Evaluating the long-run impacts of the 9/11 terrorist attacks on US domestic airline travel

Abstract

Although the US airline industry began 2001 with 24 consecutive profitable quarters, including net profits in 2000 totaling $7.9 billion, the impact of the 9/11 event on the industry was substantial. Whereas the recession that began in early 2001 signaled the end of profitability, the 9/11 terrorist attacks pushed the industry into financial crisis after air travel dropped 20% over the September–December 2001 period compared to the same period in 2000. Given the decline in domestic air travel, an important question is whether the detrimental impact of the attacks was temporary or permanent. That is, did airline travel return to the trend that existed prior to the terrorist attacks? There are theoretical reasons to the believe that it would not. Economists have long viewed travel-mode choices as the outcome of a comparison of opportunity costs and benefits. Thus, anything that permanently raises the opportunity cost of travel, holding benefits constant, should reduce the level of travel volume. To determine whether air travel was permanently reduced, we use econometric and time-series forecasting models to generate a counter-factual forecast of air travel volume in the absence of the terrorist attacks. These dynamic forecasts are compared to actual air travel levels to determine the impact of the terrorist attacks. The findings suggest that domestic air travel did not return to the levels that would have existed in the absence of the attack.

Notes

¹ Federal Aviation Administration, Aviation Policy and Plans, Aviation Industry Overview, FY 2000 and Aviation Industry Overview, FY 2001.

² Revenue passenger miles represent the product of the number of revenue paying passengers and the number of miles flown by those passengers. These data are derived from the Air Transport Association.

³ Moses and Williamson (1963) focused primarily on the choice between commuter rail versus automobile commuters, or alternatively the choice of route taken and tolls incurred for automobile commuters.

⁴ One might argue that there is another important difference between business and pleasure travellers. That is, business travellers may be unconcerned with the out-of-pocket expenses associated with the trip assuming business travel is covered by the employer. However, this logic is flawed. We assume that it is the employer, rather than the traveller who makes the travel decision. The employer considers the full opportunity cost of travel when making the modal choice.

⁵ We have compared air travel to automobile travel, yet it can be generalized to rail travel as well. The enhanced security measures for train travel did not increase the time necessary to travel by rail as much as they did for air travel.

⁶ We eliminate the period beyond 2003:01 so as to avoid any confounding influence brought on by the US war in Iraq.

⁷ The critical value of the ADF statistic for the revenue passenger miles, in level form is t = 2.88, and the actual value is 0.984, and hence we accept the null of a unit root. After first differencing, the value of the ADF is t = 5.13 which exceeds the critical value of t = 2.88. Given the strong seasonality in the data, we also test for stationarity in the seasonal difference. Again, we find that the revenue passenger miles series is stationary in the first seasonal difference (i.e. with no consecutive differencing).

⁸ The models were essentially unchanged when the estimation range was shortened with one exception. Although the income coefficient remained positively signed in the structural demand model, it became statistically insignificant.

⁹ An ARIMA model of the year-over-year change of RPM was also estimated for 1989:01–2000:04 (with an AR term at lag 1 and MA term at lag 12) and ex post forecasts for 2000:05–2001:08 generated. They are in line with those reported above; the MAPE is 1.5%, and that model over-predicts air travel by more than 1% nine times while it under-predicts RPM by more than 1% on just two occasions.

¹⁰ Note that the substantial increase in over-prediction in November 2002 may reflect the fact that the Thanksgiving holiday weekend extends into December 2002. This would be consistent with the finding of a much smaller over-prediction in December 2002. This pattern is also observed in earlier years where this phenomenon occurred.