Market Competition and Price Dispersion:Evidence. from the Airline Market along the Beijing-Shanghai. High-Speed Railway

1 Market Competition and Price Dispersion:Evidence from the Airline Market along the Beijing-Shanghai High-Speed Railway Han Lia,Lan Zhangb, Feng Wei...
Author: Douglas Hall
19 downloads 0 Views 1MB Size
1

Market Competition and Price Dispersion:Evidence from the Airline Market along the Beijing-Shanghai High-Speed Railway Han Lia,Lan Zhangb, Feng Weic, Zaichao Dud

Abstract: In this paper, we investigate how the exogenous demand intervention caused by the introduction of the Beijing-Shanghai High-Speed Railway affects airline market competition along the route. We find that: (1) increased competition among the airlines increases price dispersion; (2) the price dispersion increases because the airline companies discount more on the price-elastic consumers than on the price-inelastic consumers. Further investigation based on another exogenous demand intervention caused by the Wenzhou Train Crash reinforces our conclusions.

JEL: D43, L11, L93

a.

Corresponding author. Associate Professor, Research Institute of Economics and management, Southwestern University of Finance and Economics, Chengdu, P. R. China, [email protected]

b. Associate

Professor, Research Institute of Economics and management, Southwestern University of Finance and Economics, Chengdu, P. R. China

c.

PhD Student, Research Institute of Economics and management, Southwestern University of Finance and Economics, Chengdu, P. R. China

d. Associate

Professor, Research Institute of Economics and management, Southwestern University of Finance and Economics, Chengdu, P. R. China

2

I.

Introduction

Theoretical predictions on the impact of market competition on price dispersion are conflicting. The standard microeconomic theory argues that a competitive firm cannot price-discriminate because it is a price-taker but a firm with some monopoly power can. As more competitors enter a market, the incumbent firm will find it more difficult to charge a price over the marginal cost. Thus, as the competition increases, the market price will converge to the marginal cost, which leads to a declining price dispersion. However, some authors emphasize the dominance of brand effect in differentiated product markets (Gale, 1993; Borenstein, 1985; Holmes, 1989; Gale and Holmes, 1993). 1They observe that consumers in imperfectly competitive markets often have different demand elasticities due to brand preferences. When the competition intensifies, if a firm discounts more prices on the elastic consumers than on the inelastic consumers, the market price may become more dispersed. In this paper, we study the effects of demand interventions caused by two natural events on price dispersion in China’s airline market. On July 1st 2011, the Beijing-shanghai High-speed Railway (abbreviated as Jinghu HSR thereafter) was officially introduced between Beijing and Shanghai, which was previously dominated by airline transportation. The second event is the subsequent occurrence of the Wenzhou High-speed train crash on July 23rd, which cushioned the downward demand intervention generated by the introduction of the HSR to the airline market along the HSR. Using the data collected from June 20th to August 3rd, 2011, we find that the airline companies discount more prices to the elastic consumers than to the inelastic consumers after the introduction of the Jinghu HSR. As a result, the price dispersion in the airline market along the HSR increases as the market becomes more competitive. Subsequent occurrence of the Wenzhou accident reduced the competition in the airline market and caused the airline companies to raise the prices to the elastic consumers but keep the prices to the inelastic consumers barely changed. Consequently, the price dispersion is reduced as the competition in the airline market decreases. Overall, we established a positive relationship between market competition and price dispersion in the airline market along the Jinghu HSR. Our study is related to the literature on the relationship between market structure and price dispersion, which, so far, has focused mainly on the U.S. and European markets. Borenstein and Rose (BR, 1994), in the seminal paper on price dispersion in the airline market, find a positive relationship between market competition and price dispersion. Similarly, Starvins (2001) finds that price dispersion attributed to ticket restrictions increases as the airline markets become more competitive. More recently, using panel data on U.S. domestic airline prices from 1993:Q1 through 2006:Q3, Gerardi and Shapiro (GS, 2009) find that competition has a negative effect on price dispersion which supports the prediction of the traditional microeconomic theory. Gaggero and Piga (2011) analyze the intertemporal price dispersion in the airline markets connecting the UK and the Republic of Ireland and find competition hinders the airlines’ ability to price discriminate. 1

Alternatively, the brand loyalty may be derived from consumer search behavior as studied in the search literature. Bay, et al. (2006) provides a nice review over both the theory and the empirics of the literature.

3

As always in the literature, we have to deal with the endogeneity issue caused by simultaneity.2 Both BR (1994) and GS (2009) employ the instrumental estimation method. However, as pointed out by GS (2009), the positive relationship between price dispersion and market competition found by BR (1994) is biased due to the omitted variables issue. GS (2009) reconcile their results with that of BR (1994) by introducing a fixed effect panel model. Although GS’s approach has many advantages, the results they obtained are still contingent on the use of instrumental variables. In contrast, we use the two natural experiments----the opening of the Jinghu HSR and the Wenzhou accident—to solve the endogeneity problem arises in the paper. The two events generate exogenous demand interventions to the airline market since the occurrence of two events does not depend on the price dispersion but they do affect the market competition in the airline market along the Jinghu HSR. Therefore, our estimation results are free of instrumental variables. The rest of the paper is organized as follows. Section 2 describes the data. Section 3 first illustrates the key findings and then followed by a rigorous econometric examination. Section 4 further studies how the airlines adjust their pricing upon the demand interventions. Section 5 provides some robustness checks and Section 6 concludes the paper. II.

