CHAPTER 7

Understanding and Modeling Shock Events

Shock events are difficult to incorporate into forecasting and the planning, financial, and operational processes that rely on those forecasts. Many forecasts, including those produced within the aviation sector, are largely based on a continuation of business as usual with gradual, incremental change over the long term.

One of the primary challenges with predicting or modeling shock events is that there is often little historical data from which to draw predictive estimates. Before the start of 2020, many would be able to see the potential for another pandemic to impact the U.S. or global aviation industry using past pandemics, such as SARS, as examples of impacts and effects; however, the magnitude of the COVID-19 pandemic proved challenging to predict. When looking back through modern history for a pandemic of similar magnitude impacting the United States and the world, the 1918 influenza pandemic is the most similar event. However, historical analysis is constrained by the rapid evolution of technology, both medical and aviation. In 1918, commercial aviation was in its infancy in the United States. Medical knowledge in 1918 had not yet reached a point to recognize viruses, so the 1918 pandemic’s relevance for aviation was easy to dismiss before 2020. Thus, there were few direct impacts to draw on from the influenza pandemic to compare it with the COVID-19 pandemic (the global spread of the 1918 flu was facilitated by the maritime movement of troops toward the end of the First World War). While this is an extreme example, it does highlight the difficulty that historical events cannot always be used to predict or model future potential shock events.

This challenge of the lack of historical data or inability to apply historical analysis has been noted by various authors. Gilbert, Habibi, and Nadim (2016) assert that classical statistical approaches to managing black swan events have limited value due to there being little or no historical data with which to analyze these events, by definition. These authors instead propose a new “Decision Entropy Theory” framework specifically to work in low-information environments and the use of Bayesian analysis to quantify previously unknown risks. Arney et al. (2013) highlight this challenge in the lack of historical information where surprise and the unknown are common in the paper’s military strategy setting; they propose an evolution of scenario planning and simulation analysis combining rare event and tipping point analysis as ways to evaluate and explore surprise events when there is little historical information. From these examples, the literature points to analysis and, fundamentally, the use of creativity and imagination as the means to counter the lack of historical information on unknown shock events to bring them into the realm of the known, though still highly uncertain.

Evaluating Shock Events

Lack of historical data and inherent human biases make predicting shock events difficult and attempts to do so are rarely accurate. Attempts to predict the occurrence and timing of shock events, such as earthquakes or terrorist events, require an impractical level of information and computation (McMorrow 2009), are largely unproven, and have little applicability to airport forecasting.

While predicting shock events may not be practical, there are methods to better understand how shock events may arise, develop, and impact a business. There are established methods for risk assessment that can be applied to shock events impacting airports. For example, bow tie analysis is a structured approach to risk identification and management that gives a summary of all plausible scenarios that could exist around a certain shock event or risk and also captures what an organization does to control or mitigate those scenarios. The diagram is shaped like a bow tie as shown in Figure 22, with factors contributing to the shock event on the left side and consequences and reactions of the event on the right side. The advantage of this approach is that it provides an overview of complex scenarios through a simple visual explanation of a shock event. The UK’s Civil Aviation Authority has used this approach in understanding and managing safety and human factors risks (UK Civil Aviation Authority, n.d.). Another visual approach is influence diagrams that can be used to illustrate and understand risk and shock factors, depicting key elements including decisions, uncertainties, and objectives (Kincaid et al. 2012).

Feduzi and Runde (2014) propose a process of “eliminative induction” to elicit and evaluate potential shock events and decision-making. The authors lay out a method for collecting evidence and generating hypotheses about future states of the world, which is valuable for risk management but is posed as a method to potentially reduce exposure to shock or black swan events by “bringing into light states of the world that might not have been uncovered otherwise” (Feduzi and Runde 2014, 20). These authors also suggest that their method helps counteract confirmation bias in the search and ideation of future states of the world, particularly with respect to looking beyond future states or events, which might be a positive for an organization. In the context of shock events for the aviation industry, the conceptualization and discussion are more often negative than positive from a forecasting perspective; still, the methodology proposed

does provide at a conceptual level, a means to look beyond expected outcomes to attempt to see the unexpected.

