The structuralist school

      Bain argues that the "condition of entry is determined...by the advantages of established sellers in an industry over potential sellers", with the comparison made between the pre-entry profits of established firms and the post-entry profits of entrants. Thus, a barrier to entry exists if an entrant cannot achieve the profit levels post-entry that the incumbent enjoyed prior to its arrival. Further, he argues that

      "...the condition of entry is...primarily a structural situation...[which] describes...the circumstances in which the potentiality of competition will or will not become actual" (Bain 1956, p. 3). He suggests three kinds of behaviour by incumbents in the face of an entry threat: blockaded entry, deterred entry and accommodated entry. He distinguishes between the fundamental structural conditions which create asymmetries between firms and the strategic behaviour of incumbents which exploit them. He argues that asymmetries in outcome ultimately (i.e. in the long run) rest on structural factors.

The Chicago school of efficiency

      Stigler's view of barriers contrasts with Bain's. He proposes that: "...a barrier to entry may be defined as a cost of producing (at some or every rate of output) which must be borne by firms which seek to enter an industry but is not borne by firms already in the industry" (Stigler 1968, p. 67). The primary conceptual difference between Stigler and Bain is that, in the former case, the entrant and incumbent are compared post-entry: a barrier exists if the two are not equally efficient after the costs of entering the industry are taken into account. Bain's emphasis on the conditions of entry assigns an entry barrier to any industry in which structural conditions exist that permit an established firm to elevate prices above the minimum average cost of potential entrants. Stigler considers an entry barrier to exist only if the conditions of entry are less difficult for established firms than for new entrants. Although Stigler's definition may appear somewhat ambiguous, in terms of measuring these barriers or when accounting for revenue-based, not cost-based, sources as barriers (i.e. product differentiation), his definition remains applicable in its broader generalized sense.

      The practical distinction between Bain and Stigler lies in the evaluation of economies of scale. Economies of scale, according to Bain, are barriers to entry due to the structural advantages they provide to the incumbent, whether the incumbent operates at minimum efficient scale or not. With Stigler, such economies do not present barriers if the incumbent is forced, after entry happens, to produce at suboptimal levels together with the entrant. Stigler attributes such cost disadvantages to new conditions in demand for the incumbent after entry, and not to the existence of particular barriers to entry.

The normative school

      Von Weizsäcker's concern is not with the factors that impede the mobility of capital, but rather with "socially undesirable limitations to entry of resources which are due to protection of resource owners already in the market" (Von Weizsäcker 1980, p. 13). His definition is a normative qualification of the definition proposed by Stigler: "...a cost of producing (at some or every rate of output) which must be borne by firms which seek to enter an industry but is not borne by firms already in the industry, and which implies a distortion in the use of economic resources from the social point of view." This definition implies that there could be too little entry due to excessive protection, as well as too much entry due to too little protection.

      With new entrants likely to create excess capacity, economic surplus would increase with increased costs of entry as well. In cases where entry costs increased total economic surplus, von Weizsäcker's definition of entry barriers would not apply. Demsetz (1982) has further extended this normative approach by arguing that, in many cases, what is called an entry barrier is an endogenous response to consumer preferences and supports an efficient allocation of resources. Rather than allowing resources into high profit industries, it might be better to take into account the role of externalities, information and transaction costs, and to consider entry barriers as a valuable second best answer to real world frictions.

      The major strength of this approach - its normative focus - is also the source of its major weakness. It is difficult enough to determine and measure barriers to entry without adding an additional layer of normative complexity. Normative issues can still be discussed after a barrier has been identified and measured.

      Returning to the focus of this thesis, we then ask: "Based upon our understanding of barriers to entry (and to mobility): What types of entry barriers can we identify and how do they apply to the airline industry? And what are the likely consequences, especially in terms of firms' strategic behaviour and pricing, of such barriers?" Ultimately, it was my intention in this thesis to develop a methodology that depicts the impact of entry barriers and strategic behaviour on the airline industry's choices in pricing and output following deregulation in Europe.

      Obviously, the concepts behind the different types of barriers to entry needed to be closely examined. I also had to understand the impact of these barriers for incumbents in terms of strategic behaviour and how this applied specifically to the examined industry. As studies advanced on the subject, it became clear that eminent economists had developed different models that partially contradicted each other - models that remained theoretic and were subject to many constraints. How could I apply such theory and its potential predictive power to the real world of airlines and their changed competitive environment? Being convinced that the concept of entry barriers was highly relevant for real-world industry analysis, I decided that one way to advance was to understand the industry and to find relatively detailed operational data for capacity and prices that were used in the economists' models.

      As my studies and research advanced, I became acquainted with the succinct theory of perfectly contestable markets. The more I learned about it, the more I realized its potential for my thesis. Contestability theory was intuitive and provided general predictions, especially in terms of pricing. There were few constraints to the theory, and the most important ones were exactly the ones I was out to test - entry barriers and strategic behaviour. In addition, significant studies based on this theory had already been conducted for the airline industry in the US. Somehow, it appeared that specific industry knowledge could be reconciled with the robustness of contestability theory. To prove the non-contestability of the European airline sector despite deregulation it was essential to find very specific and relatively detailed data for parameters that were suspected of raising entry costs and thus impacting on prices.

II.3.1 General cost drivers in the industry

      Before examining some prominent economies in the industry in detail, I would like to provide a more general overview over the cost structure of the industry. This helps to situate possible economies of scale within the big picture of total costs of the industry or firm. In line with the remark already made by Spence on the role of advertising and economies of scale, I would like to show the possibility that within an industry or firm there are likely to be different sources for increasing returns to scale, which need not be aligned at all times. The introduction of several potentially important cost drivers in the industry will help to indicate such possible increasing returns, which may then be examined further on. This section will in large part be based on "An empirical study of cost drivers in the US airline industry" by Banker and Johnston (1993). The authors distinguish cost drivers that are volume based from those that are operations based. This distinction is noteworthy, seeing economies of scale (in the tradition of Bain, Spence et al.) as typically volume-based returns. Apparently, operations can also have significant influence on the firm's unit costs. Indeed, while Bain saw plant size as the key for achieving economies, operations can just as well lower unit costs to an extent that any new entrant who cannot employ the same operations without a cost will face a cost disadvantage.

      The following operations-based drivers represent choices of alternative technologies, as such choices of aircraft models, route structures, flight frequency or density, and traffic flow control. These drivers need to be regarded separately from scale-based drivers, which are based on the number of passengers transported or on the capacity of seat miles offered.

      When the examined cost drivers showed either increasing or decreasing returns, marginal costs would also change with these operations-based drivers. ¹³ 

      

Exhibit 2: Cost drivers of airlines and their impact on unit costs 14
Input category and measurement units	Volume based drivers	Operations based drivers
Fuel in gallons	Capacity seat miles by aircraft type (+)	Average stage length (-)
Flying operations, labour hours	Capacity seat miles by aircraft type (+)	Density (-) Hub concentration (-) Hub domination (-)
Passenger service, labour hours	Capacity seat miles by aircraft type (+)	Density (-) Hub concentration (-) Hub domination (-)
Maintenance, labour hours	Capacity seat miles by aircraft type (+)	Density (-) Hub concentration (-) Hub domination (-) Scale (+)
Maintenance materials and overhead, deflated costs	Capacity seat miles by aircraft type (+)	Density (+) Hub concentration (-) Hub domination (-)
Aircraft and traffic servicing, labour hours	Passengers (+)	Density (+) Hub concentration (-) Hub domination (-)
Promotions and sales, labour hours	Passengers (+)	Density (+) Hub concentration (-) Hub domination (-)
General overhead, deflated costs	Total capacity seat miles (+)	Density (+) Hub concentration (-) Hub domination (-) Scale (+)
Group property and equipment, deflated costs	Total capacity seat miles (+)	Density (-) Hub concentration (+) Hub domination (+) Scale (+)

(source: Banker and Johnston 1993, p.#580)

      

• The results for competitive and dominated hubs show large, significantly negative coefficients for labour that handles passengers, cargo and aircraft on the ground. Ground property and equipment and labour for flying operations are also significantly negative, and those for general overhead are significantly positive. Thus, by adopting a hub and spoke strategy, a carrier can achieve fairly substantial economies in the use of most inputs.
• The coefficient estimate for average stage length is significantly negative for the fuel cost category. This lends support to the hypothesis that the marginal requirements for fuel inputs diminish as average stage length increases.
• For promotions and sales labour, the coefficient for density is significantly positive, which indicates that adding flights on a given network requires additional support labour. The coefficient for passenger service labour is significantly negative. The coefficients for flying operations labour, maintenance labour, maintenance materials and overhead are insignificant. There are inherent difficulties in measuring capital inputs.
• Comparisons among older aircraft, notably between larger and smaller planes, indicated highly significant differences in coefficients and are consistent with the hypothesis and industry evidence that wide-bodied aircraft require less fuel and flying operations input than older regular-bodied aircraft. However, sharp differences between older large aircraft and newer, slightly smaller models are not indicated. This suggests that the improvements in fuel efficiency and reductions in required crew size for the newer models have made up for the size-based advantages larger aircraft had in the past.

II.3.2 Economies of scale versus density

---

---

---

II.3.3 Tracing cost curves for the airline industry

      

Exhibit 5: Cost relationships for the hub-density-aircraft complex

      The above graphical depiction provides a somewhat broad map of the relationships in the hub-density-aircraft complex. The signs (positive or negative) depict the expected relationships that empirical studies have found on marginal and average costs. In this section, I am trying to find a specific operations-based cost curve for the airline industry, which allows discussion of the trade-offs between increasing and decreasing returns and - it is hoped - identification of the shape of "minimum efficient density" in analogy to Bain's minimum efficient scale.

      As I have already pointed out, Caves, Christensen and Tretheway (1984) define returns to density as declining unit costs, with increasing numbers of passengers being transported between two points without any change in load factor occurring. Their parameters for declining unit costs (as well as marginal costs) are frequency of flights and size of aircraft (I chose to neglect the option of adding more seats to existing airplanes). Casually observed differences in unit costs for trunk airlines and for locals are largely explained by differences in characteristics of the firms, particularly by lower density of service and shorter stage lengths for the locals (p. 483). Thus, the following cost curve can be drawn:

      

Exhibit 6: A cost curve for the airline business (1)

      Encaoua (1991) identifies higher density solely with a higher load factor. He admits that the average load factor's impact on increasing, decreasing, or constant returns can hardly be shown on such an aggregated basis of overall airline operations. To him, what he calls geographical and network configurations are the source of variations in unit costs. He judges the aircraft type to be less relevant for such variations than average stage length. This statement per se does not contradict the above graph, though it lessens the significance of aircraft type for unit costs.

      Bailey, Graham and Kaplan (1991) focus on the returns of hub-and-spoke operations. Operating hub-and-spoke networks combines passengers with different origins and destinations - increasing the average number of passengers per flight and thereby reducing costs. Essentially the broader scope of operation lets the carrier take advantage of the economies of scale in aircraft (p. 74). The hubbing carrier would serve more passengers on its flights so it could use larger aircraft and/or higher load factors. This indeed is a classical argument of economies of scale: Larger, more efficient aircraft may offer lower unit cost if their capacity is sufficiently employed, just as a larger production plant may produce at lower average unit costs if its production volume is sufficiently close to its minimum efficient scale. This is highly relevant, because if Bailey et al. are right, then the key parameters of density would be nothing but simple economies of scale, similar in nature to Bain's concept of specialization. By funnelling traffic through a hub, hubs and spokes are instrumental in the exploitation of these economies of scale. Density together with hubs are a prerequisite for obtaining the minimum efficient scale, but would not be the origin of decreasing unit costs. Brueckner and Spiller (1994) build on this understanding of the density impacts of hubbing. They focus their research on marginal cost decreases when traffic on traffic spokes increases. They understand the growth of hub-and-spoke networks as an attempt to exploit economies of traffic density, according to which the marginal cost of carrying an extra passenger on a non-stop route falls as traffic on the route rises (p. 380). It is important to note that although they do not take into account the varying degrees of aircraft efficiency or load factors actually achieved, they nevertheless acknowledge their importance: " This higher density then can be exploited, analogously to Bailey, Graham and Kaplan (1991) by higher load factors or bigger, more efficient aircraft.

      

Exhibit 7: A cost curve for the airline business (2)

      If we understand hubbing and increased density as the means for exploiting economies of bigger and more efficient aircraft later on, then the marginal cost elasticities empirically tested by Brueckner and Spiller can be analogously applied not only to density on spokes but also to the associated parameters of aircraft type and average load factor. The following graph depicts this relationship:

      

Exhibit 8: A cost curve for the airline business (3)

      As a result, marginal cost falls by about 3.75% for every 10% increase in spoke traffic. The effect is stronger than the estimates by Caves, Christensen and Tretheway which would correspond to a fall in marginal cost of about 2%.

      In a follow-up study, Berry, Carnall and Spiller (1996) find no economies of density on shorter routes (less than 500 miles). In their model's specification, there are common costs across products because the same spoke can enter into the production of many demand side products. According to Banker and Johnston (1993), savings in marginal cost (av.$27) derive largely from concentrating flights through hubs. The savings associated with increasing average stage length are smaller (av.$2 per 1,000 capacity seat miles) and those associated with increasing density are the smallest (av.$1 per 1,000 capacity seat miles). Approximately 38% of the observations, however, exhibit net increases in costs with increases in density. The estimated net savings associated with all operations-based drivers average $33 and range from $10 to $132 per 1,000 capacity seat miles. If we were to aggregate effects in hubbing with those in density, decreasing marginal costs would seem to be confirmed.

      The question remains about the ambivalent effect of average stage length on average unit and occasionally observed increases in marginal costs. It is well documented that within the range of each aircraft, unit costs generally decrease with distance (Encaoua, Geroski and Jacquemin 1986), largely due to the character of planes as fixed costs and also due to the fixed cost character of take-off and landing fees. Banker and Johnston (1993) find this impact small, albeit favourable, when compared with the hubbing-density complex. Brueckner and Spiller (1994) even find increasing marginal cost with distance, according to their formula. ²¹  Decreasing marginal costs even at high densities could also not be found at short distances (Berry, Carnall and Spiller 1996). We conclude from this that the contribution of distance on decreasing marginal costs is small, and even strongly negative at distances below 500 miles, since it suffices to compensate otherwise strong economies of spoke density. This may also explain why Brueckner and Spiller had found a positive and slightly increasing relationship between overall marginal cost and the average distance in their sample.