Data Sources and Data Construction

The Beijing-Shanghai High-speed Railway connects Beijing and Shanghai, China’s two largest cities. Completed in 2010 and put to use in 2011, the rail line is 1,318 kilometers (820 miles) long. There are 22 stations along the lines, among them 7 stations (cities) are also connected by 10 pairs of air routes.3 After the introduction of the rail line, the train ride between Beijing and Shanghai is shortened from 10 hours to 5 hours.

2

The simultaneity arises in the context because, on the one hand, the price dispersion depends on the intensity of competition; on the other hand, the extent of price dispersion determines the market structure through firms’ entrance or exit decisions. 3 The 10 routes are Beijing-Shanghai, Tianjin-Shanghai, Beijing-Wuxi, Beijing-Nanjing, Tianjin-Nanjing, Jinan-Shanghai, Beijing-Xuzhou, Jinan-Nanjing Beijing-Jinan, and Nanjing-Shanghai.

4

Figure 1: Beijing-Shanghai High-speed Railway Map

In China, the train ticket price is regulated by the Ministry of Railways. During our sample period, the ticket price from Beijing to Shanghai is fixed at US$149 per person of first class and US$89 per person of second seat. In contrast, As observed by Zhang and Round, “„the pricing of air fares in China’s domestic market has, de facto, been deregulated, without a formal Deregulation Act such as in the US.” (Zhang and Round, 2008) We obtain the ticket and price information from Ctrip (http://www.ctrip.com), which is a China-based online travel agency. In 2010, Ctrip.com sold out about 36% of the tickets online, ranking No.1 nationwide in terms of market share. 4For each day between June 20th and August 3rd, 2011, we collect data at two separate time slots (i.e., 0:00a.m.-0:30a.m. and 8:00p.m.-11:00p.m.) for all direct flights along the Jinghu High-Speed Railway at Ctrip.com. The information we collected include fares for tickets 0, 1, 3, 5, 15 and 30 days booked in advance, departure time, airplane type and left-over flights, etc. The fares posted on Ctrip reflect the prices charged by the airline companies because Ctrip charges a 4-5 percent of the face value for each ticket sold through its website. Other information such as population and per capital GDP is from the Sixth National Population Census conducted in 2010. In table 2, we compare the duration and fare between air transport and High-speed train among the major cities on the Jinghu HSR. The table reveals that, comparing with air transport, HSR takes more time but costs less. Therefore, HSR is substitutable for air transport and the introduction of it is supposed to make a significant impact on the demand for air transport. In light of the above observation, we analyze how demand interventions created by the HSR affect airline price dispersion through two major events, i.e., the opening of the HSR and the Wenzhou Accident. TABLE 2 The Comparison of the Travel Time and Fare between the HSR and the Airlines Distance

Routes 4

Stops of

Number of

High-Speed train Time Fare

http://travel.people.com.cn/GB/16006927.html

Airline Time

Fare

Ratio of

Ratio of

5

(kilometers)

HSR

Airlines

(Minute)

(Yuan)

(Minute)

(Yuan)

time

fare

145 185 280

55 50 65

800 630 800

1.45 1.84 2.15

0.18 0.29 0.35

NJ-SH BJ-JN JN-NJ

273 412 579

88 128 101

1 1 1

80 92 140

BJ-XZ JN-SH TJ-NJ BJ-NJ

650 852 907 981

73 108 17 120

2 3 2 5

164 201 220 250

310 400 405 445

80 80 90 120

690 760 880 1010

2.05 2.51 2.44 2.08

0.45 0.53 0.46 0.44

BJ-WX TJ-SH BJ-SH

1120 1133 1178

34 21 90

4 5 6

310 300 288

515 515 555

100 120 150

1080 1030 1130

3.10 2.50 1.92

0.48 0.50 0.49

Note: Ratio of time=Time of HST/Time of airline;Ratio of fare= Fare of HST/Fare of airline. NJ stands for Nanjing; SH stands for ShangHai; BJ stands for BeiJing; JN stands for JiNan; XZ stands for XuZhou;TJ stands for

TianJin; WX stands for WuXi.

Data Source: The Beijing-Shanghai high-speed rail timetable (July 1, 2011 Edition); National Airport's flight schedule (2011 summer and fall version).

We study a panel data. The identity i of the panel is defined by three parts,the number of days booked in advance k(k=0, 1, 3, 5, 15, 30), the route(j) and the flight (l). Therefore, the same flight booked with different days in advance has different identities. For example, the CA176 flight departuring on June 25 at 20:50 from Beijing to Shanghai booked with three days, five days or zero days in advance in the panel is assigned with different identities. The time dimension of the panel data is defined by the date that the flight departures. With the defined data structure, after eliminating missing data, we obtain a sample with 31,075 observations. Among them, the observations before the opening of the HSR is 3598 while the observations after the opening of the HSR is 27477, accounting for 88.4% of the total sample. The sample points with the booking time before the Wenzhou accident is 18030, the sample points after the accident is 13045, accounting for 42.0% of the total sample. 5Table 3 defines the ticket prices and other key control variables in the study and the summary statistics of the variables are presented in Table 3. TABLE 3 The Control Variables

Variable category Name Event

High Accident

Running-in period Flight characteristics

Share Size_dum i*Airline

5 6

Definition Dummy variable for the Jinghu HSR,High=1 if the flight take-off after June 30, 2011 Dummy variable for the the Wenzhou accident,Accident=1 if the ticket is puschased after July 23, 2011 Dummy variable for HSR run-in period,Running-in period=1 if the ticket is purchased at the date between July 10, 2011 to July 20, 2011. Market share of a particular airline6 Dummy variable for the size of aircraft,Size_dum=1 if it is wide-body. Dummy variable for airlines

Our panel data is unbalanced because of uneven sampling and fluctuation of the demand. Share is calculated by the number of seats of the plane.

