Addressing Uncertainty about Future Airport Activity Levels in Airport Decision Making (2012)

Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

Primer on Risk and Uncertainty in Future Airport Activity P a r t I

13 As noted in Section 1.1, airport activity is subject to a significant degree of uncertainty. Some of this uncertainty derives from the fact that airports are part of the larger avia- tion industry, and some is due to the specific characteristics and circumstances of individual airports. The sections that follow provide an overview of the uncertainty and risk factors facing airports and their implications for the performance of air traffic forecasts. 2.1 Defining Risk and Uncertainty Risk has an unknown outcome, but we know what the under- lying outcome distribution looks like. Uncertainty also implies an unknown outcome, but we don’t know what the underly- ing distribution looks like. So games of chance like roulette or blackjack are risky, while the outcome of a war is uncertain. Knight said that objective probability is the basis for risk, while subjective probability underlies uncertainty. (Mauboussin, 2006, p. 36) This distinction between risk and uncertainty has been criticized on the basis that, in most real-world cases, it is very difficult to obtain accurate and complete a priori probability information (Taleb, 2007). Even in cases where extensive data exists (e.g., the stock market), probabilistic analysis based on historical data has proven to be a poor predictor of future events (such as stock market crashes). Taleb put forward the idea of black swan events, a concept that goes beyond Knightian uncertainty, referring to high-impact events that are impossible to predict or anticipate because there is no historical precedent for them (Taleb, 2007). A similar concept, the unknown unknown, was popularized following its use by former Secretary of Defense Donald Rumsfeld (in a press briefing on February 12, 2002). In this guidebook, the terms risk and uncertainty are used interchangeably. There are some variables of inter- est that reflect the characteristics of Knight’s definition of risk (such as fuel prices, for which there are extensive data and forward markets) and others that are closer to Knightian uncertainty (such as terrorist events or techno- logical advancement). There may also be unknowns that are black swans in that it is not known what these events are or when they may occur. The discussion and methodology in this guidebook are designed to enhance the robustness of airport decision mak- ing in the face of all these forms of risk and uncertainty (both positive and negative) regardless of the information available on them. Data can be incorporated into the analysis when C h a p t e r 2 Uncertainty in Airport Activity Key Takeaways Separate and distinct definitions of risk and uncertainty have been put forward, but there is not total agreement on these. In this guide- book the terms risk and uncertainty are used interchangeably to refer to the broad range of unpredictable factors, both positive and negative, that influence future airport activity. In common with much of the literature in this area, this guidebook uses the terms risk and uncertainty. Before dis- cussing their nature in the airport context, these two terms are discussed in the following, in particular as to whether they refer to different concepts. In the field of economics, Knight first formally distin- guished between risk and uncertainty. He defined risk as a situation where some quantification is possible (i.e., probabil- ities can be assigned to events), while uncertainty (sometimes referred to as Knightian uncertainty) is immeasurable and not possible to calculate (Knight, 1921). Mauboussin expands on this definition as follows:

14 available, but fundamentally this guidebook does not differ in its treatment of risk and uncertainty. 2.2 Sources and Types of Uncertainty Facing Airports • Airline restructuring or failure. A number of airports have experienced extreme changes in traffic volumes and traffic mix as a result of the restructuring or failure of an incum- bent airline (see examples in Chapter 3). • LCC growth. The entry or expansion of a low-cost carrier can result in rapid traffic growth and put competitive pres- sures on other carriers at the airport (sometimes resulting in other carriers cutting back service). • Competition from other airports. For example, LCC growth at secondary airports has placed additional com- petitive pressures on primary airports. Airport competition is arguably more pronounced for air cargo than passenger traffic—shippers have considerable flexibility to change cargo routings (modes of transport as well as airports) and will do so for relatively small cost or efficiency improvements. • Technology change. Developments in aircraft technology, air traffic control, and passenger facilitation can have impli- cations for traffic levels and airport capacity. For example, a new aircraft design that lowers unit costs can open up new route opportunities. Changes in aircraft technology can also affect air cargo. For example, the use of more narrow-body aircraft on longer haul routes could reduce the amount of belly space available for cargo. • Regulatory and government policy. Government deci- sions regarding security requirements, noise restrictions, emission standards, carbon taxes and caps, and so forth can all have implications for air traffic volumes (whether for passengers, cargo, or aircraft operations), as can changes to air service bilaterals and open skies agreements. Taxa- tion levels on the aviation sector can also affect airport traffic. • Social or cultural factors. Changes in the attitude of soci- ety and business toward the use and value of air travel can affect traffic volumes and mix—for example, a greater willingness of businesses to use Internet technologies to conduct meetings rather than flying staff to meetings. Public concerns regarding greenhouse gas emissions from air transport may lead to some consumers curtailing their air travel. • Shock events. Shock events such as the September 11, 2001 (9/11) terrorist attacks, the SARS outbreak in 2003 (which affected air traffic at Hong Kong and Toronto in particular), and severe weather events can have short- and long-term implications for air traffic development. • Statistical or model error. Often, forecasts of future airport activity are derived from analytical models of air traffic activity. For example, a model may be based on a statistical relationship between air traffic and gross domestic product (GDP) growth. Model mis-specification or errors in the data analysis can result in an erroneous Key Takeaways Both the overall volume of traffic and the mix of traffic at an airport are subject to risk and uncertainty. The causes of this uncertainty can range from the fairly global (e.g., the state of the national economy) to the local (e.g., the performance of a local air carrier). Uncertainty about future airport activity levels can mani- fest itself in two fundamental ways: 1. The overall volume of traffic: total passengers, total air- craft operations, air cargo volumes, and so forth, and their volatility over time. 2. The mix or type of traffic at the airport: domestic versus international, origin/destination (O/D) versus connecting, low-cost carrier (LCC) versus full service/legacy carrier, turboprop versus regional jet versus large jets, and so forth. In either case, there can be profound implications for the development of airport facilities and operations. For exam- ple, declines in total passenger traffic can lead to facilities that are underutilized, with high operating and capital costs, and supported by too small a traffic base; or the sudden growth of international traffic could require the airport to enhance its facilities for processing international traffic (e.g., immigra- tion control, customs inspections, security processes). Uncertainty in the volume and mix of airport activity stems from various sources. Some are fairly global, while others are specific to the airport in question. Sources of uncertainty include: • Global, regional or local economic conditions. His- torically, air traffic has more or less tracked economic conditions—increasing during periods of economic growth (generally faster than the economy) and declin- ing during recessions. • Airline strategy. Airlines’ decisions to start, expand, con- tract, or shut down service have major implications for the airport, particularly when the airline makes up a large share of airport operations (e.g., hub carriers).

15 forecast. More fundamentally, the historical relation- ships captured in the model may not continue into the future due to structural changes in the market. For exam- ple, GDP may not drive traffic growth in quite the same way as it has in the past. Some of these factors have only short-term implications for airport traffic. Traffic levels often recover and revert to trend following a recession. Traffic volumes in the United States reverted back to trend approximately 4 years after the 9/11 attacks. Other factors may have longer-lasting implications for an air- port or may trigger longer-lasting impacts. The loss of a major carrier can result in depressed traffic volumes for an extended period, and while U.S. traffic as a whole did recover from the 9/11 attacks, some airports saw long- term changes in traffic as a result of airline decisions made after the attacks. 2.3 Forecast Accuracy and Traditional Airport Planning Air traffic forecasting is a crucial building block of the air- port decision-making and planning process. The configura- tion and the size of an airport are often determined on the basis of detailed estimates of long-term airport activity. The standard airport master plan approach can be characterized as follows: 1. Determination of the forecast, and 2. Selection of a single plan that best suits this forecast. This standard practice is embedded, for example, in the guidelines for master planning issued by both the Federal Aviation Administration (FAA, 2005) and the International Civil Aviation Organization (ICAO, 1987). This approach is fairly workable in a largely stable business environment Key Takeaways Airport decision making and planning relies heavily on forecasts of future airport activity. Research has found that airport traffic is subject to greater volatility now than has been the case in the past. The accuracy of air traffic forecasts has been mixed at best, due in great part to unanticipated events and circumstances not accounted for in the forecasts. where changes in traffic patterns are slow and predictable. Certainly, this model of airport planning characterized the development of airports in the pre-deregulation era. However, since airline deregulation (in 1978 in the United States and in the 1990s in Europe), the aviation industry has arguably become more volatile and unpredictable. Empirical research confirms that airline deregulation has indeed increased the traffic volatility experienced by U.S. air- ports. For example, de Neufville and Barber (1991) found that deregulation had resulted in a more than threefold increase in volatility (measured in terms of actual traffic vol- umes versus the long-term trend). Such variability creates significant challenges when trying to predict future levels of airport activity. Maldonado com- pared forecasts and actual volumes of total annual aircraft operations at 22 airports in the six states of the FAA New England region. The forecasts were obtained from individ- ual airport master plans and the data on actual traffic vol- umes from FAA records. Ratios of forecast to actual volumes were calculated at all airports for three planning horizons: short-term (5 years), medium-term (10 years), and long- term (15 years). Overall, forecasting errors were found to be large, with ratios ranging from 0.64 (forecast traffic was two thirds of actual traffic achieved) to 3.10 (forecasts were over three times actual traffic) and tended to get larger for longer forecasting horizons. In addition, no relationship was found between forecast errors and the size of the airport (Maldonado, 1990). The challenges in air traffic forecasting are illustrated in Figure 1, which shows passenger enplanements at Hartsfield- Jackson Atlanta International Airport, the world’s busiest airport (in terms of total passengers and aircraft operations) between 2000 and 2011. Also shown are the FAA’s Terminal Area Forecasts (TAFs) of traffic at the airport for various years between 2001 and 2009. As can be seen, the forecasts were sub- ject to considerable revision since the airport traffic was af- fected by the 9/11 terrorist attacks, recession in 2001 and 2008/ 09, and Delta Air Line’s entry into bankruptcy protection in 2005 and subsequent restructuring. Many of the forecasts produced failed to track actual volumes over the period reviewed. Due to the observed unreliability of airport traffic fore- casts, de Neufville and Odoni argue that “the forecast is always wrong” (de Neufville and Odoni, 2003, p. 70) since there will always be unanticipated events and circumstances that will cause traffic to deviate from the expected trend. Thus, future traffic development, more likely than not, will be very different from the forecast. Therefore, a master plan based on a single traffic forecast and a single future is much more likely to be wrong than right.

