Are airports engines of economic development? A dynamic panel data approach

Abstract

This paper applies the dynamic panel data generalized method of moments estimator to the data on commercial passenger air traffic at all primary airports in the United States to evaluate the impact of traffic volume and number of destinations served with non-stop flights on the key indicators of regional economic development. We find that number of destinations served with non-stop flights has a much clearer and more robust impact on level of employment, number of business establishments, and average wage in the region. Passenger traffic volume affects employment and average wage, but not number of establishments. At the sample median, connecting a metropolitan statistical area with an extra destination, keeping everything else constant, creates 98 jobs and facilitates the opening of four new business establishments that employ people. The corresponding numbers for the sample mean are 223 jobs and 15 businesses. The impact of air travel on regional economic development is influenced by competition on the respective airline markets.

Keywords

airports air traffic development dynamic panel data GMM

Introduction

This paper examines the issue of the impact of air traffic on economic development, using data from the United States. While intuitively we can suspect that a well-developed and well-connected airport should facilitate attractiveness of a respective metropolitan area for business, quantifying this relationship is not a straightforward task. High-quality air services will facilitate face-to-face contacts with business collaborators, foster intercity agglomeration economies, and ultimately lead to business and job creation. This new business however will create more demand for air services. Such feedback necessitates a careful approach to data analysis. One of the contributions of this paper lies in this area: we use a data analysis technique, which has not been applied to studying the air travel–economic development link before.

Additionally, our study looks into several questions with clear practical relevance for local governments. When thinking about developing air service from a local airport, local authorities may have to prioritise either strengthening existing links (i.e. increasing frequency of service to existing destinations) or attracting service to new destinations. Our study demonstrates that, other things equal, adding a new destination creates more jobs in the area as opposed to increasing frequency to existing destinations. Another issue of relevance to the local authorities is whether to promote services of airlines already serving an airport, or to incentivise entry of new carriers. Our study examines this dimension as well, suggesting that allowing the dominant carrier to develop its presence appears to be good for business. This result appears consistent with the expected impact of airlines’ customer loyalty programmes, keeping in mind that: (a) business travellers are more likely to be active users of those programmes; and (b) passengers living in metropolitan areas with concentrated airports are more likely to benefit from customer loyalty programmes (Borenstein, 1989, 1991). We thus demonstrate that the impact of commercial passenger air travel on regional economic development indicators is influenced by competition on the airline markets. This finding is of notable interest, as previous studies (see Bilotkach and Lakew, 2014, for a recent review) have focused exclusively on the impact of airline competition on prices and other measures of consumer welfare within the airline markets. Ours is among the first works to evaluate how competition between the airlines may affect other sectors of the economy.

Our dataset is a 17-year panel covering all US metropolitan areas with airports offering commercial passenger air services. We consider three measures of economic development (total employment, number of business establishments, and average weekly wage) at the metropolitan statistical area (MSA) level. We find that a 10% increase in the number of flights is associated with a 0.1% increase in average wage, with the impact of this variable on employment and number of business establishments not being statistically significant. Air traffic volume measured by the number of passengers has a positive and significant effect on employment and wage, but not on the number of business establishments. Number of destinations served by non-stop flights is found to have a robust positive impact on each of the three measures of economic development. A 10% increase in this variable yields a 0.13% increase in employment; a 0.1% increase in the number of business establishments; and about 0.2% increase in average weekly wage. At the sample median, connecting an MSA with an extra destination, keeping everything else constant, creates 98 jobs and facilitates the opening of four new business establishments that employ people. The equivalent numbers for the sample mean are 223 jobs and 15 new business establishments. We also found a statistically significant relationship between our measures of air traffic volume and the average MSA wage. Yet, economic significance of this relationship is not remarkable, as a substantial increase in traffic volume would be required to yield appreciable changes in mean wages.

Thus, the main lessons from this work for local authorities are the following. Most importantly, adding new destinations yields a stronger impact on regional development than adding traffic to existing destinations. At the same time, while increase in air traffic can contribute to creating jobs and businesses, one should not expect a sizeable increase in average income. We should note that our results represent aggregate effect – we have not evaluated which sectors of the economy benefit more from increase in air traffic, whether it comes from extra flights to existing destinations, or from adding new services. We will leave this issue for future studies.

The rest of the paper is organised as follows. The next section reviews the relevant literature. This is followed by a discussion of the data and estimation methodology. Estimation results are presented and discussed in the following section, and the final section concludes.

Literature survey

Academic studies on the impact of aviation on economy are relatively scarce. Yet, over the recent years we have seen a renewed interest in careful examination of this relationship. The general consensus from the literature is, not surprisingly, that air services have a positive impact on regional development. The main differences between the studies are in the data and methodologies employed. It is therefore understandable that quantitative estimates of the air traffic–economic development relationships also differ across studies. Additionally, some papers find that the relationship only holds for some but not other sectors of the economy.

Studies most relevant to this paper include van den Berg et al. (1996), Brueckner (2003), Green (2007), Bel and Fageda (2008), Button and Yuan (2013), Blonigen and Cristea (2012), Chi and Baek (2013) and Sheard (2014). The first paper in the list is rather descriptive in nature, building an argument for cooperation between airports and the nearby businesses. Brueckner (2003) uses a cross-sectional approach, analysing the air traffic–employment relationship in the US metropolitan areas in 1996. The instrumental variables approach is used to address the endogeneity problem (which is another name for the feedback between the employment levels and demand for air services we mentioned in the opening paragraph). The main finding of that study is that a 10% increase in passenger enplanements yields a 1% increase in service sector employment. A somewhat similar approach (instrumental variables in a cross-sectional setting) is employed by Sheard (2014). The finding of that study is that air traffic volume has a positive impact on service industry employment; however, the effect on manufacturing employment is negative. Bel and Fageda (2008) analyse whether the number of non-stop intercontinental flights determines location of large firms’ headquarters. They find support for this hypothesis. Their dataset is also a cross-section.

Button and Yuan (2013) focus on the relationship between air freight volume and metropolitan area income and employment. Their dataset covers 35 airports from 1990 to 2009, and their estimation technique of choice is the vector autoregressive (VAR) model. They conclude that air freight is indeed a driver of local economic development. Button and Yuan’s (2013) study spans nearly the same time period as ours (our data span 1993–2009); however, our study encompasses the entire population of airports in the US, providing commercial passenger air services.