6

Weekend Days Evening Peak_time Route characteristics

Flights HHI Distance Hub

Dummy variable for weekends flight, Weekend=1 if it is the weekends flight The days booked in advance, Days={0, 1, 3, 5, 15, 30}. Dummy variable for night flight ,Evening=1if the flight takeoffs before 9 a.m or later than 21 p.m Dummy variable for peak time,Peak_time=1 if the flight takeoffs between 9 a.m-10 a.m or14 p.m-16 p.m The remaining number of flights when the data is collected Herfindahl index 7 Route mileage(kilometers) Dummy variable for Hub, Hub=1 means that at least one of end airports on the route is the airline’s base airport

TABLE 3 Summary Statistics Vriables

Mean

Std. Dev.

Min

Max

Sample Size

lnP High

6.582 0.884

0.387 0.320

4.867 0

7.029 1

31075 31075

Accident Running-in period

0 .420 0.182

0.494 0.386

0 0

1 1

31075 31075

Share Size_dum Weekend

0.371 0.329 0.284

0.217 0.470 0.451

0.1 0 0

1 1 1

31075 31075 31075

Days Evening Peak_time Flights Distance Hub

6.411 0.289

8.459 0.454

0 0

30 1

31075 31075

0.236 25.807 1040.264 0.781

0.424 18.180 209.659 0.413

0 1 273 0

1 49 1178 1

31075 31075 31075 31075

Note: lnP is the natural logarithm of the fare. Data Source: Ctrip.com (http://www.ctrip.com) and websites of the airlines. III.

Model and Results

Illustration of the Main Results Before employing an econometric model, we present some graphs and summary statistics on price levels and price dispersion before and after the key events to illustrate the main findings of the paper. Figure 4 depicts the price levels on a daily basis over the sample period. The horizontal axis represents the difference in days between June 30th (the date when the HSR is introduced) and the departuring date. The vertical axis represents the mean price. As we can observe from the two dividing lines on the figure, the mean fare decreased after the introduction of the HSR but A.

7

Herfindahl index is calculated by number of seats.

7

800 700 600

Mean_Price

900

1000

increased after the Wenzhou accident. Data on column 3 of Table 5 confirm our observation by showing that the decrease in the average fare is about 10.9% while the increase is about 4.9% of the fare.

-10

-5

0

5

10

15

20

25

30

35

The Time Difference Between Airline Take-off Date and The Event Date Note:the event date is the opening date of the Jinghu HSR(June 30th,2011)

Figure 4: The Price Levels Before and After The Key Events(June

to Aug



TABLE 5 Summary Statistics on Price and Price Dispersion Departure Time

Observations

Mean price

Std. Dev. of Price

Coefficient of Variation

Before June

3598

842

225.16

0.149

Between June to After July After July

14432

750

263.42

0.218

13045

779

262.25

0.202

Data Source: Ctrip.com (http://www.ctrip.com) Column 4 and Column 5 of Table 5 report the extent of price dispersion in terms of both the standard deviation and the coefficient of variation, respectively. As we can see from the table, price dispersion increased by 17% and 46.3% after the introduction of the HSR and decreased by 0.4% and 7.33% after the accident. Figure 6 exhibits a similar pattern using the mean of the coefficient of variation as the measurement of price dispersion. In what follows, we employ econometric models to rigorously establish the pattern we observed above.

.1

.15

Mean_CV

.2

.25

8

-10 -5 0 5 10 15 20 25 30 35 The Time Difference Between Airline Take-off Date Between Event Date Note: the event date is the opening date of the Jinghu HSR(June 30th,2011)

Figure 6: The Price Dispersion Before and After The Key Events(June B.

to Aug



Measuring Price Dispersion

Ideally, economists want to measure price dispersion in homogeneous product markets. 8However, the data that most economists could acquire are from differentiated product markets. For example, in the airline market researchers usually observe just a fare for a particular flight. If the researcher wants to use the information from other flights, they must take into account ticket heterogeneities, such as differences on the number of days booked in advance, the operating airline company, departure time, aircraft type, check-in service, etc. Therefore, the researcher who wants to study the price dispersion in differentiated product or service markets must control for the price difference caused by the product heterogeneities. In the paper, we apply a fixed effect model to control for the price differences caused by product heterogeneities. Comparing cross-sectional data, the advantage of panel data is that it can control for the unobserved product heterogeneities that do not vary with time. Evidently, in the airline ticket market there are a large amount of product heterogeneities. For example, frequency –fly programs are commonly installed in the airline industry. For travelers who are enrolled in frequency-fly programs with different companies, the service provided to them is different even if the tickets share the same departure time, class and the booking time. It is hard for the researchers to control for the heterogeneity in cross-sectional analysis since they usually cannot observe the enrollment information on frequent-fly programs. However, if we assume that the passengers’ subscription to a particular airline company’s frequent-fly programs remains constant in the sample period, then we can control for the impact of the heterogeneity on ticket prices in a panel regression. For the same reason, the panel model also can control for the impact of other 8

It should be noted that, even in the homogenous product market, one must control for potential retailer heterogeneities to get the exact measurement of price dispersion.