16 35 37 39 41 43 45 47 49 51 53 55 2000 2002 2004 2006 2008 2010 2012 2014 Pa ss en ge r E np la ne m en ts (M illi on s) Actual Traffic TAF 2001 TAF 2003 TAF 2005 TAF 2007 TAF 2009 Source: Hartsfield-Jackson Atlanta International Airport operational statistics and FAA TAFs. Figure 1. Actual and forecasted total passenger enplanements at Hartsfield-Jackson Atlanta International Airport.

17 The following sections provide examples where unfore- seen events and changing conditions, not accounted for in the original forecasts, had a significant impact on an airport, either positive or negative. These examples were selected to illustrate the difficulties airports face as a result of air traffic uncertainty and in no way are meant to suggest any deficien- cies in the decision making of the airport authorities. 3.1 Lambert-St. Louis International Airport predicted that traffic would reach 20 million enplaned pas- sengers by 2006 and 25 million enplaned passengers by 2012. In early 2001, TWA again experienced financial difficulties, which resulted in its assets being acquired by American Air- lines’ (AA) parent company (AMR Corporation), and the air- line declared bankruptcy for a third time. AA initially indicated that it planned to keep STL as a hub, in light of the conges- tion at Chicago O’Hare. However, with the severe downturn in traffic that followed the terrorist attacks of 9/11, AA began reducing its STL operations, focusing more on its main hub operations at Chicago O’Hare and Dallas/Fort Worth (DFW). In 2003, AA converted many routes to regional services, result- ing in a significant loss of total capacity. As a result of TWA’s collapse, passenger volumes at STL declined by 56% between 2000 and 2004, to 6.7 million enplanements. Traffic failed to recover significantly from this level and declined further between 2008 and 2010 as result of economic conditions and further cutbacks by AA. The cut- backs by AA have been somewhat offset by Southwest Airlines, which increased operations at the airport in 2010 and is now the largest carrier at STL in terms of departures. Due to delays in the planning process, construction of the proposed third runway did not start until 2001. While some consideration was given to delaying the construction, given the uncertainty regarding the operations of TWA/AA, it was decided to continue development. This decision was supported by the FAA on the basis of enhancing national system capacity. During the period of construction, FAA continually revised its forecasts for STL enplanements downward. By the time the FAA completed its forecasts for 2003, it was projecting STL’s pas- senger traffic in 2015 would be less than levels achieved in 1993. As a result of this traffic decline, Concourse D, previously used by TWA, has been largely empty, and large parts were closed off in the fall of 2008. In addition, Concourse B has limited traffic and Concourse C is not currently used for commercial traffic. The newly built runway, completed in 2006 at a cost of $1.1 billion, is heavily underutilized. C h a p t e r 3 Implications of Unforeseen Events and Conditions Key Takeaways: Loss of a Major Carrier The airport experienced large reductions in passenger traffic due to the collapse of its largest carrier, resulting in excess airport capacity and unused facilities. In 1982, Trans World Airlines (TWA) named Lambert- St. Louis International Airport (STL) as its principal domes- tic hub, which resulted in passenger traffic at the airport almost doubling between 1981 and 1986, from 5.3 million to 10.0 million enplaned passengers (see Figure 2). During the 1990s, TWA drove strong traffic growth again, with total enplanements at the airport reaching 15.3 million passengers in 2000, despite the carrier entering bankruptcy protection twice (in 1992 and 1995). Connecting traffic accounted for a large proportion of passenger volume during this period. In response to this growth, a 1994 airport master plan update for STL proposed the construction of a third runway. The new runway was expected to allow STL to reduce delay times (which the airport had become prone to), improve capabili- ties in adverse weather, enhance capacity, and continue to accommodate TWA’s hubbing operations. This recommen- dation was supported by FAA TAFs around that time, which

18 In recent years, there have been efforts to develop STL as an air cargo hub to take advantage of its excess capacity. In 2009, the public–private Midwest-China Hub Commission was established to develop an implementation plan for air cargo services focused on China (St. Louis Business Journal, 2009). In the fall of 2011, China Cargo Airlines started a once weekly service to STL from Shanghai having signed a 2-year lease for cargo space at the airport (Lea, 2011). 3.2 Baltimore/Washington International Thurgood Marshall Airport During the early 1980s, Baltimore/Washington Inter- national Thurgood Marshall Airport (BWI) experienced strong growth due to Piedmont Airlines selecting the air- port as a hub. (Piedmont was absorbed into US Airways in 1989.) In the late 1980s, international services started to develop at BWI, operated by US Airways and other carriers. By 1992, the airport had service to 11 international destina- tions in Europe, Canada, and Mexico. International traffic doubled between 1989 and 1991, reaching approximately 323,000 enplanements. Marketing studies conducted by the airport from the early 1990s indicated that interna- tional enplanements at BWI were forecast to reach as high as 500,000 by 2000 and 700,000 by 2010. In anticipation of the projected international traffic increases, particularly on US Airways, which was expected to connect passengers at BWI, the airport began construction of a new international terminal in 1994 (along with other projects, including a runway extension allowing BWI to accommodate larger aircraft). In 1993, Southwest Airlines launched service from BWI. Over the next several years, the number of destinations served by Southwest from BWI grew steadily. The competi- tive pressure from Southwest, as well as other industry fac- 0 5 10 15 20 25 30 1985 1990 1995 2000 2005 2010 2015 Pa ss en ge r E np la ne m en ts (M illi on s) Actual Traffic TAF 1998 TAF 2001 TAF 2002 TAF 2003 TAF 2009 TWA declares bankruptcy TWA declares bankruptcy for the second time TWA declares bankruptcy for the third time and AA buys TWA. Construction of new runway begins AA reduces services at STL AA terminates its focus city at STL Source: Lambert-St. Louis International Airport passenger statistics and FAA TAFs. Figure 2. Actual and forecasted total passenger enplanements at Lambert-St. Louis International Airport. Key Takeaways: Significant Change in Traffic Mix Downsizing by US Airways and the growth of Southwest Airlines resulted in a significant shift in traffic mix, leading to the underutilization of the international terminal and congestion in the domestic facilities.

19 tors (including the 9/11 attacks) contributed to US Airways scaling down its BWI operations and moving operations to Philadelphia. US Airways’ moving its hub to Philadelphia resulted in BWI losing about a third of its international traffic. With limited options for connecting traffic, and with opera- tions dominated by LCCs, BWI struggled to attract addi- tional international service. As a result, total international passenger enplanements dropped by half between 1991 and 2009, falling from the 1991 high of 323,000 to less than 163,000 in 2009 (see Figure 3). International traffic increased slightly again in 2010, reaching almost 190,000 enplaned passengers. The decline in international traffic left BWI with an underutilized international terminal. However, the rapid growth of Southwest led to increased demand for domes- tic facilities. Despite having an underutilized international facility, BWI had to undertake additional capital spending on its domestic facilities because the international terminal was not suitable to meet the needs of Southwest. Although international traffic failed to reach forecast lev- els, total traffic was broadly in line with the long-term fore- casts, as illustrated in Figure 4. However, the mix of traffic was quite different from the forecasts. This example also sug- gests that forecasting O/D traffic is perhaps inherently less risky than forecasting connecting traffic. Total O/D traffic at BWI developed in a manner reasonably close to the fore- cast, but the connecting traffic transferred to another airport (Philadelphia). 3.3 Louis Armstrong New Orleans International Airport 0 50 100 150 200 250 300 350 19 72 19 74 19 76 19 78 19 80 19 82 19 84 19 86 19 88 19 90 19 92 19 94 19 96 19 98 20 00 20 02 20 04 20 06 20 08 20 10 Pa ss en ge r E np la ne m en ts (T ho us an ds ) Piedmont announces hub Sept. 11, 2001, and recession March - Nov 2001 Recession Dec 2007 - June 2009 End of US Airways alliance with British Airways US Airways moves international operations to Philadelphia New flights by KLM, Icelandair, US Airways Source: U.S. DOT data and Ralph M. Parsons Company (1987). Data from 1985 to 1988 could not be obtained. Note: This figure excludes military passengers carried by the U.S. Air Mobility Command. Figure 3. International passenger enplanements at Baltimore/Washington International Thurgood Marshall Airport. Key Takeaways: Large Natural Disaster Devastation by Hurricane Katrina in 2005 resulted in an immediate and substantial loss in passenger traffic, which has not yet been recovered. In August 2005, Hurricane Katrina hit New Orleans. The storm resulted in one of the largest natural disasters in U.S. history, causing widespread flooding, billions of dollars of

