Average Fare Modelling

The modelling of changes in airline airfares as a function of changes in airline cost enables simulation of how airline cost changes impact passenger demand. As described by Waitz et al. (2006-2), this is particularly important when modelling policy measures intended to mitigate the environmental impacts of aviation. This is because passenger demand responses to increased fares resulting from policy interventions may lead airlines to schedule fewer flights. This could result in greater reductions in emissions and noise than forecast by an analysis that does not model this effect on demand.

Airline fare strategies are, however, highly complex, particularly in the United States and Europe. This has resulted primarily from deregulation of the airline industry (in 1978 in the United States and by 1997 in Europe), the development of computerised revenue management systems (in the 1980s), and the development of web-based airline ticket distribution and on-line travel agents (in the late 1990s) (Carrier, 2006). As a result of these developments, airlines do not simply apply cost-based pricing, but apply a combination of cost-, demand-, and service-based pricing, applying price discrimination and product differentiation to increase total flight revenues, with little consideration to total operating costs (Belobaba, 2008). Airlines segment markets with different levels of willingness to pay by offering different fare products to business and leisure travellers, preventing diversion by setting restrictions on lower fare products (e.g. requiring a Saturday night stay) and limiting available seats. The result is higher revenues and load factors than could be achieved through any single fare strategy.

Approaches to Modelling Fares

Models exist that can be applied directly to airline ticket pricing, such as the probabilistic decision model described by Belobaba (1989), which identifies what booking limits should be applied to the number of seats available at different prices on the same flight in order to increase airline revenue. These models are, however, complex, and are not tailored to forecasting applications, requiring information about passengers that is difficult to predict.

Detailed Modelling

Other approaches model average fare changes as a function of changes in airline cost without considering airline revenue management, and are more appropriate to forecasting applications. Fare models applied in the context of forecasting aviation growth in the future typically do not model competition effects explicitly. The Aviation System Analysis Capability (ASAC) Air Carrier Investment Model (ACIM) (Wingrove et al., 1998), allows the user to specify airline yields or operating profits, and how they are expected to change over time. The model adjusts fares to produce the yields or operating profits specified. This approach has the advantage of allowing differentiation between airlines, but does not explicitly model competition beyond assuming that competition maintains profits at the levels specified.

An alternative approach optimises average fares to maximise system profit, within limits, as described by Pulles et al. (2002). This approach effectively identifies the cost “pass- through” to fares that results in maximum profit, modelling the impact of fare changes on passenger demand, and the impact of passenger demand changes on capacity supply (flights), which in turn impact cost. The approach also allows for user specified cost pass-through, if desired. However, it does not capture any competition effects, nor does it model any revenue price discrimination or revenue management approaches. Waitz et al. (2006-2) describes a similar approach in which a bounding analysis is completed. A range of scenarios are modelled with different degrees of cost pass-through to fares applied in each case. For example, scenarios are run in which airlines pass 100%, 50% and 0% of cost increases through to fares. This approach is very transparent, providing a range of outcomes with bounds, but lacks guidance as to the most likely outcome. Again, airline competition effects are not modelled explicitly, despite the fact that they may drive the degree to which cost is passed through to fare.

A further approach adjusts fares to maintain an existing airline rate of return, as described and applied by Waitz et al. (2006-3). This is accomplished by maintaining a proportional relationship between fares and costs. Thus, if costs are predicted to increase by a certain percentage, fares are modelled to increase by the same percentage. This approach also does not explicitly account for competition effects.

Chapter 5

stage, Nash best-response game, identifying flight frequencies and average O-D market fares for different passenger types, as described in Section 2.3. The impact of competition on fares and frequency is modelled explicitly, and some degree of price discrimination (different passenger types are modelled) is captured. Also as described in Section 2.3, Schipper et al. (2003) and Carlsson (2002) model the effects of competition on average fares and flight frequencies by solving a two stage game separately for each market, as opposed to the whole network operated by the airline. In the first stage of the game, airlines simultaneously choose flight frequencies in the market, and in the second stage, after having observed the other airlines’ chosen frequencies, the airlines simultaneously choose fares for the market. The game is solved analytically as a function of passenger value of time, airline costs and passenger demand. This formulation was modified by Evans et al. (2008) for variable passenger demand. Both formulations, however, do not model any price discrimination or revenue management. Nor does the model distinguish between different passenger routings on the same O-D market. Fares, passenger value of time and costs may differ quite significantly for different routes in the same market, particularly between non-stop flights and connecting flights. Finally, the even distribution of flights through the day applied to define the flight schedule does not capture passenger preferences to fly at certain times of day (particularly the early morning and evening), and therefore ignores the increased demand at these times. Nero (1998) and Januszewski (2004) make similar assumptions to formulate equations for fare as a function of cost within a competitive environment in order to examine airline scheduling and the effect of flight delays on airlines’ prices.

Modelling of Fares in the Airline Response Model

Ideally, average O-D fares by passenger type should be included as a decision variable in each airline’s network optimisation objective function, described in Section 5.1. This would allow different fares to be identified for each airline within the competitive environment, subject to airline operating costs. However, this approach would add significant complexity, because the calculation of market share in each network optimisation would require passenger choice modelling based on both fare and flight frequency, as opposed to the relatively simple passenger choice model based on flight frequency only (equation 5-3). Such an approach would also be unlikely to add significantly to the model results, because in reality fares are set using complex revenue management techniques, and in a different cycle

Detailed Modelling

to the more strategic decisions of flight frequency and network choice. Instead, the modelling approach adopted for the Airline Response Model is to adjust base year fares to maintain the existing rate of return. This is identical to the approach used by Waitz et al. (2006-3), and is accomplished by maintaining a proportional relationship between fares and costs. Thus, any percentage change in cost is applied directly to base year fares.

Base year fares in the United States are obtained from published fare lists (DOT, 2007), while base year operating costs by O-D city pair are estimated by running the Airline Response Model with fares fixed to base year values. This ensures that modelled fares deviate from observed base year fares according to how modelled operating costs deviate from operating costs that are consistent with the observed base year fares.