Studies by Green (2007) and Blonigen and Cristea (2012) focus on the effect of air service on growth rates in key economic performance indicators. Green (2007) finds that a 10% increase in passenger enplanements per capita leads to a 3.9% higher population growth and a 2.8% increase in employment growth over the 1990–2000 period. Green’s dataset is a cross-section of 100 largest US airports. Blonigen and Cristea (2012) exploit the airline deregulation as a quasi-natural experiment leading to an increase in air traffic. They choose to examine the impact of deregulation-induced change in passenger air traffic on long-term growth rates in over 300 US metropolitan areas. The study confirms the positive impact of air service on regional growth, with the magnitude of this effect differing across the metropolitan areas.

Chi and Baek (2013) focus on the aggregate data for passenger numbers and flight volumes for the US aviation sector. The study finds that both passenger traffic and freight volumes tend to increase with economic growth. This is in contrast to Button and Yuan’s (2013) study, which suggests that causality runs the other way.

Thus, our study expands the scope of methodological approaches used to examine the relationship between air travel and regional economic development. Our dataset encompasses all US primary commercial passenger airports over a lengthy time period, which includes all phases of the business cycles. Unlike prior studies, we also control for the impact of changes in the airline market structure, by including airport-level concentration and airline market shares at airports into our specifications.

In addition to the above studies, we should mention the rather substantial literature on the effects of air traffic noise on property values. A relatively recent contribution by McMillen (2004) includes an extensive survey of prior studies in this area.

Our paper is also related to the literature on the impact of public infrastructure investment. Indeed, airports in the United States are run as public enterprises, owned by local authorities. Here we should note that the studies of the effects of public infrastructure (most notable seminal contributions include Aschauer, 1989 and Holz-Eakin, 1994) have not reached a consensus as to whether this public expenditure contributes to the local economic growth.

Data and methodology

Our dataset is a 17-year panel of annual observations for each US metropolitan area housing a primary commercial passenger airport. Primary airport is defined by the Federal Aviation Administration (FAA) as an airport with annual scheduled commercial passenger air traffic exceeding 10,000 passengers. Our data span the years 1993–2009. We can define three broad categories of variables in our dataset. First, we have the key dependent variables, which are the indicators of regional economic development. Second, we use two demographic control variables. The third set of variables includes airport-level measures, which in turn consist of our key measures of air traffic, and airport-level controls representing the degree of airport concentration and airline market competition.

This study uses three indicators of regional economic development: total employment, number of business establishments with employees, and average weekly wage. The respective data come from the Bureau of Labor Statistics – a division of the United States Department of Labor. All the data are annual figures at the MSA level; average weekly wage is adjusted for inflation using 1995 as the base year. Employment and average earnings are clear indicators of regional economic development. Number of business establishments can be thought of as an indicator of competitive environment in the area. Also, using this variable allows us to draw parallels between our work and Bel and Fageda (2008). Natural logarithms of the respective variables will be used in all specifications.

The two demographic control variables are MSA-level population and unemployment rate. These are obtained from the US Census Bureau. In regressions, we will use natural logarithms of the respective variables. Increase in population is expected to yield higher employment and more business establishments, as well as potentially lead to an increase in average wage. Increase in unemployment rate, other things equal, should lead to a decrease in each of the three key measures of economic development.

The majority of airport-level variables are computed from the T100 Segment databank, provided by the United States Department of Transportation (DOT). This dataset includes information about all commercial airline services departing from US airports. For this paper, we are only including the services on the US domestic market. The information in the raw dataset is provided monthly at the airline-origin-destination-aircraft type level (e.g. Delta Air Lines Boeing-757 services from Los Angeles International to New York John F. Kennedy airport in January of 2000 are recorded separately from the Boeing-737 services of the same carrier on the same route in the same month) and includes the number of flights performed, seats provided, and passengers carried. We have aggregated the data at the annual level, giving us the total passenger volume and the number of flights performed from each airport. We have also computed the number of unique destinations (at the airport level) served with non-stop scheduled passenger flights from each airport in every given year. In this computation, we only included those destinations served with at least 100 flights per year – roughly two flights per week were thus required for a destination to count.

The three key measures of commercial aviation activity we have computed will be our key independent variables. The research hypotheses are straightforward – increase in both the number of flights and number of destinations is expected to increase each of the three key indicators of regional economic development. We will be interested in comparing the magnitudes of the corresponding effects, with the view of understanding whether increasing traffic volume (in terms of number of passengers or frequency of flights) at existing routes will have a larger impact on economic development than attracting non-stop flights to an additional destination. This question is clearly an important one for airport managers and local authorities, especially keeping in mind that commercial passenger airports in the United States are operated as public enterprises. As is the case with other variables, our key dependent variables will be included into our specifications in the logarithmic form. This will of course enable us to interpret the corresponding regression coefficients as elasticities.

Other airport-level variables we will use in regressions as controls are airport-level concentration, airline market shares at the airport, and average airfares for flights originating at the corresponding airport. As the measure of airport-level concentration, we will use a conventional Herfindhal-Hirschman Index (HHI). HHI is the sum of squared shares of flights, across all the airlines at an airport. Before computing HHI and airlines’ flight-based market shares, we have merged regional airlines with the respective major carriers.¹ Average annual airport-level airfares are computed by the DOT from the 10% sample of actual itineraries.² We adjust those fares for inflation using 1993 as the base year. Last but not least, each specification we will estimate will include year indicator variables, to control for the corresponding heterogeneity.

Other things equal, we can expect higher airfares to be less conducive to regional economic development, thereby yielding lower employment, number of businesses, and potentially lower wages in the metropolitan area. We can of course expect this variable to be endogenous, as an unobserved shock that increases employment will also yield higher airfares for flying to/from the area. The effect of airport concentration on economic development can potentially tell us something about the importance of airlines’ loyalty programmes. Specifically, airlines dominating a local airport will likely attract most local travellers into their frequent flier programmes. On one hand, this may be a source of the dominant carriers’ market power (Borenstein, 1989, 1991; Lederman, 2008). On the other hand, considering that frequent flier miles are essentially fringe benefits, businesses might be drawn to concentrated airports. Further, large concentrated airports are hubs, which brings us back to Bel and Fageda’s (2008) evidence that hub airports attract multinational firms.