      

Exhibit 9: A cost curve for the airline business (4)

      In order to find one general cost function for the airline sector, the counteracting slopes between density (aircraft, average load factor) and average stage length would need to be balanced and synthesized into one marginal cost curve. This can only be done by introducing a third axis (dimension) into our graph (for a three-dimensional graph, see Berry, Carnall and Spiller 1996, p. 41). Here, we choose to represent three marginal cost curves, each representing another stage length, in order to keep our cost curves in two dimensions. This representation also proved to correspond to the empirical values of table "Derivative of marginal cost with respect to density" ²²  (p. 38).

      

Exhibit 10: A cost curve for the airline business (5)

      On the other side, costs that remain unaffected by a change of aircraft type are greater than those that depend on the type of aircraft (Encaoua 1991). Encaoua regards average load factor the key driver for increasing returns to scale, to be more important than average aircraft size. Beyond the obvious trade-off of either choosing a larger, more efficient plane or keeping a smaller one and increasing its load factor, other economies "ex aircraft" are identified. The fact that such other costs could represent a larger part than aircraft-based costs would not matter if their returns were constant. At this point, it may be helpful to go back to Banker and Johnston's (1993) empirical analysis of cost drivers. Now we are only interested in cost drivers for indirect airline costs, which are not determined by aircraft type. From their table with the hypothesized cost driver relationships, we retain the following:

      

Exhibit 11: Cost drivers outside the operation of aircraft that affect indirect costs
Input category and measurement units	Volume-based drivers	Operations-based drivers
Aircraft and traffic servicing, labour hours	Passengers (+)	Density (+) Hub concentration (-) Hub domination (-)
Promotions and sales, labour hours	Passengers (+)	Density (+) Hub concentration (-) Hub domination (-)
General overhead, deflated costs	Total capacity seat miles (+)	Density (+) Hub concentration (-) Hub domination (-) Scale (+)
Group property and equipment, deflated costs	Total capacity seat miles (+)	Density (-) Hub concentration (+) Hub domination (+) Scale (+)

      The results for the four categories above show significantly increasing returns to scale with respect to hubbing. These returns were even stronger when an airline dominated an hub. Levine (1987, p. 473) sees hubs as a means of reducing the time during which traffic builds to long-term levels, thus reducing the firm-specific non- recoverable entry costs (this would not affect marginal cost). He sees these and other lower incremental costs from hub operations as close to zero. Apparently, hubbing exhibits additional economies of scale, probably due to (organizational) centralization effects, which are independent of aircraft economies of scale. Congestion effects may be the most likely constraints to these increasing returns (Kirby 1986). These economies of scale due to centralization may be more pertinent than the possible efficiencies of density and bigger aircraft.

      

Exhibit 12: A cost curve for the airline business (6)

      Only by accounting for both effects - economies due to aircraft and due to hubbing - can one general average unit cost curve be found, which represents the minimum efficient scale for all major cost functions of an airline business. A clear understanding of which parts of density and hubbing are reflected in aircraft economies of scale and which in centralisation economies is key for the following integration of both effects. In short, an airline would tend towards its own minimum efficient scale by trying to organize hub-and-spoke operations and trying to dominate them, though avoiding congestion at the airport. At the same time, it would choose the most efficient aircraft to handle the increased traffic, probably wide-bodied long-haul carriers, to direct them at high load factors to other hubs for maximum economic gain.

      

Exhibit 13: A cost curve for the airline business (7)

      The above graph suggests the combined impact of economies of scale with wide-bodied, more efficient aircraft and economies of scale due to hub dominance. We assume more efficient, wide-bodied jets to be more expensive to buy than smaller ones. That's why AC2 is initially higher than AC1. At a higher load factor, the new aircraft's efficiency starts to show, where average unit costs become smaller compared with a narrow-bodied plane. However, the big plane's minimum efficient scale is achieved at higher capacity seat miles than with the small one. That is, a hub has to funnel consistently more passengers towards bigger planes serving its spokes in order to achieve its minimum efficient scale (or even to undercut average unit costs of smaller planes). With the risk of over-concentration at hubs, the risk of traffic congestion rises. In the case of traffic congestion, we assume marginal costs for bigger planes to be significantly above those for smaller ones. This explains why the slopes at high centralization of traffic at hubs are steeper for bigger, albeit efficient, planes. We expect the shape of average cost lines to hold for distances between 600 and 1,500 miles approximately, following empirical tests by Berry, Carnall and Spiller (1996). With greater distances, the differences between low and high density are probably more important. Below 600 miles, however, we expect both AC₁ and AC₂ to flatten out significantly.

II.3.4 Economies on European routes

      We are interested in locating European airlines on these depicted cost curves. In particular, we will try to determine if certain airlines appear to be operating closer to their minimum efficient scale than others. In order to do so, we need to look at the average stage length for city pairs of European carriers and at their respective flight concentration from and to their hub airports. The latter is a strong indicator for density, in the sense of having the opportunity to employ larger aircraft with higher load factors, and also indicating the potential to centralize operations at the hub.

---

---

III.4.1 Active differentiation

---

---

---

---

---

---

III.4.2 Passive differentiation

      With active differentiation we sought to identify factors that might induce customers to pay premium prices due to their own preferences. We found that business passengers and economy passengers paying full economy fares are most likely to demand such services. But this demand driven approach is not necessarily what airlines, especially incumbents, tend to apply within their different flight classes. Exhibit 21 shows an example of price differentiation as exercised by an incumbent. It is noteworthy that a so-called Euro-budget fare (BB) is applied across all seasons, with a price level reflecting business class. This Euro-budget price is two and a half times the tariff of the lowest price offered on the same route by the same airline. For the designated economy fares, we identify clear price premiums for high season versus low season and shoulder season. It is however, arguable whether external factors, such as seasons, do really reflect customers preferences or rather present forces that leave passengers no other choice but to pay a price premium. We could argue that the passenger is still free to choose another season to fly with the same airline or to try to find a low-cost carrier with maybe less frequent flights, less service and perhaps more inconvenient flight connections. In addition, fully flexible tickets (designated by F) are sold with a premium over restricted tickets (R). ²⁹  This premium appears to be significantly higher during high season compared with low or even shoulder season. Flights on weekends (W) are also valued at a slight premium when compared with those during the working week (X). We can observe price premiums in the economy class in the following order of significance:

      (1) season, (2) ticket flexibility and (3) weekend. For simplicity, we have grouped the incumbent's lowest fares in one single group (S). The different Super Saver fares are rather insensitive to season, but can differ between weekdays and (not considered in this study) flight number. The range of these lowest fares in the given example was between GBP93 and GBP104. It is equally noteworthy that fares during low season are already on a level comparable with Super Saver fares.

      

Exhibit 21: BA pricing for a European city pair in November 1997

---

---

IV.3.1 Air transport capacity

      The most obvious point to start searching is transport capacity, usually measured in available ton kilometres or available seat kilometres: Is there any excess capacity among European former flag carriers? To answer this question, we have to account for the airline industry's idiosyncracies. Capacity can be reflected in the larger fleet of aircraft. It can also be reflected in a larger size of the average aircraft, given the same fleet. In addition, capacity would also be increased if given planes were used more frequently. While from the outside all these variables may appear to represent the capacity of an incumbent, their effects are very different.

      

Exhibit 30: Division of an incumbent's capacity

      Only new aircraft that are deployed on routes to old, or established, destinations can conceivably deter entry. Increasing capacity on new city pair routes is not viable to deter other entrants from entering older, already established routes. Any reflection on excess capacity, then, must focus on the evolution of capacity on city pairs, which was already established before liberalization started in 1992.

      As we have seen, airlines can make capacity choices for given city pairs along two dimensions: changing aircraft size and the frequency of flights to city pairs. Both dimensions impact on the average load factor: For a given demand, increasing flight frequency on a city pair lowers average load factor, as does the utilization of larger aircraft. The following table depicts the changes in average load factor of industry incumbents on European routes.

      

Exhibit 31: European traffic of national airlines
Airline	Passengers (million)	Average load factor	Passengers (million)	Average load factor	% change in no. of passengers
Aer Lingus	3.6	63%	3.9	69%	9%
Air France	25.7	57%	24.4	64%	-5%
Air Portugal	2.8	65%	3.2	63%	17%
Alitalia	17.1	61%	20.1	63%	18%
Austrian	2.2	53%	2.2	50%	-3%
BA	17.4	66%	21.0	69%	20%
Finnair	2.9	47%	4.5	60%	51%
Iberia	19.3	65%	17.7	64%	-9%
KLM	5.2	57%	6.9	67%	33%
Lufthansa	22.0	54%	26.8	58%	21%
Luxair	0.5	54%	0.6	53%	39%
Olympic	4.9	66%	5.7	66%	16%
Sabena	2.4	46%	4.3	49%	75%
SAS	13.6	58%	18.6	57%	37%
TOTAL	139.7	59%	159.7	62%	14%

(Source: Civil Aviation Authority 1998, p. 97)

      Indeed, the above table strongly suggests excess capacity among major flag carriers at the beginning of the liberalization period. This excess capacity was utilized increasingly, while output increased in most cases. Among the major European carriers, excess capacity was particularly strong with Lufthansa and Air France, less so with BA. In comparison to the European mean, Lufthansa still maintained some - albeit reduced - degree of excess capacity in 1996. Without being able to pinpoint the source of excess capacity now, we can see its relevance to what was said by Spence earlier on excess capacity and lowering marginal costs, although the overall equilibrium output and average load factors differ significantly among incumbents. The fact that double digit passenger growth was reflected in only a small increase in the seat factor shows that airlines were keen on keeping up investment in capacity, instead of using existing capacity more efficiently.

      The question has to be asked, whether such excess capacity does in fact represent commitment, provide for lower marginal costs of the incumbent and thus deter new entrants?

---

---

---

---

IV.3.2 The commitment in frequent flyer programmes

      Frequent flyer programmes (FFP) were introduced as a competitive tool by airlines to win frequent travellers by providing rewards for flights taken on a particular airline or group of airlines. These rewards usually take the form of points that count towards free flights, ticket upgrades, leisure travel and holidays, among other benefits. Many airlines view FFPs as defensive, because they were used for maintaining their existing customer base rather than gaining additional customers. FFPs offer two advantages for the airlines. First, they provide a useful marketing tool, supposedly better matching customer requirements on the basis of new marketing strategies. And second, since they are constrained by available capacity, they enable airlines to improve their load factors. The major drawbacks are the costs associated with administering the programme and providing the rewards. The airlines in the EU had introduced FFPs cautiously and with a large variation in the levels of rewards.

      By 1996, more airlines were awarding points the lowest fares and leisure travellers as well, having somewhat extended their prior focus on business or fully flexible tickets. Hence, airlines were keen to fill up economy seats of the aircraft on a marginal cost basis. There were also indications of competition on FFPs, with airlines increasing rewards and in some cases doubling them. Finally, business travellers seemed to be shifting their preferences for FFP rewards from earning free miles to being at the head of the waiting list for overbooked flights.

      Though the number of members in the major EU airlines' frequent flyer programmes was still tiny compared with US numbers, membership was rising. More than 85% of business travellers belonged to a frequent flyer scheme in 1996.

      But despite their apparent success, it was doubtful whether FFPs represented incumbents' commitment to deter new entrants. One key point of FFPs was to lock in customers and to increase an aircraft's load factor, thus allowing for lower costs due to less residual capacity available. To this extent, the potential for higher load factors and for larger, more efficient aircraft could imply lower marginal costs for the incumbent. Furthermore, European incumbents with their united FFPs (such as Lufthansa, SAS and Austrian) should then have a significant advantage. This, however, is not the case.

      For one, the administration of FFPs is expensive. Each FFP passenger has his or her own account, with the relevant administration, mailing and dedicated customer service at the airports to be taken into accounted. This increases marginal costs. Also, the low-cost approach of new entrants, especially with savings in personnel and service-related costs, is much greater than any effect of FFPs could possibly be. ⁴⁵  This holds particularly true for economy or discount frequent flyers, where administration costs for these programmes become an important factor in overall costs. Here, the incumbent's marginal costs will probably be higher than those of a new entrant. For the business frequent flyer, as already mentioned, more of them prefer to use their bonus miles for upgrades or non-flight perks (vouchers for hotels or car rentals, etc.) rather than for more flights. Again, the theory of diminishing marginal costs due to FFPs does not hold. Moreover, in order to be committed to FFPs, the incumbent must not be able to sell them. The principle of "burnt bridges", however, does not apply. Nothing prevents an incumbent firm either stopping FFPs or even selling them if they prove to be less profitable than expected. The fact that FFPs are already diluted in their flight-specific dedication shows in the integration of not specifically flight-related services (car rentals, hotels, tourist tours, duty-free vouchers, etc.). In addition, it may even be lucrative for the incumbent to take a new entrant into its FFP, if this entrant adds value to it by enlarging the route network. All these considerations contradict a classification of FFPs as sunk costs.

IV.3.3 Computer reservation systems

      The real sunk-cost content of most IT-management systems - be it capacity management, yield and revenue management or FFPs - lay in the airline's computer reservation systems (CRS). Based on their fixed cost character, CRS provided significant potential for earnings with revenues growth. These CRS could be owned by a single airline (Sabre, for example, belonged to American Airlines). In Europe, CRS were usually held by consortia of airlines. Amadeus, a system operated jointly by Lufthansa, Air France and Iberia (each holding a 29.2% stake, with Continental Airlines holding 12.4%) had linked up in 1992 with two other systems operated by American and Asian airlines. ⁴⁶  Worldwide, Amadeus counted 35 partner and owner airlines. ⁴⁷  Another European CRS was Galileo International, whose European partner airlines were Aer Lingus, Alitalia, Austrian, BA, KLM, Olympic, Swissair and Air Portugal. The most important US partners were US Airways and United. All Galileo partner airlines also held stock in the company, although to a much lesser extent than the owners of Amadeus. ⁴⁸  On a worldwide scale, Amadeus had 44,148 travel agencies linked up, while Galileo had 38,400 agencies served with its CRS. ⁴⁹ 

      In Rezendes (1988), the anti-competitive effects of airline CRS were examined. For one, such effects were found with incremental revenues. These incremental revenues arose initially because of biased display of flights on travel agents' video screens. Such a bias was somewhat tempered from the beginning in Europe, since a code of conduct for CRS was adopted as a regulation by the European Community in July 1989 (Civil Aviation Authority 1993, p. 8). As the inherent incentives persist for owners of a CRS to favour their own flights directly or indirectly, and as the technology of CRS continues to develop rapidly, these codes of conduct will need to be revised continuously to avoid such incremental revenues. For the US, a DOT report found that even after the screen bias rule had gone into effect, CRS-vendor airlines increased revenues by 9% to 15% (according to CRS vendors' data) and by 12% to 40% (according to DOT's own analyses) over what they would have been without of CRS ownership. ⁵⁰  In Europe, the aforementioned biases were not longer an issue in reports on the European airline industry. This may also be partly due to the adoption of the no host technology by both Amadeus and Galileo in the early 1990s. ⁵¹  This technology provided the agent with reliable and unbiased information, also on smaller or non-owner airlines. As we are looking for an analogy with excess capacity, we are interested in identifying diminished marginal costs after entry and not in increased marginal revenues vis-à-vis an entrant.