20 0 2 4 6 8 10 12 19 72 19 74 19 76 19 78 19 80 19 82 19 84 19 86 19 88 19 90 19 92 19 94 19 96 19 98 20 00 20 02 20 04 20 06 20 08 20 10 Pa ss en ge r E np la ne m en ts (M illi on s) Actual Traffic 1987 Master Plan Forecast (Baseline) Piedmont announces hub First Gulf War and recession Southwest Airlines launches services 9/11 and recession Recession US Airways "de-hubs" Source: Analysis of data from FAA and the 1987 Master Plan. Figure 4. Actual and forecasted total passenger enplanements at BWI. property damage, and more than 1,300 deaths. Hurricane Rita followed less than a month later, adding to the devas- tation. During the immediate time period of the storm and recovery, Louis Armstrong New Orleans International Air- port (MSY) was closed to commercial air traffic. However, MSY itself escaped sizable damage during the hurricanes and reopened after only 2 weeks. Nevertheless, the hurricane had a catastrophic impact on the airport and resulted in a 39% decline in traffic in 2006 relative to 2004 (pre-Katrina) levels, as shown in Figure 5. While nearly all the major carriers have returned to MSY, passenger traffic levels in 2009 were still 21% below 2004 levels, due to the loss of tourism and conventions, declines in the local population, economic decline, and reduced air carrier capacity. Moreover, traffic levels are well below the FAA’s TAFs produced before Hurricane Katrina, and more recent forecasts have been revised downward, suggesting a long recovery period. According to the 2010 TAF, the FAA does not expect MSY to return to 2004 passenger lev- els until 2021. Hurricane Katrina is an example of a sudden and unex- pected event. Mitigating the traffic impacts of such an event is challenging. The airport has instigated an incentive program to encourage new service to the airport. 3.4 Bellingham International Airport Key Takeaways: Unexpected Upside Traffic Growth Entry of low-cost carrier Allegiant Air resulted in much higher than forecast traffic growth. Bellingham International Airport (BLI) is located in Whatcom County, Washington, 3 miles northwest of Bellingham, a city with a metro population of approximately 200,000. The airport is approximately 21 miles south of the Canadian border and 90 miles north of Seattle. Prior to the 9/11 terrorist attacks, BLI had service to Seat- tle operated by Horizon Air and United Express/SkyWest

21 (accounting for 79% of seat capacity in 2000), plus service to the San Juan Islands off the coast of Washington (URS et al., 2004). In October 2001 the United Express/SkyWest services were terminated, in part due to service rationaliza- tion following the 9/11 attacks. Traffic declined to a low of 64,365 in 2003 due to further service cutbacks. In August 2004, low-cost carrier Allegiant Air entered the market at BLI and started service to Las Vegas. Over the next few years, the airline increased the range and frequencies of service out of BLI. By January 2008, Allegiant Air opened up its sixth base at BLI. As of December 2011, the airline oper- ated direct service to Las Vegas, Los Angeles, Palm Springs, San Diego, Oakland, and Phoenix. As a result of Allegiant’s entry, traffic at BLI increased by 374% between 2004 and 2010, an average growth rate of nearly 30% per annum. The airport also experienced the short-lived entry of two other carriers: • Western, an LCC headquartered in Bellingham, entered the BLI market (serving three destinations) on January 18, 2007, but ceased operations due to financial difficulties on February 7, 2007. • Skybus, an ultra low-cost carrier based in Ohio, operated flights to Ohio between May and October 2007 before can- celling the service. As can be seen in Figure 6, since the entry of Allegiant, traffic levels have greatly exceeded forecasts produced by the FAA and in the airport’s 2004 master plan. BLI’s expansion plans were affected by the rapid and un- expected increase in traffic. For example, the expansion of the terminal building, originally scheduled to be completed in 2018, has been accelerated to be completed by 2013. In addition, a $29 million runway resurfacing project was completed in September 2010 that will enable larger aircraft to operate at BLI (Puget Sound Business Journal, 2010). This will allow Allegiant Air to operate its larger B757 aircraft at BLI, as well as existing MD-80 services, potentially contributing to fur- ther traffic growth. This BLI example shows that upside risk can lead to a need for rapid airport expansion in order to keep airlines and passengers satisfied to ensure that airlines can continue to expand their services and to avoid congestion that may lead to a loss of passengers or the exit of a carrier. 0 1 2 3 4 5 6 7 8 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 2014 Pa ss en ge r E np la ne m en ts (M ill io ns ) Actual Traffic TAF 2001 TAF 2004 TAF 2005 TAF 2009 Hurricane Katrina Source: FAA TAFs. Figure 5. Actual and forecasted total passenger enplanements at Louis Armstrong New Orleans International Airport.

22 0 50 100 150 200 250 300 350 400 450 1985 1990 1995 2000 2005 2010 2015 2020 Pa ss en ge r E np la ne m en ts (T ho us an ds ) Actual Traffic TAF 2000 TAF 2003 Master Plan Forecast United Express/ SkyWest exits in 2001 Allegiant enters in August 2004 and rapidly develops service Allegiant opens base at BLI in January 2008 Source: Bellingham International Airport Master Plan Update (URS et al., 2004), FAA TAFs, and BLI airport statistics. Figure 6. Actual and forecasted total passenger enplanements at Bellingham International Airport. 3.5 Zurich Airport and Brussels Airport As a result of restructuring, Swiss International Air Lines cut its seat capacity at ZRH by 43% between 2000 and 2004. The airline was subsequently taken over by Lufthansa (in 2007) but continues to operate as a separate brand. The capacity cuts by its home carrier contributed to a 25% decline in total traffic at ZRH between 2000 and 2004. (Swissair accounted for 66% of traffic before its failure.) In the years following, traffic gradually recovered (by 5.4% per annum) to reach close to its pre-collapse levels by 2008 before declining in 2009 due in large part to the global recession. A similar story occurred at Brussels Airport (BRU), which was the primary hub of Sabena, the former national carrier of Belgium. In fact, the two events are connected since it was the failure of Swissair to make a scheduled payment of U.S.$200 million to Sabena in 2001 that triggered Sabena’s collapse. In November 2001, Sabena ceased operations, and many of its assets were transferred to a short-haul sub- sidiary, Delta Air Transport. In early 2002, the airline was renamed SN Brussels Airlines. The new airline cut seat capacity by 68% between 2001 and 2002. In 2007, the airline merged with Virgin Express (an LCC based at BRU) and was renamed Brussels Airlines. The airport experienced a 33% decline in traffic between 2000 and 2002 (Sabena accounted for 55% of traffic before Key Takeaways: Collapse and Restructuring of the Main Hub Airline Both airports experienced the collapse of a home carrier that was partially replaced by a smaller, restructured airline. Traffic declined sharply at both airports, which was then followed by varying degrees of recovery. Zurich Airport (ZRH) served as the hub for Swissair, former national carrier of Switzerland. Due to its central position in Europe, Swissair (and thus ZRH) profited from generating transfer passengers. However, with the deregula- tion and liberalization of the air industry in the European Union (which Switzerland participated in despite not being a member of the EU) and the economic downturn during 2000 and 2001, Swissair experienced severe financial difficulties, leading to the airline filing for bankruptcy in October 2001. Many of Swissair’s assets were taken over by a subsidiary of Swissair, changing the name to Swiss International Air Lines.

23 its failure), after which traffic grew by 4.3% per annum so that by 2008 passenger traffic levels were 21% below its pre- collapse levels. Both airports saw traffic decline dramatically, which was then followed by modest recovery. Both have yet to recover fully to pre-collapse levels. Even though some recov- ery has occurred, both airports are way off the traffic trend that was apparent prior to the airline failure. In both cases, the traffic recovery was largely a result of the home carrier rather than other carriers replacing the lost capacity. Despite the loss of traffic following Swissair’s collapse, ZRH decided to continue expansion plans started in 2000. In September 2003, ZRH completed its new Dock E. As a consequence, ZRH had considerable excess capacity. The lack of traffic led to a closure of the existing Dock B in the same year. As an example of adapting facility use, Dock B was converted into an event venue (EventDock) for a period of time, although it has since undergone reconstruction and was reopened in December 2011. BRU had started construction of a new pier (Pier A), which was completed in May 2002, before the collapse of Sabena. Fol- lowing Sabena’s collapse, the decision was made to close the satellite terminal that had originally served as the terminal for Figure 7 shows total passenger enplanement at Washing- ton Dulles International Airport (IAD) since 1990. Strong growth occurred in 1998 to 2000 as a result of a build-up of Key Takeaways: Widely Fluctuating Traffic Volumes Over the last decade, the airport has experi- enced widely fluctuating traffic volumes, due largely to market entries and exits of diverse air carriers as well as economic downturns and the 9/11 terrorist attacks. This has made forecasting challenging and resulted in large changes in the airport outlook. 0 5 10 15 20 25 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 2014 Pa ss en ge r E np la ne m en ts (M illi on s) Actual Traffic TAF 2000 TAF 2001 TAF 2003 TAF 2004 TAF 2005 TAF 2009 Source: FAA TAFs and Washington Dulles International Airport air traffic statistics. Figure 7. Actual and forecasted total passenger enplanements at Washington Dulles International Airport. the intercontinental operations of the airline. After some recon- struction, that terminal is now used as office space (BRU, 2009). 3.6 Washington Dulles International Airport

24 traffic levels, followed by Virgin America a year later. Passen- ger levels declined in 2008 and 2009, with JetBlue, Southwest, and Mesa all making significant cuts (of more than 10%) in capacity, due in large part to economic conditions. The FAA TAFs for IAD since 2000 are also shown in Fig- ure 7. These forecasts have required considerable revisions in response to changing market conditions. These revisions demonstrate how fluctuations in carrier activity and traffic volumes can create considerable uncertainty regarding the long-term outlook for traffic. In the case of IAD, most of this volatility has been the result of domestic operations, which have caused domestic passenger volumes to change by as much as -20% to +41% year-on-year. By comparison, inter- national traffic has exhibited more steady growth, averaging a 4.2% increase per annum between 2000 and 2010. competition between United Airlines and US Airways. After 2000, US Airways decided to remove considerable capacity from IAD, which combined with a similar capacity response from United Airlines, the economic recession in the United States, and the 9/11 terrorist attacks, resulted in traffic levels declining. In June 2004, Independence Air started service as an LCC based at IAD. This resulted in enplaned passenger volumes increasing significantly in 2004 and 2005. However, Indepen- dence Air ceased operations in January 2006, resulting in pas- senger volumes declining again in that year. In addition, other LCCs entered the market at IAD. JetBlue Airways commenced air service in November 2001, followed by Mesa Airlines, a regional LCC, in 2004. In 2007, Southwest Airlines began operations at the airport, further contributing to increased