Before we continue our discussion, it is necessary to explore the issue of multi-airport metropolitan areas. Some of the largest MSAs in the US are homes to several airports, such as Midway and O’Hare in the Chicago area. Where this is the case, we summed up total flights from all airports in the MSA; summed up all unique destinations served by non-stop flights from those airports; and computed passenger-weighted airport HHI and airline market shares. The airport groupings thus obtained largely correspond to those suggested by Brueckner et al. (2014), with a couple of notable exceptions. First, Baltimore airport was considered separately from Washington Dulles and Washington National, and San Jose was not grouped with San Francisco and Oakland, by virtue of being located in different MSAs.

Descriptive statistics for all the variables are presented in Table 1. The distribution of the main variables is understandably skewed, as the data include observations from both many smaller and few very large metropolitan areas. Mean employment and number of businesses are both about five times larger than the corresponding median figures. The same is true for the total number of flights from the airport. We can also see that an average US airport is quite well-connected to the US aviation network, with a median airport featuring non-stop flights to 16 US destinations. At the same time, airports are on average rather concentrated: mean HHI of 0.51 is equivalent to a symmetric duopoly, or two airlines each performing half of all the flights from the airport. Bilotkach and Lakew (2014) also report high levels of concentration in both large hubs and smaller airports in the US.

Table 1.

Descriptive statistics.

	Mean	Standard deviation	Median
Dependent variables
Total employment	565,551	1,328,718	120,366
Total number of establishments	33,885.5	85,889.1	6,693.12
Real weekly wage	496.22	104.86	474.83
Airport level variables
Total flights	25,222.1	53,841.3	5101.00
Total passengers	1,782,273	4,302,731	180,930
Number of non-stop destinations	33.02	39.12	16.00
Airport level HHI	0.5120	0.3039	0.4311
Mean airfare, constant dollars	307.07	110.31	310.75
MSA level controls
Population	1,189,265	2,854,425	254,401
Unemployment rate	5.44	2.48	4.90

Note: Sample includes all metropolitan areas with airports handling over 10,000 passengers via scheduled commercial services.

The choice of our estimation methodology will be driven by the econometric challenges we are facing. As noted above, the key issue to be addressed is that of endogeneity inherent in the data generating process. Clearly, an unobserved shock that increases the region’s attractiveness for business will also likely yield an increase in air traffic to the region’s airport. This means that the error term in the least squares regression will be correlated with the dependent variable, rendering coefficient estimates biased and inconsistent. Additionally, our estimates must take proper account of MSA-specific heterogeneity, and the possible autocorrelation within and heteroscedasticity across the cross-sections. To address these issues, we will be taking advantage of both the panel nature of our dataset and the recently developed dynamic panel data analysis techniques.

We will be using MSA fixed effects models to account for regional heterogeneity. Further, taking advantage of the panel nature of our dataset, we will lag all the dependent variables (except for year dummies) one year. Clearly, an unobserved shock this year is less likely to be correlated to last year’s level of air traffic. Yet, last year’s traffic could also contribute to this year’s unobserved shock, so that the endogeneity problem is not completely solved by simply using lagged independent variables. Additionally, we will be using second lags of total flights, number of non-stop destinations, average airfare and airport-level HHI as instruments for first lags of same. This completes the setup for MSA-level two-stage least squares fixed effects estimation. As a final note, we will be using standard errors robust to autocorrelation and heteroscedasticity.

The fixed effects estimation is a useful first step in our data analysis exercise, and provides a clear reference point. However, the issue of the dynamic nature of the underlying data generating process is not addressed by this technique. There is an inherent path-dependency in each of the key indicators of regional economic development we are using: this year’s employment and income will clearly be positively correlated with last year’s. This is a clear case of omitted variable bias, rendering estimates biased and inconsistent. Including a lagged dependent variable on the right-hand side of the regression specification does not, however, solve the problem, as this variable is now correlated with the error term.

In order to address this endogeneity threat, we will employ the generalized method of moments (GMM) estimator for dynamic panel data. Specifically, we will use the system estimator proposed by Arellano and Bover (1995), which built on and improved the Arellano and Bond (1991) GMM estimator. System GMM analysis is specifically designed to address endogeneity issues with dynamic panel data models (i.e. biases in the coefficient estimate for the lagged dependent variable). The idea behind the Arellano-Bover estimator involves using further lags (as well as lagged differences) of dependent variable to build a set of instruments to yield consistent estimates in this dynamic setting. We implement the Arellano-Bover estimator in the following way. In all cases, we will include first lag of the respective dependent variable into the econometric specifications. Instruments will be constructed based on third to fifth lags for specifications with employment and average wage as dependent variables; and second and third lags for regressions with number of business establishments as dependent variable. These lags were chosen for instrument specifications as they satisfy both of the fundamental conditions for the system GMM estimator: no correlation between the instruments and the residuals, and no autocorrelation in the residuals.

The GMM estimator we are using has gained popularity in the literature, as studies have demonstrated its desirable properties in settings similar to the one we are facing (e.g. Bun and Windmejer, 2010). We should also note that this dynamic panel data GMM estimator has been applied to studying such issues as agglomeration economies (Brülhart and Mathys, 2008) and the impact of public debt on economic growth (Panizza and Presbitero, 2014).

To sum up our discussion, we will estimate specifications of the form:

\log Y_{it} = α_{i} + \sum β_{t} I_{t} + γ_{1} \log (Traffi c_{i, t - 1}) + γ_{2} \log (Destination s_{i, t - 1}) + δ X_{i, t - 1} + ε_{it}

where $Y_{it}$ is one of the three measures of regional economic development in metropolitan area i at time t; $α_{i}$ represent MSA fixed effects; $β_{t}$ are coefficients measuring year-specific effects, with $I_{t}$ representing the corresponding indicator variables. The key regression coefficients for our purposes are $γ_{1}$ and $γ_{2}$ . For each of the three dependent variables, we will run specifications where either $γ_{1}$ or $γ_{2}$ will be restricted to zero – that is, only traffic or only number of destinations will be included. Further, $X_{i, t - 1}$ is the vector of control variables: natural logarithm of area population, unemployment rate, airport-level concentration, average airfare, and airlines’ market shares at the area airport(s). Then, $δ$ is the vector of coefficients on those control variables. The error term is denoted via $ε_{it}$ . Finally, note that both key independent variables and control variables are lagged one year to mitigate the endogeneity problem.

