Академический Документы
Профессиональный Документы
Культура Документы
Abstract
With global annual sales exceeding $65 billion, the large civil aircraft industry is an important economic and strategic element of the
European Union and US economies. Here we de"ne large civil aircraft (LCA) as those aircraft with capacities exceeding 121
passengers and dedicated to the air passenger market served by commercial airlines. This paper employs a transformed log-centered
market attraction model to forecast the US market share of LCA. This model speci"cation ensures that the predicted market share is
in the range [0,1] and that the sum of all predicted market shares is equal to 1.0, both logical process requirements. In this special case,
where there are two producers, the market attraction model becomes a logit regression model. Here we specify the logit regression
model as an autoregressive distributed lag model in which US market share is predicted by quantitative and qualitative predictor
variables, an autoregressive lag operator and a linear trend component. ( 2000 Elsevier Science Ltd. All rights reserved.
0969-6997/00/$ - see front matter ( 2000 Elsevier Science Ltd. All rights reserved.
PII: S 0 9 6 9 - 6 9 9 7 ( 9 9 ) 0 0 0 1 7 - 4
4 A. Fernandez et al. / Journal of Air Transport Management 6 (2000) 3}11
regression. The paper concludes with presentation and former Soviet Union) are not included since their market
interpretation of the results. has historically been, and continues to be, limited to
domestic producers.
Region 1 Region 2 Region 3 Region 4 World Region 1 Region 2 Region 3 Region 4 World Region 1 Region 2 Region 3 Region 4 World
1971 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00
1972 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00
1973 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00
1974 1.00 1.00 1.00 1.00 1.00 1.00 1.00 0.92 0.94 0.95
1975 1.00 1.00 1.00 1.00 1.00 1.00 1.00 0.87 0.83 0.89
1976 1.00 1.00 1.00 1.00 1.00 1.00 1.00 0.85 0.88 0.94
1977 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 0.79 1.00 0.54 0.84 0.78
1978 1.00 1.00 1.00 1.00 1.00 1.00 0.37 1.00 1.00 0.86 0.74 0.77 0.37 0.77 0.65
1979 1.00 1.00 1.00 1.00 1.00 1.00 0.16 1.00 0.84 0.75 0.77 0.63 0.38 0.69 0.61
1980 1.00 1.00 1.00 1.00 1.00 1.00 0.19 1.00 0.61 0.73 0.76 0.60 0.40 0.52 0.54
1981 1.00 1.00 1.00 1.00 1.00 1.00 0.38 1.00 0.56 0.71 0.46 0.81 0.29 0.34 0.39
1982 1.00 1.00 1.00 1.00 1.00 1.00 0.35 0.77 0.76 0.79 0.27 1.00 0.38 0.46 0.41
1983 1.00 0.76 1.00 1.00 0.96 0.96 0.30 0.67 0.79 0.80 0.34 0.64 0.93 0.60
1984 1.00 0.77 0.80 1.00 0.91 0.80 0.47 0.55 0.66 0.68 0.42 0.00 0.29 1.00 0.41
1985 1.00 0.94 0.72 0.71 0.92 0.76 0.59 0.57 0.51 0.66 0.54 0.00 0.34 1.00 0.39
1986 1.00 0.64 0.74 0.56 0.83 0.92 0.65 0.75 0.53 0.76 0.41 0.85 0.65 0.63 0.63
1987 1.00 0.46 0.69 0.18 0.69 0.97 0.82 0.85 0.58 0.86 0.44 0.93 0.74 0.63 0.68
1988 1.00 0.64 0.56 0.14 0.74 0.95 0.88 0.82 0.79 0.90 0.48 0.47 0.60 0.35 0.51
1989 0.88 0.89 0.59 0.21 0.78 0.90 0.67 0.76 0.70 0.77 0.48 0.13 0.43 0.21 0.38
1990 0.57 0.84 0.68 0.32 0.64 0.84 0.44 0.78 0.34 0.62 0.49 0.41 0.37 0.16 0.41
1991 0.31 0.82 0.76 0.35 0.55 0.87 0.53 0.84 0.13 0.70 0.38 0.73 0.43 0.28 0.46
1992 0.45 0.98 0.72 0.62 0.63 0.90 1.00 0.85 0.30 0.88 0.36 0.70 0.44 0.48 0.47
A. Fernandez et al. / Journal of Air Transport Management 6 (2000) 3}11
1993 0.63 0.77 0.59 0.44 0.63 0.92 1.00 0.77 0.67 0.88 0.47 0.38 0.31 0.55 0.40
1994 0.66 0.64 0.59 0.33 0.58 0.86 0.91 0.67 0.79 0.80 0.34 0.27 0.30 0.63 0.42
1995 0.73 0.57 0.56 0.53 0.63 0.77 0.83 0.58 0.69 0.71 0.50 0.62 0.41 0.63 0.54