25 In the aviation sector, procedures used to account for risks and uncertainties have traditionally been ancillaries to the methods used for developing demand or air traffic forecasts. This is because in the early stages of the aviation planning process, decisions are made regarding the intended use of forecasts, and a method (or set of methods) is subsequently selected for producing demand projections based on perceived accuracy, ease of use and interpretation, and adaptability or flexibility in applications. As described in ACRP Synthesis 2: Airport Aviation Activ- ity Forecasting, the techniques that are often employed in forecasting can be grouped within four categories (Spitz and Golaszewski, 2007): • Time-series methods: trend extrapolation using statisti- cal techniques that rely on lagged and contemporaneous traffic data to infer future values. • Econometric modeling with explanatory variables: sta- tistical techniques that examine the relationship between traffic and possible explanatory variables, such as the econ- omy (e.g., gross domestic product or personal incomes), population, fuel prices, and so forth. • Market share analysis: a technique used to forecast local activity as a share of some larger, aggregated forecast. For example, a forecast of airport traffic may be based on its assumed share of national traffic, as forecasted by a third party. • Simulations: a technique involving the use of com- plex models that evaluate different snapshots of a travel network. The sections that follow examine approaches for address- ing uncertainty in aviation demand forecasting. Where information from the aviation industry is limited, relevant methodologies from other industries in the transportation sector (e.g., toll road demand forecasting) and other sectors or disciplines (e.g., demographic forecasting) have also been reviewed. Readers requiring more detailed information are encouraged to review Appendix D. 4.1 Standard Procedures to Account for Uncertainty in Aviation Demand Forecasting C h a p t e r 4 Approaches for Incorporating Uncertainty into Demand Forecasting Key Takeaways Three fairly common procedures are used in air traffic forecasting to account for uncertainty: • High and low forecasts • What-if analysis • Sensitivity analysis However, these approaches provide only a cursory understanding of risk and uncertainty and are rarely incorporated into the planning process in any meaningful way. Three standard procedures commonly used to account for uncertainty in demand forecasts are described in the follow- ing sections. All three have been used in conjunction with the four forecasting techniques described previously. 4.1.1 High and Low Forecasts In developing demand projections under this approach, many (or all) of the forecasting assumptions are modified in one direction to produce an optimistic forecast, then in the opposite direction to produce a pessimistic forecast. For example, the high forecast may assume that GDP growth will be one percentage point higher per annum than the rate used in the baseline or most likely forecast.

26 This procedure is fairly common in air traffic forecasting since it can be easily incorporated into standard forecast - ing techniques, including market share analysis and econo- metric modeling. It is also one of the easiest procedures to implement once a forecasting model has been developed. In addition, interpreting the outcomes of the analysis is generally straightforward and does not require any spe- cific knowledge of probability theory—the concepts of “high” and “low,” or “optimistic” and “pessimistic,” are intuitive and generally understood. The low/high forecasts can be based on analysis of trends, judgment, or projec- tions of key input values developed outside of the forecast model. However, there are a number of limitations to the high/ low forecast approach: • The range between the low and high forecasts is often rela- tively small (e.g., +/- 25% relative to the baseline forecast). However, as illustrated in Chapter 3, traffic deviations can be much larger than this. • The forecasts provide no information on the likelihood of such outcomes, which limits their use and applicability. Indeed, the very idea that multiple assumptions will veer from baseline expectations in the same direction is itself arbitrary. • Although frequently provided in forecast exercises, the high and low forecasts have little input into subsequent planning efforts. The baseline forecasts are used for most of the planning, with little consideration given to the high and low forecasts. 4.1.2 What-If Analysis In this procedure, also known as “impact analysis,” the impact of a single event (such as an economic downturn, a rapid increase in fuel prices, or a health pandemic) is estimated and reported relative to the baseline forecast. Uncertainties are typically assumed to be event-specific and are defined as either threats or opportunities. The procedure involves the following steps: 1. Establish a baseline forecast using any of the four techniques mentioned at the beginning of this chapter and assuming that none of the identified events will materialize. 2. Determine the magnitude of the events (e.g., the sever- ity and duration of an anticipated downturn in economic activity or the percentage increase in fuel prices). 3. Estimate the effect of each event, taken individually, on the baseline forecast. This can be done with the estimated parameters of the forecasting model itself, by using infor- mation or similar events in the past, or through judgment (or a combination of the three). 4. Report the outcomes of the analysis. Impacts are typi- cally reported one at a time, with no reference to potential dependencies or correlations with other events. One of the main strengths of this approach is that it can be used with a variety of forecasting tools and techniques, including econometric methods and complex simulation models. On the other hand, the use of what-if analysis requires reference to a baseline, most likely forecast, the probability of which is typically unknown. Furthermore, as with the high/low forecast, the probability of alternative outcomes under different what-ifs often remains unknown, mak- ing the interpretation of the outcomes somewhat difficult. Furthermore, assumptions do not veer from expectations one at a time in the real world, making what-ifs difficult to translate into implications for airport planning. 4.1.3 Sensitivity Analysis In a sensitivity analysis, forecasting assumptions are var- ied one at a time, and the resulting changes in the projected outcomes (e.g., passenger demand forecast) are reported accordingly. A sensitivity analysis may serve multiple purposes, including: • Helping to identify the variables and model parameters whose variations have the greatest impact on the forecast: the critical variables. • Evaluating the impact of changes in the critical variables (i.e., of reasonable departures from their preferred, base- line values). • Assessing the robustness of the forecast. In particular, whether the general conclusions reached using the base- line assumptions are significantly altered through changes to the key assumptions. Occasionally, the sensitivity analysis will involve chang- ing multiple assumptions simultaneously. This is sometimes referred to as scenario analysis. The limitations of sensitivity analysis have been docu- mented in a number of publications. For example, sensitivity analysis forecasting assumptions are often varied by arbi- trary amounts instead of by reference to reasoned analysis of potential error (Lewis, 1995). In addition, varying one assumption at a time does not provide an accurate view of the real world, where all factors affecting forecasts are likely to vary simultaneously. On the other hand, this procedure can be useful for assessing the significance of individual fore- casting assumptions in the production of the overall demand

27 Advanced procedures have been developed to incorporate uncertainties into forecasting. These address some of the con- cerns highlighted in the previous sections, particularly with respect to quantifying the probability of alternative outcomes and forecasts. However, to date, it appears that these proce- dures are not widely used in the aviation industry. They vary in scope and complexity but can be grouped into two broad cat- egories: data-driven (objective probability) or judgment-based (subjective probability) procedures. Section 4.2.1 describes the methodologies for data-driven statistical analysis available to produce objective measures of probability. Section 4.2.2 dis- cusses the use of judgment and opinion to produce subjective measures of probability. These two categories are defined solely to assist in grouping methods for the purpose of the following discussion. In prac- tice, procedures based on historical data are often combined with judgment-based input, as is described in Section 4.2.3. 4.2.1 Data-Driven Procedures Three classes of methods have been identified where the incorporation of uncertainty relies exclusively on the analysis forecasts, and it has been used fairly extensively in aviation demand forecasting. 4.2 More Advanced Procedures for Incorporating Uncertainty into Forecasting of historical data: (1) time-series methods, (2) extrapolation of empirical errors, and (3) distribution fitting and simula- tions. These methods are summarized in the following. 4.2.1.1 Prediction Intervals from Time-Series Methods Time-series methods use statistically estimated models to estimate future values of a variable of interest (e.g., air pas- senger enplanements) based on the historical values of that variable. They include extrapolative methods, which are based solely on identifying data patterns in the variable of interest (e.g., autoregressive moving averages) and explanatory vari- able methods (e.g., regression analysis), which introduce causal variables to explain and forecast the variable of interest. Most time-series methods recognize the uncertainty asso- ciated with model specification through the inclusion of an error term and stochastic parameter values (as reflected in standard error, t-statistic, and R-squared statistics, etc.). Time-series methods allow estimation of a prediction inter- val, similar to a confidence interval, within which future estimates (or forecasts) will fall at a certain probability. For example, the prediction interval could be described as fol- lows: based on the error terms of the model, there is a 95% probability that forecast passenger traffic in 2030 will be in the range of 10 to 12 million. A common feature of predic- tion intervals is that they increase in length as the forecast horizon increases (i.e., uncertainty increases as we forecast further into the future). However, this distribution only reflects uncertainty in the model specification (the functional form of the model) and its parameter values (i.e., statistical uncertainty). It does not reflect uncertainty due to changes in the economic environment, air carrier decisions, rare shock events, and so forth. Although prediction intervals can be derived from most statistical software packages, it is rare for this infor- mation to be provided for forecasts derived through statis- tical methods. 4.2.1.2 Extrapolation of Empirical Errors This general approach consists of developing ranges of pos- sible forecast values based on observed errors from historical forecasts (i.e., the difference between actual values and prior forecasts of those values). As a simple example, if analysis found that previous forecasts had differed from actual values by +/- 20%, it could be assumed that the prediction interval related to current forecasts of future values is +/-20%. Research for this guidebook found very few applications of this extrapolation approach being used in aviation demand forecasting. However, this approach has been used in fore- casts of population growth. The National Research Council conducted an analysis of the distribution of past errors in Key Takeaways Although to date not widely used in aviation activity forecasts, there are analytical proce- dures that have the potential to enhance the understanding of risk and uncertainty, catego- rized as follows: • Data-driven procedures involving the statistical analysis of historical data (objective probability), and • Judgment-based procedures incorporating the judgment and opinion of experts and stake- holders (subjective probability). These types of procedures can be used in com- bination to develop a deeper understanding of the risk and uncertainty facing an airport. They can be enhanced by the use of computational methodologies such as Monte Carlo simulation.

