Presentation on theme: "E CONOMIC A NALYSIS OF F LIGHT D ELAY Nathan Boettcher Dr. Don Thompson Division of Natural Science; Pepperdine University April 3, 2015 A BSTRACT This."— Presentation transcript:
E CONOMIC A NALYSIS OF F LIGHT D ELAY Nathan Boettcher Dr. Don Thompson Division of Natural Science; Pepperdine University April 3, 2015 A BSTRACT This project began as an investigation into the phenomenon of flight delay. We approached this problem with two goals in mind. First, we used mathematical statistics and econometric methods to develop a predictive model of flight delay. An improved forecasting process has obvious benefits for customers, and would additionally shed light on the factors which airports and airlines should change in order to reduce flight delay. Our secondary goal was to complement this predictive research with a theoretical analysis of the incentive structure that consumers and producers face. We limited the scope of this model to delayed flights that resulted as a direct consequence of a consumer or producer’s action. Choosing to focus on the cost of wasted time, we developed an economic model of how airlines and customers interact under the imposition of flight delay. We then revisited the data and used our model to inform and refine our econometric methods. S UMMARY OF E CONOMIC M ODEL The joint surplus of flight, or the good generated by a successful flight, requires both the labor of airlines and capital of consumer’s time. Under this version of the model, the entire joint surplus is awarded to the capital (consumers), who then (or rather, have already) compensated the laborers for their production contribution. The issue then becomes: what happens when we introduce the constraint of flight delay? Cases where flight delay is a result of failure by parties to make proper prior investments are the only ones deserving analysis. If a consumer does not take proper precaution, thus inhibiting their ability to provide capital (time) at the pre-arranged time, the plane will depart without them; thus, they are forced to bear the full cost of the externality, and are properly incentivized to take the correct amount of precaution. If, however, a consumer overinvests, they receive no additional benefit, e.g. the flight will not depart ahead of schedule. As such, consumers have no incentive to over-invest. A producer’s (airline’s) over-incentive analysis is similar to that of a consumer’s lack of incentives to over-invest. The producer’s marginal benefit of reducing the likelihood of flight delay is even less than that of a consumer. An airline’s incentive structure regarding an under-investment in preventing delay now deviates significantly from that of the consumer. In the event that a flight is delayed, but not cancelled, the consumers are forced to bear the cost of wasted time without any form of compensation from the airlines. In this manner, the airlines could potentially abuse this monopoly use of consumer time to under-invest in precautionary policies to pursue higher profit-generating policies at the expense of consumers’ wasted time. It is this analysis that refined our econometric modeling. We predicted that analyzing source of delay as well as other airline- controlled factors would yield significantly strong predictive models. P RELIMINARY R ESULTS R ESULTS AS INFORMED BY THE E CONOMIC M ODEL FLIGHT DELAY PREDICTED BY MINUTES DELAYED R-Squared:.885, n=139 y= t-4.046x, where y= percent of arrivals on time per month (within fifteen minutes of schedule) t= time measured in months x= total minutes delayed per month (measured in 10,000s) D ISCUSSION As we refined our analysis with minutes delayed, we found several interesting results. The first was that our earlier regression of mean level of delay on airfare becomes irrelevant when we conducted regression of minutes delayed on airfare. We could now reconcile our earlier confusion at consumers effectively paying for higher or lower delay, and also suggest that the factors determining a flight’s price do not include the expected level of delay. Furthermore, we found the above model of predicting minutes delayed. The negative coefficient on this variable, combined with the positive coefficient on scheduled arrivals, suggests that a flight delayed by an airline will experience fewer minutes delayed than a flight delayed by other causes outside of the airlines’ control. This, and other findings not shown here, led us to the conclusion that flight delay occurs mostly as a result of factors outside both airline and consumer control. These factors include the available airspace above an airport, the technological capability of the control tower to predict and route flights, as well as the competency of the air traffic controllers, etc. This conclusion is similar to that found in auto transportation. Most commuter delays occur as a result of factors outside the commuter’s control, such as construction, traffic, the configuration of highways, the time of day, and the weather. Thus, when we integrate the airline and consumer’s incentives, we find that they align and are most often affected by factors outside of individual parties’ control. This would suggest that in order to improve future performance, the best solutions would be those of increased technology, such as improved air traffic management or improved ability of airplanes to operate in poor weather. Similarly, solutions that revolve around better allocating airspace and planning of air traffic effect a reduction of minutes delayed more effectively than any individual policies imposed upon airlines or their customers. I NTRODUCTION At the outset of our research, we conducted a comprehensive preliminary data analysis in order to determine both the plausibility of predicting flight delay and the existence of relationships between various FAA and economic factors. Using monthly data from , we investigated a wide variety of possible relationships, and found several key results that propelled the remainder of our research. Our first key result was a rejection of the normal distribution for levels of mean flight delay (measured as % of total flights) for SEA (p- value=.0379), MIA (.001), and ORD (.023). This surprising result arose from a significant correlation (Pearson:.678, P-value:.001) between the average flight delay and the variance of flight delay in multiple airport datasets. We then sought to find the best-fit distribution, and concluded that a gamma distribution was both the optimal model for each dataset, as determined by Q-Q plots. The parameterization of this distribution explains the correlation between mean and variance for each dataset. Secondly, we found a significant relationship between mean airfare and mean flight delay at SEA, MIA, and ORD. This relationship is positive at SEA, but negative at ORD and MIA. This further confirmed the unique nature of flight delay. These two key findings, in addition to a variety of other promising results, confirmed our preliminary hypotheses: that there are statistically significant relationships between FAA and economic factors, and that it is plausible to develop a predictive model of flight delay. R-Squared:.611, n=139 y= x+.008x 2, where y= minutes delayed per month (measured in 10,000s) x= share of delay cause due to carrier fault (measured as percentage) MINUTES DELAYED PREDICTED BY CARRIER FAULT R EFERENCES Zamir, Eyal, and Doron Teichman. The Oxford Handbook of Behavioral Economics and the Law. N.p.: n.p., n.d. Print. Wooldridge, Jeffrey M. Introductory Econometrics: A Modern Approach. Mason, OH: South Western, Cengage Learning, Print.