the inherent di$culty in collecting and quantifying this The individual aircraft data for COST and EFF were
type of data for the LCA industry did not allow their grouped according to US or foreign manufacture and
consideration in the present study. averaged on an aircraft type basis, e.g., yearly values of
COST for US Type 7 aircraft. A potential problem with
the direct use of COST and EFF is the high degree of
4.3. Endogenous and exogenous explanatory variables correlation among the respective predictor variables of
the producers (Owen et al., 1995). To overcome this
Adopting an LCA producer's perspective, some of the problem, these two explanatory variables were evaluated
explanatory variables in this study may be classi"ed as in the model development as ratios of US to foreign
endogenous and exogenous depending on a producer's average values. The variable COST was calculated as the
capability of directly impacting the respective factor. The ratio of cost e!ectiveness of US aircraft of a particular
two exogenous variables included in this study are the type to the corresponding cost e!ectiveness of foreign
relative strength of the US dollar relative to the Euro- aircraft (therefore unitless). The coe$cient for COST is
pean Currency Unit (ECU), and the global average jet expected to be negative, re#ecting decreasing US market
fuel cost. The variable $/ECU is the average US dollar to share as US aircraft become more expensive relative to
ECU exchange rate, a factor impacting the competitive- their foreign competition. The EFF variable was sim-
ness of US producers in the global market. Yearly values ilarly calculated as the ratio of the average operational
of $/ECU are taken from the The European Commission e$ciency of US aircraft of a particular type to the corre-
(1998). A weak dollar relative to the ECU is conven- sponding operational e$ciency of foreign aircraft (there-
tionally thought to bolster the position of US producers fore unitless). The coe$cient for EFF is expected to be
in the global market. The sign of the coe$cient for the positive, re#ecting increasing US market share as US
variable $/ECU is thus expected to be positive. The aircraft become more fuel e$cient relative to their foreign
second exogenous variable, FUEL, is the average global competition.
costs of jet fuel (expressed in units of constant 1995 The "nal group of endogenous explanatory variables
dollars per imperial gallon) and is taken from The Avi- re#ects the potential market share impact associated
ation and Aerospace Almanac (1997). This variable is with the introduction of a new aircraft model by LCA
a proxy for the fuel e$ciency of US aircraft relative to producers. According to the US International Trade
that of foreign producers, since the price of jet fuel is Commission (1998, pp. 2}8), `the "rm that makes a
thought to drive purchasers towards more e$cient air- successful &"rst move' typically garners the largest sharea
craft. The sign of the coe$cient for the variable FUEL is of a new or existing market. Two dummy variables,
expected to re#ect the relative e$ciency of US aircraft NEWUS and NEWFOR, were included to control for
with respect to its foreign counterpart, where a positive the impact of new market introductions by US and
value would be indicative of perceived greater opera- foreign producers, respectively. The variables NEWUS
tional e$ciency of US aircraft. and NEWFOR take on the value 1 in the year in
The second set of candidate explanatory variables are which any US or foreign LCA producer, respectively,
considered endogenous to LCA producers, that is, con- introduces a new aircraft model of a particular type.