28 population forecasts by the United Nations over two decades to define predictive intervals for more current UN popula- tion projections (Bongaarts and Bulatao, 2000). A similar concept is called reference class forecasting, which aims to address optimism bias and general uncertainty in demand and cost forecasting for public works. Reference class forecasting involves evaluating (or even developing) a forecast for a particular project by referencing it against actual outcomes from a group of similar projects. This process is composed of the following steps (Flyvbjerg et al., 2005): 1. Identify a group of similar past projects, called the refer- ence class. 2. Using data from projects within the reference class, estab- lish a probability distribution for the variable of interest (e.g., traffic levels). 3. Compare the specific project with the reference class dis- tribution in order to establish the most likely outcome for the new project. The approach is designed to provide an outside view of the project, without the need to identify and forecast the impact of specific events (Flyvbjerg et al., 2005). This approach has been used in the UK, Netherlands, Denmark, and elsewhere for forecasts of highway traffic and project costs. No formal applications of reference class fore- casting for aviation demand have been identified. However, informal use of this approach likely occurs, for example, by comparing a forecast against traffic development at other similar airports. 4.2.1.3. Distribution Fitting and Simulation Under this group of methods, a probability distribution is defined for the variable of interest on the basis of past growth rates or activity levels. Simulation techniques are used to combine multiple realizations of this distribution over time and develop probable growth paths. The process is composed of three steps (summarized in Figure 8): 1. Historic, annual growth rates or activity levels for the vari- able of interest for a specific airport (e.g., total passen- ger volumes) are used to identify a distribution through goodness-of-fit evaluation tests; 2. Monte Carlo simulations (see Monte Carlo text box) are run to produce the entire distribution of possible growth rates or levels over the forecasting horizon, using the dis- tribution function identified in step 1; and 3. Simulated growth rates or levels and associated probabili- ties are converted into annual forecasts. Bhadra and Schaufele (2007) applied this approach to fore- casting traffic at the top 50 commercial airports in the United States. As noted by the authors, the approach has a number of limitations: • It is assumed that the distributions of annual growth rates or levels for the variable of interest are independent. (This can be formally tested to rule out the possibility of correlation.) Probability Distribution for One Period of Traffic Growth Monte Carlo Simulations of Distribution Combined with Forecast Point Forecast with Fitted Distribution Actual Data Forecast Historical Traffic Data Distribution Fitting Source: Based on Bhadra and Schaufele, 2007. Figure 8. Distribution fitting and simulation.

29 • The same fitted distribution of the relevant variable is used throughout the entire forecasting period (i.e., time-invariant property of the distribution) resulting in the range of prob- able outcomes widening over time. • Since each variable is being simulated separately at each airport, the method ignores network dependencies (i.e., competition or interactions between airports). • The sources and the nature of uncertainty remain unknown, making the interpretation of possible outcomes and the practical use of the forecasts difficult. 4.2.2 Judgment-Based Procedures Under this group of procedures, the probability of an event is viewed as the degree of belief sustained by an informed person (or group of stakeholders) that it will occur, rather than any property of the physical world. Sub- jective probabilities can be blended with objective data (e.g., historical variations in enplanements) and/or market data (e.g., crude oil futures prices) to arrive at a distribu- tion for future outcomes. In aviation demand forecasting, as in other fields, forecast- ers often use experts’ opinions and judgments to assess traffic outlook and uncertainties. This is particularly true when point forecasts and/or prediction intervals are difficult to obtain through objective methods or when the variable of interest may be affected by rare events. There are a number of techniques for eliciting expert opin- ions and judgments on forecast outlooks and probabilities. One of the most well established techniques is Delphi forecast- ing. The Delphi technique is an elicitation technique defined by four key features: anonymity, iteration, controlled feed- back, and the statistical aggregation of group responses (Rowe and Wright, 1999). Delphi forecasting typically begins with the planner select- ing a group of experts and preparing a questionnaire, which Introduction to Monte Carlo Simulation Monte Carlo simulation (or the Monte Carlo method) is a computerized simulation technique that makes use of randomization and probability statistics to investigate problems involving uncertainty. Typically, it involves a computer model of a system or project (e.g., air traffic at an airport). The inputs to the model, instead of being fixed numbers or variables, are specified as probability distributions. For example, in the model described in Figure 8, rather than traffic growth being set at X% per annum, it may be defined as having normal (bell-curve) distribution with a mean of X% and a standard deviation of 1.0%. Using computer software, the model is run multiple times, each time randomly sampling from the input distributions, resulting in different outcomes each time. Often, the model will be run thousands or tens of thousands of times (known as iterations), and the results are collected from each run. With enough iterations of the model, the output can demonstrate the range of possible outcomes and provide statistical estimates of the probabilities of various outcomes. Depending on the complexity of the model and input distributions assumed, the range of outcomes can be large and not always linear. Expected or most likely values can also be generated. Monte Carlo can be seen as a powerful what-if or scenario-generating exercise where every possible what-if or scenario is generated (within the confines of the model specification), including interactions between the various input factors. Another way of looking at it is that each iteration of the model represents one possible future for the system being modeled. By running the model thousands of times, the user can view whole sets of possible futures, assess which are most likely to occur, and identify areas of greatest downside or upside. Monte Carlo is used extensively in a wide range of fields. One of its first applications was in designing the shield- ing for nuclear reactors at the Los Alamos National Laboratory in the 1940s. (The name Monte Carlo was coined as a code name by scientists at the laboratory in reference to the Monte Carlo casino resort.) Monte Carlo simu- lation has since been used in finance, project planning, engineering studies, traffic modeling, cancer radiation therapy, and telecommunications network design, among many other applications. Monte Carlo simulations are also discussed in Part II of the guidebook.

30 in the context of airport planning may include a series of projections. The questionnaire is distributed to each respondent separately. The answers provided by the experts are summarized and tabulated, and the results are returned to the experts for a second round. In this second round, the experts are asked to assess the group responses and justify their choices. During subsequent rounds, group averages and comments are provided, and experts are asked to re- evaluate their choices. The rounds continue until an agreed level of consensus is reached. The literature suggests that by the third round a sufficient consensus is usually obtained. One of the most important factors in Delphi forecasting is the selection of experts: The persons invited to participate must be knowledgeable about the issue, and represent a variety of backgrounds. The number must not be too small to make the assessment too nar- rowly based, nor too large to be difficult to coordinate. It is widely considered that 10 to 15 experts can provide a good base for the forecast. (Rodrigue et al., 2009, p.14) The Delphi technique is listed as a qualitative forecasting method in the 2006 ICAO Manual on Air Traffic Forecasting (ICAO, 2006). Other elicitation techniques are: • Statistical groups: where individuals give their forecasts without interacting with each other; • Unstructured interacting groups: where individuals can interact freely with each other; and • Nominal group technique: using a Delphi structure, but allowing face-to-face discussions between rounds. Defined broadly, the elicitation process helps experts con- struct a set of carefully reasoned and considered judgments. Specifically, elicitation is conducted with a range of available or circumstance-specific protocols (such as Delphi) to obtain people’s subjective but accurately specified quantitative expres- sions of future probability. A number of shortcomings of the elicitation approach have been highlighted in the academic literature: • Existing research has found that experts may often be too confident and attach too high a probability to their predic- tions (Keilman et al., 2002). • Experts may encounter problems in determining the exact probability bounds associated with a given prediction inter- val [e.g., the difference between a 90% and a 99% probabil- ity (Keilman et al., 2002)]. • There is evidence to suggest that objective methods are gen- erally more accurate than subjective methods as changes in the environment increase and the forecasting horizon is lengthened (Armstrong and Grohman, 1972). An alternative to the use of formal elicitation techniques, which combines judgment—often informally—with data and statistical modeling, has been presented in the academic literature as “the Poor Man’s Bayesian Analysis” (Armstrong and Grohman, 1972). This alternative method can be as simple as changing the parameters in an econometric model to reflect effects not captured in previous analysis (e.g., low- ering the GDP elasticity to capture the idea that economic growth provides less stimulus to traffic growth as the econ- omy becomes wealthier). Rather than changing the model parameters, the same effect can be achieved by adjusting the forecast values directly. The use of elicitation techniques in aviation planning is not well documented but is fairly common, although on a more informal basis: Frequently, a group of professionals knowledgeable about aviation and the factors affecting aviation trends are assembled to examine forecasts from several different sources, and compos- ite forecasts are prepared in accordance with the information in these sources and the collective judgment of the group. (Horonjeff et al., 2010, p.153) 4.2.3 Combining Data-Driven and Judgment-Based Analysis The objective information from data analysis can be com- bined with subjective information obtained through elicita- tion techniques to form an even richer assessment of risk and uncertainty. Lewis proposes a risk analysis elicitation frame- work to best achieve this aim. The framework uses data analysis to initially estimate the probabilities’ values and distributions, which are then modified and expanded upon using elicitation techniques to obtain risk and probability beliefs from experts and stakeholders. A final set of probability distributions can then be developed, which are a combination of objective and subjective probability information. These are combined using Monte Carlo simulation techniques to produce forecasts of future activity together with estimates of the probability of achieving alternative outcomes (Lewis, 1995). This sort of framework has been used to provide decision support in a number of areas, including traffic and revenue forecasting for toll roads, the quantification of airport invest- ment risk, the estimation of construction cost, and the sched- uling of large infrastructure projects. The Risk Uncertainty Analysis text box provides an example of a similar form of risk analysis used in air traffic forecasting at Portland Inter- national Airport (PDX).

