Click for next page ( 19


The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement



Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.

OCR for page 18
19 CHAPTER 3 Statistical Model Development The Airport Forecasting Risk Assessment Program is soft- vice models and apply that percentage change to the user fore- ware designed to assist airports with anticipating changes in cast. If, for example, the air service models show only a 2 per- air service due to external shocks (particularly fuel price and cent increase in operations due to the 10 percent fuel oil price income changes). Because every airport is different, this soft- increase, then the scenario 2011 forecast for the existing user ware is meant to assess risk in existing forecasts. Such a fore- model would be 102,000 operations. In this way, the user can cast might be an internal projection made by or for airport assess various "what-if " scenarios and how they might affect staff or it could be from an external source such as the FAA's the baseline forecast. TAF, which provides long-term projections of operations and enplanements for over 3,000 U.S. airports. The latest available 3.1 Air Service Models TAF for air carrier/air taxi operations and enplanements are used as baseline projections for the next five years in the soft- To develop the air service models, annual airport-level data ware described here, but the user may replace the TAF with his from 1990 through 2009 have been collected and analyzed. or her own baseline forecast (or adjust the TAF), if desired. The data vary both cross sectionally (across airports) and lon- The software program is based on statistical air service gitudinally (over time), resulting in a "panel" set of data. The models that are intended to provide a plausible description of FAA's hub classification system was used to categorize airports the major factors that may affect observed changes in activity into the following groups:1 at U.S. airports. As will be discussed, the activity metric used in the models is actually seat departures; the resulting predic- Large hub airports tions of seat departures then are translated into predictions of Medium hub airports operations and enplanements to match the metrics used in Small hub airports the TAF or user-supplied forecast. An overview of the logic Non-hub airports behind the software is provided in Exhibit I-22. The findings Non-primary commercial service airports from earlier tasks described in Chapter 2 formed the basis General aviation airports and other airports for designing the overall structure of, and inputs to, the air service models. Based upon feedback from the ACRP Project 03-15 panel, It is important to understand that the results from the air the scope of the analysis was limited to the first four cate- service models are used only to project changes to an existing gories, which together comprise over 99 percent of scheduled forecast that may be expected to result from user-specified vari- commercial service; airports in Alaska and Hawaii were also ations in the explanatory variables of the models. So, for exam- excluded from the analysis. In addition, large hub airports that ple, suppose an existing baseline forecast projected 100,000 serve as primary connecting hubs for major airlines were bro- operations in 2010 and 105,000 operations in 2011, and was ken out and treated separately from other large hub airports based on the underlying assumption that fuel oil prices would because their observed activity levels will depend not only on increase by 3 percent. The user could input these baseline fuel prices, income changes, and other determinants of air assumptions and forecasts into the software, and then run a scenario where fuel oil prices increase by, say, 10 percent instead. The software then will forecast what the change in 1 The analysis accounted for the possibility that an airport could change hub clas- operations from 2010 to 2011 would be based on the air ser- sification over the 20-year period.

OCR for page 18
20 Exhibit I-22. Overview of how the software works. Macro Macro Economy Economy Including Including Fuel Fuel Price Price && Variation Variation in in Existing Forecast Air Air Service Service Flights Flights Airport Airport Planning Planning Local Local Factors Factors TAF & & User Provided Including Including Existence Existence of Enplanements Enplanements Development Development Population, Population, Air Air Service Service Income, Income, Proximity Proximity toto Other Other Airports Airports Airport Airport Operating Operating Environment Environment service in local markets but also on carriers' decisions about difficult to model successfully from a statistical standpoint. how to flow traffic through the hubs and across their networks. Instead, a simpler approach was pursued that incorporates Through the modeling development process and subse- airport-specific average seat size as an exogenous variable that quent statistical testing, the non-connecting large hub airports may help to explain variations in total seat offers. This approach were combined with medium hub airports into a single cate- is discussed in more detail below. gory. Minimum activity requirements were also imposed for To moderate the data collection effort, Official Airline the non-hub airport category,2 resulting in a total of 271 air- Guidescheduled seat departures for the combined months of ports that were included in the final analysis, broken out as February and July for each year between 1990 and 2009 were follows (as of 2009): utilized as reasonable measures of average daily seat offers at Large connecting hub airports: 17 each airport included in the analysis. There will be a wide vari- Other large/medium hub airports: 43 ation of activity levels at individual facilities within each airport Small hub airports: 63 category over time. Given this background, a large airport-level Non-hub airports: 148 database was assembled that includes many data items that may help to explain the observed changes in airport-level Some consideration was given to how best to measure and domestic seat departures over the past 20 years. Exhibit I-23 define air service levels at these airports. For modeling purposes, provides a description of the explanatory variables examined average daily scheduled domestic seat departures were utilized in the work program and their expected effects on seat offers at as the appropriate measure. It is recognized that changes in seat individual airports.3,4 offers may be accomplished either by changing frequency or Standard statistical regression techniques for panel data were aircraft gauge, and that the impacts of such changes, particu- utilized to assess how some or all of these variables may help larly at small airports, may be quite different between the two explain variations in airport-level domestic seat departures alternatives. As mentioned previously, results from the air over the past 20 years. To help account for trend effects, a one- service models then are translated into predictions of opera- year lag of the dependent variable (daily seat departures) was tions and enplanements to match the metrics used in the TAF also included as an explanatory variable. As will be seen, not or user-supplied baseline forecast. all of the variables listed in Exhibit I-23 were statistically sig- The software only considers the effects of external impacts nificant contributors to the estimating equations. on domestic scheduled operations and enplanements. Any international activity at an airport is accounted for but held constant throughout the analysis. Because most scheduled 3 As seen in Exhibit 23, an attempt was made to account for variations in tech- international activity is affected by bilateral or multilateral nology and fleet mix that might help explain activity variations across airports. Admittedly, the metric used for this (average seat size) is a crude measure. Also, agreements between countries, the likely response to external the Leisure Destination Index was defined based on the notion that resort areas shocks would be difficult to assess. (such as Las Vegas and Florida airports) will likely have a much higher percent- Consideration was given to modeling changes in both seat age of traffic that originates elsewhere with the airport as a final destination, as opposed to non-leisure areas where the traffic would exhibit a more even split offers and flight offers simultaneously; however, such an between origin-destination trips that either start or end at the airport. approach would be fairly sophisticated econometrically and 4 The initial exploratory analysis also incorporated other efforts to improve the model, including testing for time dependence (so-called "autocorrelation"), alternative formulations of the explanatory variables (including different time 2 Any Essential Air Service (EAS) locations, airports without at least three years lag structures), separating out fuel price and airline cost impacts (since, as dis- of three or more flights per day, or airports where average daily seats were less cussed above, airlines may undertake measures to mitigate the effects of fuel than 100 averaged over the entire time period were excluded from the analysis. price increases), and capturing additional airport-specific effects.