We will apply to the above specification both the conventional two-stage least squares MSA fixed effects estimator, and the Arellano-Bover estimator as described earlier in this section (technically, application of Arellano-Bover estimator will require including at least one lagged dependent variable in addition to the key variables and various controls). The estimation results are presented and discussed in the next section.

Results and discussion

Results of our data analysis exercise are presented in Tables 2 –8. Tables 2 –7 show results using number of flights as the measure of airline traffic, with the aim of demonstrating, inter alia, the difference in results between the two stage least squares (2SLS) fixed effects and Arellano-Bond dynamic panel data GMM regressions. Tables 2 –4 report the fixed effects estimation results, whereas Arellano-Bover dynamic panel data GMM results are in Tables 5 –7. Each table corresponds to specifications using the same indicator of regional economic development as the dependent variable. Results for total employment are in Tables 2 and 5; output for number of establishments with employees are in Tables 3 and 6; and results for average weekly wage can be found in Tables 4 and 7. Within each table, we report six specifications. First, we report regressions which only include one of the two key independent variables, as well as a specification that includes both. Further, each of the three specifications is run excluding and including the airline effects (individual airlines’ airport market shares, lagged one year). As a reminder, we lag all the independent variables except the year dummies one year. We also treat number of flights, number of destinations, airport HHI and average airfare as endogenous variables, as use second lags of those as instruments. Thus, fixed effects results reported in Tables 2 –4 are obtained using the two-stage least squares estimation technique.

Table 2.

Regressions for total employment, fixed effects.

	(1)	(2)	(3)	(4)	(5)	(6)
Constant	3.9860** (0.8575)	3.0291** (0.6856)	3.9953** (0.8419)	3.1147** (0.7683)	3.9940** (0.8407)	3.1111** (0.7691)
Log(total flights)	0.0093* (0.0054)	0.0063 (0.0052)	–	–	0.0041 (0.0064)	0.0041 (0.0061)
Log(number of destinations)	–	–	0.0222* (0.0141)	0.0112 (0.0139)	0.0204 (0.0154)	0.0095 (0.0150)
Log(population)	0.6370** (0.0703)	0.7182** (0.0564)	0.6385** (0.0700)	0.7132** (0.0643)	0.6361** (0.0690)	0.7110** (0.0636)
Log(unemployment rate)	-0.1025** (0.0085)	-0.1015** (0.0082)	-0.1054** (0.0092)	-0.1031** (0.0089)	-0.1050** (0.0094)	-0.1027** (0.0091)
Log(average airfare)	-0.0096 (0.0153)	-0.0158 (0.0136)	-0.0118 (0.0130)	-0.0163 (0.0125)	-0.0113 (0.0136)	-0.0161 (0.0130)
Log(airport HHI)	0.0230** (0.0101)	0.0273** (0.0104)	0.0251* (0.0129)	0.0265** (0.0130)	0.0275** (0.0119)	0.0290** (0.0120)
Airline effects	No	Yes	No	Yes	No	Yes
Adjusted R-squared	0.9994	0.9994	0.9994	0.9994	0.9994	0.9994
Durbin-Watson statistic	0.5250	0.5977	0.5493	0.5980	0.5505	0.6007
Sargan statistic (p-value)	0.0584 (0.8089)	0.1435 (0.7047)	0.1196 (0.7294)	0.1234 (0.7253)	0.1282 (0.7202)	0.1330 (0.7153)

Notes: Dependent variable is the natural logarithm of total employment at the MSA level.

Number of observations = 4412.

Estimation methodology = two-stage least squares with MSA fixed effects.

All independent variables are lagged one period. Second lags of total flights, number of destinations, airfare and HHI are used as instruments.

Year indicator variables are included into all regressions.

Airline effects correspond to individual carriers’ market shares at the airport, lagged one period.

Conventional significance notations are used: * < 10%; ** < 5%.

Table 3.

Regressions for number of establishments, fixed effects.

	(1)	(2)	(3)	(4)	(5)	(6)
Constant	-0.8419 (1.0428)	-0.8951 (1.0391)	-0.8370 (1.0638)	-0.8923 (1.0546)	-0.8416 (1.0434)	-0.9042 (1.0366)
Log(total flights)	0.0144** (0.0051)	0.0135** (0.0052)	–	–	0.0142** (0.0053)	0.0137** (0.0053)
Log(number of destinations)	–	–	0.0068 (0.0077)	0.0046 (0.0085)	0.0007 (0.0076)	-0.0010 (0.0084)
Log(population)	0.7843** (0.0853)	0.7890** (0.0846)	0.7926** (0.0861)	0.7969** (0.0850)	0.7843** (0.0853)	0.7898** (0.0845)
Log(unemployment rate)	-0.0467** (0.0087)	-0.0465** (0.0086)	-0.0480** (0.0087)	-0.0478** (0.0085)	-0.0468** (0.0087)	-0.0464** (0.0086)
Log(average airfare)	0.0004 (0.0113)	0.0009 (0.0109)	-0.0011 (0.0114)	0.0004 (0.0110)	0.0004 (0.0112)	0.0009 (0.0109)
Log(airport HHI)	0.0111 (0.0086)	0.0112 (0.0087)	0.0027 (0.0085)	0.0030 (0.0086)	0.0112 (0.0088)	0.0111 (0.0089)
Airline effects	No	Yes	No	Yes	No	Yes
Adjusted R-squared	0.9993	0.9993	0.9993	0.9993	0.9993	0.9993
Durbin-Watson statistic	0.4365	0.4537	0.4230	0.4446	0.4363	0.4542
Sargan statistic (p-value)	0.1808 (0.6706)	0.1914 (0.6617)	0.2153 (0.6426)	0.2009 (0.6539)	0.2038 (0.6516)	0.2081 (0.6482)

Notes: Dependent variable is the natural logarithm of the number of establishments with employees at the MSA level.

Number of observations = 4412.

Estimation methodology = two-stage least squares with MSA fixed effects.