      The other potential for anti-competitive effects came from booking fees linked to CRS. Such booking fees, if they exceeded costs, could transfer cash flows made by one airline to another airline. Competing airlines, who pay booking fees, have little alternative but to pay those fees if they wish to remain competitive in the air travel business. ⁵²  The option of being listed in CRS, but not paying booking fees and thus being excluded from CRS bookings is not considered viable. The Rezendes (1988) report states: "...An airline competing with other airlines in major routes, particularly a new entrant, could not use this strategy..."(p. 11). In Europe, probably the most notable exception to this rule is easyJet, where flights cannot be booked over CRS and reservations are made directly with the airline.

      

Exhibit 35: The marginal cost benefits of owning a CRS

      The above graph depends critically on several parameters. The difference in marginal costs between the CRS owner and non-owner airline is a function of booking fees exceeding unit costs. Without any unit profits, there is no cross subsidy possible from non-owners to CRS owner airlines. With per unit losses, the presented relationship may in fact become inverted. Also, the more non-owner airlines and their networks participate in another airline's CRS, the more of such booking fee unit profits are likely to be distributed to the CRS owner airlines. We could have displayed this profit transfer relationship as a constant MC-curve for the owner airline or an increasing marginal profit function originating from the non-owners (implying that unit profits are likely to increase with a wider network of airlines joined into it). This, however, would have strained our requirement for marginal costs too much. Instead, we preferred to assume that such unit profits are transferred to the CRS owner airline, where they will be strategically used to diminish marginal costs on the owner's routes in order to deter new entrants. Any airline wishing to establish its own reservation and sales system can theoretically escape from this mechanism. A direct sales approach may even save between 7% and 15% of the regular travel agent's commission on tickets sold (see below). This, however, is likely to work only for low-cost entrants on specific low- to middle- density city pair routes.

      

Exhibit 36: Distribution of owner and non-owner potential passengers for booking on CRS
Airline's stake in corresponding CRS	Amadeus	Galileo*	Non-owner incumbents	Non-incumbents**
Aer Lingus (0.1%)		3,90
Air France (29.2%)	24,40
Air Portugal (0.1%)		3,20
Alitalia (8.7%)		20,10
Austrian (0.1%)		2,20
BA (14.7%)		21,00
Finnair			4,50
Iberia (29.2%)	17,70
KLM (12.1%)		6,90
Lufthansa (29.2%)	26,80
Luxair			0,60
Olympic (1%)		5,70
Sabena			4,30
SAS			18,60
TOTAL (m passengers)	68,9	63	28	80
% of total passengers	28,72%	26,26%	11,67%	33,35%
* Galileo numbers does not include Swissair figures. Including Swissair, Galileo's total would approximately equal Amadeus' total of passengers moved by CRS owner airlines. ** There were no exact figures for total non-incumbent traffic in Europe. The given figure is an approximation from the CAP685.

      If we assume that both CRS in Europe have similar market presence (and that cross bookings for owner airlines on another CRS would be mutually balanced), we can observe a residual 45% of European scheduled passenger traffic, which may significantly alter the profits generated by the competing CRS. However, we have no reason to believe that these smaller airlines would favour one CRS over the other. The salient feature of entry-deterring effects of CRS in Europe then would be their differing ownership stakes. Iberia, Lufthansa and Air France are likely to lower their marginal costs significantly, due to profit transfers from other airlines' paid booking fees. It is interesting to note that Swissair has both a relatively high participation in Galileo (13.2%) and transports relatively few passengers in Europe. The combination of these factors is likely to reduce Swissair's marginal costs even more than Lufthansa's much higher stake in Amadeus reduces its marginal costs. KLM's situation is comparable to Swissair's.

      The interest in selling its ownership and buying another one for an airline in the competing CRS is not very pronounced. A shareholder would lose its stake, and it would be unlikely that as a new comer to an established system it could get the same percentage share as it had before with the former CRS. It is also likely that clauses within the CRS agreement would make it difficult and expensive for shareholding incumbents to bail out. We conclude that ownership stakes in CRS do present commitment. The higher the stake and the bigger the underlying route structure, the more the incumbent will be apt to deter new entrants, in analogy with Spence's kinked cost curve.

IV.3.4 Alliances: Code sharing and franchising

      The most rapid transformation of the industry structure was happening through alliances, both within the EU and around the globe. These alliances took a number of forms, from equity investment to code-sharing agreements and franchising. The latter two will be discussed in more detail

      Code sharing

      Code sharing was a form of cooperation between airlines. It let airlines place their flight codes on flights that were actually operated by other airlines. Conceptually, two forms of code sharing were distinguished. Interlining involved placing an airline's code on a connecting service that was in reality operated by another airline. This meant that an airline was able to offer to the customer a wider network in its own name than it actually operated. The other form of code sharing went further and adjusted flight operations of both airlines involved. With shared codes, formerly competing airlines might have chosen to operate only one flight for a certain city pair, where before they had one flight each. ⁵³ 

      

Exhibit 37: National carrier code shares with other EU airlines in 1992 and 1997
Airline	1992	1997
Aer Lingus	Sabena	Sabena, Finnair, Hamburg Airlines, KLM
Air France	Luxair, Austrian, Sabena, Tyrolean	Luxair, Alitalia, Brit Air, CityJet, Eurowings, Finnair, Flandre Air, JEA, Maersk Air, Regional Airlines
Air Portugal	LAR Transregional	Aerocondor, Hamburg Airlines, Portugalia, SATA Air Acores
Alitalia	Austrian	Air France, Alpi Eagles, Azzurra Air, Brit Air, British Midland, Eurofly, Finnair, Luxair, Maersk Air, Meridiana, Minerva
Austrian	Finnair, Air France, Alitalia, KLM, Olympic, SAS	Finnair, Iberia, Lauda Air, Lufthansa, Sabena, Tyrolean
British Airways	CityFlyer, Deutsche BA, Brymon	CityFlyer, Deutsche BA, Brymon, Air Liberté, British Regional, GB Airways, Loganair, Maersk Air (UK), Sun-Air
Finnair	Austrian, Lufthansa	Austrian, Aer Lingus, Air France, Alitalia, Braathens, Maersk Air, Sabena
Iberia		Air Nostrum, Austrian, British Midland, Lauda Air, Luxair, Regional Airlines
KLM	Tyrolean, Austrian	Tyrolean, Aer Lingus, Air Exel, Air UK, Eurowings, Maersk Air, Regional Airlines, Transavia
Lufthansa	Cimber Air, Lauda Air, Tyrolean, EuroBerlin, Finnair	Cimber Air, Lauda Air, Tyrolean, Air Dolomiti, Air Littoral, Augsburg, Austrian, Britrish Midland, Business Air, Contact Air, Luxair, Rheintalflug Seewald, SAS, VLM
Luxair	Air France	Air France, Alitalia, Conseta, Iberia, Lufthansa
Olympic	Austrian
Sabena	Air Lingus, Maersk Air, Air France, British Midland, Brymon	Aer Lingus, Maersk Air, Austrian Finnair, Regional Airlines, Tyrolean, Virgin Express, VLM, Sobelair
SAS	Austrian	British Midland, Cimber Air, Falcon Aviation, Lufthansa, Regional Airlines, Skyways, Spanair Wideroe's

      (source: Official Airline Guide 1993 and 1998)

      Besides the overall growth in code sharing, this table also shows the volatility of such partnerships. Most of the more important partnerships of 1992 had been replaced by new, more promising ones in 1997 (see partners for Austrian or Air France, etc.). Unfortunately, we cannot tell from this table which kind of code shares prevailed: simple interlining or the more collusive adjustment of mutual flight operations.

      At this point, we shall continue with an approximation, albeit a realistic one: Code shares between incumbents tended to fall overwhelmingly into the second category. If Lufthansa, for instance, shared codes with SAS, we know that both networks had covered the most important routes for Europe from their own hubs independently before. Routes from Sweden, Norway or Denmark to Germany (and vice versa) were based on former bilateral agreements. We can infer from this historical evidence that these newly formed code shares do not intend to exploit new routes that have previously never been served by one of the carriers. Nor is it likely that the routes concerned would be low-density routes (within Europe), which were unprofitable for an incumbent to exploit by himself with his regular flight operations.

      

Exhibit 38: European incumbents and their code-sharing incumbent partners
	Incumbent partners in 1997	Code-sharing potential*
Aer Lingus	Sabena, Finnair, KLM	Low
Air France	Luxair, Alitalia, Finnair	Average
Air Portugal		None
Alitalia	Air France, Finnair, Luxair	Average
Austrian	Finnair, Iberia, Lufthansa, Sabena	High
British Airways		None
Finnair	Austrian, Aer Lingus, Air France, Alitalia, Sabena	High
Iberia	Austrian, Luxair	Low
KLM	Aer Lingus	Low
Lufthansa	Luxair, SAS, Austrian	Average
Luxair	Air France, Alitalia, Iberia, Lufthansa	High
Olympic		None
Sabena	Aer Lingus, Austrian, Finnair	Low
SAS	Lufthansa	Average
* This rating was directly related to the partner's network size in Europe. The bigger the network, the more route operations could be adjusted between both incumbents. Another factor was the density of key routes between the partners: Dublin-Amsterdam (AL/KLM) had less potential for adjustment than Frankfurt-Helsinki (LH/SAS).

(Derived from Exhibit 37)

      This rating was directly related to the partner's network size in Europe. The bigger the network, the more route operations could be adjusted between both incumbents. Another factor was the density of key routes between the partners: Dublin-Amsterdam (AL/KLM) had less potential for adjustment than Frankfurt-Helsinki (LH/SAS).

      The question at this point is whether such shared codes (not interlining) represented commitment of the incumbent. At first glance the answer would appear to be negative, due to the short duration of most such alliances. Indeed, those airlines that appeared to extract the greatest advantage were relatively small players (see Luxair, Austrian or Finnair). Looking at the table we find the most important European players (see BA, LH, Air France or KLM) were not particularly disposed to these kinds of alliances. The market values intra-European routes, especially high-density ones, and, a priori, nothing prevents an incumbent from changing partners if he believes this to be more advantageous. The definition of sunk investment does not fit with this type of code sharing. But maybe there are decreasing marginal costs as a function of the number of codes shared? Can shared codes present a form of excess capacity? This would imply that airlines that shared many codes would present lower marginal costs than those that held fewer code shares. This is not the case: The potential for reducing costs by combining, and consequently by eliminating, flights on high density-routes aims to reduce excess capacity, not create it.

      Interlining is another dominant form of code sharing. In Europe, it most often took the form of regional alliances and usually involved a national airline which formed a partnership with a regional carrier. ⁵⁴  The regional airline was presumed to have a more appropriate fleet as well as greater expertise in serving the smaller markets, consisting of low density routes. The proliferation of such partnerships was impressive. ⁵⁵  and Lufthansa was said to hold the greatest number of regional alliances in 1997.

      

Exhibit 39: Lufthansa's regional partners in 1997 56
	Home country	Number of code shares
Air Littoral	France	26
Air Dolomiti	Italy	12
Rheintalflug Seewald	Austria	3
Tyrolean	Austria	3
VLM	Belgium	3

      (Source: Civil Aviation Authority 1998, p. 154)

      The system of interlining maintained access for incumbents to routes, that the incumbent could not economically serve by himself (due to his cost structure and operational constraints). In this respect, marginal costs for an incumbent serving such routes with his own operations would not be sufficiently low to undercut an entrant's average costs. However, in order to determine, whether these interlining arrangements presented a viable commitment to deter new entrants, we need to compare the average cost curves of the regional, interlined airline with those of a potential new entrant. Regional airlines do not reason on a marginal cost basis, as they seek to break even and to compete head to head against low-cost entrants. A priori, we cannot say whether the cost structure of the regional code-share partner is lower than that of a new entrant. But we know that incumbents use these agreements to feed a certain number of passengers on to these code-shared flights. ⁵⁷  Regional flights from or to airports where the code-sharing incumbent maintains a dominant share are especially powerful in providing a minimum average load factor for the partner airline. This enables the regional partner to plan with lower residual capacity (higher load factor) or even to employ larger aircraft than the new entrant, hence exploiting advantages in efficiency when operating larger planes. Due to the regional partner's dependence on the incumbent's steady provision of passengers, and due to the geographical dependence on an incumbent with hub dominance, it is very unlikely that these partners will leave for other incumbents. For the same reasons it is unlikely that the incumbents would transfer shared codes to other incumbents. Their value is simply too closely linked to the incumbent's hub dominance to be worth selling. Interlining then must be considered as commitment to deter entry on these regional routes. Employing "Jumbolinos" or Canadair jets on such routes could be interpreted as a form of excess capacity when new entrants would most likely serve such routes with turbofan-powered smaller aircraft. ⁵⁸  However, as we have stated, the lever for such cost advantages of the interlined partner airline is linked to the incumbent's hub dominance. That is, interlining arrangements for undominated airports, especially for seventh freedom routes, can hardly present such commitment and are unlikely to deter new entrants effectively.

      

Exhibit 40: Average costs (per passenger) for regional carriers with interlining

      Franchising

      The concept of code sharing for interlining became more fully developed with the franchising approach. The basic rationale, however, remained the same: to maintain a presence on such routes (mostly low-density ones) where the franchisee was more suited to serve than the incumbent with his high-cost operations. The incumbent's interest was to transfer such routes, which he had previously operated by himself, to the franchisee. Usually, the partner operated under the incumbent's brand and this allowed the incumbent to maintain his own profile on such routes (Flight International, 7-13 May 1997). Unlike most interlining arrangements, the franchisee could easily operate beyond an only regional scale. BA's partners, like CityFlyer, Brymon or GB Airways, served the domestic market and even some international routes. Air France's partners, like JEA or Brit Air, clearly went beyond a merely regional level with their international routes between France and the UK.

