Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

An application of a gravity model to air cargo at Vancouver International Airport Turner, Sheelah Anne 2002

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
831-ubc_2002-0274.pdf [ 6.21MB ]
Metadata
JSON: 831-1.0090306.json
JSON-LD: 831-1.0090306-ld.json
RDF/XML (Pretty): 831-1.0090306-rdf.xml
RDF/JSON: 831-1.0090306-rdf.json
Turtle: 831-1.0090306-turtle.txt
N-Triples: 831-1.0090306-rdf-ntriples.txt
Original Record: 831-1.0090306-source.json
Full Text
831-1.0090306-fulltext.txt
Citation
831-1.0090306.ris

Full Text

A N APPLICATION O F A GRAVITY M O D E L T O AIR C A R G O A T VANCOUVER INTERNATIONAL AIRPORT by SHEELAH ANNE TURNER M.A.(Economics), University of British Columbia, 2000 B.Sc. (Statistics, Operations Research), University of South Africa, 1999 B.S. (Pure Math, Economics), Southern Methodist University, 1995 A THESIS SUBMITTED I N PARTIAL F U L F I L M E N T O F T H E REQUIREMENTS F O R T H E D E G R E E O F M A S T E R O F SCIENCE I N BUSINESS ADMINISTRATION in T H E F A C U L T Y O F G R A D U A T E STUDIES ( Faciil ty of Commerce & Business Administration) We accept the thesis as conforming to the required standard T H E UNIVERSITY O F BRITISH C O L U M B I A November 2001 © Sheelah Anne Turner, 2001 In presenting this thesis in partial fulfilment of the requirements for an advanced degree at the University of British Columbia, I agree that the Library shall make it freely available for reference and study. I further agree that permission for extensive copying of this thesis for scholarly purposes may be granted by the head of my department or by his or her representatives. It is understood that copying or publication of this thesis for financial gain shall not be allowed without my written permission. of ( j ^ i y n i / i ^ j ^ c ^ . CWA The University of British Columbia Vancouver, Canada Date / O b l K ^ U ^ 2O0[ DE-6 (2/88) ABSTRACT There has been very little research in the area of air cargo demand analysis and forecasting. This thesis attempts to investigate the application of gravity models to air cargo. Using international export volumes from Vancouver International Airport in 1998, a gravity model was built. The inclusion of tariffs as an impedance factor allowed testing of the effect of tariffs as predicted by gravity models. The results were consistent with international trade theory that tariffs provide a barrier to international trade. Further, a comparison is made between aggregate and disaggregate models (across commodities). It was found that aggregation eliminates commodity specific characteristics. In using the gravity model, there are two adjustments which need to be made to reduce the bias in the model: firstly, adjustment is necessary to the bias inherent in the constant term of a log-linear model; and a further adjustment is required when forecasting actual levels rather than log levels. Even after adjustments for both types of bias, the gravity model did not produce accurate forecasts. The aggregate model produced better forecasts than the disaggregate model, but both sets of forecasts did not accurately predict the actual volumes transported. This could be as a result of the stable nature of the variables included in the model, which tend to change very slowly over time. Further, it is apparent that other additional explanatory variables should be included in the models to better capture the short-term changes in air cargo. © Sheelah Anne Turner i i T A B L E O F C O N T E N T S A B S T R A C T ii LIST OF T A B L E S v LIST OF FIGURES vi A C K N O W L E D G E M E N T S vii 1. INTRODUCTION 1 2. A I R C A R G O 3 2.1 Classification of Air Freight and Air Cargo 3 2.2 Origin/Destination of Air Cargo 3 2.3 How Air Cargo is Carried 4 2.4 Modal Competition 5 3. U N D E R S T A N D I N G TRANSPORTATION D E M A N D 6 3.1 The Influence of Time on the Demand for Transport 6 3.2 Price as an Influence on Demand for Transport 7 3.3 Quality of Service and Demand for Transport 9 3.4 Exogenous Variables affecting Demand for Transport 10 3.5 Other Exogenous Factors 10 3.6 Comparison between Passengers and Cargo 11 4. A N A L Y S I S OF INTERNATIONAL EXPORTS VIA AlR F R O M V A N C O U V E R INTERNATIONAL AIRPORT (YYR) 12 4.1 International Exports from Y V R 13 4.2 Cargo by Destination 16 4.3 Cargo by Commodity 17 4.4 Preparing for the Future 19 5. A P P R O A C H I N G A I R C A R G O FORECASTING 20 5.1 Forecasting using Growth Rates 20 5.2 Forecasting using Historical Data 21 5.3 Forecasting between Two Nodes 21 5.4 Comparison between Cargo and Passenger forecasting 23 6. G R A V I T Y M O D E L 25 6.1 Examining the Role of Tariffs 26 6.2 Aggregate versus Disaggregated Models 26 6.3 Estimating the Model 27 6.4 Implications of Logarithmic Transformations 28 6.5 Data 28 © Sheelah Anne Turner i i i 6.6 Transforming the Variables 33 7. RESULTS A N D FINDINGS 35 7.1 Data Analysis 35 7.2 Testing of Assumptions 36 7.3 Estimating the Model 38 7.4 Results for the Aggregate Model 39 7.5 Results of the Disaggregate Models 40 7.6 Empirical Results Comparison 43 7.7 Discussion of Objective Findings 44 7.8 Criticisms of the Model 45 8. F O R E C A S T I N G USING G R A V I T Y M O D E L S 47 8.1 Forecasts using Aggregate and Disaggregate Models 47 8.2 Problems in Forecasting Demand 51 9. F U T U R E IMPACTS O N A I R C A R G O 52 9.1 Canada's International Air Policy 52 9.2 The Internet 53 9.3 Electronic Data Interchange 54 9.4 Electronics Industry Growth 54 9.5 Impact of September 11th 2001 Terrorist Attack 55 10. DIRECTIONS FOR F U R T H E R R E S E A R C H 57 10.1 Time Series 57 10.2 Other Variables 57 10.3 Transformations on the Data 58 10.4 Inclusion of other Origin Points 58 10.5 Air Cargo as a Derived Industry 57 10.6 Bias Estimation 58 10.7 Non-Linear Estimation 58 11. C O N C L U S I O N 59 REFERENCES 61 A P P E N D I X 1: D A T A 64 A P P E N D I X 2: REGRESSION RESULTS 68 A P P E N D I X 3: FORECASTS 81 © Sheelah Anne Turner iv LIST O F T A B L E S Table 4.1: Largest International Export Partners by Volume (in Tonnes) 16 Table 4.2: International Export Commodities by Volume (in Tonnes) 17 Table 6.1: Disaggregate Commodity Groups 30 Table 7.1: Descriptive Statistics of Natural Logs of Data 35 Table 7.2: Correlation Matrix of Explanatory Variables 37 Table 7.3: Eigenvalues of Centred Correlations 37 Table 7.4: Results of Initial Aggregate and Disaggregate Gravity Models 38 Table 7.5: Results of Final Aggregate and Disaggregate Gravity Model Estimation 39 Table 7.6: Coefficients of Gravity Model Estimation 43 Table 8.1: Gravity Model Predictions - 1998 49 Table 8.2: Gravity Model Predictions - 1999 50 Table A l .1: Aggregate and Disaggregate Export Data (in Tonnes) 64 Table A1.2: Explanatory Variable Data 66 Table A3.1: Forecasts for 1998 81 Table A3.2: Forecasts for 1998 83 Table A3.3: Forecasts for 1998 85 Table A3.4: Forecasts for 1998 87 Table A3.5: Forecasts for 1999 89 Table A3.6: Forecasts for 1999 91 Table A3.7: Forecasts for 1999 93 Table A3.8: Forecasts for 1999 95 © Sheelah Anne Turner V LIST O F F I G U R E S Figure 4.1: Total Air Cargo by Sector 12 Figure 4.2: International Export Volumes 14 Figure 4.3: International Growth Comparison 15 Figure 4.4: International Export Partners by Volume (in Tonnes) 16 Figure 4.5: International Export Commodities by Volume (in Tonnes) 18 © Sheelah Anne Turner vi A C K N O W L E D G E M E N T S I would like to acknowledge a number of individuals who contributed to the completion this thesis. Vancouver International Airport Authority, particularly Craig Richmond and John Korenic, are acknowledged for providing a challenging and stimulating project, part of which has developed into this thesis research. Acknowledgement also goes to the COE, who provided resources for the completion of the research. Thank you to Dr Derek Atkins for valuable guidance and feedback. I would like to acknowledge U B C for the Fellowship funding received during part of my studies. Last but not least, I would like to acknowledge my friends and family. Thank-you to my parents who provided continuous support across the miles, in particular to my mother Maylene Turner for her critical proof reading. Thanks also goes to Nicole Franjic and Sarah Colby for assisting on the technical aspects of the modelling; and to Oyvind Helgerud for being a patient sounding board and critical proof reader. Kathryn Kolbuch and Ellen Fowler are thanked for their continuous support, and Johnny Yeung for the constant supply of library books from SFU libraries. © Sheelah Anne Turner vii i . INTRODUCTION Air cargo has evolved substantially since the first cargo was flown in the 1940's. No longer merely an isolated form of transportation, the air transportation industry has become a link in the chain of global logistics. This change in the nature of air transportation has accompanied the changing business environment. With the world becoming smaller on a daily basis, and international transactions being carried out as effortlessly as local transactions, the logistics industry has bloomed. It is providing smooth intermodal transitions, and is connecting businesses worldwide in an efficient, cost effective manner. Despite the merging of different modes of transportation, each mode still has its own niche in the logistics environment. In particular, air transport is still able to provide the fastest transportation time between intercontinental cities, and the increasing volume of global air cargo is testament to the fact that it remains an indispensable part of the logistics industry. Growth in international air cargo in 2000 was up 8% from 1999, with 18.8 million tonnes of cargo being carried (Folley). However, the fragility of the air transportation industry has once again been highlighted by the terrorist attack on New York and Washington on September 11 t h, 2001. The general economic downturn, together with the terrorist attack have resulted in a sharp decline in cargo volumes in an already depressed industry. Second quarter showed a decline of almost 6%, and the third quarter is expected to show an even more substantial decline (Folley). Since the air transport industry changed on September 11 t h, gaining an understanding of the demand for air cargo is vital to stimulate the industry and restore consumer confidence in the industry. Insight into the nature of air cargo can assist air transportation carriers in providing the best, most competitive services. Furthermore, forecasting air cargo will also assist cargo transportation providers in providing the necessary infrastructure (from aircraft to storage facilities) to match the changes in the air transportation industry, and continued smooth operation of the global logistics flow. Combining these issues, this research undertakes to examine the international demand for air cargo. Vancouver International Airport (YVR) was chosen as the reference point, with the analysis of international goods flowing between Y V R and its trading partner countries. Knowledge of the characteristics of air cargo is important to provide the correct air cargo services. The types of commodities typical in air cargo, and how these commodities are carried is addressed in section 2. In order to provide a competitive, efficient transport process, it is vital that the factors affecting the demand for transportation are recognised and considered. Section 3 © Sheelah Anne Turner 1 addresses factors such as time, price and service which may influence decisions on transportation mode. Turning attention to Vancouver International Airport, section 4 analyses international air cargo exports from YVR. This includes analysis of the specific commodities which are transported, as well as the primary destinations of the air cargo leaving the airport. Before building any forecasting models, a review of past forecasting methodologies is undertaken. Section 5 discusses the roles played by different forecasting methods. In order to examine the