This document reports the findings of a review of the economics and business literature on empirically-estimated own-price elasticities of demand for air travel for Canada and other major developed countries. It does not explicitly measure the impact of the Air Travellers Security Charge on air travel in Canada, nor does it assess how the enhanced air travel security system has affected the demand for air travel.
The demand for a particular good or service depends on a variety of factors. Key influences include the tastes of consumers, the levels of consumer income, the price and quality of the product in question and the prices of other goods, especially goods that are close substitutes. In order to obtain useful estimates of the price sensitivity of demand for a product, researchers must carefully control for all the factors affecting the demand.
As a general rule, when other influences on demand remain unchanged, a higher price for a product results in a lower quantity demanded. However, the price responsiveness of demand varies from one good to another and from one market to another. The own-price elasticity of demand measures the responsiveness, or sensitivity, of the demand for a good to changes in its price when other influences on demand are held constant. It is defined as the percentage change in quantity demanded resulting from a given percentage change in price.
For example, if a 1% increase in price results in a 1.2% decrease in quantity demanded, the own-price elasticity of demand is 1.2. In this case, since the percentage fall in demand is greater than the percentage rise in price, total spending on the good will decline, and the demand for it is said to be "elastic". If, on the other hand, a 1% price rise causes a smaller percentage decline in the quantity demanded, the own-price elasticity will be less than one, and demand is said to be "inelastic".
This report identifies six distinct markets for air travel. Specifically, it is observed that studies of the demand for air travel demand should distinguish among markets for: business and leisure travel; long-haul and short-haul travel; and international and North American long-haul travel. Accordingly, to examine the sensitivity of the demand for air travel to its price, separate estimates of the own-price elasticity of demand are gathered for each of these distinct markets.
Since the availability of alternative modes of transportation that are reasonably close substitutes for air transport diminishes with distance travelled, it is expected that the demand for air transport will be less elastic for longer flights than for shorter flights. Further, international travel tends to be spread over more time than domestic travel, so that the airfare is a smaller proportion of overall trip costs, which makes international travel less sensitive to changes in ticket prices. In addition, leisure travellers are more likely to postpone trips to specific locations in response to higher fares, or to shop around for those locations offering more affordable fares. Consequently, it is expected that the demand for air transport for leisure reasons will be more elastic than business travel.
Standardized information on 21 Canadian and international empirical studies of own-price elasticities of demand for air travel is provided in this report. This information includes the author(s), study date, country/countries considered, model type and functional form, type and source of data, and markets examined (e.g., business, long haul, and international). A scoring system based on desirable characteristics of demand studies is applied to each of the 21 studies to improve the level of confidence regarding the practical use of own-price elasticity values in assessing air travel demand. For example, recent studies are more highly rated, as are studies that control for changes in consumer income, and, for short-haul markets, the availability of alternative travel modes. This methodology identifies 11 studies that are expected to provide particularly relevant estimates for different markets.
Summary results from the report are reproduced in Chart 1. Median elasticity values differ significantly by type of traveller, travel distance and domestic/international routes, and confirm the existence of distinct air travel markets. They generally accord with expectations that the demand for air transport should be less elastic for business than for leisure travellers, and less elastic for flights of longer distances, notably for international travel. Based on available studies, however, markets for business travel do not show a consistently increasing responsiveness as distance travelled becomes shorter. The chart also shows, for each market, the range of elasticity values within which one-half of the estimates lie.
There are two important caveats to the interpretation and use of these estimates for purposes of assessing the responsiveness of air transport demand in Canada to the ATSC.
Chart 1 - Own-Price Elasticities of Demand
The ranges of values shown capture the middle one-half of the estimates and encompass the median, which is represented by a black dot.
Consequently, the median value for a particular market is not necessarily representative of the corresponding market in Canada. It is argued in the report that analysts should instead consider the broader range of values around the median value when examining the impact of price changes on the demand for air travel in Canada.