      Although data on franchising between European incumbents are scarce, some observations in the case of BA can be made.

      

Exhibit 41: Scheduled traffic before and after franchise arrangements
	Total			Domestic
	1992	1996	Change	1992	1996	Change
Brymon	293,439	461,158	57%	190,119	350,898	85%
Maersk	311,671	443,100	42%	116,643	154,760	33%
CityFlyer	234,818	849,145	262%	145,815	330,659	127%
Loganair	590,727	251,214	-57%	573,029	244,186	-57%
British Regional - Manx	643,932	1,589,154	147%	524,520	1,346,835	157%
GB Airways	233,837	484,400	107%	-	-	-
Total of BA franchisees	2,308,424	4,078,171	77%	1,550,126	2,427,338	57%
BA	27,387,460	32,340,654	18%	5,765,337	6,132,523	6%
Other UK carriers	8,546,604	14,710,257	72%	4,379,964	6,425,249	47%

      (source: Civil Aviation Authority)

      It was assumed that these significant increases in traffic were largely due to BA's feed traffic in connection with its airport hubs (Heathrow and Gatwick). About 500,000 passengers a year were estimated to connect between BA and its franchisees (Flight International, 25-31 March 1998). A 1996 Civil Aviation Authority origin/destination survey compared the level of connecting traffic on a sample of routes operated by a franchisee and by an airline independent of BA. In all cases the franchisees had a higher proportion of traffic connecting on to BA services than their rivals and, generally, a higher overall connecting proportion, even when the fanchisee was based at Gatwick and its rival at Heathrow.

      We can infer that the connecting traffic from or to incumbents had a decreasing effect on the franchisee's average costs, similar to the effects of interlining. We are tempted to apply the same average cost curve for franchisees as we have done for the regional partners with interlining agreements. However, these franchise partners quite often serve other airports, where the incumbent has only low local presence and hence little traffic to feed on to the partner's connections. Especially for the fifth and seventh freedom routes, there is no reason why an independent carrier should have higher average costs than a franchised one. The franchise agreement can also be considered as relatively durable in its nature. In any case, franchise agreements are more durable than most large-scale intercontinental alliances between major incumbents' networks (say KLM and Northwest). The reasons for such durability are several: For one, franchise contracts cannot be resold. The franchisee's interest in getting connecting services from the incumbent prevents it from looking for other partners with geographically less interesting hub airports. On the incumbent's side, his potential gains in finding another new entrant that might serve the same routes even less expensively while maintaining a minimum standard of quality, are remote. We can conclude that franchise agreements represent an incumbent's commitment, as long as they involve a significant amount of the incumbent's connecting service.

IV.3.5 Advertising

      Schmalensee (1983) shows how advertising is to be considered differently from sunk costs. In particular, over-investment in advertising before entry is not optimal to deter entrants: "Despite the obvious resemblance to... the use of investment in production capacity to deter entry, here the incumbent monopolist never finds it optimal to advertise more if entry is possible than if it is not... These results and others presented make plain the dangers of analysing investments in advertising largely by analogy to investments in production capacity. They also suggest that the strategic implications of investment in advertising are highly sensitive to the effects of advertising on buyer behaviour and to the nature of post-entry equilibrium..."(p. 636). The issue of advertising is more profoundly treated in chapter III, which deals with product differentiation.

IV.4.1 Entry deterrence by committing to flight frequency

      The incumbent can commit to serving the city pair with a very high frequency, such as a shuttle service. As shown, high flight frequency presents commitment, lowering marginal costs in the case of entry. Such commitment is independent of efficiency gains due to operating bigger aircraft. With this logic, deterrence would be successful if such high frequency lowered marginal costs (when operating below full capacity after entry), so that prices would drop enough to prevent the entrant from achieving a sufficiently high load factor to operate at least one flight profitably on the given route. In analogy with Dixit, the incumbent would operate high frequency flights (with smaller aircraft) at lower average load factors and at comfortable prices before entry occurs. Excess capacity in both seats and frequency would allow the incumbent to jack up prices on special occasions, such as peak periods, holidays, etc. Once entry happens, pricing can be done on a marginal cost basis, undercutting the entrant's average unit costs and thus preventing him from breaking even. In addition, the incumbent may even maintain his full service range, maintaining premium passenger service and thus finding potential for subsidizing losses from marginal cost pricing at the bottom end of the passenger spectrum. A low-cost entrant is not likely to propose such a full range of services to his customers. Therefore, in order to fill up his entire capacity, he cannot discriminate on price, but will need to counter the incumbent's lowest price, which will be based on his marginal costs.

      The empirical part of this thesis (see section VII) will focus on the significance of flight frequency as a barrier to entry and proceed to measure its impact on ticket prices. An empirical assessment of price and output changes with respect to entry or exit of competitors into established routes, as suggested by the Dixit model, shall not be conducted. Although the Dixit model may apply for sunk costs, other types of barriers to entry may have the same or even a greater impact. Section VI presents a framework that is more universally applicable for our purpose of identifying all relevant barriers to entry and measuring their impact on prices, independently of entry/ exit dynamics.

IV.4.2 Consequences of strategic behaviour with CRS

      The issue of cross-subsidies arising from ownership in CRS is not easy to apply to the Dixit rationale. If we assume that profits derived from booking fee profits are used strategically to diminish existing marginal costs for the incumbent's own operations, we can only allocate these lowered (or rather subsidized) marginal costs to the entire route network of the CRS-owner incumbent. Evidently, the smaller the incumbent's route system for a given ownership stake, the greater the cross-subsidization effect on a per city pair basis.

      Applying Dixit's graph, we can discern two main effects:

• A shift of the incumbent's reaction function to the right. This depends on the incumbent's ownership stake in the CR system, the size of his own route structure and, of course, the extent to which the CR system produces profits.
• Whether the entrant will join a CRS or not. This determines the entrant's own marginal costs, and thus may lower his own Cournot-reaction function.

      The hypothetical case that extremely expensive booking fees might alter the demand for tickets and may change the slope of the depicted reaction functions can be ignored. This yields an adapted Dixit graph:

      

Exhibit 42: An analogy to Dixit's model with commitment in Computer Reservation Systems

      This model shows two extremes of Cournot equilibria: A and F. With A as post-entry equilibrium, the lowest degree of deterrence appears for an entrant. First, the incumbent does not obtain significant cross-subsidies from CR systems, because his ownership stake is relatively low in comparison with his overall route network and the degree to which he uses the CR system himself. Second, the entrant's own marginal costs do not increase with the CR system if the entrant bypasses such systems. The other extreme, at F provides maximum deterrence to new entrants because an entrant subscribed to a CR system would compete against an incumbent benefiting from CR system due to his high ownership stake relative to the size of his own route structure (and his own bookings). Under these circumstances (at equilibrium F), CR systems develop their highest degree of deterrence. In other words, if an entrant were to chose a city pair to compete against an incumbent, it should pick, ceteris paribus, those routes with competition from BA or Alitalia, without having to do bookings on computer reservation systems. Between the extremes shown, multiple equilibria (B, C, D, E) are all possible, with their differing degrees of deterrence.

      It is, however, not impossible, that entry may be successful at point F: If booking fees were modest, and cross-subsidization from CRS low (due to low profitability of such systems, especially in the first years of operation), the differential of an entrant's and an incumbent's marginal costs may become very small. If CRS were to become unprofitable, due to the emergence of alternative channels of distribution, or, due to competition from other CR systems, then indeed, deterrence might collapse altogether and entrants might actually benefit from competing on costs by advertising their lower costs on incumbents' CRS.

      The question whether computerized reservation systems act in the way that the analogy of the Dixit model would suggest, or in some other way that might reflect a barrier to entry, shall be empirically assessed in chapter VII. If CRS are found to be a significant factor, its impact on price changes will be measured.

IV.4.3 Consequences of interlining behaviour (post-entry)

      Applying Dixit's model to strategic behaviour with interlining seems obvious. The action to cover low-density city pairs from dominant hub positions by applying the interline partner's lower cost structure can be interpreted as the post-entry reaction along models with excess capacity. In fact, interlining agreements appear even more subtle than the "classical" excess capacity approach of entry deterrence: Instead of having to invest in excess capacity, and thus incurring higher average costs, and taking the risk of under-utilization of this spare capacity, interlining leaves this risk entirely to the interling partner. Also, instead of threatening to use this excess capacity, the interlining partner actually already operates on these routes with adapted operations. The effects of entry deterrence due to excess capacity seem to be obtained, while limiting sunk costs for the incumbent airline.

      

Exhibit 43: Kinked cost curve for interlining agreements

      The incumbent's capacity choice is based upon service on medium- and high-density routes. For lower-density routes, his operations are less adapted and marginal costs are not sufficently low (see v₁ compared with v₂+s₂, which are average unit costs for a regional airline). Operating at lower-densitiy city pairs than the incumbent can still be profitable for the regional partner. If entry occurs on this regional level, the interlined regional airline can still lower prices to its idiosyncratic marginal costs (v₂) to drive out or deter entrants. In addition, feeder service with the interlined partner incumbent will provide, ceteris paribus, more traffic for the interlined regional airline than for any entrants.

      The constraints of such deterrence by interlining are, however, clear: The incumbent needs to "recruit" an existing airline for this partnership. If several low-cost airlines are already competing on a given city pair, the advantage of such interlining agreements may not be substantial enough to force the other airlines to exit the route, as they in turn may use pricing based on marginal costs. The scenario for entry deterrence with interlining then must be: The incumbent code shares with the only low-cost operator on given routes (which feed into the incumbent's hubs), or the incumbent sets up his own low-cost company as the first operator serving these routes (also by eventually withdrawing existing route service by the incumbent's own operations).

      In fact, the incumbent is not trying to undercut low-cost competitors' prices on regional routes, but to capture as much existing demand as possible along the lines of Cournot competition by attaining a similarly competitive cost structure as potential entrants. It is questionable whether such a Cournot share of a low-density market, plus additional demand for feeder services into the incumbent's hub, will suffice to deter or drive out other competitors. If this Cournot share plus additional demand due to feeder advantages (shared codes, schedule alignments, connecting services, etc.) allow the interlined carrier to use bigger, more efficient aircraft at equally high load factors as the non-interlined competitors, then average costs can decrease further than those of the regional competitors. In such an optimal case, non-interlined competitors could become unprofitable and eventually be driven out of the market.

IV.4.4 Consequences of strategic behaviour with franchising

      The effects of post-entry Cournot competition with franchising are similar to what has already been described for interlining arrangements. In fact, the incumbent tries to lower his cost structure to a level comparable with low-cost operators by finding franchisees, as a first step. As a second step, a uniform branding and assured quality control by the incumbent franchisor, as well as a high percentage of connecting services between the incumbent franchisor and the franchisee, may allow for higher load factors than would be the case with interlining partners. Also, due to the usually longer flight distances - as compared with mostly regional interline arrangements - larger, more efficient aircraft for the franchisee can be employed. The impact on average costs could be depicted as follows:

      

Exhibit 44: Average cost effects of franchising on low-density routes

      In analogy to Dixit's post-entry Cournot competition with excess capacity, we would derive the following post-entry equilibria under franchising agreements:

      

Exhibit 45: Cournot equilibria with franchising partners

      As shown above, the incumbent can influence the low-cost competitor's post-entry output by working with franchisees. Although this does not necessarily mean that entrants would be deterred, or that low-cost competitors would be driven out of the market, their output would be reduced. We do not assume that at point B the competitor would be sufficiently harmed to incur losses. The reason is obvious: Ceteris paribus, the franchisee himself would then incur losses, too. The real question is whether point C lies far enough to the right on R^ent to make competition unprofitable. Again, the decrease in average costs due to more efficient, usually bigger, aircraft at equally high load factors as the entrant is crucial. It would allow for lower pricing post-entry, but could only be maintained with advantages in demand for the franchisee. Otherwise, the larger, more efficient plane (which is not a sunk cost) might not obtain a sufficiently large load factor to operate profitably and the entrant might serve his residual demand with a smaller plane. To obtain such a surplus in demand, compared with an independent competitor, the incumbent must provide a powerful brand with the notion of quality, as well as connecting services through hubs.

      The empirical validation of the above predictions for strategic behaviour based on interlining and franchising agreements, however, is outside the scope of the research design of this thesis: our sample of 35 city pairs contains mostly high and medium- density routes. On such routes, interlining and franchising are extremely rare - too rare to be tested statistically. In these cases, code sharing with other incumbents was more prevalent (the impact of such code-sharing on prices was empirically tested, although we had not confirmed commitment value of such code sharing, see IV.3.4).

      Beyond the practical aspects of sampling the "right" city pairs, empirical testing of such strategic behaviour should focus on intertemporal aspects of price changes before and after entry of competitors on such low-density routes. Our research design, however, uses contestability theory that allows us to empirically test for entry barriers and to measure their impact on prices, independently of actual entry. At last, interlining and franchising partners cannot necessarily be considered a full part of the incumbent's network. Even if their efficient cost structure may deter other low-cost entrants from competing on certain routes, it would not necessarily imply that the incumbent would reap the financial rewards of such behaviour.

V.2.1 Learning by doing

      This concept is associated with the notion of lowering production costs by increasing experience on the job (Lieberman 1984). This notion can be considered as comparable to Bain's access to superior management and working techniques. Such experience-related economies can be the origin of cost asymmetries favouring an incumbent over new entrants. The interesting aspect of such experience-related economies is that the incumbent can maintain this cost asymmetry by simply imitating the entrant's actions. Production decisions are the means to develop such learning effects (increase output, extend life cycles of production, etc.). This, however, implies that learning by doing will not diffuse easily to competitors.

      There is no empirical study to this day that has suggested learning effects in the airline industry subject to higher output, capacity or increase in flights, etc. We assume the business of airlines is shaped by routines that are easily replicated and transferable to other airlines, be it incumbents or entrants. Most airline employees are rather narrowly skilled with highly specific tasks. These job characteristics suggest a rather flat learning curve.

V.2.2 Raising a rival's costs

      Along the lines of Bain's second condition for the absence of absolute cost advantages "...(b) that the entry of an added firm should have no perceptible effect on the going level of any factor price", Salop and Scheffman (1986) argue that behaviour intended to increase industry costs can benefit established firms (despite increasing their own costs) because it causes rival firms to reduce their output. Efforts to increase union wages would figure as part of such behaviour.

