One might argue that the Launch is not completely exogenous, as is the Crash. Even so, we can still justify our results using a regression discontinuity design (RDD) argument,²⁵ as we restrict the sample to a relatively narrow time window. In RDD, assignment to the treatment is determined by a running variable—in this case, time—on either side of a fixed threshold, the Launch date. Unlike a traditional RDD with a time trend, our analysis here can be viewed as an RDD coupled with a DID method, see, e.g., Davis (2008) and Chen and Whalley (2012) for similar arguments.

Next, we study how our results will change with dispersion measures based on airfares directly instead of based on residuals from . To that end, we construct dispersion measures directly from all the available airfares for a given carrier–route combination I and departure date t. We then re-estimate the effects of the two events using but with the new dispersion measures.

Tables C9 and C10 report the estimated effects of the Crash and the Launch on measures of price dispersions based on airfares directly using the full sample. Comparing the coefficients of |$Even{t_t}Trea{t_I}$| in the first four columns of Table C9 with those of Table C7, one can see that Crash has a more negative effect on dispersion with the new measures. Similarly, the corresponding results in Tables C8 and C10 generally show a more positive effect of Launch on dispersion with the new measures (except one measure about the same as before). These results are not surprising as the dispersion measures on airfares directly also contain dispersion coming from the cost of serving different passengers.

Taken together, we identify a positive relationship between competition and price dispersion in China’s airline markets; our estimation results are consistent with those of Borenstein and Rose (1994). However, our study differs from Borenstein and Rose’s in at least two respects: first, our estimation is based on panel data, whereas Borenstein and Rose use cross-sectional data. Second, they adopt instrumental variables to solve the issue of simultaneity bias between price dispersion and competition intensity, whereas we rely on two exogenous demand shocks to identify the effects of competition on price dispersion. Our results stand in contrast to those of Gerardi and Shapiro (2009), who find a negative relationship between competition and price dispersion. Next, we examine price discrimination across the distribution of consumers across different price quantiles.

5. How do airlines price discriminate across consumers in different price quantiles?

To shed more light on the relationship between competition and price dispersion we established in the previous section, we employ a fixed-effect panel quantile regression to further examine how airline companies price discriminate among consumers in different price quantiles when facing the demand shocks induced by the two inter-industry events. Compared to linear regression, quantile regression could capture a more detailed relationship between the conditional distribution of ticket price and the intensity of competition among the airlines because we can run a quantile regression on any quantile of the price distribution. In addition, estimators of quantile regression are more robust because it minimizes the absolute deviation (Koenker and Bassett, 1978).

Because the unobserved effect term |${u_{ijks}}$| may be correlated with the independent variables, we first estimate |${u_{ijks}}$| in Equation 2 using the between equation after estimating first the αs with the fixed-effect estimator and then subtract it from the log price to get the residual price.²⁶ We then apply a quantile regression to estimate the effects of the events on different quantiles of the residual prices:

where |${\widetilde {\ln {p_{ijkst}}}^q} = \ln {p_{ijkst}} - {\hat u_{ijks}}$| are the residual prices and q = 10, 20, 25, 75, 80, and 90 are the |${q^{{\rm{th}}}}$|percentile of the residual prices. As before, explanatory variables include event dummies, interaction terms, and time-varying variables such as flight and weekend. The previous two-step fixed-effect panel quantile regression method is justified by Canay (2011).

5.1. Effect of the crash

Table C11 reports the estimation results from for the Crash. As we observe in the third row of the table, coefficients for the interaction between the Crash and treatment dummies are significantly positive and decrease sequentially as the quantiles increase, except for that of the 50th percentile (or the median quantile). Further tests show that differences in coefficients for different percentiles are statistically significant. Thus, after the Crash, airlines on the HSR route increase fares by more for consumers at lower quantiles than for those at higher quantiles. In particular, airfares for passengers in the |${10^{{\rm{th}}}}$|quantile increase by 40%, whereas those for the |${90^{{\rm{th}}}}$|increase by 22%. The Crash alleviates competitive pressure on airlines from HSR, at least temporarily. Therefore, our results reveal that when competition is reduced, airlines charge higher prices for price-elastic consumers than for price-inelastic consumers, which leads to more dispersed airfares.

5.2. Effect of the launch

Turning to Table C12, the third row shows that coefficients for the interaction term between Launch and treatment dummies are significantly negative and increase as the quantiles increase after the Launch.²⁷ In particular, airfares for passengers in the lower |${10^{{\rm{th}}}}$|quantile decrease by 53.6%, whereas those for the |${90^{{\rm{th}}}}$|only decrease by 1.4%. A similar statistical test also finds significant differences in coefficients for different quantiles. Therefore, after the Launch, airlines along the HSR discount more steeply for consumers in the lower quantiles than for those in the higher quantiles. As the prices paid by price-elastic (leisure) passengers are mostly located in the lower quantiles, whereas the prices paid by price-inelastic (business) passengers are mostly located in the higher quantiles, we can interpret the previous relationship as demonstrating that when competition becomes more intense due to HSR’s entry, airlines discount more steeply for consumers with elastic demand than for consumers with inelastic demand.

To summarize, results from Tables C11 and C12 show that airlines adopt a consistent price strategy when facing demand interventions caused by the two inter-industry events. On the one hand, the airlines lower prices more for price-elastic consumers than for price-inelastic consumers when competition is intensified and charge more for price-elastic consumers than for price-inelastic consumers when competition is relaxed. These price-discrimination strategies give rise to the positive relationship between competition and price dispersion established in the earlier section—i.e., airfares become more dispersed after the Launch (more competition) and less dispersed after the Crash (less competition).

