Sitraer 7 (2008) 100-111 Tr.

312

DEMAND FORECASTING IN REGIONAL AIRPORTS:

DYNAMIC TOBIT MODELS WITH GARCH ERRORS
Matthew G. Karlaftis, Ph.D.
Department of Transportation Planning and Engineering
National Technical University of Athens
Tel: +302107721280, e-mail: mgk@central.ntua.gr

ABSTRACT
In this paper we discuss the general issue of forecasting highly seasonal demand in regional airports, where peak
flows approach airport capacity. For this, we propose a modeling combination, dynamic Tobit models with
GARCH errors/disturbances, that is able to capture many of the shortcomings of most traditional models.
Models are calibrated using monthly passenger and flight data for a 20 year period for the airport of Corfu in
Greece, where traffic over the summer approaches airport capacity and seasonal fluctuations in demand are very
intense. Results show that: i. Not explicitly accounting for seasonal variations in demand or for traffic
approaching capacity may significantly bias model parameter estimates and affect demand predictions; and, ii.
Improved demand model specifications are an invaluable tool in obtaining more accurate demand estimates.

1. INTRODUCTION
Forecasting is at the heart of the planning and design process at airports. Airport terminals,
runways, freight storage facilities, parking lots, and even the roadway network to and from an
airport are all based on the forecasts for the airport. Forecasts of passenger volumes are
translated to space requirements for the terminal building facilities, while forecasts of aircraft
movements are translated to the runway, taxiway, and apron needs, as well as to the need for
air traffic control systems.
When planning and designing airport terminal there is a significant number of traffic
characteristics that must be forecasted: total number of passengers for the design period,
domestic, commuter and international passenger ratios at peak hours, seasonal variations in
demand, volumes of transfer and/or transit passengers for each type of traffic, number of
passengers, bags and well wishers, distribution of dwelling times of passengers, and
sometimes origin and destination of flights for immigration, customs, and health control
purposes (Tsic, 1992). These forecasts are used to determine the space requirement for new
terminals or the expansion of existing facilities.
It is evident that the forecasting process can be the most critical factor in the development of
the airport (Howard, 1974). Mistakes made in this phase of the process can be very costly and
damaging for local economies. Underestimating demand can lead to increased congestion,
delay, and lack of storage facilities, as it happened in Venezuela in 1974. The discovery of oil
resulted in a dramatic and unforeseen increase of the freight volumes handled by the Caracas
airport. The planned storage facilities were insufficient to handle this increased demand, and
so the cargo was stored in areas where it was either destroyed or stolen. Overestimating
demand could also create significant problems. Forecasts of passenger demand for the
Newark airport were so high that the newly constructed airport remained empty for a number
of years (de Neufville, 1976). Errors in the forecasting process can lead to long delays or to
empty terminals: both these cases can cause significant damage to the economy of an airports
hinterland that depends on the successful operation of that airport.

100

Sitraer 7 (2008) 100-111 Tr. 312

There are three main approaches for forecasting civil aviation traffic: market surveys, trend
projections and econometric relationships (models). Particularly the former, econometric
methods, with regression as the main tool, have enjoyed years of extensive use
internationally; however, there are cases where traditional regression time-series or ARIMA
analyses yield questionable forecasts, mainly as a result of their inherent assumptions. Such a
case is in forecasting heavily seasonal passenger traffic in regional airports, where intense
fluctuations (volatility) exists in daily and monthly passenger traffic, and where both
regression and classical time-series analyses cannot successfully capture the nature of
demand.
In this paper we propose a modeling combination, dynamic Tobit models with GARCH
errors/disturbances, that is able to capture many of the shortcomings of traditional models. A
tobit model assumes a dependent variable with censoring (where, for example, passenger
traffic in certain months approaches airport capacity). The GARCH extension for the Tobit
model disturbances allows us to model the volatility in seasonal passenger traffic without
jeopardizing model robustness. Models are calibrated using monthly passenger and flight data
for a 20 year period for the airport of Corfu in Greece, where traffic over the summer
approaches airport capacity and seasonal fluctuations in demand are very intense.