      Williamson (1968) examines the Pennington case in the coal industry from such a perspective: Given structural differences in the employment of labour versus capital between big firms and smaller ones, the "manipulation" of wage rates might favour the firm with a low labour ratio by increasing the average cost curve for the high-labour content firm disproportionately. Although this logic was developed from the perspective of driving competitors out of a market, the same rationale may be applied to deter entrants by incumbents whose output is not determined so much by costs involving labour.

      Provided that the incumbent had a lower labour factor in production than the entrant, he would aim to gain output and decrease the entrant's output. This would prevent the entrant from investing in capital to replace the high labour content in his production function and would also lower the entrant's profits. The incumbent would at least maintain his pre-entry output choice later. This can be encompassed by pricing below the entrant's average costs.

V.2.3 Long-term contracts with customers

      Such strategies are based on agreements between established firms and/or customers that limit access to particular markets by actual or potential competitors. ⁶⁰  They are aimed at limiting the size of a rival's market and increasing the rival firm's average costs if the rival's production technology has increasing returns to scale. One particular case of such contracts is described by Aghion and Bolton (1987, p. 389) with their "optimal contracts between the buyer and the incumbent seller". In this model, an established firm can enter into long-term contracts with customers that specify prices conditional on the entry of a competitor. Although Geroski, Gilbert and Jacquemin (1990, p. 46) consider such strategies are related to absolute cost advantages, we do not follow this analogy for two reasons: (1) demand-reducing strategies are not cost-increasing strategies: cause and effect need to be treated separately; and (2) as Gilbert says, these strategies are based on the assumption of increasing returns to scale for the entrant. Only then will the entrant experience higher average costs than the incumbent. However, as Bain has defined, absolute cost advantages provide lower average costs independent of the scale of output. This second argument reinforces the first one.

      Since such strategies may be relevant for the discussion of frequent flyer programmes, we have chosen - even though they may not be considered absolute cost advantages - to mention them.

V.3.1 Airport slots

---

---

---

V.3.2 High wage rates due to unionization

      The view of Williamson (1963) on wage rates as a competitive tool against new entrants needs to be discussed.

      It is a given that labour-related cost functions for new entrants are largely inferior to those of incumbents. New entrants generally have cost advantages of at least 20% over incumbents, especially for passenger services, passenger embarkation and crew costs, as well as administration. If incumbents could coerce new entrants to apply the same high-pay policies as themselves (through joint union memberships, for example), then, indeed, the entrants' cost advantage may disappear. It is difficult to give safe estimates about the capital to labour ratio of incumbents and entrants. If the capital to labour ratio of incumbents were significantly higher than that of entrants, Williamson's approach would provide an argument for incumbents to negotiate higher wages with unions, although increasing their own labour costs by it.

      The relationship between union power and market competition was examined for the period between 1976 and 1994 (Neven and Röller 1996). Their research found significant rent sharing between flag carriers and labour. Combined with government subsidies, such excessive wages were found to "...influence the ability of firms to stay in business, thereby inducing excessive exit " (p. 14). If new entrants were to continue to face such high-pay competition from former flag carriers in the labour market, especially with continued subsidies towards the big carriers, their market presence based upon low-cost competition would seriously be endangered. It is interesting to note that in contrast with the US rationale, which found potential for anti-competitive behaviour in asymmetrical capital-labour ratios between big and small firms, the European context before 1994 appeared to favour outright rent sharing, regardless of "hidden efficiencies" due to the employment of capital investment.

      Another study, by Alamdari and Morell (1997) shows that labour costs per available ton-kilometre were reduced by around 23% for European carriers between 1991 and 1994. Of individual airlines only Iberia and Sabena witnessed rising unit costs during this period. It was found that gains in productivity were offset to a degree by some increase in real wages. The paper also recognizes that the trend towards outsourcing may inflate the productivity improvements. Both studies suggest high wages until 1994 for incumbent airlines, despite apparent progress in overall productivity. Such increased productivity can originate, as Williamson suggests, in a replacement of the labour force by capital investment or by increased outsourcing to other suppliers. ⁶³ 

      Within the scope of this paper, we are particularly interested in the period between 1993 and 1997. Both studies mentioned overlap only with the beginning of the period of interest. Between 1993 and 1997, in particular, important initiatives were launched by national carriers to increase productivity as a whole and to decrease labour costs in particular: Their success was not evident. Air France's efforts to cut labour costs were hindered by labour unrest. The airline's proposal to introduce a two-tier pay system for cockpit crew in 1997 caused a number of strikes. ⁶⁴  Olympic also had problems with labour unrest. Management proposals including a wage freeze, a 45-hour work week, a cut in the number of cabin crew and route reductions prompted staff to stage a three-hour strike in March 1998. The Greek Government said that Olympic's 1994 recovery plan (intended to stretch over five years) had been overturned due to union action and mistakes by management. ⁶⁵ 

      However, other national carriers used this period to significantly reduce labour costs: BA, Lufthansa and KLM are the most prominent carriers that lowered labour costs over the period (partly by lowering wages and replacing crew by younger, less expensive employees). In 1996, BA unveiled a plan that included eliminating 5,000 jobs and cutting GBP 1 billion off its annual costs by the year 2000. In general, we can infer that for the period after 1993 the overall pressure was to cut jobs and to freeze or even lower wages, not to increase them in the way Williamson suggested or as incumbents had actually done before 1993. It is also noteworthy that most employees of new-entrant airlines were not unionized, and possible union membership was still positioned on a national level. As a result, quasi- "collusionary" behaviour between incumbents and airline unions aiming to damage new entrants' cost structures on a European scale was not successful. In addition, airline subsidies were phased out due to EC directives after 1995, so the excesses of rent sharing had to be financed by the carriers themselves. These factors help to explain why the major efforts of former flag carriers in the period in question were aimed at increasing productivity, especially by cutting down on labour costs.

V.3.3 Cross-subsidies from intercontinental routes

      If incumbents were to obtain subsidies from governments, one could interpret them as analogous to absolute cost advantages: No matter how high the prices for fuel, or how inefficient the organization, etc., the government would guarantee debts and would regularly reconstitute the incumbent's equity if need be. Independently of the scale of output, the incumbent might thus use these subsidies as if he had lower factor costs or lower interest rates for procured credit than the entrant. But we have seen that the European Commission forbade government subsidies after 1995.

      The same logic of subsidization on a strictly European scale can be applied to intercontinental flights, especially North American routes. The following exhibit shows the percentage of such intercontinental flights (as measured in millions of revenue-passenger-kilometres) for some of the European incumbents:

      

Exhibit 49: Share of intercontinental flights out of total flights sold in 1996
RPK 1996	SAS	Iberia	KLM	Lufthansa	BA	Air France*
Intercontinental	7,441	13,873	40,061	44,330	77,800	33,440
Total	19,487	25,931	45,531	58,050	96,163	43,079
% of total RPKs	38%	53%	88%	76%	81%	78%

(Source: compilation of the airlines' various annual reports.) * For Air France only 1992 values available

      We know that long-haul routes, especially to North America are very profitable for incumbents. Based on bilateral agreements between the various nation states and the US, entrants to the European market are usually excluded from such routes (the two major exceptions being Virgin to the US and Lauda Air to Asia). We have already shown that long-haul, dense routes can optimize the efficiency of wide-bodied carriers serving major international hubs, thus lowering average unit costs. This is a second source of super-natural profits on intercontinental routes compared with intra-European traffic. A geographical analysis for the BA group showed that in 1996, 96% of operating profit was made outside Europe, ⁶⁶  although 65% of turnover was made in the UK and the rest of Europe!

      Such protected profit streams may be diverted to subsidize other non-profitable operations in Europe, and thus work as absolute cost advantages. Obviously, those carriers having the highest exposure to intercontinental flights would benefit the most. As with regular absolute cost advantages, the incumbent could lower his ticket price to drive out entrants from specific city pairs while cross-subsidizing apparent losses with excess profits from intercontinental routes. We would categorize the incumbents as: (1) heavily exposed to intercontinental routes (KLM, BA, Lufthansa, Air France and Virgin); (2) medium exposed to intercontinental routes (Alitalia, SAS, Iberia and Olympic); (3) little or not exposed at all (Sabena, TAP and entrants).

V.3.4 Frequent flyer programmes ⁶⁷ 

      Since the early 1990s, FFPs have spread to the EU from the US. All significant European airlines now have their own FFP or participate in another airline's programme. US experience suggests that once FFPs have been established, it would be extremely difficult, if not impossible, for airlines to abolish them (Civil Aviation Authority 1993, p. 9). In the EU, smaller airlines have been just as quick as their larger competitors to realize the importance of FFPs. Some have sought to overcome their inherent size disadvantage by combining to market joint frequent flyer programmes (such as Virgin and British Midland, and previously Dan Air) or participating in larger airlines' programmes (for example Air UK in KLM's), or both (British Midland participates in SAS's FFP as well as Virgin's).

      Since late 1998, four global alliances have formed, allowing for unified frequent flyer programmes within one alliance. These alliances vary significantly in their composition, as the following exhibit shows, with varying degrees of involvement of European incumbents. ⁶⁸ 

      

Exhibit 50: Frequent flyer alliances in 1997
Alliance	Participating airline
Star Alliance	Lufthansa, United Airlines, Varig, SAS, Thai Airways, Air Canada
One World	British Airways, Quantas, American Airlines, Air New Zealand, Ansett
Wings	Alitalia, KLM, Northwest Airlines, Continental
Qualiflyer Group	Swissair, Sabena, TAP, Delta, Turkish Hava Yollari, AOM, Austrian, Lauda Air, Tyrolean Air, Air Littoral, ANA, Malaysian Airlines, Singapore Airlines*
Note: Air France announced its intention to develop a separate alliance along with Delta and other yet to be identified Asian airlines by the year 2000**
*Due to talks between Air France and Delta (see Le Figaro, 23 June, 1999), Swissair and Sabena threatened to withdraw from Qualiflyer and join other partnerships. ** See "Le Figaro" of 22 June, 1999: section BI.

      FFPs have proved popular with passengers, particularly those travelling on business, but less so with the companies who pay for business travel. It is claimed that the plans encourage unnecessary travel, travel at a higher fare than strictly required or travel by a circuitous route so as to use a particular airline. FFPs tend to favour larger airlines at the expense of smaller carriers: In the case of a flag carrier dominant in a particular city, its more extensive network of routes will encourage passengers wanting to collect frequent flyer points to use its services rather than those of a smaller competitor (Civil Aviation Authority 1993, p. 9). Similarly, the larger airline is likely to have more holiday destinations in its route network for programme members to use their points to visit. ⁶⁹  Following this rationale, the FFPs of Star Alliance, One World and the Qualiflyer Group seem to have both a stronger base with business travellers, and an extremely wide network to cover as many city pairs as possible, for business and holidays alike. In contrast, airlines such as Air France, Iberia, TAP, Olympic and Sabena appear to have had a significant disadvantage as of 1997.

      Frequent flyer programmes encourage passenger loyalty through the award structure. The General Accounting Office (1990, p. 62) states: "Because the award structures encourage passengers to fly regularly on a single airline, a frequent flyer plan helps a well-established airline to discourage its passengers from flying on other airlines that offer new service to the same destinations. The dominant airline at an airport generally offers service to the most destinations and will, therefore, offer participants in FFPs the most opportunities to earn and redeem awards" The same report found, based on a survey of 520 US travel agents, that business customers chose flights to accumulate additional frequent flyer miles more than half the time. It also concluded that FFPs were heavily used and that the airline providing the most flights from a particular city was likely to attract the most frequent flyer participants.

      Indeed, FFPs represent long-term contracts with both customers and other incumbents. They aim to prevent entrants from gaining market share by locking in customers. The levers for effective FFPs appear to be: (1) an award structure that is powerful enough to build customer loyalty (we assume no asymmetries among the given FFPs, ceteris paribus); (2) a business class passenger willing to pay above-normal ticket prices to accumulate bonus miles; ⁷⁰  (3) a wide choice of destinations; with direct flights preferred; and, consequently, (4) a strong presence at an airport hub. Due to their "global network" approach, however, FFPs may be less effective in deterring entrants from specific point-to-point competition on intra-European short- or medium-distance routes.

      

Exhibit 50(a): A comparison of marginal revenues between FFP and non-FFP airlines

• FFPs imply significant administration costs: FFP programs need to be advertised, potential frequent flyer passengers need to be induced to participate in such programmes, FFP passengers need to open an account, these accounts need to be administered with monthly or quarterly account balances sent to the customer, etc.
• FFPs require, by definition, excess capacity on a broad range of city pairs. This is a clear trade-off between unit yields and unit costs. FFP airlines need to choose larger aircraft if they want to accommodate additional frequent flyer customers who seek to cash in their bonus miles (ergo: marginal unit costs become very small, average unit costs per flight drop). Conversely, these frequent flyer passengers will not add to revenues on such flights (ergo: marginal yield will be zero, average unit yields will drop).

      

      _{The fact that business class tickets might provide the necessary yield premium over economy class tickets may render FFPs profitable, though not necessarily profit maximizing. In order to be profit maximizing, the following equation must hold:}

      After simplification we get:

      

      The incumbent increases demand in economy class, while the entrant loses ticket sales (especially for holiday city pairs).

1. The incumbent's economy fares remain unaffected by this, whereas the entrant is under pressure to lower his prices.
2. The incumbent's business class ticket prices would need to increase in order to finance the bonus miles plus FFP administration, ceteris paribus. Or, if business class ticket prices remained unchanged, the incumbent's number of business class tickets would need to increase due to FFPs. Otherwise the incumbent would maximize his profits by dropping the FFP altogether.

VI.3.1 Contestable markets and fixed costs

      According to contestability theory, fixed costs may make incumbency sustainable. To present the power of fixed costs for sustainable prices, Baumol, Panzar and Willig (1988) use an intuitive example based on the idea that the incumbent could choose between fixed and variable costs in order to produce a certain output.

      

Exhibit 53: Price sustainability with fixed costs

      With an alternative cost function, which duplicates C but also adds into it fixed costs F, the situation changes fundamentally. With a new revenue function meeting curve C+F at point y^*, the new cost function remains above the new revenue function at all outputs lower than y^*. There is no opportunity for entrants for profitable entry below y^*. The reason is that an increased fixed cost raises the average cost of smaller outputs relative to those of larger outputs. By placing smaller firms at a relative disadvantage, fixed costs make sustainable prices possible for the incumbent (pp. 286).