9

unobservable variables on the ticket prices, such as the differences in passengers' willingness to pay for the check-in service, the differences of flights operating cost, etc. Similarly, Lach (2002) and Lewis (2008) use panel data to measure the price dispersion in the grocery and gasoline market, respectively. To summarize, we obtain the price residuals that cannot be explained by the individual ticket characteristics after controlling for the observed and unobserved characteristics by a fixed effect model. The model is postulated as below:

ln pit  0   X it  i   it

(1)

In the above model, subscript i denotes the identity variable, which is defined by the product of the pre-booking days (k), the route (j) and the flight (l). t (  k) is the departure time, t-k is then the date the ticket is purchased. Therefore, the dependent variable ln pit is the logarithm of the ticket price purchased k days before departure time t for flight l at route j. We use the logarithm form of airfare as dependent variable because the effect of some independent variables depends on the base level of ticket price. Besides, Using the logarithmic form of price facilitates the comparison with the results in existing literature (Borenstein,1989; Stavins,2001).

The control variables X it include the dummy variables for the two key events, High t and

Accident k t . High t equals one if the ticket is scheduled to departure after June 30th when the Jinghu HSR is officially introduced and zero otherwise. Similarly, Accident k t equals one if the ticket is purchased after July 23rd when the Wenzhou Accident occurred and zero otherwise. The Wenzhou accident is an unexpected event and the airlines companies are unable to adjust their price strategies in advance in response to the demand intervention generated by the event. As a result, what restricts the value of the Accident k t dummy is the booking date of ticket t-k. In contrast, the opening date of the Jinghu HSR is predetermined and airlines companies could adjust the ticket price in advance. As a result, the value of the dummy High t is contained by the take-off date t. In addition, we also control for the other demand and supply factors that affect price and may vary over time. The demand factors we controlled for include the dummy variable for flights that take-off between July 10th and July 20th ( Running-in period it ) and the dummy variable for weekend flights ( Weekendit ). For the former, from July 10th to July 20th, the frequent occurrence of rail glitches along the Jinghu HSR makes travelling by the HSR less attractive to the passengers

10

who value safety and punctuality. Consequently, the impact of the demand intervention generated by the opening of the Jinghu HSR is lower comparing with the time off the period. For the later, the market demand on weekends is supposed to be more elastic since the passengers who travel on weekend are mainly for leisure purposes. The supply factors include the number of remaining flights when the ticket was booked ( Flight it ) and the mode of the outbound aircrafts ( Size_dumit ). Generally, the greater the number of the remaining flights, the lower the ticket price. However, the effects of aircraft mode on price are ambiguous: On the one hand, a larger aircraft is more likely to provide a safer and more comfortable service and the passengers who value more on quality are willing to pay a higher price premium. On the other hand, the average fixed cost of a large aircraft is lower than that of a smaller one; as a result, the large aircrafts are more likely to charge lower prices. It is worth to point out that the fixed effect term ( i ) contains not only the fixed effect of the flight ( l ), the fixed effect of the route (  j ) and the fixed effect of the number of the days booked in advance (  k ), but also includes possible interactions among them. Consequently, the fixed-effects model not only controls for the unobserved and time constant product heterogeneity, such as the differences in consumer brand loyalty; but also the impact of any interaction, such as the effect of consumer brand loyalty on ticket price may vary with different routes and the number of days booked in advance. The main control variables are summarized in Table 3. 

Estimating the fixed effect model (1), we get the the predicted value of the error term ( 

it

), i.e.,

the price residual. The price residual is the price that cannot be explained by the differences in ticket characteristics after controlling for both observable time-varying factors (including both demand and supply factors) and unobserved and time constant product heterogeneities. We interpretate the price residuals as the price dispersion charged by different airline companies for the homogeneous products or services as in Barron et al. (2004) and Lewis (2008). In our sample, the median of the price residuals is 0.017, the minimum is -1.384, the maximum is 1.035, and the standard deviation is 0.225. C.

Econometric Model

Since the price residuals reflect the portion of the prices that cannot be explained by the ticket characteristics, we therefore construct the measurements of price dispersion based on the price residuals. The measurements we adopt include the variance of residual, standard deviation and coefficient of variation, which are commonly used in the literature. 9Our study shows that the estimation results differ quantitatively rather than qualitatively under different indexes of price dispersion, we thus demonstrate the results using the variance of the price residual as the measure 9

See Baye , Morgan and Scholten (2006) for more indexes of price dispersion.