31 Risk and Uncertainty Analysis Example: Portland International Airport The 2008 Airport Master Plan Update for PDX provides forecasts of unconstrained passengers, cargo, general aviation, and military aircraft operations for the years 2012, 2017, 2027, and 2035. The forecast incorporates probabilistic elements and is based on a review of the FAA’s TAFs, historical demand data, and inputs from airport planning stakeholders and representatives from the City of Portland. The sources of uncertainty considered in the passenger demand model include coefficient estimation errors and uncertainty in demand determinants such as per capita income, oil prices, non-fuel costs, carbon taxes/climate change policies, and load factors. Probability distributions for each determinant were tested and fitted inde- pendently to available historic data. Additionally, sensitivity tests were conducted for the 2035 results based on probable changes in technology (e.g., 5% substitution of video conferencing for business travel), changes in the propensity to travel by age and income group (e.g., 65 and older persons will take an additional trip), oil shocks (e.g., 20% increase in oil prices), and other potential changes. The cargo model was developed using oil prices and uncertainty around those prices. The figure that follows, extracted from the master plan update, illustrates some of the outcomes of the risk and uncertainty analysis. The lines indicate the traffic forecasts at different levels of probability. For example, the top 90% line indicates that there is a 90% probability the traffic will be at or below the line. Source: Jacobs Consultancy, 2008. High-impact events for which there is a sparse or nonexis- tent historical record pose two challenges for airport demand forecasting. One is anticipating them: will a volcano erupt and shut down the air traffic system, as occurred in Europe in 2010? The other is anticipating their impact on demand: how long will volcanic ash keep the system down and what effects will that have on demand? Some rare events have a high but short-lived impact on demand, while others have longer-term implications. The literature on anticipating rare, high-impact events cuts across a range of professional disciplines, including sta- tistics, forecasting, and decision sciences, and addresses a range of events (e.g., weather and environmental, economic, crime, terrorism, and political), and it deals with a variety of 4.3 Is It Possible to Predict and Forecast the Impact of Rare or High-Impact Events? Key Takeaways Lack of historical data and innate human biases make the prediction and forecasting of rare events difficult—and often unsuccessful. How- ever, there are techniques airports can employ that will better prepare them to manage rare, high-impact events should they occur.

32 stakeholders (e.g., human safety and welfare, infrastructure enterprises, business firms, and governments). Goodwin and Wright (2010) conclude that existing fore- casting methods, including statistical approaches and expert judgment (i.e., Delphi forecasting), are not effective in antici- pating and estimating the impact of rare events on enterprise outcomes. Bonabeau further explains that our inability to predict low-frequency, high-impact events is due to two fun- damental cognitive biases that affect human decision mak- ing: availability and linearity. Availability heuristics guide us toward choices that are easily available from a cognitive per- spective (i.e., if it is easy to remember, it must make sense). Linearity heuristics make us seek simple cause-and-effect rela- tionships in everything (Bonabeau, 2008). Bonabeau makes the case that anticipating rare events requires augmented paranoia—that is, the rejection of both cognitive biases: availability and linearity. His corrective strat- egies include tapping the collective intelligence of people in a group (i.e., the wisdom of crowds) and tapping the creative power of evolution (i.e., considering a more-or-less randomly generated population of solutions and selecting, altering, and/ or breeding the fittest until a satisfactory solution emerges) (Bonabeau, 2008). Rather than attempting to forecast rare events, methodolo- gies have been put forward that attempt to develop a capability to manage such events. These include scenario planning. Sce- nario planning abandons the assumption that rare events can be predicted or given a meaningful probability of occurrence. Instead, scenario planning assumes that the best that can be done is to identify critical outcome uncertainties and plan for the range of futures that could plausibly unfold: Essentially, scenarios highlight the causal reasoning under- lying judgments about the future and give explicit attention to sources of uncertainty without trying to turn an uncertainty into a probability. (Granger and Henrion, 1990, p. 363) A similar concept is protective strategy, the systematic means of protecting an organization from the occurrence of events with negative impacts while allowing it to benefit from the occurrence of events with positive impacts. Open and wide- ranging discussion forums (sometimes called “devil’s advo- cacy” or “dialectical inquiry”) can be combined with Delphi approaches and scenario planning to enhance the anticipation of, and robustness to, rare events (Goodwin and Wright, 2010). While the previous discussion and broad conclusions pertain to forecasting rare events for organizations in gen- eral, the aviation literature suggests a convergence of ideas. Horonjeff et al. report the increased use of techniques that rely less on mathematical modeling and more on an analysis of different scenarios, human judgment, and protective strat- egy: “Although judgment has always played a role in demand forecasting, it is becoming more important as a subjective test of the reality associated with forecasting outcomes.” (Horonjeff et al., 2010, p.168). Furthermore, Horonjeff et al. state that in the wake of the 2008/09 financial crisis and recession, airport planning pro- cesses are becoming more phase-oriented and continuous in recognition of uncertainty about rare events and their impacts on demand forecasts. The authors point to an increasing use of ongoing sensitivity, trade-off, and scenario analysis in the planning and design of airport facilities and operational procedures.

33 The sections that follow describe some of the approaches that have developed in academia and industry to better address risk and uncertainty in airport decision making, including theoretical methodologies, practical applications, and diversification strategies. 5.1 Flexible Approaches to Airport Planning with greater ease or lower costs than if no flexible options were considered (McConnell, 2007). Different authors have proposed slightly different steps and procedures or variations: • Dynamic strategic planning (de Neufville and Odoni, 2003); • Flexible strategic planning (Burghouwt, 2007); • Adaptive airport strategic planning (Kwakkel et al., 2010). As the names suggest, these three approaches are fundamen- tally very similar, although they differ in detail. They are largely conceptual, although based on real-world experience, and have not been fleshed out into detailed planning procedures. While there are examples of airport planning that have contained some elements of these approaches, they have so far not been applied in practice. The three approaches are described in more detail in Appendix E. The contrast between these approaches and more traditional airport master planning is characterized in Table 1. One of the ways these approaches differ from traditional master planning is that rather than have most of the planning developed around a single forecast, the plan considers a range of forecasts. The approach allows for plans that can be rela- tively easily adjusted over time as events unfold and conditions change. While not all aspects of uncertainty can be eliminated or mitigated, it is possible to reduce or mitigate some uncertain- ties through demand management techniques (i.e., uncer- tainties that are caused by market fluctuations) (de Neufville, 2004). Airports can impede certain traffic types and facilitate others through pricing or direct controls (e.g., encourag- ing general aviation traffic to move to other airports, freeing capacity for commercial operations). One example of this is Kansas City International Airport, where the passenger ter- minal was impractical to serve transfer traffic. As a result, the planning team encouraged the locally based airline to estab- lish a hub at another airport (de Neufville and Odoni, 2003). C h a p t e r 5 Addressing Risk and Uncertainty in Airport Decision Making Key Takeaways A number of alternative approaches to airport planning have been proposed by practitioners and academics that seek to incorporate much greater flexibility into the planning process. To date, these approaches are largely theoretical and have not been applied in practice. A concept that offers considerable promise in making airport planning more flexible is real options. Drawing from the use of options in finan- cial markets, real options is the right, but not the obligation, to take a certain course of action. 5.1.1 Conceptual Frameworks Given the traffic uncertainties facing airports and the difficulty in addressing them in traditional airport master plans, a number of alternative, adaptable approaches to airport planning have been proposed in the literature. A key element of these proposed approaches is building far greater flexibility into the planning process. While many definitions of flexibility exist, what all of them have in com- mon is that flexibility allows a system to undergo change