The purpose of this study is to report on all or most of the economics and business literature dealing with empirically estimated demand functions for air travel and to collect a range of fare elasticity measures for air travel and provide some judgment as to which elasticity values would be more representative of the true values to be found in different markets in Canada.
While existing studies may include the leisure – business class split, other important market distinctions are often omitted, likely as a result of data availability and quality. One of the principal value added features of this research and what distinguishes it from other surveys, is that we develop a meta-analysis that not only provides measures of dispersion but also recognizes the quality of demand estimates based on a number of selected study characteristics. In particular, we develop a means of scoring features of the studies such as focus on length of haul; business versus leisure; international versus domestic; the inclusion of income and inter-modal effects; the age of the study; data type (time-series versus cross section) and the statistical quality of estimates (adjusted R-squared values). By scoring the studies in this way, policy makers are provided with a sharper focus to aid in judging the relevance of various estimated elasticity values.
Elasticity values in economic analysis provide a "units free" measure of the sensitivity of one variable to another, given some pre-specified functional relationship. The most commonly utilized elasticity concept is that of "own-price" elasticity of demand. In economics, consumer choice theory starts with axioms of preferences over goods that translate into utility values. These utility functions define choices that generate demand functions from which price elasticity values can be derived.
Therefore elasticities are summary measures of people's preferences reflecting sensitivity to relative price levels and changes in a resource-constrained environment. The ordinary or Marshallian demand function is derived from consumers who are postulated to maximize utility subject to a budget constraint. As a good's price changes, the consumer's real income (which can be used to consume all goods in the choice set) changes. In addition the goods price relative to other goods changes. The changes in consumption brought about by these effects following a price change are called income and substitution effects respectively. Thus, elasticity values derived from the ordinary demand function include both income and substitution effects.
Own-price elasticity of demand measures the percentage change in the quantity demanded of a good (or service) resulting from a given percentage change in the good's own-price, holding all other independent variables (income, prices of related goods etc.) fixed. The ratio of percentage changes thus allows for comparisons between the price sensitivity of demand for products that might be measured in different units (natural gas and electricity for example). 'Arc' price elasticity of demand calculates the ratio of percentage change in quantity demanded to percentage change in price using two observations on price and quantity demanded. Formally this can be expressed as:
represent the observed change in quantity demanded and price
represent the average price and quantity demanded. The elasticity is unitless and can be interpreted as an index of demand sensitivity; it is measuring the degree to which a variable of interest will change (passenger traffic in our case) as some policy or strategic variable changes (total fare including any added fees or taxes in our case).
In the limit (when are very small) we obtain the 'point' own-price elasticity of demand expressed as:
Q(P,S) is the demand function
P = a vector of all relevant prices
p = the good's own-price.
q = equals the quantity demanded of the good
S = a vector of all relevant shift variables other than prices (real income, demographic characteristics etc.)
We expect own-price demand elasticity values to be negative, given the inverse relationship between price and quantity demanded implied by the 'law' of demand, with absolute values less than unity indicating 'inelastic' demand: a less than proportionate response to price changes (relative price insensitivity). Similarly, absolute values exceeding unity indicate elastic or more sensitive demand: a more than proportionate demand response to price changes (relative price sensitivity).
The ratio of change in quantity demanded to change in price [equation (1)] highlights that elasticity measures involve linear approximations of the slope of a demand function. However, since elasticity is measuring proportionate change, elasticity values will change along almost all demand functions, including linear demand curves. Estimation of elasticity values is therefore most useful for predicting demand responses in the vicinity of the observed price changes. As a related issue, analysts need to recognize that in markets where price discrimination is possible aggregate data will not allow for accurate predictions of demand responses in the relevant market segments. In air travel, flights by a carrier are essentially joint products consisting of differentiated service bundles that are identified by fare classes. However the yield management systems employed by full-service carriers (FSCs) also create a complex form of inter-temporal price discrimination, in which some fares (typically economy class) decline and some increase (typically full-fare business class) as the departure date draws closer. This implies that ideally, empirical studies of air travel demand should separate business and leisure travellers or at least be able to include some information on booking times in order to account for this price discrimination, and that price data should be calibrated for inter-temporal price discrimination: for example, the use of full-fare economy class ticket prices as data will overestimate the absolute value of the price elasticity coefficient. Within the set of differentiated service bundles that comprise each (joint product) flight, the relative prices are important in explaining the relative ease of substitution between service classes. Given the nature of inter-temporal price discrimination for flights, the relative price could also change significantly in the time period prior to a departure time.