      However, fixed costs are not necessarily considered entry barriers because they need not prevent optimal welfare performance by the industry (p. 289). Employing a normative criterion of welfare effects to define entry barriers, Baumol, Willig and Panzar follow von Weizsäcker (1980).

VI.3.2 Contestable markets and cross-subsidies

      Cross-subsidies within a multi-product firm ocur when one product produces above-normal profits that are used to compensate for below-normal returns of other products. Prices below marginal costs for a certain product/service indicate the presence of some form of cross-subsidization. This means that, at current prices, users are paying too much for some of the incumbent's services. Consumers would save money on these products/services if a separate firm were to provide only these products/services. Contestability theory argues that "...prices cannot be sustainable if they involve any cross-subsidies..." (Baumol, Willig and Panzar 1988, p. 351). Any product of an incumbent must yield incremental net revenues at least as great as his incremental net costs. Otherwise, an entrant may choose to compete for above-normal-returns by undercutting the incumbent's prices. The entrant would earn more than the incumbent previously did, as it would not be subsidizing other products/services. The incumbent would be left with the below-normal-return products. Therefore, prices that enable an incumbent to cross-subsidize other products/services cannot be sustainable.

VI.3.3 Contestable markets and sunk costs

      Without sunk costs and with identical technologies, the incumbent firm and a potential entrant bear the same cost at each level of output. There is no strategic asymmetry between an entrant and an established firm because each faces exactly the same cost and revenue function. To illustrate further the concept of perfectly contestable markets, we will now touch on their relationship with sunk costs. Caves and Porter (1978) see sunk costs as "the difference in incremental cost and incremental risk" between an incumbent and an entrant. The full cost of an investment to be sunk, already foregone by the incumbent, would need to be made anew by the entrant. The risk of losing this investment is then higher for the potential entrant than the incumbent (Baumol, Willig and Panzar 1988, p. 291). From this perspective, we can understand that the risk of losing unrecoverable capital needed for entry can render markets less contestable.

      This asymmetric risk exposure between incumbents and entrants, however, becomes relevant only in the case of the entrant exiting the market. If the entrant manages to stay in the market sufficiently long, his sunk costs, as the incumbent's before, will eventually become zero (p. 279). This observation is critical to the understanding of the role of post-entry behaviour of the incumbent. If the incumbent were to accommodate the entrant, and if this were known to the entrant, markets could become contestable again because the incumbent's only way to keep an entrant out would be to lower prices to sustainable, that is unprofitable, levels.

      Therefore the importance of the incumbent reacting after entry is evident: "...Sunk costs are not an entry barrier in itself, but permit a wide variety of other influences to affect and to increase the entry barrier." For example, the autors quote economies of scale or "the threat of retaliatory strategic or tactical responses of the incumbent" (p. 290).

VI.3.4 Contestable markets and strategic behaviour

      From the above we infer that the monopolist's response can produce entry costs. The authors write in one proposition: " Where no other entry barriers and frictions are present, either the absence of sunk costs or the prevention of post-entry responses by incumbents are sufficient for markets to be contestable" (p. 301). With different models for entry deterrence, the critical role of the incumbent's countermoves to determine post-entry market equilibrium has already been highlighted outside the scope of contestability theory in previous chapters.

      Baumol, Panzar and Willig (p. 296) constructed a model to establish the relationship between sunk costs, entry costs and welfare, without imposing restrictions on the incumbent's post-entry behaviour. In their findings, sunk costs produced entry costs, which was intuitively explained by their shorter depreciation period during a finite disequilibrium period after entry occurs. Only after such a disequilibrium period was the incumbent allowed to react to drive the entrant out of the market. The possibility of such a reaction by the incumbent accounted for a shorter depreciation period than normally expected (if the capital invested had no salvage value, thus was sunk). But this shorter depreciation period was pointless if the incumbent failed to react against the entrant, because this would extend the shortened depreciation period and thus lower entry costs.

      "In the absence of other cost disadvantages to the entrant, entry costs are zero if either (a) there are no sunk costs, or (b) the monopolist is never permitted to respond" (p. 300). According to this model, both structural (such as entry barriers) and behavioural elements (for example, constantly undercutting the entrant's prices) are needed to produce entry costs within otherwise contestable markets.

VII.1.1 Data and sample selection

      After having tried unsuccessfully to obtain back copies of the OAG from Reed or from travel agencies, I discovered that no university or public institution in Switzerland (including IATA in Geneva) had them listed in their catalogues either. By coincidence, I discovered the library of the Civil Aviation Authority next to Gatwick airport in London. From there, all the raw data required were found and copied. In order to select my sample, it was clear that data needed to be ordered along the lines of city pairs - this was the first major crucial assumption for ordering and checking for completeness of my data. Then, I needed to decide which city pairs should be retained within my sample in order to make a valid statement for pan-European development after deregulation. This proved relatively tricky because preferences of, say, 80% of the passenger traffic in Europe (or 80% of passenger kilometres flown) might have biased the results towards high-density routes alone and neglected lower density-ones, which might still be very relevant for many European member states.

VII.1.2 Variable selection and analysis of raw data

      

Exhibit 55: Summary description of variables
Independent variable	Acronym	Type	Unit of measure	Meaning of measure
Average aircraft size	dAircra	Interval	Passenger seats of plane	Change in size of plane from '93 to '97
Daily flight frequency	dDepd	Interval	Number of departures per day	Change in daily frequency ('93-'97)
Weekly departures	dDepw	Interval	Number of departures per week	Change in weekly frequency ('93-'97)
Business ticket	dBusine	Interval	Number of types of business tickets	Ticket differentiation within premium class (between '93 and '97)
Season ticket	dSeas	Interval	Number of types of season tickets	Ticket differentiation due to season ('93-'97)
Weekday ticket	dWeek	Interval	Number of types of weekday tickets	Ticket differentiation due to day of the week (between '93 and '97)
Restricted ticket	dFlexibi	Interval	Number of types of restricted tickets	Ticket differentiation due to new restricted tickets ('93-'97)
Special discount	dSpecia	Interval	Number of types of low-fare tickets	Ticket differentiation due to lowest fares tickets ('93-'97)
Daily flight frequency	Frequ	Interval	Number of daily departures, if > ;6	Committed to high frequency in 1997
Computer reservation	CRS	Ordinal	{-1;0;1}	{no booking via CRS; airline uses CRS for booking; airline is CRS owner}
Code-sharing	Codesh	Interval	Number of daily code-shared flights	Influence of code-sharing as of 1997
Frequent flyer programme	FFP	Ordinal	{0;1;2}	{no FFP; FFP with average network size; FFP with important network and partners}
Intercontinental flights	Interco	Ordinal	{-1;0;1}	{solely intra-European traffic; average presence of intercontinental routes in network; strong presence of intercontinental routes}
Hub dominance	Hubdo	Ordinal	{0;1;2}	{secondary airport; presence at dominated airport; dominance at airport}
Dependent variable
Price Business class	dPriceBus	Interval	Local currency	Change in fare between '93 and '97
Price Economy class	dPriceHWPX	Interval	Local currency	Change in fare between '93 and '97
Price cheapest ticket	dPricelow	Interval	Local currency	Change in fare between '93 and '97

VII.1.3 Estimation

      

      regression equation would be rather unwieldy, with a true risk of obscuring rather than highlighting drivers for sustainable prices. In addition, the equation's outlined general structure would fail to account for interaction between the categorical variables, a possibility that cannot necessarily be excluded following our assumptions made in the theory part before. We have to look for alternative formulations, even at the risk of increasing the sum of squared errors of a somewhat reduced regression equation. Factor analysis appears to be the proper tool for reducing our set of multi-variates. Then the (reduced) regression equation is:

      

VII.2.1 Cluster analysis versus factor analysis

      Cluster analysis seeks to classify data into relatively homogeneous groups according to their similarities. These classes of distinct cases can be visually represented, usually as clusters in two-dimensional space, or as branches of a tree. In this study, we based our similarity measure on distance. ⁷⁶  although it could also be based on correlation coefficients. With factor analysis, components of variance are the central concept (Karni and Levin 1972). The correlation coefficient is a measure of the variance component common to two variables. Factor analysis can be used to examine the underlying patterns or relationships among a large number of variables and to determine whether or not the information can be summarized into a smaller set of constructs. Factor analysis identifies underlying linear variates that draw out as much sample variance as possible, and usually considers only those variables with "high" loadings. With cluster analysis, all of the variables are considered when measuring similarities and classifying them. Cluster analysis ignores the variance (or distances) of variables with "low" loadings. However, cluster analysis will not interpret how these variables relate to each other.

      As we expect valuable insight from applying both procedures (and later from regression analysis to measure sustainability), we shall first of all form groups (or cases) that describe the various behavioural patterns of airlines serving different city pairs in Europe in the 5 years after deregulation.

---

---

VII.2.2 Factor analysis of measures associated with entry barriers

      Four distinct types of entry barriers were of particular interest: economies of scale, differentiation, sunk costs and absolute cost advantages. As shown before, we assumed these barriers to be partially multidimensional, not unidimensional, and that some entry barriers (especially sunk costs and absolute cost advantages) could not be measured with uni-dimensional parameters. Since I had multiple measures for each construct, we wanted to determine whether the multiple measures were more closely associated to one another than they were to other measures which were intended to measure a different type of entry barrier. The appropriate analysis procedure was to do a principal component factor analysis of all the measures, with varimax rotation, and to allow the analysis procedure to determine which measures loaded together (Kim and Mueller 1978).

      A principal component analysis explains as much of the total variation in the data as possible with as few factors as possible. The analytical goals of parsimony and independence are achieved through this method. However, the first component often represents an overall measure of the information contained in all the variables, and it is difficult to interpret whether the factor is more related to one subset of variables than another. Rotation methods alter the initial factors so that we can more easily interpret the factors, and the varimax rotation method rotates the axes orthogonally. The variables which load highly (near +/- 1) on a factor do not necessarily describe an underlying concept: this task is left to the individual researcher (Kleinbaum, Kupper and Muller 1988).

      If all our measures for each construct loaded together, then the measurement model was well specified and reflected each specific domain of content; but if they did not load together, then we would learn where the discrepancies were and which data were less trustworthy. A scree test, with the minimum eigenvalue set at 1.00, clearly suggested three factors. ⁸⁰  From the following factor analysis it appeared that some of my measures reflected the constructs as we had anticipated, but other measures did not seem to differentiate between types of entry barriers.

      Whereas Factor 1 clearly reflected differentiation, based on ticket classes, the remaining two factors lumped together different measures from different kinds of entry barriers. However, this aggregation into the two remaining factors is not inconsistent with the theoretical assumptions outlined before: Factor 2 essentially loads highest on capacity-related choices, such as flight frequency and aircraft size. Interestingly, Factor 2 also loads as well on code sharing, an indicator that is likely to be positively correlated with flight frequency. These measures are associated with economies of scale (although frequency and aircraft size would go in opposite directions) and sunk costs (for flight frequency in absolute terms and code sharing). Factor 3 loads high especially on absolute cost advantages (such as hub dominance, intercontinental flights and FFPs). It also shows a high loading on CRS (sunk costs), possibly due to leverage effects that CRS can exercise when combining the three identified absolute cost advantages.

      

Exhibit 59: Components matrix after rotation

VII.2.3 Regression analysis and entry barrier measures

      Multiple regression analysis is by far the most widely used multivariate technique in strategy management research, since it can be used to examine the relationship between a single dependent variable (in this case, sustainability of prices) and a set of independent variables such as the ones discussed above. Its popularity often results in simple ordinary least square estimations with little regard to testing the assumptions underlying regression analysis or alternative specifications to determine a better fit with the data.

      General specifications

      We applied the regression to selected clusters of city pairs. However, when examined separately, different clusters are likely to have significantly different intercepts and slopes for their respective independent variables. We defined prices as a linear function of a constant Y-intercept and seven regressors, which were selected from our factor analysis due to their heavy loading on a particular factor and due to their relevance for identifying the dominant types of entry barriers:

      

      

      with the prefix d describing the differences between 1997 and 1993 in the independent interval variables. For meaningful interpretation of the impact of each independent variable on price sustainability, an intercept term ≠ 0 was not necessarily helpful: A positive Y-intercept already would have included at least some of the price increases that had occurred. In other words, a positive intercept on the Y-axis alone already suggests increases in prices, thus a priori unsustainable prices. As our research question concentrates on entry barriers alone and their respective contribution to price changes, any potential Y-intercept would conveniently cancel out due to our subtraction of two linear equations, which had used an identical term of a_j for both in 1993 and 1997.

      Choice of clusters

      Going back to our clusters, we first of all eliminated clusters 5 and 6 for obvious reasons: Cluster 5 described city pairs where airlines had decided to quit service between 1993 and 1997. Again, per definitionem, there was no point in examining price sustainability (variable for price changes equals -1). Cluster 6 was also eliminated, since it contained only one case. This left us with clusters 1 through 4. More clusters then were withdrawn: Cluster 2 initially contained five observations, but needed to be corrected "manually" due to imperfect clustering. After this reassignment of cases, only three cases remained within this cluster. Cluster 3 only contained ten cases, and it was not considered fruitful to run a regression analysis with 7 explicative variables on only 10 cases. An alternative would have been to lump cluster 3 with cluster 4, since both clusters showed city pairs expanding on capacity. However, there were excellent reasons to cluster these two groups apart: By consolidation, mutually opposite signs in ticket differentiation and CRS would have further watered down our difficult analysis to identify the driving barriers in each cluster.

      Testing the assumptions

      

VII.3.1 Results for incumbent routes (Cluster 1)

      The sample of city pairs within cluster 1 involved those routes where the carrier already had a strong presence at either the origin or destination airport. These routes were frequently served and indicated high traffic densities. One could describe these routes as dominated by former flag carriers.