demand for air cargo, a gravity model has been applied to the international air cargo export data. This model has been used in the demand for many forms of urban travel of passengers, but has not been used in air cargo. The model is discussed in section 6, and the results and findings are discussed in section 7. Section 8 uses the estimated model to forecast air cargo exports. The future drivers of air cargo will provide a further indication of potential growth prospects. With the global economy becoming more integrated, and consumers demanding customised goods and services, the nature of air cargo is changing dramatically. These, and other factors, are discussed in section 9. The findings of this research have highlighted some areas of further investigation. The area of air cargo demand analysis has received limited attention in the past, and as such provides significant opportunities for research in the future. Some ideas are addressed in section 10. The paper concludes with section 11, a brief summary and conclusion of the current research. © Sheelah Anne Turner 2 2. A I R C A R G O 2.1 Classification of A i r Freight and A i r Cargo Air freight consists of shipped goods, mail and separate paid passenger luggage. Paid passenger luggage consists of excess luggage a passenger may wish to take, above and beyond the normal allocated luggage. This luggage takes first priority on a passenger aircraft after normal passenger luggage is included. Mail consists of all letters, parcels and other containers that are sent via the postal system from a sender to a receiver. Mail is prioritised next, receiving space allocation after all passenger luggage has been loaded onto the aircraft. Shipped goods are goods that are sent from a shipper to a consignee via a transportation provider. These goods vary in size and value from machine parts to electronic components, clothing to toys. On passenger aircraft, shipped goods receive last priority, and are allocated the remaining space and weight after all other freight is loaded. Thus, capacity is stochastic in nature. In general, the types of goods that are shipped can be classified into five broad categories: i . High degree of obsolescence, for example computer equipment and electronics i i . Short lived life cycle, for example apparel and clothing i i i . Perishable, for example seafood, fruits and vegetables, and flowers iv. Time sensitive, for example newspapers and magazines v. Valuable products, for example jewellery and pharmaceutical products Each of these commodity categories has different processing and storage requirements, and consequently may have different infrastructure needs. For the purposes of further discussion, air cargo will refer to shipped goods only, excluding mail and separate paid passenger luggage. 2.2 Origin/Destination of A i r Cargo Air cargo in Canada is classified according to the origin (for imported cargo) or the final destination (for exported cargo). There are three sectors that have been defined: i . Domestic: this includes all air cargo that is transported within Canada i i . Transborder. this includes all air cargo that is transported between Canada and the United States i i i . International: this includes all air cargo that is transported between Canada and other countries (excluding the United States) © Sheelah Anne Turner 3 2.3 How Air Cargo is Carried Air cargo carriers transport air cargo from the origin to destination airports, which may form a part of the total transportation process from shipper to consignee. The cargo may be packed into containers or onto pallets, or it may be bulk loaded. Containers and pallets are packed and secured before loading onto the aircraft, while bulk loading involves the loading of individual pieces of cargo. The most common form of transportation is belly hold cargo. In Canada, more than half of all air cargo is carried in this manner. Belly hold cargo is containerised or bulk loaded, depending on the size of the aircraft and the loading facilities loaded into the bellies of passenger aircraft, filling any available space. Due to the extensive passenger networks already in existence, this is a convenient means of transporting air cargo. A significant disadvantage with this means of transportation is that air cargo has the lowest priority in the loading and unloading of the aircraft: passenger luggage is loaded and unloaded first, followed by mail. This may delay the processing of air cargo. Air cargo may also be transported by all-cargo aircraft, also known as freighters. A l l air cargo is containerised and loaded on the main deck of the aircraft. These freighters may be operated as independent airlines, or in conjunction with a passenger airline. The benefit of using this form of transportation is that there is no priority restriction in the loading and unloading of the cargo. Thus, freighters offer an improvement over belly hold cargo in terms of processing time of cargo. Airlines have different options in the ownership and operation of freighters. They may be owned and operated by the airline themselves; they may be "wet leased" from a third party company who operates the aircraft; or they may be leased from a third party but self operated. The advantage of leasing aircraft, particularly freighters, is that it provides flexibility in responding to changing demands, without the significant capital investments that are necessary in the ownership of aircraft. Aircraft can be added or removed from the fleet more easily. A very important player in cargo transportation is integrated carriers. Integrated carriers, or integrators, specialise in time definite and express shipment of parcels and packages, typically smaller in size and weight than normal cargo. The cargo is containerised and loaded on the main deck of the aircraft, in the same manner as freighters. An important difference between integrators and other airlines is that integrators operate all components of the transportation service: from ground handling of the aircraft, to loading and unloading the aircraft, to final delivery. This ensures a smooth transition from one mode of transport to another, and an efficient service from shipper to consignee. © Sheelah Anne Turner 4 2.4 Modal Competition A i r transportation is not the o n l y f o r m o f transport available to shippers i n Canada . In most cases, there are other c o m p e t i n g modes o f transport between the shipper and consignee, dependant o n the locat ion o f the shipper and consignee. T h e p r i m a r y competitors are r a i l , t r u c k i n g and marine shipping. F o r domestic cargo, c o m p e t i t i o n arises f rom trucking and r a i l . W i t h the l o w e r costs i n v o l v e d w i t h t r u c k i n g and r a i l transportation, p l a n n i n g can save s ignif icant costs o n transportation. . Transborder cargo also faces c o m p e t i t i o n from trucking , and to a m u c h lesser degree, r a i l . T r u c k i n g i n part icular is a s ignif icant form o f compet i t ion for destinations that are w i t h i n a two-day truck dr ive away. I n these cases, the s i m p l i c i t y o f not h a v i n g to wai t for a speci f ic f l ight, or for less p h y s i c a l h a n d l i n g o f the cargo m a y result i n t r u c k i n g b e i n g a faster f o r m o f transport. T h i s , together w i t h the lower cost associated w i t h t r u c k i n g , makes it a strong competi tor to air transportation. International cargo o n l y faces c o m p e t i t i o n from marine shipping. F o r some c o m m o d i t i e s , the large difference i n transit t ime between air and marine s h i p p i n g tends to carry m o r e importance than the difference i n s h i p p i n g rates. O n e disadvantage o f the lack o f compet i t ion is that international markets m a y be inaccessible to some l o w e r v a l u e d goods: it m a y be too expensive to send relat ively l o w e r v a l u e d goods b y air, but the goods m a y perish or b e c o m e obsolete i f sent b y sea. C o m p e t i t i o n from various modes o f transport m a y be m o r e appl icable to certain c o m m o d i t i e s than others. Perishables, such as fresh vegetables and cut f lowers , are less l i k e l y to consider marine s h i p p i n g a v iab le alternative for l o n g distances. S i m i l a r l y , raw materials , such as lumber and coa l , are u n l i k e l y to consider air transport a v i a b l e alternative to r a i l or marine s h i p p i n g . © Sheelah Anne Turner 5 3. U N D E R S T A N D I N G T R A N S P O R T A T I O N D E M A N D It is vital to know and understand the characteristics that are incorporated in decisions regarding modal choice. These characteristics may be classified as either endogenous or exogenous, depending on the amount of control transportation firms have over them. This section will attempt to identify some endogenous factors of the firms both demanding and supplying transportation; exogenous variables beyond the influence of these firms are given somewhat less attention. Some differences between demand for passenger transportation, and demand for cargo transportation will also be discussed. There are some broad general categories of variables affecting the demand for transportation. These categories are by no means exhaustive, but give an idea of considerations taken when considering different transportation modes. i . Time - this includes both the time horizon and time cycles in demand; i i . Price - the freight rate charged by different modes of transport for transportation of goods; i i i . Quality of Service variables - these include a variety of variables such as damage rates and journey time which may affect the goods being transported; iv. Exogenous variables - typically these variables are exogenous to the suppliers and demanders of transport, but will still have an effect on decisions made with respect to transportation, although they may be subject to influence by other agencies, such as governments. Examples are income levels, location of population and business centres; v. Other Exogenous factors - these are variables exogenous to all decision makers which are still influential in the demand for different transportation modes, such as historical and geographic influences. 3.1 The Influence of Time on the Demand for Transport The time horizon under consideration is one of the most important factors influencing both the demand for transportation, and the effect other factors will have. The time horizon can be divided into 2 periods: the short-term and the long-term. In the short-term, the reaction to changes in other factors may be strongly inelastic. The relative unresponsiveness to price changes, for example, may be due to a belief that the changes are not of a permanent nature, or technical considerations may constrain an immediate reaction. In the short-term, it is unlikely that shippers of cargo wil l search for transportation mode substitutes in the face of increasing prices. They may rather wait to © Sheelah Anne Turner 6 see i f the price change is permanent, or may not want to be inconvenienced by the search for a new mode of transport. In the long-term, however, there are numerous possibilities for the firm facing rising prices, making the price elasticity once again more elastic. These options include switching to alternative modes of transport, or relocation of the firm closer to the retail markets. Thus, as the time horizon under consideration starts to lengthen, the demand for transportation becomes more elastic. The high costs involved in switching from one alternative to another - be it another mode of transportation or relocation of the firm - are too high to allow potential temporary changes in the short term, but can be incorporated into the long term strategy of the firm, and thus could be considered as viable alternatives in the long term. Similar arguments apply to other factors. One of the most pronounced characteristics of the demand for transportation is the fluctuation over time - sometimes more than doubling from a "trough" to a "peak?'' in the cycles. These cycles may be short (such as passenger demand for daily urban transportation) to very long cycles (such as international freight transportation). There is a very limited capability of transportation operators to control these cycles in demand, and yet they must ensure that sufficient capacity exists to accommodate the peaks. Daily passengers commuting will place a high demand on transport systems during mornings and evenings, with relatively lower demand during the remainder of the day. For international cargo transportation, there is a strong peak during the summer months, a smaller peak in November (in anticipation of Christmas consumer demand) and a low trough in the early months of the year. 3.2 Price as an Influence on Demand for Transport The price referred to is the freight rate that is charged by the supplier of the transportation mode (or carrier) to the shipper of the goods. There are a number of ways that this price may influence the decisions that the shipper makes with respect to choice of transportation mode. This sensitivity of the shipper to price is captured in the price elasticity of the demand for transportation. 