2. BACKGROUND
The goal of the review is to identify the work done for airport-specific forecasting, which is
the focus of this paper, rather than for corridor forecasting. Corridor forecasting is concerned
with the demand for air-travel in a given travel route (corridor), say the Rome (Italy) - Athens
(Greece) corridor. While this type of forecasting is certainly of great interest to airline
companies, it is not as useful to airport planners who are mainly concerned with the aggregate
passenger forecasts for a given airport (Whitford, 1990). Knowledge of the aggregate
passenger forecasts will provide planners with the necessary guidelines for the planning and
design of the entire airport facilities.
The key issue in the development of successful airport forecasts is the ability to identify those
factors that influence the change in the volume of passenger traffic at an airport. There are
two major approaches in the development of forecasting models: Simple Time Series (STS)
and Causal Modeling. STS methods are the most widely used methods for predicting charter
air-travel demand, and assume that history repeats itself, in that the underlying stochastic
structure of the data does not change with time. The common characteristic, and weakness, of
all time-series methods is that they ignore the determinants of demand such as fares, income,
GNP, and do not attempt to explain the causes of change in demand. Causal modeling
attempts to determine, from given data, causality or least explanation when analyzing the
relationship between certain data; that is, the relationship between two or more variables is
used to predict one variable (the dependent) from the others (independent).
One of the first published causal modeling efforts concerned with airport-specific forecasting
was by Jacobson (1970). His linear regression based model predicted the trips generated at an
airport (dependent variable) based only on the average airfare per mile for all routes in the
United States, and the total income per capita for the airport catchment area. The model was

101

Sitraer 7 (2008) 100-111 Tr. 312

calibrated with eighteen years of data from the airports of Virginia. A rather similar
regression model was developed by Haney (1975) to analyze airport specific demand for the
St. Louis, Missouri, airport. The model is a typical airport forecasting model in that the
explanatory variables are socio-economic in nature and refer to the metropolitan area served
by the airport.
A different type of forecasting model was developed by Thomet and Sultan (1979) for the
Riyadh International Airport in Saudi Arabia. The model used imports and exports of crude
oil and petroleum products as the sole independent variable, to account for the fact that most
travel to and from Saudi Arabia was related to the oil emporium.
Another interesting example of placing a special effort in the direction of identifying key
growth factors unique to an area is the case study of Mexico citys airport traffic by Zuiga et
al. (1978). Approximately 20% of the population and wealth of Mexico are concentrated in
the capital, and the fast growth rates of these characteristics certainly influence the future
airport traffic. The models that were developed for domestic traffic included as variables the
population of Mexico City and the cost of transportation in constant pesos. Mellman et al.
(1980) developed a model for Bostons Logan airport. Unlike many other airports where
transit passengers are a significant portion of the airports traffic, Logans traffic is mostly
Boston based. Explanatory variables in this case related air-travel to the socio-economic
characteristics of the Boston region.
Using a different approach, several models were built using STS modeling for air-travel
demand forecasting. STS modeling has been used extensively in the literature in those cases
where causality could not be established, or where economic information on the airports
catchment area was not readily available (ICAO, 1991). As an example, passenger forecasts
for the Cornavin Airport in Geneva, Switzerland, have been obtained by STS methods (ICAO,
1985); STS methods were also used extensively in the forecasts for the fourth airport in the
Chicago region (al Chalabi and Mumayiz 1991, al Chalabi 1993, Whitford 1990). Moreover,
forecasts of air-travel growth for the Airbus industries are often obtained using STS methods
(Lenormand, 1989).
Finally, Karlaftis et al. (1996) examined the predictive ability and forecasting accuracy of airtravel demand models. In particular, they developed an analytical framework for developing
econometric models, and used post-fact analysis to test the accuracy of the models. Statistical
data describing air-travel patterns for two major international airports, the Miami and
Frankfurt International Airports, were used to demonstrate the effectiveness of the proposed
models. In addition, the effect of external factors such as the deregulation of the air-transport
industry was examined. The results suggested that simple models with few independent
variables perform as well as more complicated and costly models, and that external factors
have a pronounced effect on air-travel demand.
3. DATA AND METHODOLOGY
3.1. The Data
The Greek airport system comprises of 35 airports that service commercial flights. From these
airports approximately one third are international; that is, they are open to direct flights from
airports outside of Greece. The airports that were mostly affected by the increase in direct