The partial derivative in (2) indicates that elasticity measures price sensitivity independent of all the other variables in the demand function. However when estimating demand systems over time, one can expect that some important shift variables will not be constant. It is important that these shift variables be explicitly recognized and incorporated into the analysis, as they will affect the value of elasticity estimates. This will also be true with some cross-sectional studies or panels. In particular changes in real income and the prices of substitutes or complements will affect demand. In air travel demand estimations, income and prices of other relevant goods should be included in the estimation equation. Alternative transportation modes (road and rail) are important variables for short-haul flights, while income effects should be measured for both short and long-haul. The absence of an income coefficient in empirical demand studies will result in own-price elasticity estimates that can be biased. With no income coefficient, observed price and quantity pairs will not distinguish between movements along the demand curve and shifts of the demand curve.
The slope of a demand function, which affects the own-price elasticity of demand, is generally expected to decrease (become shallower) with:
Given the implied relationships above, any empirical demand study should carefully define market boundaries to include all relevant substitutes and complements and to exclude products that might be related through income or other more general variables.
In air travel, ideally market segment boundaries should be defined by first separating leisure and business passengers and second long-haul and short-haul flights. The reason is that we expect different behaviour in each of these markets. Within each of these categories, distinctions should then be made between the following:
In addition, for the North American context, long-haul flights should be further divided into international and domestic travel (within continental North America). These market segment boundaries are illustrated in figure 2.1 below, which also highlights the relative importance of intermodal competition for short-haul travel.
While distinctions in price and income sensitivity of demand between business and leisure or long and short-haul travel are more intuitive, other distinctions are perhaps less obvious. If available, data that distinguishes between routes, airlines and airports would provide important estimates of how price sensitivity is related to the number of competing flights and the willingness to pay of passengers utilizing a hub-and-spoke network, relative to those traveling point-to-point, more commonly associated with low cost carriers. To the extent that existing studies assume that each passenger observation represents O-D travel, they will not be capturing fare premiums usually associated with hub-and-spoke networks and full service carriers, nor will they necessarily capture the complete itinerary of travellers utilizing a number of point-to-point flights with a low cost carrier. For example, a passenger who travels from Moncton to Vancouver with Air Canada, and utilizes the hub at Pearson International airport, is being provided with a number of services that includes baggage checked through to the final destination and frequent flyer points as well as a choice in flights and added flight and ground amenities. The fare for Moncton-Vancouver includes a premium for these services. Now consider a passenger that is travelling with WestJet from Moncton to Hamilton, and then with JetsGo from Toronto Pearson Airport to Vancouver. In this case there are no frequent flyer points to be attained and baggage has to be collected and re-checked after a road transfer between Hamilton and Pearson International. Although the origin and destination is the same for these passengers, the itineraries are significantly different. In many cases data used for demand estimates would not able to account for these differences.
Route-specific data can also capture competition that may exist between airports and the services they offer as well as airlines. This may be especially true for certain short-haul routes where intermodal competition (road and rail) can play an important role in shaping air travel demand.
Market segments in air travel demand.
Oum et al. (1992) provide a valuable list of pitfalls that occur when demand models are estimated and therefore affect the interpretation of the elasticity estimates from these empirical studies.
1. Price and Service Attributes of Substitutes: Air travel demand can be affected by changes in the prices and service quality of other modes. For short-haul routes (markets) the relative price and service attributes of auto and train would need to be included in any model; particularly for short-haul markets. Failure to include the price and service attributes of substitutes will bias the elasticity. For example, if airfares increase and auto costs are also increasing, the airfare elasticity would be overestimated if auto costs were excluded.