      

Exhibit 60: Model summary for cluster 1
R	R-two	R-two adjusted	Standard error of estimation	Changes in the statistics
CLUSTER = 1 (selected)				Variation of R-two	ddl 1	ddl 2
0.880	0.774	0.754	7.17485	0.774	7	78
0.867	0.752	0.730	4.44081	0.752	7	82
0.900	0.810	0.791	4.26189	0.810	7	70

1. measures the proportional variance of the dependent variable relative to the intercept as determined by regression. This type of R two cannot be compared with models that include a constant for intercept.
2. Independent variables: dFlexibi, Hubdo, dAircra, Codesh, dDepw, CRS, InterCo
3. All statistics are based upon observations, for which CLUSTER = 1.
4. Dependent variable: PriceBus, PriceHWPX, Pricelow
5. Linear regression through the origin
6. Regression by least squares estimate - Weighting by WEIGHT2

      The regression was run on three ticket classes: business class, fully flexible economy and the lowest available fares. Exhibit 64 summarizes the outcome for each ticket class in a particular line. R² is in a range between 75% and 80%, but cannot be compared with R² which would include a constant factor. ⁸² 

      The following three exhibits present the coeffcients for price changes in three different ticket classes: business, economy and lowest available fares.

      

Exhibit 61: Regression results for business, economy and lowest fares in cluster 1
	Non standardised coefficients		Standardised coefficients	t	Significance level	Confidence interval at 95% of B
	B	Standard error	Bêta			Lower limit	Upper limit
dDepw	4.043E-03	0.005	0.067	0.873	0.385	-0.005	0.013
dAircra	1.360E-03	0.002	0.055	0.839	0.404	-0.002	0.005
Hubdo	4.957E-02	0.159	0.299	0.313	0.755	-0.266	0.365
CRS	7.251E-02	0.070	0.334	1.029	0.307	-0.068	0.213
Codesh	-7.836E-06	0.022	0.000	0.000	1.000	-0.044	0.044
InterCo	2.581E-02	0.145	0.155	0.178	0.859	-0.263	0.315
dFlexibi	-3.032E-02	0.008	-0.241	-3.795	0.000	-0.046	-0.014

      a Dependent variable : PRICEBUS

      

	B	Standard error	Bêta			Lower limit	Upper limit
dDepw	-6.231E-03	0.003	-0.171	-2.184	0.032	-0.012	-0.001
dAircra	8.328E-05	0.001	0.006	0.083	0.934	-0.002	0.002
Hubdo	0.186	0.094	1.855	1.980	0.051	-0.001	0.372
CRS	3.695E-02	0.043	0.283	0.849	0.398	-0.050	0.123
Codesh	3.296E-02	0.014	0.161	2.386	0.019	0.005	0.060
InterCo	-0.129	0.085	-1.278	-1.521	0.132	-0.297	0.040
dFlexibi	-7.854E-03	0.005	-0.103	-1.588	0.116	-0.018	0.002

      a Dependent variable : PriceHWPX

      

	B	Standard error	Bêta			Lower limit	Upper limit
dDepw	1.184E-02	0.003	0.315	4.249	0.000	0.006	0.017
dAircra	-2.142E-03	0.001	-0.139	-2.129	0.037	-0.004	0.000
Hubdo	2.945E-02	0.111	0.276	0.266	0.791	-0.192	0.250
CRS	1.278E-02	0.047	0.092	0.275	0.784	-0.080	0.106
Codesh	-2.746E-02	0.017	-0.110	-1.572	0.120	-0.062	0.007
InterCo	2.347E-02	0.105	0.219	0.224	0.823	-0.185	0.232
dFlexibi	-3.576E-02	0.005	-0.456	-7.265	0.000	-0.046	-0.026

1. Dependent variable : PRICELOW
2. Linear regression through the origin
3. Regression of weighted least squares - Weighting by WEIGHT2
4. Exclusive selection of observations for which CLUSTER = 1

VII.3.2 Interpretation of results (Cluster 1)

      Business Class

      For business class, a single variable representing differentiation has a significant impact on price changes. Unlike our theoretical predictions suggested, the introduction of new ticket classes (here represented by fully flexible tickets) had negative impacts on business class fares. It is not difficult to find a reason for such negative correlations: Fully flexible tickets, even in economy class, provide direct competition for business class, especially on short-haul European routes. These newly introduced and lower priced ticket classes threatened to cannibalize the business segment. However, as shown in the previous chapter, on average, business fares had increased by 37,3%. This means that other factors must have more than compensated for the negative impact of ticket differentiation. Indeed, all remaining six regressors show positive signs, with CRS and Hubdo the most important ones. However, none of these variables is significant on a 90% confidence interval, at least not on a univariate level. ⁸³ 

      Fully flexible economy

      Increases in weekly flight frequency show significantly negative impacts on ticket prices, and the extent of code sharing in 1997 shows a significantly positive correlation on the dependent variable. These results must be interpreted in the light of our theoretical predictions: Increasing flight frequency led to higher average costs. Especially on high-density routes as in cluster 1, larger aircraft could have been used instead of increasing flight frequencies. Code sharing was expected to reduce excess capacity and thus to provide lower average costs, which would allow for lower prices. These findings suggest considerable market power: Increasing frequency and lowering prices could deter entrants along the lines of the Dixit model (see IV.4.1). Sharing codes with other airlines and increasing prices indicated market power of both incumbent airlines over such routes.

      By far the most influential independent variable is hub dominance, which is also significant on a 95% confidence interval. Its positive impact on prices had been predicted by theory. The impact of hub dominance on higher ticket prices outweighs the negative impact of more frequent flights by around 10 times (compare the respective b coefficients). Both intercontinental exposure and increases in fully flexible tickets, our proxy for differentiation, contributed to explaining variances from the mean towards lower ticket prices, but failed to be significant on a 90% confidence interval. The negative coefficient for airlines also serving intercontinental routes indicates that for economy class prices may be lowered, possibly to induce passengers to transfer at the incumbents' hubs onto intercontinental routes. This would suggest cross-subsidization on European routes, a clear contradiction to the theory of perfectly contestable markets. This effect is shown to be significant on a 85% confidence interval only.

      The effect of lower prices with economy class tickets through newly introduced ticket classes was shown to be significant on a 85% confidence interval only. Its impact to reduce fares was much smaller than with business class.

      Lowest available fares

      Our third price class that we had examined in cluster 1 contained the lowest available fares for the given city pairs. Most importantly, we note that the independent variables related to size, which we assumed to be relevant for keeping premium fares expensive, no longer played a significant role. Hub dominance, intercontinental exposure and CRS not only did not show any significance, but their coefficients were much smaller compared with other classes, i.e. economy class. It is also noteworthy that intercontinental routes did not appear to cross-subsidize lowest available fares on dominated routes of cluster 1. The absence of such a relationship would correspond to the theory of perfectly contestable markets. In this context of low margin fares, frequency and capacity related variables behaved very much along the lines that had been suggested in the theory part: Increasing aircraft size tends to lower average costs, which may manifest itself in lower prices. More weekly departures do not lower costs: Prices tend to increase. Code sharing, although not significant on a 90% confidence interval, shows a clear tendency to lead to lower prices, probably through lowering average costs. The significantly negative coefficient of differentiation through new ticket classes suggests that new ticket classes had been introduced which allowed a further reduction of lowest fares.

VII.3.3 Results for new entry routes (Cluster 4)

      Cluster 4 describes the situation where new routes were started after deregulation, either by industry incumbents or by new entrants. These city pairs did only exceptionally involve airports where the airline held a strong position. In most of the cases, traffic was started from secondary airports, such as Luton or City airport for London. New service typically was started with around four daily flights for the given city pair (each way). This cluster described an environment that was expected to be more competitive than cluster 1, as the industry incumbent held no city pair-specific advantage compared to any other entrant.

      

Exhibit 62: Model summary for cluster 4
R	R-two	R-two adjusted	Standard error of estimation	Changes in the statistics
CLUSTER = 4 (selected)				Variation of R-two	ddl 1	ddl 2
0.997	0.995	0.993	0.34499	0.995	7	27
0.998	0.997	0.996	0.18836	0.997	7	29
0.998	0.997	0.996	0.22371	0.997	7	27

1. For a regression through the origin (model without a constant for intercept), R two measures the proportional variance of the dependent variable relative to the intercept as determined by regression. This type of R two cannot be compared with models that include a constant for intercept.
2. Independent variables: dFlexibi, Hubdo, dAircra, Codesh, dDepw, CRS, InterCo
3. All statistics are based upon observations, for which CLUSTER = 4.
4. Dependent variable: PriceBus, PriceHWPX, Pricelow
5. Linear regression through the origin
6. Regression by least squares estimate - Weighting by WEIGHT2

      As indicated before, a value for R² as high as the one shown in our model summary should be interpreted with caution. It is mainly due to the fact that we had defined our Y-intercept as passing through the origin of the Y-axis. A control run of our regression equation, where we introduced a constant term into the equation, yielded a value for R² of around 35%. The following exhibit presents the coeffcients for price changes in three different ticket classes: business, economy and lowest available fares.

      

Exhibit 63: Regression results for business, economy and lowest fares in cluster 4
	Non standardised coefficients		Standardised coefficients	t	Significance level	Confidence interval at 95% of B
	B	Standard error	Bêta			Lower limit	Upper limit
dDepw	2.239E-03	0.003	0.289	0.760	0.454	-0.004	0.008
dAircra	6.064E-03	0.001	3.044	6.297	0.000	0.004	0.008
Hubdo	-0.342	0.041	-4.367	-8.360	0.000	-0.425	-0.258
CRS	4.452E-02	0.106	0.525	0.420	0.677	-0.173	0.262
Codesh	4.482E-03	0.013	0.118	0.355	0.725	-0.021	0.030
InterCo	3.918E-02	0.089	0.502	0.441	0.663	-0.143	0.222
dFlexibi	-4.908E-02	0.016	-1.069	-3.010	0.006	-0.083	-0.016

      a- Dependent variable : PRICEBUS

      

	B	Standard error	Bêta			Lower limit	Upper limit
dDepw	-3.824E-03	0.002	-0.662	-2.463	0.020	-0.007	-0.001
dAircra	2.902E-03	0.001	1.953	5.687	0.000	0.002	0.004
Hubdo	-0.131	0.022	-2.252	-6.015	0.000	-0.176	-0.087
CRS	2.791E-02	0.054	0.441	0.515	0.611	-0.083	0.139
Codesh	2.177E-02	0.007	0.771	3.225	0.003	0.008	0.036
InterCo	2.455E-02	0.045	0.422	0.549	0.587	-0.067	0.116
dFlexibi	-1.584E-02	0.009	-0.463	-1.793	0.083	-0.034	0.002

      a- Dependent variable : PriceHWPX

      

	B	Standard error	Bêta			Lower limit	Upper limit
dDepw	5.838E-03	0.002	0.903	3.058	0.005	0.002	0.010
dAircra	4.974E-03	0.001	2.992	7.965	0.000	0.004	0.006
Hubdo	-0.174	0.026	-2.663	-6.562	0.000	-0.228	-0.120
CRS	0.151	0.069	2.136	2.204	0.036	0.010	0.292
Codesh	-1.221E-02	0.008	-0.387	-1.492	0.147	-0.029	0.005
InterCo	-0.292	0.058	-4.487	-5.070	0.000	-0.410	-0.174
dFlexibi	-8.579E-02	0.011	-2.239	-8.113	0.000	-0.107	-0.064

1. Dependent variable : PRICELOW
2. Linear regression through the origin
3. Regression of weighted least squares - Weighting by WEIGHT2
4. Exclusive selection of observations for which CLUSTER = 4

VII.3.4 Interpretation of results (Cluster 4)

      Business class

      Three variables showed significant impact on price changes: Aircraft size (positive), hub dominance (negative) and ticket differentiation (negative). As for aircraft size, we found that the use of bigger aircraft (such as a Boeing 737-400 compared with a Fokker 50) was significantly linked, on a 95% confidence interval, to higher ticket prices. This empircal finding supports the theory of higher average unit costs for larger aircraft, if no trade-off is made with flight frequency for given city pairs.

      Economy class

      As with business class, aircraft size increased, hubdominance and ticket differentiation both decreased fares in economy class. These relationships were all significant at a 95% confidence interval for both aircraft size and hub dominance, and at a 90% confidence interval with ticket differentiation.

      Lowest fares

      As with business and economy class, aircraft size increased, hub dominance and ticket differentiation both decreased the price category of lowest available fares (99% confidence interval). Three variables are significant and idiosyncratic in their effect on lowest prices: changes in weekly flight frequency are positively correlated with prices. This corresponds with what we would expect from higher costs being reflected in ticket prices. CRS are significant (95% confidence interval) in raising prices. Only within lowest fares on newly served city pairs were CRS significant for higher ticket fares. Unlike the theory (see IV.3.3 and IV.4.2) suggested, did we not find any transfer of profits derived from booking fees towards CRS stakeholding airlines. CRS appear to contribute to higher ticket fares for airlines that hold important ownership stakes in CRS. ⁸⁴  The most obvious explanation would be a possible control over distribution channels by incumbents that happen to be the owners of such CRS.

VII.3.5 Synthesis and comparison of the results

      Exhibit 68 synthesizes our empirical findings. It shows all significant correlations between changes in ticket prices and the respective independent variables that represented potential entry barriers. Positive coefficients mean that a higher value of the independent variable increased ticket prices, negative coefficients stand for decreases.

      

Exhibit 64: Summary of findings
	Cluster 1			Cluster 4
	Business	Economy	Lowest	Business	Economy	Lowest
Ticket differentiation	-	-	-	-	-	-
Weekly frequency		-	+		-	+
Code share		+	-		+	-
Hub dominance		+		-	-	-
Intercontinental		-				-
Aircraft size			-	+	+	+
CRS						+

      We shall now review the impact of the various selected independent variables across all three ticket fare classes within the two clusters and highlight their conformity or non-conformity with the predictions made by perfectly contestable markets.

      Ticket differentiation

      In cluster 1 and cluster 4 the introduction of new ticket classes led to lower ticket prices. This applied across all price segments that were examined, so one can therefore not suggest a price premium that may have been expressed through new ticket classes. If such new ticket classes had better segmented the market and added value to the flight service, differentiation would have allowed for such price premiums. Our empirical data show that differentiation was not a significant barrier to entry. In this sense, pefectly contestable markets are not violated.

      Weekly frequency

      Increases in weekly flight frequency were empirically significant for economy fares and lowest available fares. For lowest fares (in both clusters 1 and 4) the higher costs of operating more frequent flights were reflected in higher prices. This is not surprising, as determining lowest fares is often based upon marginal cost pricing.

      For economy class, increases in weekly frequency would lower prices. This empirical finding confirms the hypothesis made earlier (see IV.3.1.3 and IV.4.1): Increased flight frequencies on certain routes with lower fares would deter entrants. The characteristics of excess capacity and Cournot competition along the lines of Dixit can be confirmed for economy class. With regard to changes in weekly frequency, perfectly contestable markets were not impeded for the segment of lowest available fares. In economy class (for both clusters 1 and 4) markets were not perfectly contestable due to entry barriers and strategic behaviour linked to flight frequency.