3.2.1 Transportation Demand as a Derived Demand One of the major differences between the economic demand for transportation and the economic demand for normal economic goods is that the demand for transportation is a derived demand, and is thus dependant on the underlying commodity. Demand for transportation is usually not a demand for movement for its own sake, particularly in the case of transportation of freight. The presence of price differentials for the same good at different locations, or the existence of different goods at different locations, indicates the potential demand for movement of goods, and hence a demand for transportation. Differing relative prices for goods at different locations enables these locations to achieve higher consumption levels through trade of goods. However, trade in goods will only occur i f the price differentials exceed the cost of transporting the goods. © Sheelah Anne Turner 7 3.2.2 The Effect of Price Since the demand for transportation is a derived demand, the demand for transportation of freight is related to the final demand for the goods being transported and existing price differentials. Consequently, changes in the price of the transported good, as well as the price of transporting the goods, will affect the potential for transportation. Thus, the elasticity for transport demand is influenced by the elasticity of demand for the final product. In general, the elasticity of demand for transport is more elastic than the elasticity for the demand of the final product transported. Another important factor in the sensitivity of shippers to the freight rate is the relationship between the freight rate and the value of the product being shipped. In cases where the freight rate is very high in comparison to the value of the product (such as for grain and vegetables) the shipper is more sensitive to changes in the freight rate, and thus the elasticity of demand is very elastic. Conversely, in the case of shipping high value products (such as for jewellery and electronic components) the freight cost is a relatively small fraction of the total cost of placing the product on the market, and thus the shipper will be less sensitive to changes in the freight rate. Hence, the elasticity of demand is more inelastic for higher valued products. The price and availability of viable transportation alternatives will also affect the demand for any particular mode. If alternative transport modes are highly substitutable for the particular goods being shipped, a shipper will be more sensitive to changes in freight rate, and thus demand for a particular mode of transportation will be more elastic. If there are no viable alternative transportation modes, then changes in the freight rate will not drive the shipper to seek another transportation mode. Thus the elasticity of demand is relatively more inelastic. As mentioned earlier, this should be viewed in conjunction with the time horizon under consideration - it may not be a strategic move to change transportation modes i f the freight rate increases are merely temporary and expected to return to original levels in the short term. The size of the shipper could influence the impact price has on the demand for transportation. Larger shippers who tend to have larger shipments and tend to utilise transportation more frequently can afford to seek out better transportation alternatives. Due to the larger volumes of cargo, there are more alternatives with respect to routes, transport modes and frequency of shipping available to the shippers. Thus, larger shippers are more sensitive to price fluctuations, and the elasticity of demand is more elastic. Small shippers are somewhat constrained by lot sizing required for some transport modes. They may not have sufficient volumes of cargo to make other transport modes more cost effective, and are thus limited to a few routes and transport modes. Thus, smaller shippers are less sensitive to price fluctuations, and consequently their elasticity of demand is more inelastic than for larger shippers. © Sheelah Anne Turner 8 3*3 Quality of Service and Demand for Transport In addition to price, service quality also affects demand for transport. These relate to the service and care taken of the cargo both while the cargo is in transit, and while the cargo is waiting at the origin and destination points awaiting transfer from one mode to another. It is sufficient that the quality of care at only one of the abovementioned locations is sub-standard for that mode of transport to be disregarded as a potential choice. In some cases, these service variables may be more important than price or shipment time in explaining the behaviour of shippers and their choice of transportation mode. One of the most important service variables is the time taken for goods to be transported from origin to destination. The journey time or average speed of shipment affects the decision of transport mode for a number of reasons. Goods-in-transit represent a capital investment which cannot be invested elsewhere in the firm. Further, these goods do not form part of inventory or buffer stock for the firm's operations, and thus can not be utilised when needed. The length of journey time is closely related to damage and pilferage. These can vary dramatically between modes of transport, and also among carriers within the same mode of transport. It seems likely that the longer the journey time, the longer the time the goods are out of the control of the owner, and thus the higher the potential for deterioration. In the case of perishable goods requiring refrigeration, the carriers may pay less attention to controlling the environment than the owner would, and consequently there is a higher probability of damage to the transported cargo. Similarly, the longer the journey time, the more opportunity there is for pilferage of the goods to occur. Reliability of the transportation service provided is another important consideration. As more companies continue to move towards "just-in-time" production processes (with inventories kept at a minimum) the ability of a carrier to deliver goods within the stated time becomes critical. To optimise such processes, it is vital that carriers are reliable, and often firms are willing to pay additional costs to ensure this service. Related to reliability is the convenience of service. Aspects of convenience refer to the pickup and delivery of goods, as well as special handling, customs clearance and storage. Door-to-door service offered by motor carriers and integrators has long been recognised as an important service advantage, despite the higher cost associated. The frequency of transportation is a related convenience factor. As can be expected, the more frequently the goods can be transported, the more likely the shipper is to utilise the particular mode of transport. Frequency of service is also related to convenience: a more frequent service is more convenient, and is more likely to be utilised by shippers of goods. The option of sending shipments more than once a day means that goods do not need to wait long for the next service time, but can be sent more frequently and reach the market earlier. The minimum shipment size is a further service factor. Smaller lot sizes reduce the firm's inventory requirements since smaller, more frequent shipments can be utilised in © Sheelah Anne Turner 9 the managing of inventory. Smaller lot sizes also result in a smaller amount of capital being tied up in goods-in-transit. 3.4 Exogenous Variables affecting Demand for Transport In addition to the above mentioned endogenous factors (price, time and quality of service), there are numerous exogenous factors influencing the demand decisions faced by shippers. Not all of these factors are completely exogenous, random or accidental, but may be influenced either directly or indirectly by governments or other organisations. One factor regarded as being exogenous is consumer tastes. Tastes are not completely exogenous, since firms can advertise differentiated products in order to attempt to alter the tastes and preferences of consumers. This is true not only for consumer products, but for transportation as well. Large transportation companies can use advertising in an attempt to lure shippers to alter their choice for mode of transportation. The income level of consumers is influenced by various forces, but is an exogenous factor as far as individual firms are concerned. Income levels affect both the types and quantities of goods consumed. Per capita income can be influenced by the government through tax and employment policies, and also by general economic conditions and price levels. These all ultimately affect the demand for transportation, but are beyond the control of carriers and shippers. The industrial structure of the economy influences transportation demand. Through the promotion of various secondary industries, tax incentives and infrastructure investments, governments may introduce public policies which affect industrial composition. These decisions are frequently made without a direct understanding of the impact on the transportation system. For example, promotion of the lumber industry will have a vastly different impact on transportation than incentives for the electronics industry. However, these policies are not implemented by the transportation sector, and thus are treated as exogenous influences on carriers and shippers. 