102

Sitraer 7 (2008) 100-111 Tr. 312

140000

1000000

120000
800000

100000
80000
International

400000

Domestic

40000
200000

20000

12000

4000
3500

10000

3000

8000

2500
6000

International

2000
1500
1000

4000
Domestic

2000

500
0

19
8
19 9
9
19 0
9
19 1
92
19
9
19 3
9
19 4
9
19 5
9
19 6
9
19 7
9
19 8
9
20 9
0
20 0
0
20 1
02
20
0
20 3
0
20 4
0
20 5
06

4500

19
89
19
90
19
91
19
92
19
93
19
94
19
95
19
96
19
9
19 7
98
19
99
20
00
20
01
20
02
20
03
20
04
20
05
20
06

60000

600000

14000

5000

International AIrcraft Arrivals

1200000

Domestic Aircraft Arrivals

Domestic Passenger Arrivals

160000

International Passenger Arrivals

charter flights were the major airports located in popular resort areas. These airports include
the airports of Zakynthos and Kerkyra (Corfu) islands (both located on the Ionian Sea off the
west coast of mainland Greece), the airports of Kos and Rhodes (both located on the Aegean
Sea off the east coast of mainland Greece), and the airport of Chania (in the island of Crete, at
the south end of Greece). The airports off the west (east) coast serve as destinations to
passengers visiting the west (east) islands, because tourists can either remain on those
islands or take small boats to visit the neighboring islands (a type of intermodal hub-andspoke system). Similarly, passengers to the airport of Chania are visitors of Crete, the largest
Greek island. Figures 1 and 2 depict the evolution of annual and monthly data for a variety of
traffic categories (domestic flight and passenger arrivals, international passenger arrivals, and
so on). What is of particular interest is the notion (Figure 2), of very intense seasonal
characteristics for all the measures examined, coupled with similar peak values on passenger
traffic. Seasonality suggests that international passenger arrivals is almost zero in the winter
mnths, while it approaches airport capacity in the summer months.

Year

Year

Figure 1: Annual Data on Domestic and International Arrivals

250.000
International

20.000

Domestic

200.000

15.000

150.000

10.000

100.000

5.000

50.000

3
3
8
6
1
9
4
8
4
1
7
0
99 1 00 1 00 10 1 10 1 1 02 1 03 10 3 04 10 4 10 5 1 05 10 6
92
96 r 9 7
89 r 9 0
95
1
t9
t9
b 9 ep 9 pr 9 ov 9 u n
n 9 ug
n 8 ug
y 9 ec 9 Ju l
y 9 ec 9 Ju l
l
a
a
g
r
p
b
v Ju n Jan
Ja
J
Ja
ar Oct
S
Fe
ay Dec
Ju
D
Oc Ma
A
A
D
Oc Ma
A
M
N
M
Se
Fe
Au M
Ap No
M

Year

Figure 2: Monthly Data on Domestic and International Arrivals

103

International Passenger Arrivals

Domestic Passenger Arrivals

25.000

Sitraer 7 (2008) 100-111 Tr. 312

The origin of the tourists visiting Greece is important in attempting to identify the key factors
affecting the dramatic increase in island airport traffic. From Figure 3 (OECD, 1994), it is
evident that approximately 90% of the tourists originate in Europe. Therefore, explanatory
variables were selected that account for socio-economic characteristics for Europe.
Interestingly, the Greek economy is closely tied to the revenues generated by the tourist
industry; the ratio of private consumption in foreign currency to total private consumption, a
strong indicator of the importance of tourism to the National Economy (Figure 2), clearly
indicates this importance.