      Code share

      As with flight frequency, the cost effects of code sharing were directly reflected in the lowest available prices. Code sharing consolidated operations between two airlines and reduced average unit costs. These lower costs were empirically shown to decrease ticket fares for the lowest available fares. In this respect, we cannot observe any impediment to perfectly contestable markets.

      Within the economy segment, code sharing tended to increase prices. Unlike flight frequency or aircraft size, code sharing could not create excess capacity and would not allow for strategic behaviour along the lines of Dixit. Instead, the argument of market power may provide a valid explanation. In economy class (both clusters 1 and 4), code sharing impeded conditions of perfectly contestable markets.

      Hub dominance

      Dominance at airport hubs presented barriers to entry in the sense of absolute cost advantages. The empirical analysis showed that in cluster 1, hub dominance induced significantly higher prices only for economy class. We may rationalize this observation by considering that margins for business travel are already high, and incumbents have little interest in raising lowest available fares to above-normal profits (signalling low profits to entrants, etc.). Due to the hub dominance of incumbents, perfectly contestable markets did not exist in economy class for cluster 1.

      For city pairs (Cluster 4) newly entered, hub dominance showed significantly and had a strong negative impact on prices. This was explained by the fact that most of these newly served city pairs originated from secondary airports in the city where incumbents held dominant hub positions; thus, hub specific advantages could not be exercised by the incumbent. On the contrary, these secondary airports proved less congested, had less expensive landing fees and, especially, were more accessible for new entrants. This shows that once incumbents had to compete with entrants on the same level, their competitive advantage due to grandfathered rights disappeared and prices had to drop. In this respect, hub dominance of incumbents did not preclude perfectly contestable markets.

      Intercontinental

      As cross-subsidization would be generated within the airline, the principle of profit-maximization suggested subsidizing only the most strategic and competitive routes: In cluster 1, intercontinental exposure served to lower prices in economy class. It is interesting to note that no subsidies were allocated to lowest available fares in cluster 1. For economy class in cluster 1, conditions of perfectly contestable markets were impeded.

      For newly entered markets (cluster 4), competition between industry incumbents and new entrants was more intense and concentrated on the lowest available fares: Cross-subsidies were directed only at the lowest fares segment. Industry incumbents' lowest available fares were not expected to cover costs compared with new entrants who offered a favourable cost structure. Therefore, subsidies were significant if the industry incumbent wanted to stay in business on such city pairs against no-frills competitors. Again, cross-subsidization derived from intercontinental routes prevented perfectly contestable markets.

      As with subsidizing economy class - and not lowest available fares - in Cluster 1, one could expect this price segment, at least partially, to feed traffic into the incumbent's intercontinental routes. Lowest available fares were not structured to provide a feeder service for such intercontinental routes. Therefore, subsidizing such intercontinental feeder services could eventually increase the airline's total margin (compared with using other airlines and then transferring onto the intercontinental route at the hub).

      Aircraft size

      Theoretically, larger aircraft should provide lower average costs than smaller aircraft given a sufficiently high load factor. This reasoning links aircraft size to economies of scale and the expected relationship was empirically confirmed for lowest available fares in cluster 1. In such a low-margin setting, prices would closely follow the actual costs of operations. This relationship accorded with perfectly contestable markets.

      Our empirical findings are the opposite for newly started routes (cluster 4) across all price segments: Here, the use of bigger aircraft increased prices significantly in all price classes. As both entrants and incumbents increased daily frequency in order to pursue dominance on these new routes, many aircraft were operated - often smaller ones. The principle of economies of scale applies hardly in such a setting, where demand is new and preference is given to frequency over aircraft size in order to gain market share quickly. If larger aircraft were used, they would be unlikely to obtain a sufficiently high seat factor if frequency increased as well. This would mean higher unit costs for both incumbents and entrants than would be necessary if excess capacity could be avoided. ⁸⁵  This behaviour was contradictory to the principles of perfectly contestable markets, but was linked to flight frequency and not aircraft size: If entrants obtained slots, they could cause substantial losses to industry incumbents by operating fewer flights and filling up larger aircraft.

      CRS

      Computer reservation systems were the only barrier to entry we identified that presented technology-driven sunk costs. They had a significantly positive impact on prices only for the lowest fares in cluster 4. Despite careful regulation of the standards of such equipment, CRS owner airlines could still benefit from a "preferred client" status when bookings were made. Although competing offers from some entrants might also be offered through the CRS, the system might still privilege incumbents over entrants. Also, once potential passengers had enquired about flights through CRS, direct offers from entrants (i.e. by telephone or through the Internet) could not easily be made anymore. We observed that control over the distribution channel in the form of CRS precluded perfectly contestable markets for the segment of lowest fares on newly started routes.

      Exhibit 69 summarizes the factors that were empirically shown to have impeded perfectly contestable markets in the European airline industry after deregulation.

      

Exhibit 65: Entry barriers that impede perfectly contestable markets
	Cluster 1			Cluster 4
	Business	Economy	Lowest	Business	Economy	Lowest
Ticket differentiation
Weekly frequency		No PCM			No PCM
Code share		No PCM			No PCM
Hub dominance		No PCM
Intercontinental		No PCM				No PCM
Aircraft size
CRS						No PCM

VIII.1.1 Exploiting existing barriers to entry through strategic behaviour

      Flight frequency

      On routes with medium to high density, the incumbent will prefer more frequent traffic with smaller planes, rather than fewer flights using bigger aircraft. As increasing flight frequency makes markets less contestable on both hub dominated routes and routes that involve secondary airports, the incumbent can pursue this strategy universally. If his cost structure renders the incumbent too expensive, he can leave these routes to franchising partners. The strategic impact is leveraged by lowering prices in economy class in order to deter entrants.

      Code sharing

      For routes on which the incumbent increases flight frequency, code sharing among incumbents may serve as a complement: Instead of an incumbent operating new flights by himself in order to better control certain routes, other incumbents serve the same route in a coordinated manner. The notion of market power allows both partners to raise prices in economy class. For less dense city pairs and for lowest available fares, code sharing often leads to gains in efficiency and thus to lower prices.

      Hub dominance

      Market power at airport hubs is the most important impediment to contestable markets on established routes. Incumbents will be interested in maintaining control of as many available slots as possible at the given airport. If any extension of such airports were planned, it would be in the incumbent's interests to occupy these newly created slots as well, instead of providing an opening to entrants. As suggested above, the physical constraints of airport hubs can effectively be exploited by increasing flight frequency and by extending control over airport slots through code sharing agreements with other incumbent airlines.

      Intercontinental

      The other crucial element for dominating a hub-and-spoke network was intercontinental flights and the excess profits they generated due to near monopolies on such routes. On the one hand, hub and spokes allowed airlines to funnel enough passengers on to high-capacity, long-distance carriers, contributing to high profit margins on such routes. On the other hand, part of these excess profits was used to subsidize economy class on certain spokes. The incumbent was interested in subsidizing those routes where traffic was high and entrants threatened to open service. As slot openings for volume traffic were difficult for entrants to obtain, the scarcity of no-frills fares was not really a threat to incumbents at these hubs. Cross- subsidies could therefore be minimized by concentrating only on economy class.

      Outside of dominated hubs, airport slots were more easily available for entrants. In order to sell capacity (and to gain market share through increasing flight frequency), incumbents had to react to the entrant's no-frills fares. Therefore, cross-subsidies were directed at lowest available fares and not at economy class.

      CRS

      Incumbents were concerned about controlling distribution channels. Computer reservation systems allowed them to leverage on passengers' habits of booking through travel agents. Incumbents should further use this tool and try to persuade no-frills airlines that took only direct bookings (by telephone or through the Internet). CRS were likely to remain a powerful tool, especially if customers wished to transfer to other flights, for example long-distance ones. If no-frills airlines subscribed to CRS, this might give the incumbent the opportunity to further optimize his own network, and might also influence the entrant's price policies (by adding commission fees for the CRS).

VIII.1.2 Creating new barriers to entry

      Beyond the mere exploitation of entry barriers, the incumbent may also attempt to create new barriers to his advantage. We see the potential for the following new barriers:

      Slot allocation

      Because the way the incumbent conducts business is largely enabled by hub dominance, a change in the regime that allocates airport slots is not in the interests of the incumbent. As different studies have suggested, the incumbent will defend this barrier to entry against changes in regulation such as auctioning off all available slots to the highest bidder. Instead of allowing the market mechanism to efficiently allocate the scarce resource of airport slots, it is in the incumbent's best interests to defend the status quo of grandfathered rights. Only if airport capacity were expanded might the incumbent be interested in bidding for newly created airport slots, in order to maintain his market power.

      Bilateral agreements

      The incumbent has a strong interest in maintaining bilateral agreements on air traffic for as long as possible. Such bilateral agreements are the source of the supra-normal profits that can be earned on many intercontinental routes. Bilateral agreements often facilitate code sharing between incumbents since they de facto tend to split up markets between partners.

      If bilateral agreements are dissolved, due to liberalization of air traffic, the incumbent will look for new barriers to create. As long as such agreements persist (as on intercontinental routes), the incumbent can maintain collusive arrangements with his partner airline flying into his hub.

      Additional hubs

      In order to expand their route network and to relieve the dominated hub from congestion effects, many incumbents form additional hub airports, which they dominate. For British Airways this would be Gatwick, for Lufthansa it would be Munich, for Air France Orly or Lyon Satolas. Knowing that sufficiently dense traffic existed for these airports, incumbents might also preempt entrants from developing a home base there by dominating such hubs. In such cities, it may become impossible for entrants to find alternative airports, which would render the advantages due to hub dominance all the more effective for the incumbent.

      Control over secondary airports

      Secondary airports in major cities present ideal airports for new entrants: They are not dominated by incumbents, airport slots are available, landing fees tend to be much lower and airport congestion is less acute. With the customer base as close as for dominated hubs, entrants can better exploit their operational efficiency and cost advantages at such airports. It is in the incumbent's interests to gain influence over secondary airports in order to neutralize these advantages: The local incumbent may, for instance, enter the same secondary airports (as BA did with City airport in London), either with its own equipment or through franchisees. The operator of the dominated hub may be the same as the operator of the secondary airport (for example BAA for both Heathrow and Luton). Pressure from the dominant client or conflict of interest may lead the operator to increase landing fees at the secondary airport as well. Finally, European-wide air traffic control may privilege dominated hubs over secondary airports with regard to navigation.

      Air traffic control

      Air traffic control is a critical bottleneck in air traffic and provides the parties that control it with the opportunity to exploit the limitations of air space. The inefficiencies that result from airport congestion may be either alleviated or exacerbated by air traffic control. If incumbent airlines, or a consortium of incumbents, were to operate air traffic control in Europe, conflicts of interest may appear. Incumbents may tend to privilege their own airlines by making them fly fewer loops before landing or by letting them take off before other airlines' planes on the runway. For no-frills entrants, where operational efficiency and quick turnaround of planes are so critical, having to wait longer on the ground or having to fly an extra waiting loop may become a critical disadvantage. Incumbents, therefore, will have a strong interest in controlling air traffic.

VIII.3.1 Liberalization of European air traffic has failed

      As we have tried to show throughout this thesis, the liberalization of European air traffic did not yield the benefits that were put forward by advocates of liberalization: The entry of new competitors was neither pervasive nor lasting; the quality of service did not improve significantly, ticket prices did not come down; and convenient point-to-point connections with larger, comfortable and modern aircraft remained elusive. In addition, bottlenecks in air traffic worsened, with passengers spending record times in the air circling around airports or on the ground waiting for ever scarcer take-off slots to become available.

      The post-deregulation period in the US after 1978 had experienced quite the opposite effects. At the least in the short run, massive entry from new airlines occurred, creating real and serious competition for the major carriers. More than that, the industry incumbents competed against each other in a cut-throat manner, with pricing policies regularly based on marginal costs. Although the industry progressively consolidated, the profound impact of deregulation in the US of lowering ticket prices and enabling true competition on high-density routes cannot be denied.

      It can be argued that gradual deregulation may have its advantages over a total, one-off procedure. However, the objective goal of any liberalization effort must ultimately be to make markets at least contestable, no matter how gradual the approach chosen for deregulation. One may also stress that market contestability is already very much towards the laissez faire end of the regulatory spectrum, when compared with the equilibrium ideal of perfect competition. With perfect competition requiring at least two, obviously non-colluding firms, contestability would tolerate it if only one was presently operating in the market. However, as long as markets do not become contestable, incumbents can act as if entry would not occur, and competition (or contestability in this case) will not deploy its forces to increase public welfare. It is noteworthy that these conclusions, which appear modern and derived from a very topical market inefficiency, have already been formulated by Bain (1956, p. 206): "What is suggested here is that a somewhat more general and comprehensive attention might be given under the law to the preservation of a socially desirable condition of entry to our industries - the preservation of an effective degree of potential competition."

VIII.3.2 Economic barriers to entry replace regulatory barriers

      As we have tried to show in this thesis, entry barriers were at the origin of market imperfections that rendered the different routes less contestable. Before deregulation, new airlines were prevented from entering city pairs, because they were governed by bilateral agreements. Thus regulatory barriers to entry were replaced by new "economic" barriers, created by the more able among the industry incumbents. The strategic behaviour of these incumbents leveraged on existing barriers, which were not subject to explicit deregulation. These barriers were solidified and extended, notably by organizing operations around dominated airports and creating tremendous scales of operation in such a setting. Such strategic behaviour and the economic barriers that it created caused industry structure to change with it.

VIII.3.3 Why privatization will not resolve these market imperfections

      With these market inefficiencies becoming worse in the years after 1998, stakeholders in air traffic, including advocates of liberalization, could not continue to ignore them. Along the lines of what has been said above, it would be surprising to see such stakeholders proposing the elimination of economic barriers to entry to produce contestability in the European market. They may, probably rightly, argue that new entrants at the most important European hubs would probably not reduce welfare losses in the form of airport congestion. In the light of a profoundly changed market structure, it is almost impossible to reverse hub-and-spoke networks now, both practically and economically. Nevertheless, with welfare losses so obvious, these advocates are pressed to propose solutions. They are proposing the same sort of remedy that was the cause of the current situation: further liberalization, especially with regard to air traffic control (Fixing America's airlines 2001).