3.5 Other Exogenous Factors Lastly, there are some variables affecting the demand for transportation which are purely exogenous. These are factors which are a result of the geography of the region, a result of historical events or a result of natural resources within the area. The position of a natural harbour and the occurrence of mineral deposits are such examples. Since these factors affect the location of people and industries, they too will affect the demand for transportation. Although their truly exogenous nature may seem to imply that they are of little interest to the demand for transportation, they may indeed provide explanation for particular demand patterns for transportation. © Sheelah Anne Turner 10 3.6 Comparison between Passengers and Cargo The above discussion does not distinguish between the demand for passenger transportation, and the demand for cargo transportation. Instead, it is of a more general nature, and deals primarily with the similarities between passenger demand and cargo demand. There are, however, characteristics specific to each which provide additional understanding of transport demand. In general, passengers are more sensitive to the time of day they travel than cargo. Passengers prefer to travel at convenient times of the day, and may require substantial incentives to travel either very early in the morning or very late at night. On the other hand, cargo is more insensitive to the time of day. Cargo is more concerned with reaching the final destination, and not as concerned with the time of day the transportation occurs. A very significant difference between passengers and cargo is in the routing. Passengers generally will not tolerate very circuitous routing, and prefer a more direct routing with as few stops and detours as possible. Cargo, on the other hand, has no preference for direct or circuitous routing; it is more concerned with reaching the final destination by the time required. This allows more flexibility on the part of the shipper, as consolidation of smaller shipments for parts of the route may realise cost savings. Another important factor to consider in transport demand is the demand for one-way transport versus return transport. Typically, passengers require return travel, as trips are usually between residence and work, between home and vacation or a business trip. In contrast, cargo seldom requires a return trip - once the goods have reached the final destination, they generally will be employed for their intended purpose. Thus, cargo carriers may be faced with potentially empty backhaul, and charge higher rates to compensate. © Sheelah Anne Turner 11 4. ANALYSIS O F INTERNATIONAL E X P O R T S V I A A I R F R O M V A N C O U V E R INTERNATIONAL AIRPORT ( Y V R ) Having gained an understanding of how choices are made between transport modes, the focus is now on the transportation of cargo via air through Vancouver International Airport. As is typical of other trade, Canada's air cargo industry is tied very strongly with the economic health of both Canada and her trading partners. With the Asian Crisis in 1998, and the corresponding economic slump in Asia, international air cargo volumes, especially export volumes, were depressed. The economic recovery starting in 1999, resulted in a similar recovery in international air cargo volumes. As from the end of 2000 and into 2001, North America has experienced a dramatic economic slowdown. With both Canada and the United States similarly affected, an impact is being felt on both transborder and domestic air cargo volumes. Vancouver International Airport (YVR) has experienced a similar pattern in its air cargo. The high dominance of Asia as an international export region resulted in Y Y R experiencing a much larger drop in international export volumes than was evident at other airports during the 1998 crisis. The dramatic effect of the North American slowdown is having a direct effect on both transborder and domestic export cargo volumes for all Canadian airports, including Y V R . International exports are also experiencing a decline as a result of the slowdown. This is a more indirect effect, as the North American economic slowdown overflows into the global market, causing a global slowing. The following figure shows a breakdown of total air cargo traffic at Y V R by sector. FIGURE 4.1: TOTAL AIR CARGO BY SECTOR 14,000 12,000 at 10,000 z § 8,000 H 2 6,000 H u S 4,000 2,000 0 1997 1998 1999 2000 2001 Total domestic air cargo, as a percentage of total air cargo, processed by Vancouver has been declining steadily from a high of 49.3% in 1997 to current level of 42%. © Sheelah Anne Turner 12 Furthermore, the actual volumes of domestic air cargo passing through Vancouver Airport have also seen a decline, even after accounting for seasonality. Transborder air cargo has seen the opposite pattern to that displayed by domestic air cargo, with an increase in both volumes and percentage of total air cargo. There appears to have been a stabilisation of transborder air cargo at 16% of total air cargo, but this is an increase from 13.3% of total air cargo in 1997. The strong decline in export volumes from the latter part of 2000 through into 2001 is as a result of the economic slowdown mentioned above. International air cargo has seen an increase since 1997. As a percentage of total air cargo, international air cargo has increased from a low of 36.2% to a peak of 47.9%, but is stabilising around 40%. However, the effect of the North American slowdown has resulted in a strong decline in international export volumes from the latter part of 2000 into 2001. 4.1 International Exports from YVR The concentration of this research is on International export volumes, and will be the centre of discussion for the remainder of the section. International exports do not include exports to the United States, only exports to other international destinations. International exports from Y V R have averaged a little over 40% of total exports from Y V R over the last 4 years. There is great fluctuation, however, from lows in the winter months of 38% to highs in the summer months of 50%. In 2000, 56,000 tonnes of cargo from Y V R were destined for international markets, out of a total of 123,300 tonnes of export goods from Y V R . This is declining sharply in 2001, and is not expected to show recovery until late in 2002 and early 2003. International exports from Y V R display seasonal patterns in volumes. The highest peak is in July, with lowest volumes in December / January. The summer peak is primarily as a result of the large volume of fresh fruit and vegetables destined for Asian markets. The winter lows are typical for post-holiday period and the start of the new year (machinery and instrument exports are low), as well as mid-winter period (perishable exports are low). © Sheelah Anne Turner 13 This can be seen in the following figure. The actual volume of international air cargo exports is shown together with the percentage of total exports that international exports comprises. F I G U R E 4.2: I N T E R N A T I O N A L E X P O R T V O L U M E S International Exports from YVR 8,000 7,000 _ 6,000 « c b 5,000 o > O 3,000 2,000 1,000 0 Export Volumes Export Percent 80.0% 75.0% 70.0% 65.0% o n 60.0% g o CL 55.0% t o a. 50.0% UJ 45.0% 40.0% 35.0% There appears to have been consistent growth in international export volumes, with higher and higher peaks being reached in the summer months. The exception is 1998, which had lower volumes due to the Asian crisis that impacted most Asian countries. However, 1999 and 2000 continues an upward trend from 1997. Another noticeable trend is the increasing share of total exports that international exports are capturing, particularly in 2000 with the slowdown in the US economy reducing the transborder exports from Y V R . © Sheelah Anne Turner 14 The figure below shows the annual GDP growth for international export partners1 (quarter over same quarter in previous year), together with international export cargo growth at Y V R (quarter of same quarter of previous year). F I G U R E 4.3: I N T E R N A T I O N A L G R O W T H C O M P A R I S O N Growth in exports is expected to be more reliant on the economic growth of the receiving country than the exporting country. A comparison between economic growth data for Canada and international export partners with international export growth at Y V R confirms this. There appears to be a strong relationship between the GDP growth of international partners, and the growth in international exports from Y V R . The high R value of 0.83 between lagged International GDP growth and international export growth indicates the strength of this relationship. Thus, it is apparent that the strongest force driving international export growth is the economic growth of the receiving country. This further indicates that other factors may have a smaller impact on affecting international air cargo exports. 1 The GDP of international export partners was created by weighting the GDP growth of individual export partners by their contribution to international exports to YVR. Only the largest 15 countries were used, accounting for 79.1 % of exports from YVR. © Sheelah Anne Turner 15 4.2 Cargo by Destination The figure below indicates the primary international export countries, based on the volumes of air cargo from Y V R . This shows the dominant destinations which Y V R exports air cargo to. F I G U R E 4.4: I N T E R N A T I O N A L E X P O R T P A R T N E R S B Y V O L U M E (IN TONNES) International Export Partners The largest 12 international export partners, together with the export volumes received are shown in the table below. T A B L E 4.1: L A R G E S T I N T E R N A T I O N A L E X P O R T P A R T N E R S B Y V O L U M E (IN TONNES) A I R C A R G O V O L U M E (TONNES) P E R C E N T A G E Japan 10,049 23.1% Hong Kong 8,015 18.4% Taiwan 2,888 6.6% Germany 2,606 6.0% Australia 1,944 4.5% United Kingdom 1,784 4.1% China 1,615 3.7% Singapore 1,548 3.6% Netherlands 1,216 2.8% France 824 1.9% Thailand 773 1.8% Other 10,207 23.5% T O T A L 43,469 100.0% © Sheelah Anne Turner 16 The largest international export partners are Japan and Hong Kong, which together comprise 41.5% of the total International export volumes. There are many other important trading partners in Asia, which highlights the reliance on the region. In total, Asia receives over 60% of international exports. The remaining export partners are very small in comparison. Germany and the United Kingdom, the dominant European partners, comprise a smaller part of exports. Europe in total receives approximately one quarter of all international exports. 4.3 Cargo by Commodity The primary commodity groups exported internationally from Y V R are listed in the following table. TABLE 4.2: INTERNATIONAL EXPORT COMMODITIES BY VOLUME (IN TONNES) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ ^^^^^^^^^^^^ Confidential Commodities 18,400 42.3% Machinery & Mechanical Appliances 8,554 19.7% Animals & Animal Products 6,511 15.0% Instruments - Measuring, Musical 2,631 6.1% Vegetable Products 1,640 3.8% Wood Pulp Products 1,343 3.1% Base Metals & Articles Thereof 594 1.4% Miscellaneous 589 1.4% Chemical Products 563 1.3% Transportation Equipment 479 1.1% Prepared Foodstuffs 416 1.0% Mineral Products 312 0.7% Plastics & Rubber 299 0.7% Wood & Wood Products 226 0.5% Articles of Stone, Plaster, Cement, Asbestos 218 0.5% Textiles & Textile Articles 198 0.5% Animal or Vegetable Fats 165 0.4% Pearls, Precious or Semi-Precious Stones, Metals 148 0.3% Hides & Skins 137 0.3% Works of Art 23 0.1% Arms & Ammunition 20 0.0% Footwear, Headgear 3 0.0% TOTAL 43,469 100.0% © Sheelah Anne Turner 17 The large volume of confidential commodities influences the distribution of the internationally exported commodities. Confidential commodities do not have any additional detailed information as to the nature of the commodities it comprises, and thus no further breakdown can be obtained. Furthermore, it does not seem reasonable to assume that the confidential commodities follow the same commodity structure as the rest of international exports and thus cannot merely be divided proportionally among the other commodity groups. This data is depicted graphically in the following figure: F I G U R E 4.5: I N T E R N A T I O N A L E X P O R T C O M M O D I T I E S B Y V O L U M E (IN TONNES) Wood Pulp Other Confidential Commodities 4 2 % Machinery & Mechanical Appliances Other includes all groups not explicitly indicated This data reveals that, aside from confidential commodities, machinery and instruments form the largest portion of international exports, followed by perishable products -animal products and vegetable products. Combining the information on international export partners and export commodities allows an analysis and understanding of the demand for particular commodities by specific countries. © Sheelah Anne Turner 18 4-4 Preparing for the Future The different commodities may require different handling and thus different infrastructure for processing. Whereas perishable products require refrigeration, machinery and instruments may require larger storage and packing areas. Understanding the volumes of cargo flowing, as well as the commodity breakdown, can assist in the planning of future infrastructure requirements. As such, forecasting of air cargo plays a very important role in supporting capital investment decisions at Y V R . © Sheelah Anne Turner 19 5. A P P R O A C H I N G A I R C A R G O F O R E C A S T I N G The forecasting of air cargo has been an area of aviation forecasting that appears to have received the least amount of research attention, in spite of its growing importance in the twentieth and twenty first century. Most research has been more focussed on the forecasting of passengers or overall aviation demand. As such, investigation into forecasting models from a variety of areas has been necessary in order to identify potential models that could be applied to air cargo forecasting. The types of models that have been used to forecast aviation trends are varied in nature, depending ultimately on the desired usage of the forecasts. The models developed and used range from forecasting using growth rates to using historical data and cross-sectional data. The models range from relatively simple exponential smoothing methods, to more complex abstract and gravity models. Despite the wide range of models, however, it appears that there is still no reliable means of forecasting air transportation data, and thus the research into the field for improved forecasting methods is continuing. Interestingly, the early research in the field of transportation forecasting with gravity models, particularly during the late 1960's and 1970's, has provided a solid backbone to much of the research and developments during the late 1990's. It is likely that these methods will remain the backbone as further research occurs in the twenty first century. 