Figure 3: Origin of Tourists Visitng Greece

Figure 4: Ratio of Private Consumption in Foreign Currency to Total Private Consumption

104

Sitraer 7 (2008) 100-111 Tr. 312

The sample period for the monthly analysis presented is the period 1989-2006. The dependent
variables in the models were: i. Domestic flight arrivals; ii. Domestic passenger arrivals; iii.
International flight arrivals; iv. International passenger arrivals. Data were provided by the
Greek Civil Aeronautics Board; data are collected on a monthly basis for all Greek airports.
The independent variables that were considered for inclusion in the models were the Gross
Domestic Product (GNP) per capita for Europe, the Personal Consumption Expenditures
(PCE) per capita for Europe, the Disposable Income (DI) per capita for Europe (all measured
in constant 2006 ), Regional Product for Corfu, GDP for Greece, Number of Tourist arrivals
on the island, population, a time trend (taking the value of 1 for 1989, 2 for 1990 and so on),
and monthly dummy variables (to capture monthly variability in the dependent variables).
The above mentioned explanatory variables were selected because they are idely considered
as some of the best descriptors of the ability (and desire) for tourists to travel. The
development of air-travel is also affected by a large number of external economic factors,
government policies and regulations (such as travel advisories and strikes of the air-traffic
controllers), physical phenomena (Greece has faced a large decrease in the number of
international tourists the years when earthquakes have occurred), technological developments
and the operating economics of the air-transportation industry. Unfortunately, these
occurrences cannot be forecasted and, as a result, the models have to be built on measures that
have not only been diligently observed in the past, but also for which forecasts exist
(Kanafani 1981). In preparation and application of the forecasts, these uncertainties have to be
recognized; nevertheless, a forecasters task is to use the data sources readily available, and
apply methods that minimize the range of uncertainty.
3.2. The Methodology
The initial thought and approach in modeling demand for th Corfu airport was to develop
Simple Time Series (STS) models (Karlaftis et al. 1997). But, while STS modeling has been
extensively used in charter passenger travel, questions still remain as to the factors that affect
its demand, as well as its elasticity with respect to those factors. Further, and besides
modeling external influencing factors, the demand characteristics for regional Greek airports,
i.e. intense seasonal variation in demand and peak demand approaching airport capacity,
necessitated the development of a new approach. In this respect, we propose a modeling
combination, dynamic Tobit models with GARCH errors/disturbances, that is able to capture
many of the shortcomings of traditional models. A tobit model assumes a dependent variable
with censoring (where, for example, passenger traffic in certain months approaches airport
capacity); rhe GARCH extension allows us to model the volatility in seasonal passenger
traffic without jeopardizing model robustness.
Limited dependent variables were introduced in econometrics in Tobin (1958) and Amemiya
(1973). Let
yt* = xt + t

(1)

be the underlying latent equation before censoring. The xt is a vector of exogenous variables.
The observed variable is

105

Sitraer 7 (2008) 100-111 Tr. 312

yt = max{ yt* , ct },

(2)

where the ct s are known constants. To allow serial correlation in t and dynamic structure
in the model, one can consider the general formulation
yt* = ( yt 1 , yt* , xt ) + t ,

(3)

where yt denotes a vector of y consisting of its observed current and lagged values from time
1 to t , yt* is the vector consisting of all latent-dependent variables of y* preceding and up to
period t , and xt is defined accordingly. For simplicity, unknown parameters in of the
model are suppressed. As one knows whether yt* has been censored or not at each t , it is
convenient to define a dichotomous indicator I t such that I t = 1 if yt* > ct , and I t = 0 if
yt* ct . If I t = 0 , yt* has been censored and its value is not observed. When I t = 1 , yt* is
observed. A model with serially correlated disturbances such as
yt* = xt + ut , ut = ut 1 + t