trollable by either US or foreign LCA producers. Accord- Data for the NEWUS and NEWFOR variables were
ing to the US International Trade Commission (1998), collected from Jane's All the World's Aircraft (1997). The
a decisive factor in airline purchase decisions is seat-mile coe$cient for NEWUS is expected to be positive (re#ect-
operating costs over the aircraft lifecycle. Two variables ing an increased US market share during the "rst year
were included to represent aircraft operating costs over a new US aircraft model is available), and negative for
its lifecycle. The variable COST proxies for aircraft NEWFOR (re#ecting a relative decrease in the US
acquisition cost and is de"ned as an aircraft's initial market share during the year a new foreign aircraft is
purchase cost divided by the product of its range introduced).
and passenger capacity (i.e., has units of $'s per
passenger-mile). The variable COST is a measure of an 4.4. Special control variables
aircraft's cost e!ectiveness in ful"lling its operational
mission, where lower values are preferred. The variable A variable was included to capture the deterministic
EFF proxies for aircraft direct operating costs and is trend components of US market share over the sample
de"ned as an aircraft's available seat-miles (ASM) period. The variable takes on the values 1, 2,2, where
divided by its fuel capacity in imperial gallons (i.e., has "1 for the "rst year in which the global US market
units of passenger-miles per gallon). Data regarding share for a particular aircraft type is not 1.0. For
COST and EFF were collected from Jane's All the World's example, for Type 7 aircraft the variable "1 for the
Aircraft (1997), NASA's Aviation System Analysis year 1973. The coe$cient for is expected to be negative
Capability (ASAC) system (Wingrove et al., 1998), and for those aircraft types in which there was US market
Fernandez (1997). share erosion over the sample period.
8 A. Fernandez et al. / Journal of Air Transport Management 6 (2000) 3}11
The variable SH is included as a predictor vari- The market attraction model shown in Eq. (3) satis"es
US,(t~1)
able to model any autocorrelation exhibited by the the logical consistency requirements imposed by the pro-
dependent variable. Lagged market share as an indepen- cess but does not lend itself to model development.
dent variable is cited in other market share models, and Cooper and Nakanishi (1988) have used a log centering
may be interpreted as a proxy for lagged values of other transformation to linearize this equation, reducing it to
explanatory variables in the model (Owen et al., 1996; a form amenable to the application of standard regres-
Chang, 1984). sion techniques. Allowing the response coe$cient b in
Lastly, for each aircraft type a set of four dummy Eq. (3) to represent the di!erential e!ect of producer j's
variables is included to identify the four world regions, action in terms of its kth predictor variable, the log
controlling for di!erences in global and regional US centered market attraction model can be written as
market share for a particular aircraft type. The dummy
variables REG1 through REG4 assume a value of one m K
ln SH"aH# + + b ln X #eH, (4)
according to the world region being considered. The US it i kij kjt it
j/1 k/1
share of the global market results whenever all four
where, SH"S /SI , SI is the geometric mean of S i,
dummy variables take on the value of zero. The respect- it it t t it
aH"a !a6, a6 is the arithmetic mean of a i, and
ive coe$cient signs are expected to vary according to i i i
eH"ln (e /e8 ), e8 is the geometric mean of e i.
whether the US market share in a particular region is it it t t i
In an application such as this in which there are two
greater or smaller than that for the world as a whole.
producers (i.e., US and foreign producers), Eq. (4) degen-
erates into a logit regression model of the form
A B C D
5. Model speci5cation S m K
ln it "2 aH# + + b ln X #eH . (5)
1!S i kj kjt it
Three market share models forms have been widely it j/1 k/1
cited in the literature: the linear, multiplicative and mar-
ket attraction models (Lilien et al., 1992; Owen et al., 5.1. Model adaptation to the LCA market
1996). The general form of these market share models,
respectively, is given by Previous studies have indicated that LCA producer
market shares exhibit strong autocorrelation and have
K a deterministic time trend component (Fernandez, 1997;
S "a # + b X #e , (1)
it i k kit it Wingrove et al., 1998). For model development we thus
k/1
used Eq. (5) within a general autoregressive distributed
K
S "eai < (X )bk e , (2) lag (ADL) framework, similar in form to other market
it kit it share regression studies (Owen et al., 1995). The ADL
k/1
eai <K (X )bk e model incorporates the lag e!ects of the predictor
S " k/1 kit it , (3) variables on market share, and allows consideration of
it +I [eai <K (X )bk e ]
i/1 k/1 kit it autocorrelation.