5.1 Forecasting using Growth Rates Logit models are one method used in the forecasting of the growth in data. Vedantham and Oppenheimer (1998) present a model used by the Environmental Defense Fund which uses a long-term non-linear dynamic systems approach to forecast the demand growth in aviation. This model breaks the demand for aviation into different sectors (personal travel, business travel and freight, military and general) and analyses the differing underlying dynamics of each sector. When looking at the market evolution in each sector, a logit model is used to capture the changes in the growth rates as eventual saturation is reached. In this research, the growth of aviation demand is forecast under different scenarios. A n innovative forecasting model was developed by Ashley and Hanson (1995) to forecast (amongst others) the demand of passengers and air cargo at Schiphol Airport. Once again, the developed models forecast the growth in the data. For passenger demand, the growth rate was related to various economic variables, levels of air fares and demand elasticities. Incorporating competition from other airports and other modes of transport extended this model further. A hierarchical logit model was based on information on passengers' air route choices and preferences for air over other transport modes. It was used to forecast how passengers were likely to choose between modes and between alternative air routes via competing airports. The forecasting process for air cargo followed a similar process. © Sheelah Anne Turner 20 This process of forecasting growth rates seems to have been adopted by Boeing (1999). Although the details of the forecasting process are difficult to obtain, growth rates were forecast for both mail and air freight. 5.2 Forecasting using Historical Data Most of aviation forecasting has been undertaken on historical data rather than using growth rates. One of the earliest methods used by forecasters in the air transportation industry was regression analysis. Taneja (1978) provides an in depth discussion on the building and fitting of regression models to air transport data. His discussion starts with simple regression, and is extended into a multiple regression equation. Taneja also discusses typical model assumption violations that frequently occur in air transportation data, the potential effects on both fitting and forecasting, and the possible remedies to eliminate the violations. Taneja extends the single equation model to a multi-equation model, with the introduction of two-stage least squares, three-stage least squares and simultaneous equation systems. These more complex systems allow the analysis of both demand and supply in forecasting. Econometric methods have been employed by Sletmo (1971). In the estimation of various elasticities of demand for air cargo, he used econometric techniques for the estimation of air cargo demand. These demand functions depend on underlying economic demand theory. Sletmo investigated both the use of a basic static demand equation and also dynamic competitive models. He concluded that the dynamic competitive models were more suited to explaining the demand for air cargo than the static model. Further, he concluded that dynamic models provided insights that went beyond the basic static model. Another of the simplest methods employed in forecasting is that of exponential smoothing. This was the approach taken by Kasilingam (1996). In developing a cargo revenue management model, the capacity available for air cargo was forecast. Air cargo capacity was calculated as the residual capacity after passenger luggage and mail were accounted for. Exponential smoothing models on historical mail data forecast the volume of mail. Using the expected number of passengers, the expected number of bags was calculated. From these forecasts, the available capacity for air cargo was calculated. Kasilingam also used exponential smoothing to forecast different cargo products at the market level. 5.3 Forecasting between Two Nodes Some of the oldest models that have been used in the field of forecasting travel demand are those which look at flows between two points - an origin and a destination. These models may include both cross-sectional as well as time series data - adding additional information to the model. Models that fall into this category include abstract models and gravity models. © Sheelah Anne Turner 21 5-3-1 Abstract Models Abstract models formulate the demand for a particular mode of transport between an origin and a destination not in terms of demand for the mode itself, but in terms of an abstract mode with a specific set of attributes or characteristics. Each mode is characterised only in terms of features. In specifying the demand equation of a basic abstract model, Quandt and Baumol (1966) postulated that the demand for travel along an "arc utilising a particular mode of transport depends on the characteristics of the mode, and on a set of exogenous variables. These exogenous variables include socio-economic, geographic and demographic characteristics of the origin and destination nodes. In their paper, Quandt and Baumol conclude that the abstract model draws satisfactory explanation of the current situation and is able to draw strong inferences about the future. One strength of this model is the ability to evaluate future modes of transport merely on their characteristics, without defining the mode itself. Another strength of the abstract model is the ability to estimate the future demand for each mode and also the total demand for travel. This basic abstract model received much attention, and a large number of variants of the basic mode have been formulated and estimated subsequently, all in the interest of improving the basic theory. Quandt and Young (1969) researched the inclusion of different formulations of the basic abstract model with the inclusion of different explanatory variables, different model formulations and different estimation methods. However, in attempting to narrow the choice of models, it appeared that city pairs might have intrinsic characteristics that may cause travel on some arcs to be different to what might be predicted. 5.3.2 Gravity Models Gravity models forecast demand purely on empirical grounds, rather than based on economic theory of choice, as indicated by Oum (1980). The various forms of these models have essentially the same basic characteristics: the volume of traffic between an origin and a destination is an increasing function of trip generating factors at the origin, trip attraction factors at the destination and a decreasing function of the impedance factors between the origin and destination. Quandt and Baumol (1969) used the populations at the two nodes as generation and attraction factors, and the distance between the nodes as the impedance factor. However, as Oum (1980) indicates, other variables that could be included are production (generation), consumption (attraction) and impedance factors such as cost of transportation and travel time. Alcady (1967) uses a simple form of the gravity model to test whether aggregation across different modes of transportation provides better performances by the gravity model. His analysis is based on 16 city-pair routes of the California city-pair grid. He uses air, rail and highway as the three modes of transport. Based on his results, he concludes that aggregation does provide more consistent results. Gravity-type models appear to have been used in the air industry itself. Airbus (2000), in their forecasting for future freighters, forecast the demand of air cargo flow in 120 submarkets. The precise details of the model developed are not readily available. © Sheelah Anne Turner 22 However, the model used has the same generation and attraction characteristics that are typical in gravity models. With the development of more sophisticated models - in the areas of both abstract and gravity - Howrey (1969) researched which models were preferable for the prediction of air travel. In his paper, he compared predictions of the various models, based on a set of empirical data. He concluded that it was difficult to improve dramatically on the predictions of the simple gravity model. Another interesting conclusion was that the model that fits the sample data best may not necessarily be the best model to use for forecasting and prediction. 5.3.3 Inventory Approach Developing out of the abstract model was an inventory approach to freight transport demand. Baumol and Vinod (1970) extend the idea that modes of transport can be represented by a set of attributes. In making the choice between different modes of transport, indifference curves are constructed in modal space. This involves understanding the trade-off between various attributes of the different modes. In particular, when looking at the speed and economy of a mode, inventory plays a role in that different combinations of speed and economy necessitate different amounts of in-transit inventory, and different levels of safety stock. Thus, in minimising the total cost of shipping, the inventory costs should also be included as relevant to the decision. 5.3.4 Disaggregate Demand Forecasting The field of urban transportation has approached the problem of forecasting demand for trips between two nodes using disaggregated demand forecasting. McFadden (1978) explains the need for more flexible demand forecasting methods, particularly those which are able to incorporate behavioural decisions that individuals make in transportation choices. The disaggregate model has a unified concept, in which a trip generation model and the modal split model are dependant on each other. This is different from conventional models in which they are treated independently of each other. The advantages of disaggregate forecasting is the flexibility in modelling: it is not based on one model, but is rather a concept or method which can be used in different contexts. 