All independent variables are lagged one period. Second lags of total flights, number of destinations, airfare and HHI are used as instruments.

Year indicator variables are included into all regressions.

Airline effects correspond to individual carriers’ market shares at the airport, lagged one period.

Conventional significance notations are used: * < 10%; ** < 5%.

Table 4.

Regressions for average wage, fixed effects.

	(1)	(2)	(3)	(4)	(5)	(6)
Constant	−3.0564** (1.3077)	−2.4603** (1.1146)	−2.2156* (1.2154)	−1.0983 (1.1343)	−2.0382* (1.2428)	-0.8870 (1.1533)
Log(total flights)	−0.0100* (0.0055)	-0.0160** (0.0057)	–	–	−0.0326 (0.0076)	−0.0352** (0.0074)
Log(number of destinations)	–	–	0.0902** (0.0154)	0.0874** (0.0159)	0.1015** (0.0181)	0.0992** (0.0183)
Log(population)	0.8069** (0.1030)	0.7604** (0.0869)	0.7133** (0.0965)	0.6205** (0.0893)	0.7189** (0.0973)	0.6241** (0.0894)
Log(unemployment rate)	−0.0444** (0.0135)	-0.0526** (0.0138)	−0.0531** (0.0136)	-0.0601** (0.0135)	−0.0554** (0.0140)	−0.0626** (0.0138)
Log(average airfare)	−0.0907** (0.0367)	−0.0807** (0.0337)	−0.0911** (0.0304)	−0.0752** (0.0312)	−0.0921** (0.0360)	−0.0736** (0.0372)
Log(airport HHI)	-0.1003** (0.0144)	−0.0886** (0.0141)	−0.0426** (0.0141)	−0.0411** (0.0134)	−0.0606** (0.0141)	−0.0600** (0.0135)
Airline effects	No	Yes	No	Yes	No	Yes
Adjusted R-squared	0.8832	0.8939	0.8888	0.8947	0.8871	0.8935
Durbin-Watson statistic	0.8468	0.8462	0.8332	0.8411	0.8420	0.8571
Sargan statistic (p-value)	0.2559 (0.6128)	0.2416 (0.6230)	0.2894 (0.5905)	0.2990 (0.5844)	0.3229 (0.5698)	0.2751 (0.5999)

Notes: Dependent variable is the natural logarithm of the mean weekly wage at the MSA level.

Number of observations = 4412.

Estimation methodology = two-stage least squares with MSA fixed effects.

All independent variables are lagged one period. Second lags of total flights, number of destinations, airfare and HHI are used as instruments.

Year indicator variables are included into all regressions.

Airline effects correspond to individual carriers’ market shares at the airport, lagged one period.

Conventional significance notations are used: * < 10%; ** < 5%.

Table 5.

Regressions for employment, dynamic panel data GMM.

	(1)	(2)	(3)	(4)	(5)	(6)
Lagged dependent	0.4511** (0.0705)	0.4457** (0.0702)	0.3919** (0.0791)	0.3861** (0.0791)	0.3939** (0.0784)	0.3884** (0.0784)
Log(total flights)	0.0034 (0.0025)	0.0035 (0.0025)	–	–	0.0005 (0.0025)	0.0006 (0.0025)
Log(number of destinations)	–	–	0.0130** (0.0040)	0.0132** (0.0040)	0.0127** (0.0039)	0.0128** (0.0039)
Log(population)	0.3294** (0.0584)	0.3328** (0.0575)	0.3672** (0.0629)	0.3719** (0.0625)	0.3657** (0.0624)	0.3702** (0.0620)
Log(unemployment rate)	-0.0564** (0.0085)	-0.0561** (0.0085)	-0.0631** (0.0095)	-0.0628** (0.0095)	-0.0628** (0.0093)	-0.0625** (0.0094)
Log(average airfare)	-0.0067 (0.0054)	-0.0045 (0.0056)	-0.0071 (0.0056)	-0.0052 (0.0059)	-0.0070 (0.0055)	-0.0052 (0.0058)
Log(airport HHI)	0.0110** (0.0035)	0.0109** (0.0035)	0.0149** (0.0035)	0.0147** (0.0035)	0.0152** (0.0039)	0.0151** (0.0039)
Airline effects	No	Yes	No	Yes	No	Yes
Jansen’s J-statistic (p-value)	53.82 (0.1478)	52.90 (0.6270)	45.88 (0.3941)	45.06 (0.8536)	46.03 (0.4293)	45.24 (0.8750)

Notes: Dependent variable is the natural logarithm of total employment at the MSA level.

Number of observations = 4057.

Estimation methodology = dynamic panel data GMM with MSA fixed effects.

Third to fifth lags of dependent variable are used to construct the instruments utilised by the estimator.

All independent variables are lagged one period. Second lags of total flights, number of destinations, airfare and HHI are used as instruments.

Year indicator variables are included into all regressions.

Airline effects correspond to individual carriers’ market shares at the airport, lagged one period.

Conventional significance notations are used: * < 10%; ** < 5%.

Table 6.

Regressions for number of establishments, dynamic panel data GMM.

	(1)	(2)	(3)	(4)	(5)	(6)
Lagged dependent	0.7010** (0.0512)	0.6625** (0.0509)	0.6439** (0.0531)	0.6080** (0.0533)	0.6527** (0.0541)	0.6187** (0.0539)
Log(total flights)	0.0024 (0.0025)	0.0026 (0.0026)	–	–	0.0008 (0.0027)	0.0007 (0.0026)
Log(number of destinations)	–	–	0.0103** (0.0040)	0.0103** (0.0040)	0.0099** (0.0041)	0.0096** (0.0040)
Log(population)	0.2689** (0.0817)	0.2979** (0.0830)	0.3269** (0.0825)	0.3539** (0.0839)	0.3147** (0.0837)	0.3397** (0.0853)
Log(unemployment rate)	-0.1292** (0.0240)	-0.1315** (0.0235)	-0.1304** (0.0240)	-0.1310** (0.0236)	-0.1307** (0.0243)	-0.1305** (0.0236)
Log(average airfare)	-0.0456** (0.0117)	-0.0462** (0.0115)	-0.0472** (0.0119)	-0.0470** (0.0118)	-0.0467** (0.0120)	-0.0462** (0.0117)
Log(airport HHI)	0.0133** (0.0054)	0.0137** (0.0054)	0.0156** (0.0056)	0.0157** (0.0055)	0.0159** (0.0057)	0.0160** (0.0057)
Airline effects	No	Yes	No	Yes	No	Yes
Jansen’s J-statistic (p-value)	38.32 (0.1803)	33.87 (0.8351)	34.79 (0.2917)	31.76 (0.8968)	34.94 (0.3274)	32.06 (0.9107)

Notes: Dependent variable is the natural logarithm of of the number of establishments with employees at the MSA level.