where, S is the market share of producer i in period t, The "nal model speci"cation is of the form:
it
a the constant term of producer i, i"1, 2, 2, I, b the
i k K
coe$cient for kth predictor variable, X the kth pre-
kit ln SH "aH # + b ln XH #ln SH #j#eH ,
dictor variable for producer i in period t, and, e the error US,t US k kt US,(t~1) US,t
it k/1
term for producer i in period t. (6)
The linear model, Eq. (1), speci"es a producer's market
where,
share as a linear function of predictor variables, whereas
A B
the multiplicative model, Eq. (2), speci"es a producer's S
ln SH "0.5 ln US,t ,
market share as the product of predictor variables. US,t 1!S
An important consideration when modeling market US,t
share is that there should be logical consistency between S the market share of US producers in period t,
US,t
the predicted market share and the realities of the phys- XH the kth explanatory variable in period t, j the coe$c-
kt
ical process. In particular, the predicted market share ient for the time trend, with units of per year, and, the
must be in the range [0,1] and the sum of the predicted index for time.
market shares for all producers must be equal to 1.0. The form of Eq. (6) is such that SK , the predicted
US,t
Neither the linear nor the multiplicative models guaran- value of the US market share, is always in the range [0,1].
tee satisfaction of these requirements, although there are Furthermore, with two producers (viz., US and foreign
techniques for ex-post modi"cation of predictions to producers) the logit regression model also satis"es the
ensure logical consistency (Beckwith, 1973; Lee#aung requirement that the sum of all producer market shares
and Reuyl, 1984). be equal to 1.0.
A. Fernandez et al. / Journal of Air Transport Management 6 (2000) 3}11 9
6. Estimation and results ket share of LCA using a multiplicative model, of the
form shown in Eq. (2), and obtained an R2 value of 47%
6.1. Overall model results (Wingrove et al., 1998).
Separate logit regression models were developed for 6.2. Coezcient estimates of parameters
each of Type 5, 6 and 7 aircraft, respectively. The corre-
sponding parameters for the three logit regression mod- Coe$cient estimates for the logit regression para-
els, as speci"ed in Eq. (6), were estimated by ordinary meters, with associated t-values in parenthesis, are shown
least squares (OLS). Table 4 presents a summary of the in the "rst 15 data rows of Table 4. With relatively few
results for the most parsimonious logit regression model exceptions, all parameter estimates yield the expected
for each aircraft type. The explanatory power of the three coe$cient sign and are signi"cant at the 95% level or
regression models is not exceptional, with a coe$cient of above. In their most parsimonious form, none of the three
determination, R2, ranging from 58.4% to 75.2%. How- models included all of the candidate predictor variables.
ever, given the volatility of the aircraft market and the The two exogenous variables, $/ECU and FUEL, are
di$culty in collecting data, these relatively low values of included in only two of the models. The variable $/ECU
R2 are not disturbing. In all three cases the F-test results is included in the Type 6 model (where its coe$cient
(as shown in the bottom row of Table 4) provide further estimate is statistically signi"cant at the 95% level) and,
con"dence that the regression estimates are statistically as expected, has a positive coe$cient. A positive value
signi"cant as a whole. Furthermore, the resultant values implies an increase in the US market share as the dollar
of R2 exceed previously reported results by other re- weakens relative to the ECU. The variable FUEL is
searchers. A study by the US International Trade Com- included in Type 5 and 6 models, but its coe$cient is
mission (1993) modeled the US share of the global LCA signi"cant at the 95% level only for Type 5 aircraft
market using a linear model, of the form shown in Eq. (1), (although the Type 6 coe$cient estimate has a p-value of
and obtained R2 values in the range of 13}61%. A more 0.94). The positive estimates in both instances seem to
recent report prepared for NASA modeled the US mar- indicate that US aircraft are more fuel e$cient than their
foreign counterparts, an observation con"rmed for Type
5 and 6 aircraft during most of the study period.