5.4 Comparison between Cargo and Passenger forecasting In an attempt to forecast both air cargo and passengers, an understanding of the available capacity may be necessary. Defining air cargo capacity differs from passenger capacity in numerous aspects. One of the most critical aspects for air cargo is that the available capacity in belly load aircraft is uncertain, and tends to be stochastic in nature. This is in sharp contrast to passenger capacity, where there is a fixed and known seat capacity on any given aircraft. Firstly, air cargo depends on the expected number of passengers on the flight, in addition to the accompanying passenger baggage. This introduces significant variability into the available space for cargo, as the exact number of passengers and volume and weight of baggage are unknown until shortly before the aircraft departs. A second source of © Sheelah Anne Turner 23 variability is the payload of the aircraft. The payload depends on several factors such as weather, fuel weight and season (Kasilingam 37). Since the payload differs for different flights and different aircraft, the remaining available capacity - assuming passenger and baggage information is known - will still tend to be uncertain and stochastic in nature. Another critical aspect of forecasting air cargo is that cargo capacity is 3-dimensional in nature: volume, weight and number of container positions all play a role in the available cargo carrying capacity (Kasilingam 38). This is in stark contrast to the 1-dimensional nature of passenger forecasting. In the latter case, the only factor to consider is the number of seats on the aircraft. The arising issue for air cargo is that the factor constraining the available capacity may be different for each flight. In the case of high-density shipments, capacity in terms of weight may be the constraining factor. Additional volume may still be available for cargo, although no additional weight may be lifted. In other cases, the weight and volume may be available for a given shipment, but the necessary shaped container may not have a position in the aircraft. Thus, the relative simplicity of the passenger forecasting becomes more complex in the case of air cargo. © Sheelah Anne Turner 24 6. G R A V I T Y M O D E L The gravity model used in transportation is based on Newton's Law of Universal Gravitation as is used in Physics. Newton's Law postulates that the force of gravitational attraction between any two massive bodies is proportional to their masses and inversely proportional to the square of the distance between their centres (Pogge). Mathematically this can be expressed as follows: d where F = the force exerted between the two masses; G = the Gravitational Force Constant; Mi = the mass of the first object; M2 - the mass of the second object; d = distance between the objects' centres. This model has been modified for use in transportation research, and primarily for urban transportation research. In the modification, the model attempts to explain that the volume of traffic (force exerted) between two areas (two bodies) is proportional to generation and attraction factors (factors equivalent to mass) and inversely proportional to impedance factors (factors equivalent to distance). Thus, the gravity model can be formulated in transportation as follows (Alcady 1967): Ta = K - ^ -<j jy ij where Ty = the traffic volume from i to j; d = the forces of generation in i; Aj = the forces of attraction in j; Iy = impedance factors between / and j; and K, a, j5, y = constants to be estimated. In the most "nai've" model, the populations of the two cities (P, and Pj) are used as forces of generation and attraction, and the distance between the two cities (Dy) is representative of the impedance factor. © Sheelah Anne Turner 25 Thus, the model becomes pa pP However, there are numerous other factors that can be used as factors in the model. Generation factors can include the personal disposable income of the people living in the area, the gross domestic product of the area, or the level of business activity. Attraction factors can include the number of tourists to the region, the gross domestic product of the area, or the level of business activity. Impedance factors could include the cost of travel, the time taken to travel, and regulatory issues. Any of these additional factors could be included easily in the model through the inclusion of additional terms, provided there is no strong correlation between the factors in the model. Specifically, both distance and cost are important determinants of travel, but since they are likely to be highly correlated, only one can be included in the model (Howrey p218). 6.1 Examining the Role of Tariffs One factor that has not been examined as an impedance factor in gravity models is that of tariffs. There are two main reasons for this, both factors somewhat related to the other. Firstly, there seems to be limited existing research on the application of gravity models to the transportation of cargo - most research has been in the application of transportation of passengers, whose personal travel decisions are not in any way affected by the presence of tariffs. Secondly, most of the existing literature examines domestic transportation patterns - very little research appears to have been undertaken across borders, where tariffs would start to play a role. Alcady (1967) used 16 city-pair routes from a cross-section of the California city-pair grid, with passenger traffic across 4 transportation modes. Howrey used 30 of the most heavily travelled passenger routes by air emanating from Cleveland, Ohio in his study. Thus, it is evident that the effect of tariffs has not been examined in the context of the gravity model. International economics focuses on the role of tariffs in international trade, and highlights the negative effect tariffs have on the volume of imports. This paper will attempt to examine the effects of tariffs in the gravity model through the use of tariffs as an impedance factor in this model. 6.2 Aggregate versus Disaggregated Models Another interesting application of the model is that the model can be applied to both aggregated and disaggregated data. This is due to the formulation of the gravity model, which does not include variables reflecting the specific characteristics of the disaggregated data. For example, in the disaggregation of data according to transportation mode, Alcady (1967) noted that no variables are included which reflect modal peculiarities. Only variables which represent more general factors are included. Alcady (1967) further hypothesized that aggregation over different transportation modes tends to eliminate some of the peculiar characteristics of travel which are evident in the © Sheelah Anne Turner 26 individual modes. Thus, the aggregated model provides better performances of the gravity model. The reasoning is that individual modes are less stable, and less responsive to the variables that are included in the model, and more sensitive to those variables that have been excluded from the model. Studies such as Alcady's have focussed on the disaggregation over transportation mode. This paper will address the issue of disaggregation across commodity types. The working hypothesis here is similar to the hypothesis which Alcady proposed: that aggregation across commodities tends to eliminate some of the specific characteristics of each commodity type. 6.3 Estimating the Model For the purposes of this research, the basic gravity model was extended in some ways to include other economic variables which were believed to have a significant effect on the demand for air cargo. The model used is as follows: E, =K ' j ' j ea where Ey = the volume of exports (in tonnes) exported from i to j; Pi = the population of exporting region z; Pj = the population of receiving region j; It = the per capita GDP of exporting region i; Ij = the per capita GDP of receiving region j; Dy = the distance between i and j. Cy = the consumer price inflation differential between i and j. Tj = the tariff barriers in receiving region j; and K, cti, 0:2, #?, ou, as, a^, a.7 = constants to be estimated. The current model is non-linear, and so natural logarithms are taken of both sides to transform to a linear model that can be estimated by ordinary least squares. This gives the following model for estimation: InEij = K' + ajln(Pi) +a2ln(Pj) + ajn(l) + a4ln(Ij) + a5ln(Dij) + a6ln(Cij) + a7ln(Tj)+s'y As discussed earlier, the model is concerned with the volume of air cargo exports from Vancouver International Airport to other countries in the world. So, let i represent © Sheelah Anne Turner 27 Vancouver International Airport, and j represent the export receiving country. Thus, since i remains unchanged for all observations, the subscript can be dropped, and the model rewritten as follows: InEij = K'+ a,ln(P) + a2ln(Pj) + a3ln(I) + a4ln(Ij) + a5ln(Dj) + a6ln(Cj) + a7ln(Tj)+£'j Since ln(P) and ln(I) are constants, this model can again be rewritten as follows: In Ej = K" + a2ln(Pj) + a4ln(Ij) + a5ln(Dj) + a6ln(Cj) + a7ln(Tj)+£) For ease and without loss of generality, this is rewritten as follows: In Ej = K" + B,ln(Pj) + PMIj) + B3ln(Dj} + BJn(Cj) + B5ln(Tj)+s'j In this estimation, the coefficients can be interpreted as elasticity coefficients of cargo volume with respect to each of the variables. This final form of the model is that which has been used for estimation and analysis in the following sections. 6.4 Implications of Logarithmic Transformations Goldberger (1968) has indicated that in the estimation of log-linear functions there is no bias with respect to the estimation of the elasticities of the original function: the estimates of Bi,...,B5 in the log-linear function are the same as the estimations of the elasticities in the non-linear function. However, he points out that is some bias in the estimation of the constant term, K, in the original formulation. Although it seems intuitive that this is incorrect. For a random variable u ~ N(/J, cr2), the variable U = e" has a lognormal distribution with mean E{u)+-V(u) p+-v2 E(U) = e 2 = e 2 Thus, eE(K)is a biased estimator of K, and should be adjusted in order to obtain the correct level of the original function. 6.5 Data By far the most challenging part of this research was the collection of sufficient, accurate data. Access to traffic volume data was extremely difficult, especially pooled data, which was the "ideal" data set required for the building of the above model. © Sheelah Anne Turner 28 The export data was based on the volume of air cargo data that was exported from Vancouver International Airport to its various international2 destination countries during 1998, as supplied by Statistics Canada. The data is measured in tonnes of air cargo transported, and is further broken down by HS04 codes3. This gives the volume of air cargo by commodity for each destination country during 1998. The original number of partner countries included in the sample set was 127 countries and territories. This was the total number of countries for which Statistics Canada had data. This number was reduced to 116 countries with the elimination of countries for which no volume of exports was recorded. Finally, the explanatory variables under consideration were collected for as many countries as possible, with countries eliminated where data was not available. This reduced the original sample size to 61 countries4 in the final sample size. This reduction in sample size is not desirable. However, upon closer examination, the majority of the countries that were eliminated from the original sample were mainly smaller trading partners, and the remaining countries comprised 89.4% of total international export trade volumes. For this reason, the sample has been deemed sufficient for the purposes of this research. 6.5.1 Cargo Volume: Ej The cargo volume, Ej, is the aggregate volume of data that is transported from Vancouver International Airport to each receiving country j. This is measured in tonnes. Volumes were chosen rather than the dollar value of the exports since over time the "real" measurement of exports is more desirable (measured in tonnes) than the nominal measurement of exports, which is prone to changes in valuation. Furthermore, since previous models have been based on number of trips, or number of passengers, both of which are "real" measures, measuring air cargo in tonnes is consistent with previous models and research. For the disaggregate model, the commodities were broadly categorised into 5 groups, as shown in the table below. 