Number of observations = 4057.

Estimation methodology = dynamic panel data GMM with MSA fixed effects.

Second and third lags of dependent variable are used to construct the instruments utilised by the estimator.

All independent variables are lagged one period. Second lags of total flights, number of destinations, airfare and HHI are used as instruments.

Year indicator variables are included into all regressions.

Airline effects correspond to individual carriers’ market shares at the airport, lagged one period.

Conventional significance notations are used: * < 10%; ** < 5%.

Table 7.

Regressions for average wage, dynamic panel data GMM.

	(1)	(2)	(3)	(4)	(5)	(6)
Lagged dependent	0.6214** (0.0870)	0.7032** (0.0570)	0.5959** (0.0891)	0.6906** (0.0582)	0.6314** (0.0794)	0.7029** (0.0559)
Log(total flights)	0.0162** (0.0056)	0.0107** (0.0037)	–	–	0.0069* (0.0038)	0.0045 (0.0030)
Log(number of destinations)	–	–	0.0291** (0.0076)	0.0196** (0.0046)	0.0235** (0.0061)	0.0170** (0.0040)
Log(population)	0.2136* (0.1135)	0.1651** (0.0954)	0.2793** (0.1137)	0.2191** (0.0941)	0.2570** (0.1072)	0.2101** (0.0936)
Log(unemployment rate)	-0.0568** (0.0237)	-0.0536** (0.0192)	-0.0547** (0.0228)	-0.0509** (0.0186)	-0.0535** (0.0216)	-0.0505** (0.0183)
Log(average airfare)	-0.0429** (0.0160)	-0.0258** (0.0110)	-0.0416** (0.0152)	-0.0246** (0.0107)	-0.0374** (0.0140)	-0.0236** (0.0105)
Log(airport HHI)	-0.0127* (0.0065)	-0.0063 (0.0046)	-0.0100* (0.0061)	-0.0039 (0.0044)	-0.0042 (0.0051)	-0.0008 (0.0041)
Airline effects	No	Yes	No	Yes	No	Yes
Jansen’s J-statistic (p-value)	27.11 (0.9636)	43.45 (0.8485)	27.95 (0.9520)	46.19 (0.7723)	32.90 (0.8675)	48.28 (0.7265)

Notes: Dependent variable is the natural logarithm of mean weekly wage at the MSA level.

Number of observations = 4057.

Estimation methodology = dynamic panel data GMM with MSA fixed effects.

Third to fifth lags of dependent variable are used to construct the instruments utilised by the estimator.

All independent variables are lagged one period. Second lags of total flights, number of destinations, airfare and HHI are used as instruments.

Year indicator variables are included into all regressions.

Airline effects correspond to individual carriers’ market shares at the airport, lagged one period.

Conventional significance notations are used: * < 10%; ** < 5%.

Table 8.

Dynamic panel data GMM results with total passenger volume.

	Dependent variable
	Log(employment)		Log(establishments)		Log(wage)
Lagged dependent	0.4257** (0.0727)	0.3799** (0.0797)	0.6458** (0.0511)	0.5987** (0.0536)	0.6090** (0.1001)	0.6360** (0.0970)
Log(total passengers)	0.0087** (0.0031)	0.0058* (0.0032)	-0.0015 (0.0032)	-0.0055 (0.0036)	0.0238** (0.0094)	0.0162** (0.0081)
Log(number of destinations)	–	0.0101** (0.0039)	–	0.0133** (0.0045)	–	0.0148** (0.0063)
Log(population)	0.3425** (0.0584)	0.3742** (0.0627)	0.3298** (0.0835)	0.3730** (0.0856)	0.2072* (0.1159)	0.2206* (0.1212)
Log(unemployment rate)	-0.0558** (0.0083)	-0.0617** (0.0093)	-0.1359** (0.0050)	-0.1315** (0.0238)	-0.0579** (0.0210)	-0.0493** (0.0208)
Log(average airfare)	0.0013 (0.0058)	-0.0012 (0.0062)	-0.0500** (0.0127)	-0.0513** (0.0128)	-0.0203 (0.0126)	-0.0187 (0.0130)
Log(airport HHI)	0.0124** (0.0033)	0.0160** (0.0036)	0.0116** (0.0050)	0.0238** (0.0054)	-0.0103 (0.0062)	-0.0039 (0.0058)
Jansen’s J-statistic (p-value)	49.33 (0.7235)	43.77 (0.9009)	32.82 (0.8697)	30.76 (0.9346)	22.72 (0.9999)	30.87 (0.9955)

Notes: This table presents specifications equivalent to (2) and (6) from Tables 5 –7.

Number of observations = 4057.

Estimation methodology = dynamic panel data GMM with MSA fixed effects.

Same lags as in the corresponding GMM specifications reported in Tables 5 –7 are used to construct instruments.

All independent variables are lagged one period. Second lags of total flights, number of destinations, airfare and HHI are used as instruments.

Year indicator variables and are included into all regressions.

Airline market shares at the airport, lagged one period, are included into all specifications, but not reported.

Conventional significance notations are used: * < 10%; ** < 5%.

In specifications reported in Table 8, we use total traffic volume instead of the number of flights. To save space, we only report dynamic panel data GMM results from specifications which include airline market share variables. Table 8 reports two specifications for each dependent variable: regression where traffic volume is included as the only measure of air transport activity is complemented with a specification where number of destinations served is added to it. That is, specifications reported in Table 8 are equivalent to regressions (2) and (6) from Tables 5 –7, the only difference being the replacement of total flights with the total passenger volume as the measure of traffic.