(4)

is a member of the general case in Eq. (3), because it implies that

yt* = yt*1 + ( xt xt 1 ) + t after a quasi-difference transformation. Other relevant and
interesting examples are censoring models with ARCH or GARCH disturbances (Engle,
1982; Bollersev, 1986). In a Tobit ARCH model with yt* = xt + t , the conditional variance

t2 of t conditional on past information depends on past disturbances t 1 , t 2 , etc. For an

ARCH ( p ) process,
p

t2 = 1 + j +1 t2 j

(5)

j =1

In a GARCH ( p, q ) process, additional lagged variances introduced as

p

t2 = 1 + j +1 t2 j + p +1+ k t2 k
j =1

(6)

k =1

Dynamic structures can be incorporated into an ARCH or GARCH model. The ARCH in the
mean (ARCH-M) model introduced by Engle et al. (1987) is specified as
yt* = xt + t + t in addition to Eq. (5). A survey of ARCH models is given in Bollerslev et
al. (1995). Without censoring, popular ARCH and GARCH models are estimated by the ML
or generalized methods of moments (GMM). Additional complication is introduced with
censoring. Under censoring, the calculation of underlying residuals for censored observations
are infeasible, as exact values of these residuals are not observed. To avoid this difficulty,
Calzolari and Fiorentini (1996) modified the ARCH process by replacing the lagged t2 j

106

Sitraer 7 (2008) 100-111 Tr. 312

formulation in Eq. (6) by its conditional expected value Et 1 ( t2 j ) , where Et 1 means the
conditional expectation conditional on past information up to t 1 .

4. MODEL ESTIMATION
In Tables 1 and 2 we provide the models that provide the best fit to the data after extensive
testing (for details on model development and testing see Washington et al. 2003).
Coefficients for all variables, including GDP and tourist arrivals, have the expected signs and
are significantly different from 0 at the 90% in most cases. We note here that, in all cases,
both dependent and independent variables were taken in their log forms; this, as the literature
clearly shows (Washington et al. 2003), leads to models were coefficient estimates are
elasticities. In most cases, all dependent variables are inelastic with respect to explanatory
variables such as GDP and arrivals. For example, during the period of the analysis, a 10%
increase in Per Capita GDP for Greece increases domestic passenger traffic at the airport by
4.8%. We also note that, in all models, we included both an annual time trend (to capture
overall movement in the series), and monthly dummy variables to capture monthly variations
in overall traffic levels.
Despite the above corrections, from Tables 1 and 2, we note that neither monthly variation
nor traffic levels approaching capacity where fully captured. This is evident by the statistical
significance (almost all at the 95% significance level), of all Tobit-GARCH parameters for
both domestic and international traffic variables.

Table 1: Domestic Flight and Passenger Arrivals (Monthly Data)

OLS
Passenger Arrivals

Flights
Independent Variablesa
Constant
National GDP
European GDPb
Tourist Arrivals
Time Trend

Flights

Tobit - GARCH
Passenger Arrivals

Coefficient
4.92
0.69

t-stat

Coefficient
-0.16
0.48

t-stat
-0.49
1.41

Coefficient
13.78
0.36

14.87
4.07

Coefficient
2.03
0.45

t-stat

1.06
1.45

t-stat

0.72
0.10
0.05

4.51
3.67
1.47

0.53
0.06
-0.03

4.68
3.02
-1.45

0.59
0.04
0.11

12.1
3.75
16.6

0.31
0.03
0.018

5.17
3.49
1.86

0.0024
0.018

4.67
0.63

0.0026
0.16

4.30
2.02

1.03

10.64

0.865

4.67

1.58
3.27

Monthly Dummiesc
d

Unconditional Variance
Lagged Variance
Lagged Squared
Disturbance
a All variables in log form (double log model; all coefficients are elasticities)
b Weighted average of the GDP of countries of origin for tourists
c All dummies are statistically significant
d GARCH parameters from Eq. (6)