11

of price dispersion in the following. We study how exogenous demand interventions affect the price dispersion using the following model: 2





it

  0  1High t   2 Accident kt   X it  i  it ,

(2)

2

where

 it is the square of price residuals. The independent variables include main factors that

affect price dispersion. Among them, we are mainly interested in the coefficients of the dummy variables for the Jinghu HSR ( High t ) and the Wenzhou Accident ( Accident kt ). If the price dispersion increase (decrease) when the competition becomes more (less) intense, then should be significantly positive while

1

 2 should be significantly negative. Furthermore, if the

Wenzhou Accident partially offsets the decreased demand caused by the Jinghu HSR, and then

1   2 should be significantly positive, which indicates that the airline market competition after the Wenzhou Accident is more competitive than prior to the introduction of the Jinghu HSR on June 30th. Accordingly, we have the following hypothesis:

Hypothesis 1 : If price dispersion increases with competition, then we expect significantly positive and

1 be

 2 be significantly negative. Besides, if the Wenzhou Accident

partially offsets the downward demand intervention caused by the Jinghu HSR, then

1   2

should be significantly positive. As in model (1), we control for Running-in period, Flights, Size_dum, Weekend and other variables that affect demand and supply in the market in the regression. The frequent occurrences of rail glitches reduced the impact of the introduction of the Jinghu HSR into the airlines market along the route, which in turn softened the competition among the major airline competitors in the market. If the reduced price competition tends to decrease the price dispersion, we expect the parameter of Running-in period be significantly negative. GS (2009) find that the airline is more likely to adopt price discrimination if the aircraft is with wide-body type and the number of remaining flights is greater. We therefore expect a positive correlation between Size_dum and Flights and price dispersion. The coefficients for weekend flights dummy variable is predicted to be negative since most of passengers are leisure travelers on weekends. Similar to equation (1), we control for the fixed effect of flight, route and pre-book days through the fixed-effect term ( i ). Finally,

it is the random disturbance.

12

It’s worth to emphasize that, due to the exogeneity of the demand interventions caused by the Jinghu HSR and the Wenzhou accident, our results do not suffer from the endogeneity problem caused by the simultaneity bias as in most of existing literature.(BR,1994;GS,2009)The simultaneity bias appears in the literature because of the interdependent of the market structure and the price dispersion: on the one hand, the market concentration determines the price and price dispersion; on the other hand, the price dispersion may induce firms’ entry or exit decisions. The estimated coefficients are biased and inconsistent if the simultaneity issue is ignored. Although a feasible solution to this problem is by instrumental variable method,the estimated coefficients could still be biased if the instrumental variable method is applied in inappropriate context. GS (2009) point out that the positive correlation between competition and price dispersion found by BR (1994) with cross-sectional data is due to the use of inappropriate instrumental variables and omitting some key independent variables. Employing a fixed effect model, GS (2009) is able to avoid the estimate errors caused by the omitting variables, but they still need to rely on instrumental variables to solve the problem of simultaneous bias. In our study, the demand interventions generated by the two events are strictly exogenous, which allows us to evaluate the impact of competition on price dispersion without concerning about the endogeneity problem caused by simultaneity bias. (BR,1994;GS,2009) D.

Results

Table 7 presents the results with different measures of price dispersion. As we can see from Column 1 of Table 4, the estimated coefficients for High, Accident and Running-in period are 0.019, -0.008, and-0.011, respectively. The estimates are all significant at 1% level and correspond to 55.8%, -14.8% and -23.9% changes on the extent of price dispersion.10 The results indicate that the price dispersion increases significantly after the introduction of HSR but decreases significantly after the occurrence of the Wenzhou accident and the rail glitches. Because HSR is an alternative for the airline market, the introduction of HSR reduces the demand for the airline market while the the subsequent glitches of the Jinghu HSR and the Wenzhou Accident offset part of the reduced demand for the airline market. Therefore in that sense, our results confirm the positive relationship between competition and price dispersion in the Chinese airline industry from both positive and negative exogenous demand interventions. Our results support the implications of price dispersion theory based on the price discrimination (cf. Borenstein, 1985;Holmes, 1989; Gale and Holmes,1993). BR (1994) and Starvins (2001) obtain similar conclusions from the studies of the U.S. airline market. However, as we emphasized above, comparing with BR (1994) and Starvins (2001), our study is free of instrumental variables due to the exogenous nature of the events (GS, 2009). In addition, the price dispersion of airline tickets is still significantly higher after the Wenzhou accident than before the introduction of the Jinghu HSR since the difference of the coefficients between them is significant at the 1% level.11 Thus, the Wenzhou accident offsets a portion of the

10

The mean square of price residuals before the introduction of the HSR is about 0.034.

11

The difference is 0.011 and the p-value of the null hypothesis that

1   2  0

is equal to half of the

13

decreased demand caused by the introduction of the Jinghu HSR. With regard to other coefficients, the estimate for Weekend is positive and significant, which implies that passengers travelling on weekend differ more on demand elasticities than those travelling on weekdays. The estimate of Flights is also positive and significant, which indicates that the airline companies are more likely to discount prices when the market supply exceeds demand. However, the coefficient of Size_dum is not significant, which suggests that there does not exist any significant fare difference between different modes of aircraft. TABLE 7 The Impact of Demand Interventions on Price Dispersion Dependent Variable (1)

(2)

(3)

2

Std. Dev.

Coefficient of Variation

Other control variables Fixed Effect

0.019*** (0.002) -0.008*** (0.001) -0.011*** (0.001) 0.004 (0.007) 0.004*** (0.001) 0.000* (0.000) Yes Yes

0.033*** (0.001) -0.007*** (0.000) 0.005*** (0.001) 0.001 (0.003) 0.007*** (0.000) 0.000*** (0.000) Yes Yes

32.916*** (0.962) -1.300** (0.656) 0.158 (0.822) -5.354 (3.653) 2.702*** (0.577) 2,261*** (0.114) Yes Yes

Constant

0.012

0.169***

-80.383***

(0.010) 0.12 31075

(0.004) 0.068 31075

(5.858) 0.05 31075

Independent Variables

 it High Accident Running-in period Size_dum Weekend Flights

Adjusted N

Note: Standard errors in parentheses. *, ** and *** denote statistical significance at 1%, at 5% and at 10% level. The dependent variables in column (1) to (3) are the square of price residuals, standard deviation and coefficient of variation, respectively. IV.