2 Excluding the US 3 The HS04 codes are a part of the Harmonization Code System, a multipurpose international product nomenclature developed by the World Customs Organization (WCO). It comprises about 5,000 commodity groups, each identified by a unique six digit code. More than 177 countries and economies use the system as a basis for their Customs tariffs and for the collection of international trade statistics. 4 All export data is shown in Appendix 1. © Sheelah Anne Turner 29 T A B L E 6.1: D I S A G G R E G A T E C O M M O D I T Y G R O U P S 1 Confidential Commodities Confidential Commodities 2 Machinery & Instruments XVI: Machinery & Mechanical Appliances XVII: Transport Equipment XVIII: Instruments - Measuring, Musical 3 Perishable Products I: Animals & Animal Products II: Vegetable Products III: Animal or Vegetable Fats IV: Prepared Foodstuffs 4 Mineral, Metal & Wood Products V: Mineral Products IX: Wood & Wood Products X: Wood Pulp Products XIII: Articles of Stone, Plaster, Cement, Plaster XIV: Pearls, Precious or Semi-Precious Stones, Metals XV: Base Metals and Articles thereof 5 Other VI: Chemical Products VII: Plastics & Rubber VIII: Hides & Skins XI: Textiles & Textile Articles XII: Footwear, Headgear XIX: Arms & Ammunition XX: Miscellaneous XXI: Works of Art The disaggregate dependant variables, Ekj, are the volume of exports of commodity group k (for k = 1,2...5) exported to receiving country j. It is also measured in tonnes. There is a level of aggregation taking place even at this disaggregate level. This was necessary particularly where the data was sparse, and the analysis less meaningful. 6.5.2 Cargo Volume: Pj The population of the receiving country, Pj, measured in millions of people, was obtained from the World Bank's World Development Indicators 2001. Population is seen as an attraction factor in the gravity model. As in many of the previous gravity models, it is assumed that as the size of the population in the destination region increases, the demand for goods and services by the country will also increase. Thus, as the general level of demand within the country increases, it is assumed that there is a parallel increase in the demand for air transported commodities. 6.5.3 Per Capita GDP: Ij The per capita GDP of the export receiving country was also obtained from the World Bank's World Development Indicators. Per capita GDP has been reported in constant 1995 US dollars for all countries. The conversion to US dollars ensures comparison between countries. Although the country of origin is Canada, and thus it would appear that a base currency of Canadian © Sheelah Anne Turner 30 dollars would be preferable, converting the values from US dollars to Canadian dollars requires the same operation on all figures, and thus has no effect on the estimated coefficients, only on the constant term. Similar to the population, per capita GDP is seen as an attraction variable in the gravity model. The rationale is that a population with a higher per capita GDP is associated with higher personal disposable income. A higher income is related to higher demand for goods and services by the country, and thus a higher demand for air transported goods. A better measure of income would be personal disposable income. This takes into account both personal and corporate taxation, and gives a better indication of the spending power of the individuals within the region. This data, however, was difficult to find, and as such per capita GDP was used. Finally, converting all values to a common currency, whichever base currency is chosen, is performed using a currency exchange rate. This adjustment methodology does not account for the relative price levels between countries, which may also affect the purchasing power of consumers in different countries. A better variable would be one that adjusts by a purchasing power parity (PPP) measure, instead of merely a currency exchange. Once again, this data was not easily available. 6.5.4 Distance: Dj The distance is taken as the straight line distance between Vancouver and the city with the largest international airport in the receiving country. These distances were obtained from Indo.com, and are measured in kilometres. There is some debate as to the relationship between distance and volume of travel. Alcady (1976) indicates that the relationship is likely to be different for air travel than for other modes. He indicates that for passengers, the proportion of trips taken by air is an increasing function of distance. This research will decide whether the same relationship is true for air cargo. If the relationship is indeed an increasing function, then distance will no longer serve as a force of impedance. A better distance measure would be one that measures the actual flying distance between airports, since this value is more accurate. Firstly, this was difficult to define for non-direct routes. Secondly, this data was not readily available for all countries included in the sample. 6.5.5 Consumer Price Inflation Differential: Q The consumer price inflation is taken as the annual percentage change in Consumer Price Index (CPI) for each of the countries. The data is taken from the World Bank's World Development Indicators 2001. The differential is taken as the difference in consumer inflation between the receiving country and Canada. Consumer inflation influences consumer demand for products: higher inflation tends to erode the purchasing power of consumers, and thus reduces the demand for products. © Sheelah Anne Turner 31 This lower demand for goods translates into lower demand for imported goods, including those received by air. The inclusion of the consumer inflation differential attempts to take into account the purchasing power of consumers that was missing from the per capita GDP. 6.5.6 Tariff: Tj The tariff is the simple average tariff across all goods and services. It is expressed as a percentage, and is taken from the World Bank's World Development Indicators. Two measures of average tariff are conventionally used: the simple mean and the weighted mean tariff. The weighted mean tariffs are weighted by the value of the countries trade with each of its trading partners. As indicated by the World Bank, the simple averages are a better indicator of tariff protection than the weighted average. Weighted averages are typically biased downward because higher tariffs discourage trade and thus reduce the weights applied to these tariffs. As is consistent with economic trade theory, higher tariffs tend to reduce the volume of exports into the country. Thus, tariffs represent a force of impedance between Vancouver and its trading partners. 6.5.7 Other variables considered At this point, it is worth mentioning other variables that were considered for the current research. Some were unavailable, and others were tested but discarded. Both could provide ideas for future investigation. There were a few variables that were sought but were not available. The cost of travel and the time for travel were two such variables. The cost data was unavailable for the current research for a few main reasons. Firstly, the rates charged by different carriers vary widely, and so a true value would require a significant data collection effort. Secondly, the characteristics of a "typical" parcel are very difficult to define, due to the wide variety of commodities transported. Thus, defining the cost for a "typical" parcel is challenging. Thirdly, the fierce competition between cargo agents has resulted in the cost data being treated as extremely confidential, and is not shared very readily. The travel time was also unavailable for a few reasons. Firstly, it was difficult to find time data in appropriate units: although time was available in days, finer units (such as hours) would be more useful. Secondly, there are not always direct flights between the origin and destination airports. There are many different possible routes, each with a different flight time and transfer time. This complicates the collection of data. There were two other variables that were initially tested, but were discarded due to insignificance. The percentage of urbanisation of the receiving country was originally included in the model. It was hypothesized that the percentage urbanisation would change the types of commodities demanded: highly urbanised countries would have a © Sheelah Anne Turner 32 higher demand for electronics, machinery and instruments. However, urbanisation was not significant in any of the models. The currency exchange rate was also included in an earlier model. Although it is still maintained that the exchange rate is important in affecting the demand for traded goods, the model did not find significance in a static, one period model. Perhaps with the incorporation of time series data, exchange rates would play a more significant role. 6.6 Transforming the Variables Variable transformation through the natural log function was straight forward for most variables. Some variables (consumer inflation differential, tariffs and disaggregate exports) were more difficult due to the occurrence of non-positive values. There seems to be no standard methodology for dealing with this problem. One method involves truncating the data and removing non-positive values. In this research, truncation of the data would introduce bias into the data. Firstly, removing all non-negative values would significantly reduce the sample size, which is undesirable. Secondly, removing non-positive variable values would remove countries with inflation rates lower than Canada, as well as countries with no tariff barriers, which would bias the sample. As such, transformations have been used which do not reduce the sample size, but do linearly map the data into the positive real numbers. The transformations discussed below are by no means perfect, but provided a potential solution to the problem. 6.5.1 Consumer Price Inflation Differential: Cj Due to taking the consumer price differential between the receiving country and Canada, there are instances of negative values of the variable. This was evident in 13 out of the 61 countries. Since this research is addressing the differential between two countries, and thus is more of an ordinal measure than a cardinal measure, the proposition is to add Kc to all Q's, where Min{Cj\ This transformation preserves the ordinal nature of the variables, while ensuring all instances of the variable are positive. 6.5.2 Tariff: 7} In the case of tariff data, three countries out of the sample of 61 countries had no tariff barriers, and so had 0 for tariff rate. Due to the absence of negative numbers, the transformation used for consumer inflation differential was not applicable in this case. Instead, the following transformation was used: © Sheelah Anne Tinner 33 Tj '=T. +0.01 The intention was to preserve the ordinality, while adjusting the values as little as possible. 6.5.3 Disaggregate Exports: Ekj Disaggregate exports had occurrences of 0 values. Again, due to the absence of negative numbers, the transformation used for consumer inflation differential was not applicable. Instead, a transformation similar to that used for tariffs was used: Ekj'=Ekj +1 Again, this preserves ordinality with minimum disturbance to the values. © Sheelah Anne Turner 34 7. R E S U L T S A N D FINDINGS A l l model estimations were carried out using ordinary least squares in NCSS. The full results of the regressions are included in Appendix 2. The results of the aggregate and disaggregate models were compared with each other. The standard criteria were used in evaluating the performances of the models. In particular, the agreement of the signs with the results expected on the basis of a priori knowledge, the statistical significance of the coefficients, the explanatory power (R2) and significance of the entire relationship were all considered in judging the estimated equations. 