To further examine the differences in modeling airport demand with the two different models
used (OLS versus Tobit-GARCH), we examine the models highlighted in Table 2 (monthly

107

Sitraer 7 (2008) 100-111 Tr. 312

data on international flight arrivals). As is obvious, differences in coefficient estimates for the
two models (and hence of the elasticity estimates for international flights with respect to the
parameters examined) are vastly different. For example, if analysts had used OLS with
monthly corrections (the most frequently used approach for modeling airport demand), there
would have been severe underestimation of both parameter significane (i.e. what the
important parameters for modeling airport demand are) and of the magnitude of elasticity
estimates with respect to a number of parameters. Interestingly, while in the case of
international flights modeled via OLS, the only statisticaly significant parameter is the yearly
trend, indicating an annual 0.5% increase in international flights, in the Tobit-GARCH
estimation approach the results are different. First, the annual trend is decreasing (by 0.2%
anually); second, European GDP is statistically significant, suggesting that an increase in
European GDP increases Corfu Airport International flights (a 10% increase in per Capita
GDP for Europe increases Corfu Airport International flights by 10.2%); and, third, increases
in monthly tourist arrivals to Corfu also help in increasing international flights (an expected
result; a 10% increase in tourist arrivals increase Corfu Airport International flights by
0.17%).

Table 2: International Flight and Passenger Arrivals (Monthly Data)

OLS
Passenger Arrivals

Flights
Independent Variablesa
Constant
National GDP
European GDPb
Tourist Arrivals
Time Trend

Coefficient
-0.18

t-stat

t-stat

-0.32

Coefficient
11.89

0.13
0.056
0.044

0.22
1.06
2.91

-0.76
0.12
0.15

Tobit - GARCH
Flights
Passenger Arrivals
t-stat

0.94

Coefficient
-10.86

t-stat

-9.27

Coefficient
5.79

-0.56
1.01
1.96

1.24
0.017
-0.022

10.77
1.68
-3.49

0.24
0.08
0.04

2.48
2.69
6.552

3.93
2.63

0.01
0.006

4.01
1.79

4.27

2.92

4.35

6.02

Monthly Dummiesc
Unconditional Varianced
0.00025
Lagged Variance
0.105
Lagged Squared
Disturbance
2.939
a All variables in log form (double log model; all coefficients are elasticities)
b Weighted average of the GDP of countries of origin for tourists
c All dummies are statistically significant
d GARCH parameters from Eq. (6)

5. IMPLEMENTATION ISSUES
When planning and designing airport terminals there is a significant number of traffic
characteristics that must be forecasted: total number of passengers for the design period,
domestic, commuter, and international passenger ratios at peak hours, seasonal variations in
demand, volumes of transfer and/or transit passengers for each type of traffic, and sometimes
origin and destination of flights for immigration, customs, and health control purposes. The
aggregate passenger forecasts, such as those presented in this paper, are used to determine the
space requirements for new terminals or the expansion of existing facilities.