6. Conclusions

The effect of competition on pricing structure/strategy is intriguing for policymakers, academics, and market players. Market competition often evolves gradually as firms enter and exit; however, sometimes radical competitive changes take place via, for instance, the introduction of a substitute. We use such an event, namely, the introduction of high-speed rail in China, to examine the quantitative and qualitative pressures on the Chinese airline market in terms of the impacts on price dispersion. Besides focusing on China, the analysis enables us to add to the broader literature on competition and price dispersion.

Specifically, in this paper, we study how competition affects price discrimination in the Chinese airline market. Using different fares posted online on different days before departure for 26,860 flights operated on and off the Beijing–Shanghai HSR during the period of June 20 to August 9, 2011, we find that after the launch of HSR, airlines discount airfare more for price-elastic consumers than for price-inelastic consumers, which leads to more dispersed airfares in the market. In contrast, after the Wenzhou bullet-train collision, airlines raise airfare more for price-elastic consumers than for price-inelastic consumers, which decreases price dispersion in the market. Overall, we establish a positive relationship between competition and price dispersion in the Chinese airline market. These findings are consistent with empirical results for US airlines, such as Borenstein and Rose (1994) and Stavins (2001).²⁸

Two novel aspects of our study are worth emphasizing. First, we take advantage of the exogeneity of the two inter-industry events—the launch of HSR and the Wenzhou bullet-train crash—for identification, which avoids the simultaneity bias inherent in most of the previous studies on price dispersion and market competition (Gerardi and Shapiro, 2009). Second, while the HSR launch diverts some passengers from the airline market and increases airline competition, the Wenzhou crash offsets the downward demand intervention caused by HSR and, to some extent, reduces competition among airlines. The examination of the two reverse events cross-validates our main findings from two different directions.

Our study of competition between HSR and air travel could use further refinements. For example, we only observe scrapped posted fares of economy fares prior to departure at various points (30, 15, 7, 5, 3, 2, 1, and 0 days) but not the distribution of tickets actually sold, and thus, we do not observe the true distribution of fares as Borenstein and Rose (1994) did in their study of US airlines. In addition, our study does not capture the interaction between the two transportation modes because the Chinese railway industry is heavily regulated and train fares are fixed during our sample period. Recently, the Chinese railway industry began to introduce and expand ticket discounts to “adapt to market demand.”²⁹ Furthermore, one could argue that the impact of the crash is capturing short-term effects, and long-term effects could well be different. The applicability of similar analyses to other jurisdictions with changing competition in the transportation industry and elsewhere will inform market participants and policymakers.

Acknowledgments

We benefit from the comments of Yongmin Chen, Rajeev Goel, Li Gan, Myongjin Kim, Bill Nelson, Imai Susumu, Matthijs R. Wildenbeest, and seminar participants at Beijing University, Shandong University, Xiamen University, and the 2015 International Conference of Industrial Organization. All errors remain ours.

Funding

This work was supported by the Humanities and Social Sciences Fund, Ministry of Education of China (20YJA790087); National Natural Science Foundation of China (72173029); Innovative Research Groups Project of the National Natural Science Foundation of China (72121002); and the Innovation Program of Shanghai Municipal Education Commission (2023SKZD01).

Footnotes

References

Appendix A: Chinese Airline Pricing: A Hedonic Regression

To see how Chinese airlines respond to market factors, we consider the following hedonic regression model:

where |${P_{ijkst}}$| is the log price of the flight that departs at clock time s on date t by carrier i on route j and booked k days in advance (k = 0, 1, 2, 3, 5, 7, 15, and 30). In the full specification of the model, we control for a host of important ticket, flight, and route characteristics related to airline pricing, in addition to day-of-week fixed effects, weekly fixed effects, carrier fixed effects, and route fixed effects. Detailed definitions of control variables |${X_{jikst}}$| are listed in Table C1.

Equation (A4) was estimated with ordinary least square (OLS), and the results are presented in Table A. As the results with different specifications are qualitatively the same, we focus on the last set (column) with full specifications. As we observe, Chinese airlines set prices in response to important ticket, flight, and route characteristics. Like US passengers, on average, Chinese travelers are expected to pay less (RMB 49) for evening/overnight tickets and more (RMB 36) for peak-time tickets, which represents 6.2% and 4.5% of the sample mean (RMB 794), respectively. Also as in the United States, passengers in China who travel on a more concentrated route or with an airline that has a larger market share are expected to pay more, and the earlier a passenger books the flight, the less she pays for it. A simple calculation shows that Chinese travelers pay the lowest price for flights that are booked about 15 days prior to departure.³⁰Finally, as expected, distance and income generally have positive effects on airline pricing in the Chinese airline market, which is similar to what Stavins (2001) found in the United States.

However, unlike in the United States (Mantin and Koo, 2010), Chinese passengers pay less (RMB 20) for weekend flights. Interestingly, airfares for Chinese flights that have a hub in the origin or destination are more expensive but statistically insignificant, in contrast to the United States, where they are 2% cheaper (Stavins, 2001). As shown in the first column of Table A, passengers in China on a non-state-owned airline are expected to save up to 23% or RMB 183. This is probably because as new entrants to China’s airline market, non-state-owned airlines compete with state-owned airlines by adopting a low-price strategy similar to that of low-cost carriers in the United States.

Turning to key event variables, our estimations reveal strong evidence for the two events’ effects on airfare. Ceteris paribus, before the Launch, the average airfare for airlines along the HSR is about RMB 158 (19.8%) higher than for control-group airlines. The launch of the HSR lowers the average airfare by about RMB 192 (about 24.2%). In contrast, before the Crash, the average airfare for airlines along the HSR is about 8.7% lower than for control-group airlines; the Crash triggers airlines to raise airfares by RMB 188 (about 23.7%). Our analysis thus establishes that the two events represent significant demand interventions for airline pricing.