How do the Airline Companies Price Discriminate?

In this section, we intend to provide an explanation for the empirical results established in the last section. To that end, we study how the demand interventions affect the airline companies’ price discrimination strategies, which results in the price dispersion observed in the airline market along the Jinghu HSR.

p-value of the null hypothesis that

1   2  0 .

14

We employ a quantile regression to examine the effects of demand interventions on price discrimination by the airline companies. (Borenstein,1989;GS,2009;Dai et al.,2011)12 We classify the consumers into groups with different elasticities of demand based on the following observation: the passengers who travel in the off-peak time or book the ticket in advance are more likely to be price-elastic and the prices paid by them are mostly located at the lower percentiles (such as 10th percentile or 20th percentile); however, the passengers who travel in the peak period or book the ticket close to departure are more likely to be price-inelastic and the prices paid by them are mostly located at higher percentiles (such as 90th percentile or 80th percentile). Thus, by observing the effects of demand interventions on different percentiles of the conditional ticket price distribution, we can infer how the airline companies price discriminate differently among the consumers with different elasticities of demand after the demand interventions. We adopt a fixed effect panel data quantile model to estimate the impact of demand interventions on different percentiles. Similar to Canay (2011), our approach involves two steps: in the first step, we estimate the fixed-effect term and subtract it from the log-prices to get the residual prices; in the second step, we regress the price residuals on a set of independent variables. The first stage estimation is the same as equation (1) and the second stage of the quantile regression is as follows, q

ln pit  0q  1q High t  2 q Accident kt   q Xit   it q

Where q=10, 20, 25, 75, 80, 90 are the

(3)

q th percentile of the price distribution.

q

ln pit  ln pit  i with i being the estimated fixed effects term from equation (1). The explanatory variables include both the dummy variables for the events and other time-varying variables.13 For the detailed information on variables, please refer to Table 2. After the negative demand intervention of the HSR, if the airline companies decrease more prices on the price-elastic consumers than on the price-inelastic consumers, the coefficients of High t in equation (3) on the high quantiles will be greater than those on the lower quantiles. That is, we have

1L  1H  0 , where H =90, 80, 75 and L =10, 20, 25. If otherwise, we have

1H  1L  0 . Correspondently, after the positive demand intervention of the Wenzhou accident, 12

Comparing with linear regression, the quantile regression could provide more detailed relationship between the

conditional distribution of ticket price and the intensity of competition among the airlines because we can run quantile regression on any quantile of the price distribution. In addition, the estimators of quantile regression are more robust because it minimizes the absolute deviation (Koenker and Bassett, 1978; Xing, 2008). 13

We include only the time-varying independent variables in the regression because the fixed effect has been removed from the fare as the dependent variable.

15

if the airline companies increase more prices on the price-elastic consumers than on the price-inelastic consumers, we would have

2 L  2 H  0 . Otherwise, we have 2 H  2 L  0 .

We thus obtain the following hypothesis: Hypothesis 2: In face of the negative (positive) demand intervention, if the airline companies decrease (increase) more prices on the price-elastic consumers than on the price-inelastic consumers, we then expect that

1L  1H  0 and 2 L  2 H  0 , where H =90, 80, 75 and

L =10, 20, 25. Table 8 contains results from equation (3). As we can see from the first row of the table,

110th  120th  125th  175th  180th  190th  0 , that is to say the coefficients of the dummy variable High t are negative and increase as the percentiles increase after the introduction of the HSR.14 Since the prices paid by the price-elastic (leisure) passengers are mostly located at the lower percentiles while the prices paid by the price-inelastic (business) passengers are mostly located at the higher percentiles, we can interpretate the above relationship as that the airline companies discount more prices on the consumers with elastic demand elasticity than on the consumers with inelastic demand elasticity responding to the negative demand intervention from the introduction of the HSR. On average, the price discount to the price-inelastic consumers is about 9% while the price discount to the price-elastic consumers is 23%, which is 14% higher.

The coefficients for the dummy variable Accident kt are positive and decrease as the percentiles increase

after

the

Wenzhou

accident,

that

is

210th  220th  225th  275th  280th  190th  0 . Thus, the Wenzhou accident has the opposite impact on the coefficients to the introduction of the HSR. The Wenzhou accident can be regarded as the reverse event of the opening of HSR because it alleviates the downward demand intervention caused by the HSR. Our results indicate that the airline companies withdraw their price promotion for the consumers after the accident. Furthermore, the companies increase more prices on the price-elastic consumers than on the price-inelastic consumers. On average, the increased price for the price-elastic consumers is about 8%, while for the price-inelastic consumers it is about 2.3%. Overall, the empirical results from equation (3) demonstrate that the airline companies adopt a consistent price strategy when facing with demand interventions caused by the two opposite exogenous events. When facing with downward demand intervention, the airline companies cut more prices on the price-elastic consumers than on the price-inelastic consumers; in contrast,

14

The p-values of the null hypothesis for

1L  1H are 0.004, 0.012 and 0.025, respectively. We thus reject the

null hypothesis and establish that the difference between

1L and 1H

is significant.