108

Sitraer 7 (2008) 100-111 Tr. 312

Forecasting is also very important for the economic development of airline fleets which have
grown significantly in the last few decades. Orders for commercial transport aircraft have
reached an all time high, with delivery rates now backed-up into the next century for some
aircraft types. Future fleet development has an important role in the design of terminals since
the aircraft size is critical to geometric characteristics, such as the main deck elevation and the
lateral clearance at gates due to the aircraft wing-span.
The basis for the development of new and for the expansion of already existing terminal
facilities in the Greek airport system is based on these forecasts. The data necessary for these
forecasts is readily available through the Greek Civil Aeronautics Board and our analysis has
attempted to establish causal relationships between airport demand and a set of explanatory
variables. These explanatory variables are statistically adequate; that is, the data is collected
by the Greek government in a consistent manner at equally spaced time intervals. It is also
important to note that the analysis of time-series data should be done cautiously to account for
the presence of: i. error term serial correlation, ii. intense variability in monthly evolution of
the variables, and iii. cases were demand approaches capacity. If any of these three issues is
systematically ignored, conclusions could be reached that parameter estimates are more
precise than they actually are (loss of parameter efficiency) and, as shown in this paper,
biased parameter estimates could also be obtained (Washington et al. 2003). It is evident that
until actual values of the future air-travel demand are known, a model accuracy cannot be
precisely ascertained. Nevertheless, it is important to ensure the statistical validity of the
forecasting models that will be used in practice, so that a rational basis for the selection of the
appropriate model exists.

6. CONCLUDING REMARKS
The goal of this paper was to examine and analyze the demand for, and the ability to, forecast
passenger traffic and flights at regional airports. In this paper we proposed a modeling
combination, dynamic Tobit models with GARCH errors/disturbances, that is able to capture
many of the shortcomings of traditional airport demand forecasting models. A tobit model
assumes a dependent variable with censoring (where, for example, passenger traffic in certain
months approaches airport capacity). The GARCH extension for the Tobit model disturbances
allows us to model the volatility in seasonal passenger traffic without jeopardizing model
robustness. Models were calibrated using monthly passenger and flight data for a 20 year
period for the airport of Corfu in Greece, where traffic over the summer approaches airport
capacity and seasonal fluctuations in demand are very intense.
Given the nature of the data base and the rather uncertain and variable conditions that exists at
regional airports, the results are surprisingly good. The model is statistically significant and
exhibits good fit to the historical trends. Also, and very important from a statistical
perspective, there is clear evidence that not explicitly accounting for both demand
approaching capacity and for intense monthly variation, affects the efficiency of parameter
estimates and leads to biased and erroneous elasticity measures. These errors could lead
to significant underestimation of future demand at airports.
Further work on charter passenger traffic is certainly warranted, given the generally diverse
and volatile nature of travel at regional airports with high tourist related travel. Forecasts are

109

Sitraer 7 (2008) 100-111 Tr. 312

of course uncertain, yet the establishment of causal relationships gives a more rational and
scientific basis for model development, and makes airport planners more secure in planning
and designing for airports based on the available forecasts. Moreover, careful work should be
done in the direction of international chartered traffic to identify the factors that affect this
type of demand; this will become increasingly important as passenger volumes approach
airports capacities and low cost carriers become an even more dominating force in
international tourist travel.