Gilbert, Habibi, and Nadim (2016) propose a new framework for risk management, called Decision Entropy Theory, to help create better information for decision-making when there is limited or no information. The framework applies to the question of shock events as it is specifically designed as a starting point to assess probabilities and manage risks when there is little or zero prior information. While the mathematical approach applied is relatively advanced, the basic methodology involves exploring alternatives to existing risk-mitigation strategies or risk events—even if there is no meaningful knowledge of how likely a risk event may occur—and considering the impact of the existing or preferred methods of managing risks. So even if one cannot put an exact probability on a shock event or a known unknown happening, being aware of a potential event is still helpful in developing risk-mitigation strategies and plans. The authors highlight the importance of exploring how the possibilities for shock events and risks can help inform preferred alternatives and strategies for managing risk.

Kim (2012) develops a methodical process of evaluating potential shock events through a classification matrix. The process is designed for a project management audience, but the conceptual process is similar to other risk identification strategies discussed in other literature [e.g., ACRP Report 76 (Kincaid et al. 2012)]. The classification approach proposes to extend the analysis of known unknowns to consider what unknown unknowns there may be throughout a project’s lifetime. The author provides sample metrics for how unknown shock events may be categorized providing a useful guide to conceptualizing shock events and considering what kind of shocks could befall a project. The author provides brief case studies applied to two environmental disasters to major infrastructure/energy installations that are relatable to the capital-heavy and network-focused aviation sector.

Modeling Shock Events in Aviation Forecasting

This section examines methods used in air traffic forecasting that can be used to incorporate shock events. These include methods that are used to consider risk and uncertainty more generally.

High- and Low-Scenario Forecasts

One of the most common methods to model alternative forecasting scenarios is the generation of high and low scenarios developed by modifying forecasting assumptions to produce a more optimistic or pessimistic outlook, respectively. This procedure can easily be incorporated into standard forecasting techniques, and its ubiquity has led to the concept of a triad of scenarios with a central base case (also referred to as the medium or most likely forecast) and two alternatives of higher and lower growth or traffic levels. This is widely understood and intuitive for forecasters, planners, and other stakeholders.

High and low scenarios are developed by varying assumptions or inputs to the forecast, such as higher or lower economic growth driving higher traffic growth or more optimistic or pessimistic outlooks on future air service development supporting passenger growth in the forecast. However, the development of these high and low scenarios provides no information on how likely a given scenario is to occur nor do these scenarios typically have much input to further planning activities (Kincaid et al. 2012). While they may be used in some aspects of future strategic planning or financial analysis, low and high scenarios are not typically used to

COVID-19 pandemic. While IATA’s 2021 forecast has a more pessimistic outlook on 2021 than its projections a year prior, their forecasts maintained a widely held assumption that global air traffic would return to and likely surpass 2019 levels by 2023.

As a further example, the International Civil Aviation Organization (ICAO) also developed a series of potential recovery pathways for the aviation industry from the impacts of the COVID-19 pandemic (ICAO 2020). By this time in the pandemic, it was clear that the duration of the COVID-19 pandemic was not going to be a short; transitory impact, like SARS in 2003, and the scenarios developed by ICAO in mid-2023 reflected a longer recovery pathway. ICAO, like many organizations and observers, identified that there was significant uncertainty in the future recovery path of the aviation industry and developed a series of scenarios based on varying assumptions of factors influencing air traffic recovery, demand, and supply given the information available to the forecasters at that snapshot in time.

Sensitivity Testing

Sensitivity testing, or sensitivity analysis, is a process in which forecasting assumptions are varied one at a time, with the resulting changes in forecast output recorded and compared as assumptions change versus the base case. A sensitivity analysis requires an existing baseline forecast with a set of assumptions ranging from input variables to an econometric model to market share projections or air service development growth outlooks.