Table 4 The two quantitative endogenous variables, EFF and
Logit regression coe$cients COST, are included in the three logit regression models,
Parameter Type 5 Type 6 Type 7
in all cases exhibiting the expected sign for the coe$cient
estimate. The variable EFF is included in all models,
aH 1.217 !2.3152 0.868 being statistically signi"cant at the 95% level in all three
US
(1.19) (!2.69) (2.07) cases. Interestingly, the signi"cance levels for the Type
b HUS,(t~1) 0.2829 0.51483 0.520 5 and 6 aircraft is much greater than that for Type 7. This
S
(1.98) (6.07) (6.01)
b$ 0.6196
may perhaps explain why FUEL (a proxy for aircraft
@ECU
* *
(2.25) #ight-time operating costs) is also signi"cant for only
b 0.9691 0.6838 these two aircraft types. The variable COST is only used
FUEL
*
(1.82) (1.55) in the models for aircraft Types 5 and 6; however, it is
b 2.392 2.882 3.59 only statistically signi"cant at the 95% level for the Type
EFF
(2.499) (2.40) (1.77)
b !1.868 !2.496
6 regression model.
COST
*
(!1.23) (!2.26) The results for the two dummy endogenous variables,
b 0.100 0.262 NEWUS and NEWFOR, are somewhat more puzzling.
NEWUS
*
(0.42) (2.91) NEWUS is only included in the models for aircraft Types
b 0.291 !0.4669 !0.429 5 and 7, although only statistically signi"cant at the 95%
NEWFOR
(1.95) (!2.37) (!2.61)
b !0.2735 0.1097 !0.044
level for Type 7. In both these models the respective
REG1 coe$cient estimates for NEWUS are positive (as ex-
(!1.47) (0.98) (!0.43)
b 0.1386 !0.02574 0.0738 pected), implying an increase in US market share during
REG2
(0.97) (!2.33) (0.63) the year a new US aircraft model is introduced. The
b !0.0802 !0.0943 !0.0828 variable NEWFOR is included in all models, and is
REG3
(!0.54) (!0.91) (!0.84)
b !0.5138 !0.2341 0.0513
statistically signi"cant at the 95% level in all three cases.
REG4 However, the coe$cient estimates are negative for air-
(!2.71) (!2.28) (0.50)
j !0.02401 0.08936 !0.0454 craft Types 6 and 7 but positive for the Type 5 model.
(!0.40) (2.52) (!2.34) The latter result is surprising, as the introduction of
R2 64.2% 75.2% 58.4% a new foreign aircraft model was expected to result in
F 6.37 16.50 12.46
(0.000) (0.000) (0.000)
a competitive disadvantage to US producers and a result-
ant decrease in US market share during that year. A
10 A. Fernandez et al. / Journal of Air Transport Management 6 (2000) 3}11
possible explanation may be that in some cases the mar- signi"cant market share. The emergence of a strong
ket may perceive that a producer is prematurely accept- foreign competitor and the large capital investment asso-
ing orders for a new aircraft, and thus postpone purchase ciated with LCA production has been the catalyst for
decisions to a later year. major structural changes in the industry. The most recent
All three models resulted in statistically signi"cant change is the merger of Boeing and McDonnell Douglas,
positive coe$cients for SH , the value of the single which reduced the global LCA industry to Boeing and
US,(t~1)
year lagged US market share. These results corroborate Airbus Industrie. These two producers are vying for
previous studies citing a high degree of autocorrelation relative competitive position in the various LCA market
for the US market share of LCA, and highlight the segments, with market share used as an indicator of
importance placed by producers on capturing and retain- their relative competitive strength (March, 1990; US
ing market share (Fernandez, 1997; US International International Trade Commission, 1998).
Trade Commission, 1993,1998). The coe$cient estimates In this paper we have developed models for the US
for , the deterministic trend component for the US market share of Type 5, 6 and 7 aircraft, respectively,
market share are included in all models, but are statist- within di!erent LCA market segments. A market attrac-
ically signi"cant at the 95% level only for aircraft Types tion model has been speci"ed in order to allow for the
6 and 7. The negative coe$cient for Type 7 re#ects the required characteristics of the market share measurement
general erosion of US market share for this type aircraft metric. Some clear patterns and corroboration of prior
(over the study period), while the positive coe$cient "ndings have resulted from the model development.