2. Functional Forms: Most studies of air travel demand use a linear or log-linear functional specification. Elasticity estimates can vary widely depending on the functional form. The choice of functional form should be selected on the basis of statistical testing not ease of interpretation.
3. Cross-Section vs. Time-series Information: In the long run demand elasticities for non-durable goods and services are larger in absolute terms, than in the short run. This follows because in the long run there are many more substitution possibilities that can be used to avoid price increases or service quality decreases. In effect there are more opportunities to avoid these changes with substitution possibilities. Data tends to be cross-sectional or time-series although more recently panels have become available. A panel is a combination of cross-section and time-series – information on several routes for a multi-year period is a panel. Cross-sectional information is generally regarded as indicating short run elasticities while time-series data is interpreted as long run elasticities. In time-series data the information reflects changes in markets, growth in income, changes in competitive circumstances, for example. Policy changes should rely on long run elasticities since these are long run impacts that are being modelled. Short run elasticities become important when considering the competitive position of firms in a highly dynamic and competitive industry.
4. Market Aggregation/Segmentation: As the level of aggregation increases the amount of variation in the elasticity estimates decreases. This occurs because aggregation averages out some of the underlying variation relating to specific contexts. Since air travel market segments may differ significantly in character, competition and dominance of trip purpose, interpreting a reduction in variation through aggregation as a good thing would be erroneous. Such estimates might have relatively low standard deviations but would be also be relatively inaccurate when used to assess the effect of changes in fares in a specific market.
5. Identification Problem: In most cases only demand functions are estimated in attempts to measure the demand elasticity of interest. However, it is well known that the demand function is part of a simultaneous equations system consisting of both supply and demand functions. Therefore, a straightforward estimation of only the demand equation will produce biased and inconsistent estimates. The problem of identification can be illustrated by describing the process by which fares and travel, for example, are determined in the origin-destination market simultaneously. To model this process in its entirety, we must develop a quantitative estimate of both the demand and supply functions in a system. If, in the past, the supply curve has been shifting due to changes in production and cost conditions for example, while the demand curve has remained fixed, the resultant intersection points will trace out the demand function. On the contrary, if the demand curve has shifted due to changes in personal income, while the supply curve has remained the same, the intersection points will trace out the supply curve. The most likely outcome, however, is movement of both curves yielding a pattern of fare, quantity intersection points from which it will be difficult, without further information, to distinguish the demand curve from the supply curve or estimate the parameters of either.
Earlier we identified sources of bias that can arise from problems with aggregation, data quality, implicit assumptions of strong separability among others. Almost all demand studies have an implied assumption of strong separability in that they only consider aviation markets in the analysis. Such studies in effect constrain all changes or responses in fares or service to be wholly contained in the aviation component of people's consumption bundle. The paper by Oum and Gillen (1986) is the one exception where consideration of substitution with other parts of consumption was included in the modelling. It would be difficult to extract a conclusion from this one study as to existence, degree and direction of bias in elasticity estimates when other parts of consumption are and are not included in the modelling. However, having said this, an inspection of the elasticity estimates from this study shows they are not significantly different than other time-series estimates.
Elasticity estimates depend critically on the quality and extent of the data available. Currently, the best data for demand estimation is the DB1A 10 percent ticket sample in the US, but even this data has some problems. The DB1A sample represents 10 percent of all tickets sold with full itinerary identified by the coupons attached to the ticket. However with electronic tickets, as more and more tickets are being sold over the Internet, there is a growing portion of overall travel that may not be captured in the sample. This means that the proportion is not 10 percent but something less. Other important considerations are the amount of travel on frequent flyer points, by crew and airline personnel.