Sensitivity testing can be helpful in the development of a forecast for a variety of reasons. Testing individual assumptions on how the forecast will react to a given change in a single variable provides forecasters with information to determine how robust the forecast is. Sensitivity testing can also be useful to understand the role of critical assumptions driving the forecast and can be explored to see how large of a departure from the baseline assumption would lead to some other decision or action resulting from the forecast. Forecast reports may include sensitivity tests as a way to demonstrate how the forecast reacts to specific changes in inputs or to allow forecast users to better understand how changes in specific assumptions (e.g., macroeconomic conditions or airline networks) could impact future traffic volumes beyond the presented scenarios.

Sensitivity testing could be used for assessing shock events as these events may result in changes to other assumptions and inputs to the model. For example, the departure of a major carrier from an airport may be due to an economic recession or major regulatory changes that could impact multiple assumptions in a model. However, as shocks often have interconnected effects across the aviation industry and economy, changing only a single variable or assumption may underrepresent the total potential impact by leaving linked or correlated assumptions unchanged.

Monte Carlo Simulation (Stochastic Forecasting)

Monte Carlo simulation (or the Monte Carlo method) is a form of simulation analysis for forecasting designed to incorporate uncertainties into the forecasting process. It makes use of randomization and probability statistics to generate a wider range of possible traffic outcomes than conventional base-case master plan forecasts and provides estimates of the probabilities of such outcomes. By modeling various uncertainties and risks to future traffic levels—such as if and when a recession might occur, what the impact of a pandemic is if one occurred, or how annual jet fuel prices may differ from the baseline forecast prices—Monte Carlo simulation allows for quantification of the uncertainty of project traffic volumes and to capture “out-of-model” factors that traditional air traffic forecasts cannot. The application of Monte Carlo to air traffic forecasts is well covered in ACRP Report 76 (Kincaid et al. 2012).

Monte Carlo involves running the forecast model multiple times (e.g., 10,000 times) with each run generating different randomized values for the input variables. For example, one run might have normal economic performance but with high fuel costs and a terrorism event. Another might have weak economic performance and high fuel costs but with no terrorism event, and so on. The output from each of the runs (or iterations) is collected and can be shown as a distribution of outcomes with associated probabilities.

An illustrative example of the forecast probability risk range is presented in Figure 25, showing the central “most likely” (50th percentile) outcome as well as probability bands ranging from the 5th to 95th percentiles. For example, while the most likely forecast was 149,000 aircraft operations in Year 20, the 5th percentile was 59,000 and the 95th percentile was 218,000 (i.e., 90% of the simulations fell within that range).

The probability distribution of forecast airport activity levels—be it passengers, aircraft operations, cargo tonnage, etc.—provides a quantitative outlook at not just the forecast range but also the likelihood of a given level of future activity. Whereas a traditional low/medium/high or any other scenario forecast cannot provide quantitative information on how likely that scenario is to occur, the risk-based forecasting method explicitly allows forecast users to understand the forecast probability of any future activity level is within the forecast time horizon.

Shock events can be and are included in a Monte Carlo simulation by specifying a probability and impact associated with the event. For example, the impact of a terrorist attack can be included by adding a variable with the following characteristics:

• Probability of occurrence (e.g., a 5% probability in any given year).
• Impact of the event (e.g., a 10% to 30% reduction in traffic, randomly determined).
• Extent of recovery and the recovery period (e.g., recovery of all lost traffic after 2–5 years, randomly determined).

Even with this approach, there are some types (or scales) of shock events that cannot be anticipated, and by itself, Monte Carlo may not be sufficient for management planning for shock events, as it does not focus on one specific stressor. It can still be valuable in evaluating the overall risk profile of the airport and the implications of that risk, but to address shock events, they may need to be combined with scenario forecasting.

However, the perceived low-probability nature of shock events means that their impact is not prominent in the aggregate averaging of 10,000 or more runs and can get missed in the mix of other factors influencing the forecasts. This is illustrated in Figure 26, which shows the distribution of forecast traffic volumes in a specified year (Year 10) at a fictional airport. The impact of randomly including shock events is to somewhat lower the median forecast and widen the “tails” of the probability distribution for traffic in Year 10. While arguably the inclusion of shock events in Monte Carlo is beneficial in showing the wider range of outcomes, it does not highlight the significant and potentially profound effects of certain shock events on airport traffic.