Generally speaking, fixed effects specifications demonstrate remarkable fit to data, which is not unusual and is also found elsewhere in the literature. Values of Durbin-Watson statistics are reported in the corresponding tables to clearly demonstrate the presence of significant autocorrelation in the data, and motivate the use of dynamic panel data GMM technique. In Tables 5 –8, we report Jansen’s J-statistic and the corresponding p-values to demonstrate validity of instruments. The p-values demonstrate that we cannot reject the null hypothesis that the instruments we selected are valid. Overall, we observe that moving from fixed effects to Arellano-Bover specifications, the magnitude of the coefficients that retain significance in both is reduced by about a half (with some exceptions).

We will build our discussion by comparing fixed effects and dynamic panel data estimation results. For total employment, fixed effects do not show robust effects of either total air traffic or number of non-stop destinations. While those variables are significant in specifications (1) and (3) in Table 2, these effects do not survive including airline market shares into regressions. In contrast to this, dynamic panel data estimates show that number of destinations has a significant robust effect on total employment, whereas the number of flights does not. Passenger volume has a positive and significant effect on total employment. The corresponding elasticity is still quite small – a 10% increase in the number of passengers is associated with a 0.06–0.09% increase in total employment. The effect of additional destinations keeping traffic volume constant is modest as well: a 10% increase in the number of destinations will yield a 0.13% increase in total employment, based on results from Table 5. The corresponding estimate from Table 8 is a bit smaller: a 10% increase in the number of destinations leads to a 0.1% increase in employment.

Let us now translate the above reported elasticities into the number of jobs created as traffic volume and/or number of destinations served from the region’s airport(s) increases. We will report these estimates for both sample mean and median (this is appropriate as the distribution of MSAs by the key development indicators is rather asymmetric). If we connect a median MSA with an extra destination (which will be equivalent to a 6.25% increase in this variable, given that a median airport is connected to 16 other airports with non-stop services); our estimation results imply that this will create 98 jobs in the region, other things equal. The equivalent number for the sample mean is 223 jobs. Results from Table 8 imply that adding 100 passengers per day to the volume served by the local airport creates 98 jobs at the sample mean and 314 jobs at the sample median. This difference is driven by the fact that the hypothetical traffic increase corresponds to 2% of the mean, but 20% of the median passenger volume. Further, if the hypothetical additional 100 passengers per day travel to a destination which has not yet been served from an MSA, the number of jobs created at the sample mean will increase to 254, with nearly three quarters of these jobs created through the new destination rather than an increase in passenger volume. The corresponding number at the sample median will be 323.

Results for the number of establishments are quite interesting, as the key variables change their significance as we move from fixed effects to dynamic panel data results. Specifically, fixed effects results suggest that total traffic is a more important determinant than number of destinations, whereas Arellano-Bover estimates reach the opposite conclusion. Further, results from Table 8 imply that passenger volume affects the number of establishments negatively, even though the coefficients never reach statistical significance. The elasticity implied by the dynamic panel data estimates is similar in magnitude to that reported above for the total employment. That is, a 10% increase in the number of destinations served is associated with a 0.1% increase in the number of establishments with employees (0.13% if we look at Table 8). Our results suggest that an extra destination will facilitate the opening of 15 new business establishments with employees at the sample mean; the corresponding number for the sample median is four.

Fixed effects results for the average wage are quite surprising: even though both total air traffic and number of destinations are found to be significant determinants of average income, the sign of the effect of total traffic is opposite to what we expected. This anomaly disappears in the dynamic panel data estimation results, further supporting our claim that fixed effects regressions potentially suffer from the omitted variable bias, rendering the coefficients biased and inconsistent. As with fixed effects, total flights, passenger volume, and number of non-stop destinations are found to be significant determinants of average wage. However, the magnitude of the effect is much larger for the latter variable. While a 10% increase in the total number of flights, average wage increases by 0.07–0.16%; the same figure for the 10% increase in the number of destinations is 0.17–0.3%. Admittedly, the economic significance of this effect is dubious: the median wage in our dataset implies the median annual income of $24,700 (in year 1993 dollars), so that even a 1% increase in this figure will require increasing the number of destinations served with non-stop services by at least a third (if we take a higher-end estimate of the respective effect).

Our control variables behave as expected. Population has a strong positive effect on each of the three indicators of regional economic development, whereas unemployment rate affects each of them negatively. Airport concentration positively affects total employment and number of business establishments, as we expected. However, the effect of this variable on average wage is negative. This result is quite consistent with frequent flier miles being considered a fringe benefit: firms located in areas, where employees are more likely to earn and redeem miles³ appear to be able to pay lower wages to their employees. The statistical significance of this effect is however not robust, and its economic significance is dubious. Average airfare negatively affects the number of business establishments in the area, as well as the average wage, and has no effect on the total employment.

The airfare–wage effect we observe deserves some discussion, as the reader will be understandably sceptical about the direction of causality here. Indeed, we can expect lower income areas to exhibit lower demand for air travel, naturally yielding lower airfares. However, we can look at this issue from the supply side and suggest that lower airfare may induce a reduction in air traffic volume or quality of service, suppressing business activity in the area. We should also recall that airfare in our specification is lagged one year, and the second lag is used as an instrument, lessening the possibility of the causality running in the wage–price direction. Including airline market shares does not qualitatively affect dynamic panel data estimation results. Quantitatively, we only observe visible changes in values of the estimates in average wage regressions.

Summing up, we can suggest the following takeaway messages from our data analysis. First, number of destinations served with non-stop flights from the area airports comes out as a robust predictor of each of the three indicators of regional economic development used in our study. The size of the effect is small, but impact on both employment level and number of business establishments exhibits both economic and statistical significance. Impact of the number of non-stop destinations on average wage is significant statistically, but not economically. Second, air traffic volume, when measured by the number of flights, has little robust significant effect on either of the three measures of economic development. When measured by the number of passengers, however, we can conclude that volume of air traffic positively affects employment level, and the magnitude of this effect sometimes approaches that for the number of destinations served with non-stop flights.