34 • Staged investment. Staging investment as a series of out- lays, which allows abandonment of the project if conditions change. Each stage is an option on the value of subsequent stages. • Option to alter scale. The ability to accelerate or expand if conditions are favorable or contract if conditions are less favorable. At the extreme, it is the ability to halt produc- tion and restart later. • Option to abandon. If market conditions decline severely, options can be abandoned—and equipment and land sold off. • Option to switch. Develop a facility in such a way that it can change the output mix produced (alternatively, change the input mix). • Growth options. An early investment (e.g., in land, in R&D) that opens up future growth opportunities. • Multiple interacting options. Projects often involve a collection of put and call options in combination. Their combined value may differ from the sum of sepa- rate values. The greater flexibility that real options provide can have significant value to the decision maker. However, real options often (but not always) impose a cost. The trade-off between the real option’s value and its cost will determine whether to go ahead with the option. Various sophisticated analytical approaches have been developed to evaluate and value real options (and are discussed in Appendix E). Some of these have been incorporated into the systems analysis methodol- ogy described in Part II. 5.1.2 Real Options One concept that appears frequently in the literature on flexible or adaptive airport planning is real options. The con- cept of real options is based on and developed from financial options. In a financial context, options allow investors the right to acquire or to sell an asset (e.g., stock) at a specified price during a specified time frame. In short, an option is the right, but not the obligation, to take a certain course of action. There are two types of options: a call option (the right to buy, generally to take advantage of a good situation) or a put option (the right to sell, to get out of a bad situation). A remarkable feature of options is that their value increases with risk, which is the opposite of most other forms of assets (de Neufville and Odoni, 2003). As such, options are particu- larly useful in risky situations. Real options apply to the real, physical world rather than the financial world (although real options still have financial implications). The concept started to develop in the 1970s and 1980s as a means to improve the valuation of capital- investment programs and offer greater managerial flexibil- ity to organizations. Real options and real options analysis are used in many industries, particularly those undertaking large capital investments (e.g., oil extraction and pharma- ceuticals). A number of common real options are available to organizations (Trigeorgis, 1996): • Option to defer. A form of call option, where, for exam- ple, an organization may hold the lease on some land but defer building a plant on the land until market conditions are right. Traditional Master Planning Flexible Strategic Planning Passive, reactive, adaptive Re-adaptive, proactive Once-and-for-all anticipation/adjustment to change Continuous anticipation/adjustment to change Supply driven Demand driven Forecasts as predictions of the future Backcasting: Scenarios as guidelines of what may happen in the future Single-future robustness of plan and projects Multi-future robustness of plan and projects Long-term and short-term commitments Short-term commitments, long-term strategic thinking Preferred analytical tools: forecasting and net present value analysis Preferred analytical tools: scenario planning, decision analysis and real options, contingent road maps, scanning, experimenting Preferred alternative is optimal solution for a specific future Preferred alternative is best alternative across a range of possible future scenarios. Risk implicitly ignored or risk aversion Think risk culture. Risk as an opportunity Top-down/inside-out Top-down/bottom-up, inside-out/outside-in Reprinted by permission of the publishers from Airline Network Development in Europe and its Implications for Airport Planning by Guillaume Burghouwt (Farnham: Ashgate, 2007), p. 208. Copyright 2007. Table 1. Characteristics of flexible planning.

35 As previously noted, the flexible planning approaches put forward in the literature remain largely conceptual and have not been applied in any airport planning projects. Likewise, real options methodologies, while used in other industries, have not been applied to any real-world airport planning projects. Nevertheless, there are a lot of examples of airports devel- oping ways to build flexibility into the airport planning pro- cess that reflect the ideas behind real options. In general, these represent common-sense approaches based on experience in the field of airport planning rather than any formal method- ology. The sections that follow provide a summary of some of these examples, which represent industry best practice. 5.2.1 A Second Airport for Sydney, Australia During the 1970s and early 1980s, the Australian govern- ment grappled with the issue of a second airport to serve Sydney, Australia’s largest city. (The main airport in Sydney is difficult to expand due to its proximity to the city center.) Two separate studies had produced contradictory conclu- sions: one recommending the building of a new airport, and the other concluding it was not necessary. In 1985, the government embarked on a third planning pro- cess that used a decision analysis methodology. This approach recognized that the future was uncertain and therefore the plan needed to consider a wide range of scenarios rather than a single forecast. In addition, it recognized that a second airport 5.2 Real-World Applications of Flexible Airport Planning was a long-term project, and not all decisions had to be made right away. Thus, the question was changed from “should a second airport be built?” to “should land be reserved for a pos- sible future airport?” This question was considered under dif- ferent traffic growth scenarios, with the analysis finding that acquiring a site generally provided the best outcome over the scenarios, and as a result, the government of Australia did acquire a site for the second airport, which is yet to be built. 5.2.2 Toronto Pearson International Airport A critical focus of Greater Toronto Airport Authority (GTAA) is risk management, which is reflected in its approach to flexible airport planning (based on discussions with the air- port CEO). Recently, GTAA has adopted an exercise involving the management team spending one day a year considering the absolute worst-case and best-case scenarios that the airport could face (a form of scenario planning as described in Section 4.3). In both cases, the management team has to consider what actions can be taken, both now and in the future, to accom- modate such an outcome if it were to occur. GTAA has found that usually, the best-case scenario (high growth) produces the greatest challenges to the management team. The worst case can be handled by postponing or canceling development and shutting down or changing the use of a facility. However, main- taining flexibility to take advantage of the best case has proven to be more difficult. This process was integral to the authority’s response to the 2008–2009 economic downturn and in recent decisions on the development of terminal facilities. In addition, the airport has introduced other forms of flex- ible planning into its design: • In Canada, airports deal with three types of traffic: domestic, U.S. (referred to as transborder), and other international. Each traffic type has its own processing requirements. For example, at major Canadian airports, passengers to the United States are precleared by U.S. Customs and Border Protection officials, eliminating the need to go through these processes when they arrive in the United States. At Toronto, as well as other Canadian airports such as Vancou- ver and Edmonton, the terminals have systems of movable walls and internal passageways so that the gates (known as swing gates) can be switched between transborder and inter- national traffic or even domestic traffic, as required. • As well as providing the flexibility to adjust terminal space to match traffic levels on both a daily and long-term basis, the use of swing gates also reduces the overall terminal space needed to handle passenger traffic since peaks in different types of traffic flows often occur at different times of the day (e.g., the peak in international traffic does not always occur at the same time as the domestic traffic peak). Swing gates are explained in more detail in the Swing Gates text box. Key Takeaways A number of real-world examples of flexible airport planning exist that reflect some of the conceptual ideas described in Section 5.1.2. The strategies employed include: • Reserving land or terminal space for future development; • Scenario planning workshops to consider the management of best-case and worst-case scenarios; • Terminal space designed so that the same area can serve different traffic types (e.g., domestic and international) while still meeting customs, immigration, and security requirements; and • Use of trigger points: additional development is triggered by traffic reaching predetermined levels.

36 • The airport has identified terminal space that could be required for future security screening checkpoints. In the interim, this space is used for retail operations, thus allowing the airport the flexibility to convert the retail space to addi- tional security processing when traffic levels (or new security protocols) require it. The conversion takes a relatively short amount of time, and in the meantime, the space is generating income for the airport. 5.2.3 Vancouver International Airport (YVR) 2007–2027 Master Plan Between 2004 and 2007, Vancouver International Airport Authority (VIAA) undertook a master planning process to determine the airport’s development through 2027. Although the master plan was for a 20-year period, it was decided to also look at the 40-year outlook (Vancouver International Airport Authority, 2007). A major reason for this is that YVR is located on an island, and it was necessary to establish, in broad terms, whether the island had sufficient land to support the long-term capacity needs of the Vancouver market. This itself represents a form of flexible planning—ensuring that decisions made in the short- to medium-term (i.e., continue development on the island) do not lock the airport authority into a situation that is highly constrained or costly in the long-term. The 40-year analysis determined that the island was sufficient to meet the long-term needs of VIAA. The master plan places considerable emphasis on main- taining flexibility: The Master Plan is flexible in the face of changing circum- stances because it does not commit to any particular proj- ect. Development decisions are made following extensive and detailed analysis, review, and timing of future air travel needs. (Vancouver International Airport Authority, 2007, p. 1) As such, the airport authority recognizes the uncertainty facing the airport and so reviews plans regularly, monitors external events closely, favors conservative timing for capi- tal expenditures, builds infrastructure incrementally, where possible, and places great weight on flexibility and open, transparent communications of its planning activities. The master plan sets out an incremental building approach where the next stage of development only goes ahead if a pre- determined traffic level (or trigger point) is reached. Depend- ing on the facility, the trigger point can be total traffic or a particular traffic segment (e.g., domestic, U.S., other international). Only the initial development stage (the first 3 years) was tied to the forecasts, after which developments would be dictated by traffic growth. Swing Gates Swing gates provide airports with the flexibility to meet different peaks associated with different traffic sec- tors. The diagram demonstrates the concept for a gate configuration that handles three different sectors while ensuring that flows are segregated, as is commonly required in Canada. While the flows depicted are based on U.S. preclearance at foreign airports, the conceptual flows can equally apply to other jurisdictions. 1. Domestic passengers are shown as enplaning/deplaning an aircraft in one area. 2. A swing door is in place to prevent commingling with other flows. Alternate arrangements are possible to enable both gates to serve all sectors. 3. These other flows include international arrivals destined for border formalities. 4. Another swing door is in place to segregate flows. 5. Partitioning can be used to temporarily use a holdroom for transborder (U.S.) passengers or can be removed to use the hold- room entirely for domestic passengers. Other grade separation/physical barriers have been used, including aircraft bridges that can move up or down to serve different floors of an airport building.