estimate for Type 6 re#ects an increasing US market Firstly, the importance of life-cycle costs as a driving
share over time. The data shown in Table 2 con"rms factor for LCA producer market share is clearly evident.
these general time trends. The three explanatory variables in this study used as
The estimation results for the control variables REG1 proxies for life-cycle costs (FUEL, EFF and COST) were
through REG4, intended to distinguish among world statistically signi"cant in all three models. It is evident
regional markets, indicate little across-the-board LCA that aircraft purchasing decisions are driven in large
geographic market di!erentiation. Only three of the 12 measure by initial acquisition cost e!ectiveness and di-
coe$cient estimates are statistically signi"cant at the rect operational costs. Secondly, the impact of foreign
95% level. The most distinct results are for Region 4 (the exchange rate #uctuation (as measured by $/ECU, the
Rest-of-the-world) where the coe$cient estimates are dollar to ECU exchange rate ratio) was not found to be
statistically signi"cant for aircraft Types 5 and 6. The a signi"cant factor across all aircraft types. It may be that
coe$cient estimates for both these aircraft types are "nancing instruments (e.g., currency hedges, and long-
relatively large in magnitude and are negative, implying term "nancing) are to some degree insulating aircraft
a generally very weak US market position in this region. purchasing decisions from the vagaries of yearly currency
The only other statistically signi"cant estimate (at the #uctuations. Thirdly, a new aircraft model was found to
95% level) is for Region 2 (Europe) of Type 6 aircraft, impact market share in the launch year. These results
where the negative coe$cient again implies a generally con"rm the importance placed by LCA producers on the
lower US market share. However, the US producer timing of new aircraft introductions. Lastly, there do not
weakness in Europe is tempered by the relatively small seem to be signi"cant market share di!erences among the
magnitude of the coe$cient estimate. The only other coef- various LCA regional markets. This may re#ect consis-
"cient estimate of relative merit is that for Region 1 (the tent purchasing patterns across all airlines, or it may be
US) of Type 5 aircraft. This estimate has a p-value of 93% indicative of poor homogeneity amongst the geographic
and is relatively large in magnitude (with negative sign), clusters. This study adopted the industry accepted con-
implying that foreign producers have been relatively suc- vention for world regional markets, with no attempt to
cessful in the US Type 5 aircraft market. None of the other aggregate based on other considerations.
coe$cient estimates for the dummy regional market vari-
ables is statistically signi"cant. These conclusions about
the position of US producers relative to foreign competi-
tion are corroborated by the data shown in Table 2. References
Cooper, L.G., Nakanishi, M., 1988. Market Share Analysis. Kluwer March, A., 1990. The future of the US aircraft industry. Technology
Academic Publishers, Boston, MA. Review 93 (1), 27}36.
The European Commission, 1998. O$cial ECU and Euro Exchange Owen, A., Chowdury, K., Garrido, J., 1996. A market share model for
Rates. http://europa.eu.int/comm/dg02/document/ecu/xecupage.htm vegetable and tropical oils. Applied Economic Letters 3, 95}99.
Fernandez, Abel, A., 1997. Market Share Study: Commercial Air- US International Trade Commission, 1993. Global Competitiveness of
craft Industry. Unpublished technical report, Logistics Manage- US Advanced-Technology Manufacturing Industries: Large Civil
ment Institute Contract No. 9719. Old Dominion University, Aircraft. US International Trade Commission, Publication 2667.
Norfolk, VA. Washington, DC.
Jane's All the World's Aircraft, 1997. Jane's Information Group US International Trade Commission, 1998. The Changing Structure of
Limited, Surrey, UK. the Global Large Civil Aircraft Industry and Market: Implications
Jet Information Services, Inc., 1997. World Jet Inventory at Year End for the Competitiveness of the US Industry. US International Trade
1996. Woodinville, Washington, DC. Commission, Publication 3143. Washington, DC.
Lee#aung, P.S.H., Reuyl, J.C., 1984. On the predictive power of market Wingrove, E., Gaier, E., Santmire, T., 1998. The ASAC Carrier Invest-
share attraction models. Journal of Marketing Research 21, ment Model (Third Generation). NASA Langley Research Center
211}215. CR-1998-207656. Hampton, VA.