In Canada we have poor quality data because it is incomplete, even if it were accessible. Airports collect traffic statistics but these data make it very difficult to distinguish OD and segment data. Airlines report traffic data to Statistics Canada (or are supposed to) but these data do not include fare information or routing. Knowing the itinerary or routing is important because of differences in service quality and hubbing effects. Fare data is also more useful than yield information since it identifies the proportion of people travelling in different fare classes. Yet, in many cases yield information is used as a weighted average fare. There is also the problem that carriers of different size may have different reporting requirements. Some researchers and consultants have been cobbling together data sets for analysis by using the PBX clearing house information. These data are limited and apply only to those airlines that are members of IATA. The current public data available in Canada simply does not permit estimation of any demand models.
Besides demand side data it is also important to have supply side information. Elasticity estimates should emerge from a simultaneous equations framework. This data is more accessible through organizations like the OAG, which provide information on capacity, airline and aircraft type for each flight in each market. These data measure changes in capacity, flight frequency and timing of flights.
One study, which undertook an extensive survey to collect multimodal data, was the High Speed Rail study sponsored jointly by the Federal, Ontario and Quebec governments. This study, which had three different demand modelling efforts, examined the potential for High Speed Rail demand, and subsequent investment, in the Windsor-Quebec corridor. The analysis included intermodal substitution between air, rail, bus and car. The study was undertaken in the early 1980s. However, it is not possible for public access to any of the technical documents that would allow an assessment of the study. Attempts in the past to obtain access to the data have proven fruitless.
As we have stated, price elasticity measures the degree of responsiveness to a change in own or other prices (fares). However, care must be exercised in interpreting the elasticity since they differ according to how they have been estimated. Many empirical studies of air travel demand estimate a log-linear model. In evaluating such studies, it is important to keep in mind that the empirical specification implies a certain consumer preference structure because of the duality between utility functions and demand functions. It is equally important to remember that empirically estimated demand functions should contain some measures of quality and service differences or quality changes over time. Failure to include metrics for frequent flyer programs, flight frequency, destination choice or service levels in estimating an air demand function can lead to downward bias in the price elasticity estimates.
Price elasticities can be estimated for aggregate travel demand as well as modal demand. Figure 3.1 illustrates the differences between aggregate and modal elasticities. Our interest is in modal elasticities not the aggregate amount of travel but it is important ultimately that any policy analysis take account of the impact of any policy change on aggregate travel as well as modal redistribution. The impact of a change in price on aggregate demand would be measured by the –fis in Figure 3.1 whereas the Fiis would measure the impact on air travel demand. The Fiis are a composite or combination of the fis and the Miis.
The Canadian aviation industry has undergone significant change in the last several years. In 2000 Air Canada completed its takeover of Canadian Airlines, which left it with in excess of 80 percent market share. Market dominance leads to different fare and service quality levels. As a result of higher fares, for example, we should find higher absolute values of elasticities of demand simply because with higher fares we have moved further up the demand curve. In 1996 Westjet entered the market and has continued to grow each year. Canada 3000 exited the market in 2001, as did Canjet and Royal (as part of Canada 3000). Roots airline has come and gone but Canjet has reemerged in eastern Canada and JetsGo is offering some level of service on longer haul domestic flights as well as in the Montreal-Toronto market.
The entry of low cost carriers leads to lower fares for a subset of traffic and competitors will offer a supply of seats to match these fares. Lower average fares should lead to lower demand elasticity estimates, while increases in the number of competitors in the market will lead to higher demand elasticity estimates.
One should not confuse low cost carriers with a seeming lack of exploiting monopoly power. High prices or fares are not synonymous with monopoly and low fares with competition. Airlines like Westjet where they are the sole airline serving the market may still act as a monopolist but charge low(er) fares. Profit maximizing monopolists price where marginal cost equals marginal revenue, if marginal cost is low, one should expect to see lower fares but still marginal cost and revenue are equalized. Monopolists are generally viewed as being high price because they are high cost and the high costs are attributable to some degree from a lack of competitive discipline in the market. Full service carriers operating with hub-and-spoke systems have a high cost business model while low cost carriers have a low cost business model.