37 In order to ensure that the master plan could remain rele- vant over a wide range of future scenarios, and to obtain buy- in and feedback from stakeholders, the master plan involved extensive consultation processes with stakeholders, industry experts, and the community (discussions with Michael Matthews, director of the master plan). 5.2.4 Dallas/Fort Worth International Airport DFW handled 56.9 million passengers in 2010. It is a hub for American Airlines and package delivery company United Parcel Service (UPS). The airport’s most current master plan— the 1997 Airport Development Plan (ADP) update—places considerable weight on an incremental and flexible planning process: It [the ADP update] utilizes a holistic approach to airport development; providing flexibility to respond to ever changing market demands. Key elements of this approach include ongoing research, regu- lar performance reviews, and close integration of planning, opera- tions, management, and new technology at DFW Airport. (Dallas/ Fort Worth International Airport, 1997, p. 47) The ADP update sets out a phased capital plan meant to ensure the “goal of incremental or phased development that is timely and logical” (Dallas/Fort Worth International Airport, 1997). In addition, the capital improvement plan incorporates the following three ideas: continuous planning, proactive man- agement, and focus on market-based action. All investments require input from stakeholders and must consider soon-to- be-needed capacity. 5.2.5 Mombasa Airport Mombasa Airport (also known as Moi International Air- port) is the second largest airport in Kenya. De Neufville and Odoni (2003) describe the incorporation of flexibility into the airport master plan. The original master plan for the redevelopment of the passenger terminals anticipated two buildings—one for domestic traffic and another for international traffic. Each building was to be large enough to meet anticipated traffic in either sector. However, the dynamic strategic plan rec- ognized that a major risk was that the proportion of inter- national traffic could shift radically (passengers may come directly from Europe), in which case one of the buildings would be crowded and the other one underutilized. Thus, the strategy was changed to build a single passenger termi- nal with a domestic area on one side, an international area on the other, and a mixed use area in the middle to serve either type of traffic. This strategy reduced the overall size of the facility required to handle total traffic at the airport and allowed flexibility in handling different traffic mixes in the future. 5.2.6 Pease International Tradeport Karlsson (2002) describes the application of real options planning at Pease International Tradeport near Portsmouth, New Hampshire. Pease International Tradeport is a public- use joint civil–military airport. As of 2010, it has no sched- uled passenger services, but in 2002 (at the time Karlsson’s paper was published), it had scheduled passenger volumes of up to 40,000 passengers per annum. Traffic volume exhibited considerable volatility due to the entry and exit of carriers. Therefore, in the 2001 master plan, it was decided that any terminal development would be triggered by traffic reaching certain thresholds. (Specifi- cally, Phase I of the development was to be triggered by traf- fic reaching 80% of current capacity.) As the type of service was also uncertain, analysis was carried out to ensure that the terminal expansion could handle a wide range of plausible aircraft sizes. In addition, other aspects of flexible planning employed at Pease were: • A domestic/international swing gate, • Temporary use of an unutilized aircraft parking apron as an overflow car park, and • Non-load-bearing walls, allowing easy expansion or con- version of the terminal (Karlsson, 2002). 5.3 Diversification and Hedging Strategies Key Takeaways Airports can also manage risk and uncertainty by diversifying their traffic and revenue base and employing hedging strategies against certain risks. Potential strategies include: • Air service development programs to increase the range of carriers serving the airport and the scope of destinations served; • Multi-use developments such as hotels, gen- eral aviation, logistics and cargo, retail, offices, industrial parks, and leisure facilities; • Ancillary land use, such as government facili- ties, advertising, renewable energy, intermo- dal facilities, and military/civil joint use; • Airport city or aerotropolis developments. Any such diversification strategy must be backed up by a strong business case and be compliant with FAA regulations (e.g., FAA Grant Assurance 21: Compatible Land Use).

38 Addressing uncertainty and risk in the airport environment can go beyond the physical planning of the facilities to other aspects of airport strategy. Diversification involves broadening the airport’s traffic and revenue sources to avoid being heavily exposed to one particular type of risk and to reduce overall volatility. Hedging is taking a position to offset and balance against a particular or general risk. Airlines often use hedging strategies to reduce exposure to fuel price increases. One means of diversification for airports is to increase the number of carriers and destinations served. Many of the events described in Chapter 3 were the result, in large part, of the airport being heavily dependent on a single carrier. To mitigate this, airports can undertake an air service develop- ment program designed to attract other carriers to operate at the airport. Of course, any such strategy has to be balanced in such a way as to avoid undermining incumbent carriers at the airport. Further guidance on air service development strategies can be found in ACRP Report 18: Passenger Air Service Development Techniques (Martin, 2009). Increasing the number of destinations served can also have diversifica- tion benefits since it reduces exposure to risk factors at each particular destination (e.g., economic downturn, high sea- sonality, one-off disruptive events). Air service development can also provide hedging options—for example, developing legacy and low-cost carrier traffic to protect against separate developments with the two types of traffic. Revenue diversification involves an airport modifying and diversifying its products to reduce its dependence on aero- nautical revenues alone. Figure 9 shows typical aeronauti- cal and non-aeronautical revenue sources for airports. To minimize and mitigate risk, airports can focus more on the non-aeronautical revenue sources that are less dependent on traffic volumes. ACRP Synthesis 19: Airport Revenue Diversification discusses multi-use developments at airports that provide alternative revenue streams (Kramer, 2010). Examples include: • General aviation (GA) developments, • Air cargo and logistics centers, • Hotels, • Convention centers, • Offices, • Intermodal centers, • Retail malls, • Industrial parks, • Golf courses, and • Sports arenas. Similarly, ACRP Synthesis 19 also describes ancillary land use that airports can explore to diversify revenues (Kramer, 2010): • Advertising and sponsorship. An airport can gener- ate non-aeronautical revenues through advertising—for example, advertising in terminals, naming rights on air- port terminals, advertising on unpaved airfields, or ban- ners on sky bridges. • Government facilities. Government agencies are fre- quent airport tenants. These agencies also increase non- aeronautical revenues for an airport by leasing space directly from the airport. • Renewable energy. Renewable energy is another potential non-aeronautical revenue stream that has many advantages for an airport. Not only does it lower the airport’s environ- mental impact (with positive public relations implications), it is also an alternative source of power to operate the airport with the possibility to sell excess power back to the utility company (or to other users). • Intermodal facilities. Connecting rail, road, marine, and air in an intermodal facility at the airport is another poten- tial non-aeronautical revenue source. Examples include Kansas City Intermodal Business Center and Port Alberta in Edmonton. • Military/civil joint use agreements. Joint use facilities (e.g., shared air traffic control, safety and rescue, and utili- ties) can have a positive impact on the infrastructure of an airport and lead to large cost savings. Examples include Colorado Springs Airport and Peterson Air Force Base (Kramer, 2010). Source: Based on Kramer (2010) Figure 9. Revenue sources for airports.

39 It should be noted that some of these opportunities for diversification may be constrained by the FAA Grant Assur- ances (e.g., FAA Grant Assurance 21: Compatible Land Use) and similar local revenue bond ordinances. Arguably, the most comprehensive approach to revenue diversification is the airport city or aerotropolis concept developed by some larger airports, largely in Europe and Asia. Airport cities involve the development of multiple, and often complimentary, commercial and industrial activities on air- port land that may benefit from the transportation linkages that the airport offers. This can include logistics centers, free trade zones, manufacturing, offices, retail, hotels, and rec- reational facilities (e.g., golf courses, sports centers). These activities have a less direct linkage with traffic levels and can serve a wide range of customer/client types. The concept is illustrated in Figure 10. While the revenue diversification strategies described have the potential to better manage and offset uncertainty in aviation activity, simply developing new non-aeronautical activities does not guarantee an overall reduction in risk. In fact, such a strategy can expose the airport to new risks and uncertainty since non-aviation activities have their own risk profiles. The airport needs to determine that there is a strong business case for any diversification strategy and that the risks are well understood. 5.4 Assessment of the Reviewed Approaches Source: Dr. John Kasarda at Aerotropolis Schematic (2009). Figure 10. Schematic of airport city and aerotropolis. Key Takeaways Flexible airport planning, real options, diversi- fication strategies, and similar approaches can offer considerable benefits to airport decision making in the face of uncertainty. However, there are issues around the application and effectiveness of these approaches. They may require decision makers to consider politically or institutionally unpopular outcomes (e.g., the loss of a major carrier) and will require the incorpo- ration of new processes and ways of thinking. It should also be recognized that there are limits to the success of these approaches—they can reduce risk, but not eliminate it. The previous sections describe theoretical and practi- cal approaches to addressing traffic uncertainty in airport decision making. As noted, the generalized flexible plan- ning approaches described in Section 5.1.1 are untested. The

40 practical examples of flexible planning have been found to include fairly pragmatic ideas such as land banking, swing gates, and common-use facilities. Although there is little formal analysis of the benefits of these approaches, it can be argued that the success of these approaches is reflected in their increased use in airport planning. Swing gates, common- use terminal equipment (CUTE) and common-use self service (CUSS), trigger points, and movable walls have become more commonplace in airports around the world. It also appears that many of these approaches are applicable to a wide range of airports, from large hubs to small regional airports, since they have minimal resource requirements and some may even reduce costs (e.g., shared space and equipment can reduce overall facility size requirements). Nevertheless, there may be other aspects of flexible planning that may require more detailed or complex analysis. Consideration also needs to be given to some of the poten- tial issues around a more flexible approach to airport planning and decision making: • Flexible planning and risk evaluation appear to be particu- larly successful where they have taken a wider prominence in the airport organization (e.g., Toronto). Nevertheless, in some cases, it can be politically or institutionally uncom- fortable to dwell on worst-case scenarios, so they are not addressed in any significant way in planning (for example, the exit of an airline that is the airport’s largest customer). • Building in flexibility will require additional analysis to cor- rectly assess the costs and benefits of certain options. This may involve the need to adopt new analytical approaches and could impose more costs and time requirements on the planning process. • Burghouwt (2007) comments that the choice for flexibility can sometime be a “wicked problem,” which is one that exists where there are competing interests between stake- holders (or within the same organization). For example, land banking may create tensions with the surround- ing community because the land is unavailable for other uses. Equally, it creates an uncertainty for the community since it is not known whether an airport or expansion will appear on the land. • It should be recognized that these approaches, while offer- ing improvements on traditional airport planning, still have their limitations. They have the potential to reduce risk but not to eliminate it entirely. For example, trigger points allow an airport to more closely match development to exact traffic levels, but it is still possible for developments to be mistimed. For example, an airport could have com- pleted an expansion in 2007 based on the positive traffic and business environment of the time, only to see traffic drop dramatically in the 2008 recession, which could then be compounded by cutbacks by carriers or even carrier failures. • On the other hand, it is also important that there not be an overreaction to changing market conditions and that airports have the tools to evaluate the permanence of any changes (i.e., are changing traffic trends a short-term fluc- tuation or part of a long-term trend change?). An airport scaling back its development plans in response to a drop in traffic (e.g., due to recession or terrorism event) may find that it has inadequate capacity when traffic growth returns.