BIBLIOGRAFIC REFERENCES
al Chalabi, S. 1993, Aviation Demand Forecasting - Traditional and New Approaches (The application of
ground transport methodologies to enplanement forecasting), Submitted to the Transportation Research
Board.
al Chalabi, S. and Mumayiz, S. 1991, Allocation of Enplanement Forecasts to Alternative Airports, Working
paper no. 10 of the Illinois-Indiana Regional Airport study.
Amemiya, T., 1973. Regression analysis when the dependent variable is truncated normal. Econometrica, 41,
997-1016.
Bollerslev, T., 1986. Generalized autoregressive conditional heteroskedasticity. Journal of Econometrics, 31,
307-327.
Bollerslev, T., 1987. A conditionally heteroskedastic time series model for speculative prices and rates of return.
Review of Economics and Statistics, 69, 542-547.
Brocklebank, J.C. and Dickey, D.A. 1986, SAS System for Forecasting Time Series, SAS Institute Inc., Cary,
NC.
Draper, N. and Smith, H. 1981, Applied Regression Analysis, John Wiley and Sons, New York, NY.
Engle, R.F., 1982. Autoregressive conditional heteroscedasticity with estimates of the variance ofUnited
Kingdom in#ations. Econometrica 50, 987-1007.
Greene, W.H. 1993, Econometric Analysis, Macmillan Publishing Company, New York, NY.
Haney, D.G. 1975, Review of Aviation Forecasting Methodology, US DOT, Report 40176-6.
Hausman, J.A. and D.E. Taylor 1981, Panel Data and Unobservable Individual Effects, Econometrica, 49, 137798.
International Civil Aviation Organization 1985, Manual on Air Traffic Forecasting, Doc. 8991-AT/722/2,
Montreal, Canada.
International Civil Aviation Organization 1985, Airport Economics, Doc. 8991-AT/722/2, Montreal, Canada.
Jacobson, I.D. 1970, Demand Modeling of Passenger Air Travel, NASA Report CR-157469.
Kanafani, A. 1981, Transportation Demand Analysis. John Wiley and Sons, New York, NY.
Kanafani, A. and Behbehani, R. 1978, Demand Analysis for International Air Travel, Transportation Research
Record, 732, 5-15.
Karlaftis, M.G., Papastavrou, J.D., Zografos, K.G., Charnes, J.C. 1996, Methodological Framework for AirTravel Demand Forecasting, ASCE Journal of Transportation Engineering, Vol. 122, No. 2, 96-104.
Karlaftis, M.G., Papastavrou, J.D., Karlaftis, G.M. 1997, The Use of Non-Linear Time Series in Air Travel
Demand Forecasting, In Advances in Business and Management Forecasting (K.D. Lawrence ed.), JAI
Press, CT, 345-379.
Koshal, M., Koshal, R.K., Gupta, A.K., Nandola, K.N., Demand for Public and Private Passenger Transport in
the United States, International Journal of Transport Economics, Vol. XXIII, No. 2, 157-172.
Lenormand, D. 1989, European Aircraft Manufacturers Forecasting Methodology, Transportation Research
Circular, 348, 35-41.
Levenbach, H. and Cleary, J.P. 1985, The Beginning Forecaster, Lifetime Learning Publications, Belmont, CA.
Lee, L.F. 1999. Estimation of dynamic and ARCH Tobit models, Journal of Econometrics, 92, 355-390.
Mellman, R., Nelson, M. and Pino, J. 1980, Forecasts of Passenger Traffic and Air Cargo Activity at Logan
International Airport, Transportation Research Record, 768, 31-37.
Odoni, A.R. and de Neufville, R. 1992, Passenger Terminal Design, Transportation Research, 26A, 27-35.
Organization for Economic Cooperation and Development 1994, System of National Accounts 1992, OECD
Publications, Washington, D.C.
Pindyck, R.S. and Rubinfeld, D.L. 1991, Econometric Models and Economic Forecasts, McGraw-Hill, NewYork, NY.

110

Sitraer 7 (2008) 100-111 Tr. 312

Raphael, D.E. and Starry, C. 1995, The Future of Business Air Travel, Transportation Research Record, No.
1506, 1-7.
Thomet, M.A. and Sultan, S.M. 1979, Traffic Forecasting for the New Riyadh International Airport, ICAO
Manual on Air Traffic Forecasting, Montreal, Canada.
Tobin, J., 1958. Estimation of relationships for limited dependent variables. Econometrica 26, 24-36.
Vakil, F. and Russon, M.G. 1996, Income Elasticity Estimates for Short Haul Air Transport Passenger Flows,
International Journal of Transport Economics, Vol. XXIII, No. 2, 173-186.
Washington, S.M., Karlaftis, M.G., Mannering, F.L. 2003. Statistical and Econometric Methods for
Transportation Data Analysis, Chapman and Hall / CRC Press, Boca Raton, FL.
Whitford, R.K. 1990, Sketch Planning-Forecasting Future Requirements for Airports in the Chicago Region,
Working paper for the Indiana Department of Transportation, Division of Aeronautics.
Zuiga, S., de Neufville, R., Kanafani, A. and Olivera, R. 1978, Forecasting Airport Traffic: Mexico City as a
Case Study, Transportation Research Record, 732, 24-39.

111