16

when facing with upward demand intervention, the airline companies increases more prices on the price-elastic consumers than on the price-inelastic consumers. As a result, the price becomes more dispersed with more competition (after the opening of the HSR) and less dispersed with less competition (after the accident). The estimates for other variables are mostly significant and consistent with intuition. The coefficients for Running-in period are positive on high percentiles but negative on low percentiles. This indicate that during the running-in period, the airline companies only increase prices on the consumers who switch back to the airline market after the HSR glitches, but leave the fares to loyal customers almost unchanged. The coefficients for Size_dum vary from positive to negative as the percentiles increase. Thus, comparing with small crafts, the ticket prices for large aircrafts are relative high on the lower quantiles but low on the higher quantiles. Finally, the coefficients for the Weekend and Flights in table 5 are all significantly negative, which indicates the ticket prices are lower when the flights take-off in weekend or when the supply is sufficient. TABLE 8 The Impacts of Demand Interventions on the Quantiles of Price Distribution Dependent Variable: Independent Variables High Accident Running-in period Size_dum Weekend Flights constant Pseudo N

ln pit

(1)

(2)

(3)

(4)

(5)

(6)

-0.259*** (0.008) 0.083*** (0.008) 0.089***

-0.223*** (0.007) 0.079*** (0.005) 0.066***

-0.207*** (0.005) 0.074*** (0.005) 0.056***

-0.105*** (0.005) 0.025*** (0.004) -0.004

-0.093*** (0.005) 0.025*** (0.004) -0.011***

-0.067*** (0.007) 0.018*** (0.005) -0.018***

(0.009) 0.086*** (0.008) -0.037*** (0.006) -0.008*** (0.000) 6.771*** (0.008) 0.210 31075

(0.006) 0.060*** (0.007) -0.027*** (0.005) -0.008*** (0.000) 6.862*** (0.008) 0.234 31075

(0.006) 0.053*** (0.006) -0.026*** (0.005) -0.008*** (0.000) 6.905*** (0.006) 0.247 31075

(0.004) -0.034*** (0.004) -0.016*** (0.003) -0.008*** (0.000) 7.145*** (0.004) 0.324 31075

(0.004) -0.042*** (0.004) -0.015*** (0.003) -0.009*** (0.000) 7.183*** (0.005) 0.319 31075

(0.006) -0.056*** (0.009) -0.012** (0.005) -0.181*** (0.005) 7.290*** (0.006) 0.300 31075

Note: Standard errors in parentheses. *, ** and *** denote statistical significance at 1%, at 5% and at 10% level. The results are obtained by a quantile regression model with fixed effects with 300 bootstraps. V.

Robustness Check

To test the robustness of the positive relationship between price dispersion and demand interventions established in section III, we employ an Interquantile range regression model. In the

17

model, the dependent variable is the differences between quantiles of price residual (i.e.,

IQR H  L = (ln Pit ) H  (ln Pit ) L , where H=90, 80, 75; L= 10, 20, 25, respectively). As before, the control variables include Distance, HHI, LnIncome, LnPop, Share, Hub, Airline dummies, Days and Square of Days. Table 9 reports results from the regression. The results confirm that price dispersion increases with competition since the coefficients of High are positive and significant while the coefficients of the accident are negative and significant even when the dependent variable is IQR

75 25

(The narrowest differences between quantiles of price residuals).

TABLE 9 The Impact of Demand Interventions on Price Dispersion with Interquantile range regression Dependent Variables Independent

(1)

(2)

(3)

0.093*** (0.007) -0.050*** (0.005) -0.049*** (0.006) -0.037*** (0.006) 0.020*** (0.004) 0.007*** (0.000) Yes 15.848*** (0.611) 0.259 31075

0.106*** (0.006) -0.058*** (0.005) -0.058*** (0.005) -0.038*** (0.006) 0.024*** (0.005) 0.008*** (0.000) Yes 19.023*** (0.729) 0.248 31075

0.180*** (0.010) -0.066*** (0.009) -0.100*** (0.008) -0.059*** (0.011) 0.038*** (0.007) 0.011*** (0.001) Yes 24.152*** (1.071) 0.227 31075

Variables High Accident Self_accident Size_dum Weekend Flights Other control variables Constant Adjusted N

Note: Standard errors in parentheses. *, ** and *** denote statistical significance at 1%, at 5% and at 10% level. The model is estimated by an interquantile range regression with 300 bootstraps.

In addition, we introduce the interaction terms of Days*High, Weekend*High or Size_dum*High into equation (1) to test the robustness of the effects of demand interventions on the airline companies’ price discrimination strategies. Comparing with price-elastic consumers, the price-inelastic consumers are more likely to book the ticket close to departure, travel on weekdays or prefer to take wide-body aircrafts. If the airline companies discount more prices on the price-elastic consumers than on the price-inelastic consumers when the competition increases, we expect that the prices of tickets booked close to departure, in the weekdays or with wide-body aircrafts to decrease more after the introduction of the Jinghu HSR. Table 10 reports the estimates

18

from the fixed-effects panel regressions with the interaction terms. The coefficients of the interaction terms are -0.016, -0.035 and 0.044, all of which are significant at 1% level. The results suggest that after the opening of HSR, if the ticket was booked one day less in advance, the price increases by 1.6%; If the ticket was scheduled to depart in weekdays, the price is 3.5% higher; If the ticket is with wide-body aircrafts, the price is 4.4% higher. TABLE 10 Robustness Check on the Airlines’ Pricing Strategies Independent Variables High High* Days