Our study leaves several questions open at this stage, and we believe that further data analysis could shed light on them. First, we have not considered the question of air traffic–employment relationship by sectors of economy. Note that some of the previous studies have focused on the service sector, and Sheard (2014) even indicated that the relationship between air traffic volume and manufacturing sector employment is negative. The main contribution of this study is, however, in introducing a new methodological approach to analysing the air traffic–economic development relationship, which justifies, in our view, working with the aggregate data. Second, we have conducted our analysis for the entire population of US metropolitan areas. Further studies could separately focus on larger versus smaller metropolitan areas, to evaluate whether effects we found here hold for different sub-samples.

The data analysis could also be repeated in the future, to see whether the recent wave of mergers in the US airline industry had changed any of the relationships we reported here. Since 2005, there have been several high-profile mergers in the US airline industry: US Airways–America West, Delta–Northwest, United–Continental, and Southwest–Airtran consolidation events have each led to the disappearance of a major player in the industry. We have demonstrated the link between airline market structure and regional economic development, and future studies could evaluate whether recent events which appear to have reduced the extent of competition in the US airline industry could have affected regional development.

Concluding comments

This study examines the issue of the link between air traffic and regional economic development. Our approach is different from other studies in the literature in several important dimensions. First, this paper is the first to apply the dynamic panel data GMM methodology to this issue. Second, we are using a 17-year panel covering all metropolitan areas in the US, whereas previous studies have dealt with cross-sections and/or samples of metropolitan areas. Third, we examine the potential impact of both traffic volume and the number of destinations served with non-stop flights: the literature suggests that both can matter, but the two metrics have not yet been considered together in a systematic way. Fourth, we explore the relationship between airline market structure and regional economic development. At the same time, we have selected to use the aggregate employment figures in this study, without considering the impact of air traffic on employment by sectors of the economy.

We consider three measures of economic development (total employment, number of business establishments, and average weekly wage) at the MSA level. We find that a 10% increase in the number of flights is associated with a 0.1% increase in average wage, with the impact of this variable on employment and number of business establishments not being statistically significant. Air traffic volume measured by the number of passengers has a positive and significant effect on employment and wage, but not on the number of business establishments. Number of destinations served by non-stop flights is found to have a robust positive impact on each of the three measures of economic development. A 10% increase in this variable yields a 0.13% increase in employment; a 0.1% increase in the number of business establishments; and about 0.2% increase in average weekly wage. At the sample median, connecting an MSA with an extra destination, keeping everything else constant, creates 98 jobs and facilitates the opening of four new business establishments that employ people.

Our results have very clear implications for airport management and local authorities. Specifically, our results suggest that attracting services to new destinations will have a larger impact on local economy than expanding services from a local airport to existing destinations. This could potentially clash with the incentives of network carriers, especially at smaller airports – those airlines may be more interested in responding to higher demand by expanding services to their hubs than in opening new routes from an airport. Further, local authorities should not be afraid of increased airline concentration at the local airports, as long as the potential price increase from higher concentration is not too high – the two variables tend to have opposite effects on the local economy, and there is a substantial literature demonstrating the positive link between airport concentration and airfares (see Bilotkach and Lakew, 2014, for a recent review of this body of work). At the same time, it would probably be an exaggeration to call local airports ‘engines of economic development’. It is true that there is a relationship between air traffic and employment; however, the size of the corresponding effect is quite modest. Further, the effect of air traffic on average wage is significant only statistically, but not economically.

We would like to conclude by saying that this study introduces a new promising methodological approach to examining the important issue of the relationship between air traffic and regional economic development. Further work will be required to elaborate on this issue.

Footnotes

Funding

This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.

Notes

References

Arellano

Bond

(1991) Some tests of specification for panel data: Monte-Carlo evidence and an application to employment equations. Review of Economic Studies 58: 277–297.

Arellano

Bover

(1995) Another look at the instrumental-variable estimation of error-component models. Journal of Econometrics 68: 29–52.

Aschauer

(1989) Is public expenditure productive? Journal of Monetary Economics 23: 177–200.

Bel

Fageda

(2008) Getting there fast: Globalization, intercontinental flights, and location of headquarters. Journal of Economic Geography 8: 471–495.

Bilotkach

Lakew

(2014) On sources of market power in the airline industry: Panel data evidence from US airports. Transportation Research Part A 59: 288–305.

Blonigen

Cristea

(2012) Airports and urban growth: Evidence from a quasi-natural experiment. NBER working paper no. 18278. NBER, Cambridge, MA, August.

Borenstein

(1989) Hubs and high fares: Dominance and market power in the US airline industry. The RAND Journal of Economics 20(3): 344–365.

Borenstein

(1991) The dominant-firm advantage in multiproduct industries: Evidence from the US airlines. Quarterly Journal of Economics 106(4): 1237–1266.

Brueckner

(2003) Airline traffic and urban economic development. Urban Studies 40: 1455–1469.

10.

Brueckner

Lee

Singer

(2014) City-pairs versus airport-pairs: A market-definition methodology for the airline industry. Review of Industrial Organization 44: 1–25.

11.

Brülhart

Mathys

(2008) Sectoral agglomeration economies in a panel of European regions. Regional Science and Urban Economics 38: 348–362.

12.

Bun

MGJ

Windmejer

(2010) The weak instrument problem of the system GMM estimator in dynamic panel data models. The Econometrics Journal 13: 95–126.

13.

Button

Yuan

(2013) Airfreight transport and economic development: An examination of causality. Urban Studies 50: 329–340.

14.

Chi

Baek

(2013) Dynamic relationship between air transport demand and economic growth in the United States: A new look. Transport Policy 29: 257–260.

15.

Green

(2007) Airports and economic development. Real Estate Economics 35: 91–112.

16.

Holz-Eakin

(1994) Public sector capital and the ‘productivity puzzle’. The Review of Economics and Statistics 76: 12–21.

17.

Lederman

(2008) Are frequent flier programs a cause of the ‘hub premium’? Journal of Economics and Management Strategy 17: 35–66.

18.

McMillen

(2004) Airport expansion and property values: The case of Chicago O’Hare airport. Journal of Urban Economics 55: 627–640.

19.

Panizza

Presbitero

(2014) Public debt and economic growth: Is there a causal effect? Journal of Macroeconomics 41: 21–41.

20.

Sheard

(2014) Airports and the production of goods and services. Journal of Urban Economics 80: 133–152.

21.

van den Berg

van Klink

Pol

PMJ

(1996) Airports as centres of economic growth. Transport Reviews 16: 55–65.