Dependent Variables: (1)

(2)

(3)

-0.110*** (0.007) -0.016*** (0.002)

-0.143*** (0.006)

-0.165*** (0.006)

High* Weekend

-0.035*** (0.010)

High* Size_dum Accident Self_accident Size_dum Weekend Flights Constant Adjusted N

0.046*** (0.003) 0.025*** (0.004) 0.012 (0.019) -0.023*** (0.003) -0.009*** (0.001) 7.092*** (0.031) 0.120 31075

0.047*** (0.003) 0.028*** (0.004) 0.012 (0.019) 0.010 (0.009) -0.008*** (0.001) 7.012*** (0.030) 0.090 31075

0.044*** (0.010) 0.047*** (0.003) 0.028*** (0.004) -0.029 (0.021) -0.022*** (0.003) -0.009*** (0.001) 7.043*** (0.030) 0.098 31075

Note: Standard errors in parentheses. *, ** and *** denote statistical significance at 1%, at 5% and at 10% level. VI.

Conclusion

In this paper, we investigate how two opposite exogenous demand interventions affect the airline market price competition along the Jinghu HSR in China. We find that the airlines discount more on the price-elastic consumers than on the price-inelastic consumers, which leads to higher price dispersion after the introduction of the HSR. Our findings are consistent with the empirical results of BR (1994) and Starvins (2001) from the U.S. airline markets. Our paper differs from previous studies on the following aspects. First, our study builds on the two natural experiments, which are exogenous to the competition in the airline industry. Since the

19

opening of the Jinghu HSR and the occurrence of the Wenzhou accident are exogenous to the airline industry, our study therefore avoids the endogeneity problem caused by simultaneity bias between price dispersion and market competition (GS, 2009). Second, while the opening of the Jinghu HSR diverts some passengers from the airline market and lead to the more intense market competition, the occurrence of the Wenzhou Accident offsets the downward demand shock caused by HSR and softens the competition among airlines to some extent. The study of the two opposite events reinforces each other the main findings of the research. Finally, we analyze the effects of demand interventions on the price discrimination strategies adopted by the airline companies, which further support the key results of the paper that price dispersion increases with market competition.

References

John, Barron, Beck, Taylor, and John, Umbeck,"Number of Sellers, Average Prices, and Price Dispersion," International Journal of Industrial Organization 22 (2004), 1041-66. Michael, Baye, John, Morgan, and Patrick, Scholten,"Information, Search, and Price Dispersion," Handbook of Economics and Information Systems ( Amsterdam: Elsevier Press, 2006). Severin, Borenstein,"Price Discrimination in Free-Entry Markets," The RAND Journal of Economics 16(1985), 380-97. Severin, Borenstein,"Hubs and High Fares: Dominance and Market Power in the Us Airline Industry," The RAND Journal of Economics 20(1989), 344-65. Severin, Borenstein, and Nancy, Rose,"Competition and Price Dispersion in the Us Airline Industry," Journal of Political Economy 102 (1994), 653-83. Ivan, Canay,"A Simple Approach to Quantile Regression for Panel Data," The Econometrics Journal 14 (2011), 368-86. Mian, Dai, Liu, Qihong and Konstantinos Serfes , "Is the Effect of Competition on Price Dispersion Non-Monotonic? Evidence from the U.S. Airline Industry," Working Paper, 2011. Alberto, Gaggero, and Claudio, Piga,"Airline Market Power and Intertemporal Price Dispersion," The Journal of Industrial Economics 59 (2011), 552-77. Ian, Gale,"Price Dispersion in a Market with Advance-Purchases," Review of Industrial Organization 8 (1993), 451-64. Ian, Gale, and Thomas, Holmes,"Advance-Purchase Discounts and Monopoly Allocation of Capacity," The American Economic Review 83(1993), 135-46. Kristopher, Gerardi, and Adam Hale Shapiro,"Does Competition Reduce Price Dispersion? New Evidence from the Airline Industry," Journal of Political Economy 117 (2009), 1-37. Thomas, Holmes,"The Effects of Third-Degree Price Discrimination in Oligopoly," The American Economic Review 79 (1989), 244-50. Roger, Koenker, and Gilbert, Bassett Jr,"Regression Quantiles," Econometrica 46 (1978), 33-50. Saul, Lach,"Existence and Persistence of Price Dispersion: An Empirical Analysis," Review of Economics and Statistics 84(2002), 433-44. Matthew, Lewis,"Price Dispersion and Competition with Differentiated Sellers," The Journal

20

of Industrial Economics 56 (2008), 654-78. Joanna, Stavins,"Price Discrimination in the Airline Market: The Effect of Market Concentration," Review of Economics and Statistics 83 (2001), 200-02. George, Stigler,"The Economics of Information," The Journal of Political Economy 69 (1961), 213-25. Xiao, Junji and Tang, Xin,"An Empirical Study on China Airlines Price Competition," Nankai Economic Studies 2(2009), 80-90. Xing, Chunbing,"Quantile Regression,Return to Education and Income Distribution ," Statistical Research 5(2008), 43-49. Xu, Shu, Li, Han and Gan, Li,"Market Competition and China’s Airline Pricing," China Economic Quarterly 2(2011), 635-652.

Suggest Documents