7.1 Data Analysis Descriptive statistics for each of the variables was calculated. These are shown in table 7.1. TABLE 7.1: DESCRIPTIVE STATISTICS OF NATURAL LOGS OF DATA i ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ ^ llifii&: J • BG86 Exports (Total) 4.7303 2.0414 4.8283 0.0000 9.2152 Exports (Model 1) 2.9854 2.8040 4.0775 0.0000 8.4194 Exports (Model 2) 3.5894 1.9926 4.0604 0.0000 8.1047 Exports (Model 3) 1.6111 2.2308 0.0000 0.0000 8.0196 Exports (Model 4) 1.9067 1.8947 1.3863 0.0000 6.5820 Exports (Model 5) 1.8295 1.7763 1.6094 0.0000 5.7777 Population 3.0235 1.5409 2.9806 -0.9163 7.1246 Income 8.4414 1.5459 8.2744 5.2149 10.8122 Distance 9.1370 0.2893 9.0821 8.2774 9.7220 Tariff 1.8221 1.6499 2.1412 -4.6052 3.8397 Consumer inflation differential 1.6844 1.1970 1.7285 -1.8618 4.4502 The descriptive statistics reveal a few interesting points. In terms of the dependant variables, the aggregate volumes displayed symmetric behaviour, with the mean and median very close. The other dependant variables displayed more asymmetric characteristics, with the medians for models 1 and 2 larger than the mean (negatively skewed) and the medians for models 3, 4 and 5 smaller than the mean (positively skewed). This is reasonable, since the disaggregate models were more sparse than the aggregate model. In particular, the median of 0 indicates very sparse data in the model. The standard errors are larger for models 1 and 3, and fairly consistent for the remaining models. © Sheelah Anne Turner 35 The explanatory variables were more consistent, with lower standard deviations. There was also more symmetry evident, with the means and medians much closer in value. In particular, distance seemed to have a very narrow range, and very small standard error. This is intuitively correct, since the international trade partners have to be at least a continent away5, and can be no further than half a globe away. 7.2 Testing of Assumptions In the estimation of linear regression models, there are some assumptions that are made in order to ensure that the estimated coefficients are B L U E - best linear unbiased estimates. This is necessary for both hypothesis testing and model fit. 7.2.1 Normality Assumptions In all linear regression models, it is assumed that normality exists in the errors, Sj. Although the normality assumption is not required to perform the least squares regression itself, the least squares estimators perform better under normality than under conditions of non-normality on the Sj. However, the assumption of normality is required for the validity of hypothesis testing, and confidence interval estimation. Thus, the regression residuals in each estimation were tested for normality. Testing, residual plots and normality plots are used to determine whether the residuals are indeed normal. Details of the tests and plots are provided in Appendix 2. Initial expectations were that the transformed data would violate the assumptions of normality in some, i f not all, of the models. Analysis of residuals in all estimated models indicated that most models displayed normal residuals. Two of the disaggregate models did reject normality in the residuals - Perishable Products and Other Products. There was no discernable pattern in the residuals to indicate a standard transformation. These models need to be viewed more cautiously. 7.2.2 Multicollinearity Multicollinearity occurs when there are near linear dependencies between the explanatory variables. Under multicollinearity, the influence of individual explanatory variables on the dependant variable is difficult to isolate, and the coefficients are biased. The pair-wise correlations between the explanatory variables are shown in the following table. 5 US not included in the sample. © Sheelah Anne Turner 36 T A B L E 7.2: C O R R E L A T I O N M A T R I X OF E X P L A N A T O R Y V A R I A B L E S ^^^^^^^^^^^^^^^^^ 111SIIS1 ^^^^^^ 1.0000 -0.4067 0.1422 0.3586 0.1388 -0.4067 1.0000 -0.3521 -0.6137 -0.4719 0.1422 -0.3521 1.0000 0.0689 0.0303 0.3586 -0.6137 0.0689 1.0000 0.3831 CO\SI MIR IMl \IION 0.1388 -0.4719 0.0303 0.3831 1.0000 From the table it is evident that some correlations exist between some of the variables, particularly tariff and income. Although the pair-wise correlations are not extremely strong, it is still necessary to ensure there are no linear relationships between multiple explanatory variables. The eigenvalues and eigenvectors of the correlation matrix are important in the detection of multicollinearity. The nearness to zero of the smallest eigenvalue is a measure of the strength of a linear dependency. Multicollinearity can be measured in terms of the ratio of the largest to the smallest eigenvalue. This is called the condition number of the correlation matrix. When a condition number exceeds 100, multicollinearity exists. The eigenvalues are shown in the following table, together with the condition numbers. T A B L E 7.3: E I G E N V A L U E S O F C E N T R E D C O R R E L A T I O N S WSBBGSm VARIABLES 1 \ 2.2997 1.00 2 1.0231 2.25 3 0.8339 2.76 4 0.5484 4.19 5 0.2949 7.80 Since the largest condition number of 7.80 is significantly smaller than 100, the conclusion is drawn that multicollinearity is not present in the current data set. 7.2.3 Influential Observations The existence of outliers in a data set could be problematic in regression analysis, as the outlying data point could exert an undue amount of counterproductive influence on the regression results. It is vital to be able to identify these "influential" observations and determine the extent to which the estimated coefficients and predicted values are influenced by them. © Sheelah Anne Turner 37 A standard measure to test the presence of influential observations is the calculation of a D F B E T A for each regression coefficient. This measure indicates the number of standard errors that the coefficient changes i f the z'th observation were excluded from the data set. If the coefficient would change by more than 2 standard errors, the data point is influential and should be investigated further for possible exclusion from the model. Considering the wide range of countries included in the data set, it was expected that some countries would indeed be influential observations. However, the DFBETA's for each of the regression coefficients were small ( |DFBETA| <1 for most observations and models) and thus it was concluded that there were no influential observations in the data set. 7.3 Estimating the Model Since all models displayed no multicollinearity and had no influential data points, the analysis of the regression models could proceed. The evidence of normality violation in two models (Perishable Products and Other) was kept in mind when analysing the results. A l l explanatory variables were included in the original model. Using a backward selection process, variables were removed from the model until a parsimonious model was obtained: one achieving a balance between simplicity (as few explanatory variables as possible) and fit (as many explanatory variables as needed). The results of the initial models are included in table 7.4, and the results of the final model (after insignificant variables were removed) are included in table 7.5. T A B L E 7.4: R E S U L T S O F INITIAL A G G R E G A T E A N D D I S A G G R E G A T E G R A V I T Y M O D E L S ; . . : - V . • : . ' / * " E (Total) -14.4514 (-2.3504)** u.yu55 (8.1116)*** u.7olv (5.0256)*** 1.2uJ3 (2.0427)** -0.3159 (-2.5578)** -0.2421 (-1.6146) 0.6840 El (Confidential Commodities) -21.5888 (-2.5535)*** 0.8770 (5.7130)*** 1.1106 (5.3269)*** 1.5540 (1.9184)* -0.3162 (-1.8614)* -0.6387 (-3.0965)*** 0.6833 E2 (Machinery & Instruments) -13.7005 (-2.4783)*** 0.9956 (9.9190)*** 0.8425 (6.1800)*** 0.8487 (1.6023) -0.3317 (-2.9869)*** 0.0103 (0.0765) 0.7319 E3 (Perishable Products) 6.6063 (0.7800) 0.4817 (3.1323)*** 0.4814 (2.3048)** -0.9836 (-1.2121) -0.4417 (-2.5962)** -0.4295 (-2.0785)** 0.4980 E4 (Mineral, Metal & Wood) -19.5269 (-2.4045)** 0.5193 (3.5216)*** 0.7113 (3.5519)*** 1.5099 (1.9406)* -0.2585 (-1.5843) 0.3168 (1.5991) 0.3602 E5 (Other) -10.4773 (-1.7683)* 0.6838 (6.3559)*** 0.7409 (5.0707)*** 0.5032 (0.8864) -0.2230 (-1.8733)* -0.1227 (-0.8489) 0.6125 Notes: regression coefficients shown with t-values in parenthesis * * * denotes significance at 1% level ** denotes significance at 5% level * denotes significance at 10% level © Sheelah Anne Turner 38 T A B L E 7.5: R E S U L T S O F F I N A L A G G R E G A T E A N D D I S A G G R E G A T E G R A V I T Y M O D E L E S T I M A T I O N ^^ ^^ ^^ ^^ ^^ ^^ ^^ ^ • 9 1 E (Total) -16.8229 (-2.7781)*** 0.9208 (8.1617)*** 0.8501 (5.9257)*** 1.3364 (2.2589)** -0.3393 (-2.7271)*** 0.6691 El (Confidential Commodities) -21.5888 (-2.5535)*** 0.8770 (5.7130)*** 1.1106 (5.3269)*** 1.5540 (1.9184)* -0.3162 (-1.8614)* -0.6387 (-3.0965)*** 0.6833 E2 (Machinery & Instruments) -5.1961 (-4.2387)*** 0.9998 (9.9493)*** 0.7617 (6.4323)*** -0.3664 (-3.3742)*** 0.7193 E3 (Perishable Products) -3.3488 (-1.6138)*** 0.4784 (3.0985)*** 0.5822 (3.0263)*** -0.4043 (-2.4062)** -0.3944 (-1.9198)* 0.4846 E4 (Mineral, Metal & Wood) -20.0100 (-2.6236)** 0.4675 (3.1509)*** 0.7468 (4.7746)*** 1.5541 (2.0149)** 0.3075 E5 (Other) -5.9907 (-4.5930)** 0.6948 (6.4983)*** 0.7334 (5.8210)*** -0.2590 (-2.2414)* 0.6001 Notes: regression coefficients shown with t-values in parenthesis * * * denotes significance at 1% level ** denotes significance at 5% level * denotes significance at 10% level 7.4 Results for the Aggregate Model The first model estimated was the aggregate model. The model was estimated initially with all variables included. In the initial regression, only the consumer inflation differential was not significant (even at the 0.10 level) and so was removed from the subsequent model. The re-estimated model showed significance in all its independent variables. The aggregate estimated equation had the expected signs on the coefficients in the final model; namely positive for population and income, and negative for tariffs. The sign on the distance coefficient was positive. This provides an answer to the earlier discussion of whether cargo models would provide the same positive sign that passenger models provided. The significance of the coefficients in the aggregate model is very high, with 4 of the 5 coefficients significant at the 1% level. The population elasticity of the aggregate model is close to unity, implying the volume of exports is proportional to the size of the population of the receiving population. The income elasticity is also high, implying again a strong relationship between the income of the population and the demand for air cargo. The coefficient of distance is higher than unity, implying a strong propensity to consume exports via air the further the country is from Vancouver. This could be as a result of the limited modal competition for these commodities, the greater the distance from Vancouver. Possibly in the case of perishable products with potentially high obsolescence, the further the receiving market is from Vancouver, the more necessary it becomes to use air transportation. © Sheelah Anne Turner 39 Tariff had a relatively low elasticity, implying that large changes in the tariff rates are accompanied by only small changes in the volume of air exports. Thus, for commodities transported by air, the level of tariffs are not a high deterrent - consumers clearly have a demand for the commodities typically transported by air, and are willing to pay the additional price as a result of the tariffs. Finally, consumer inflation differential was excluded from the initial model due to insignificance. However, the p-value of 0.1121 shows only slight insignificance, and thus consumer inflation differential was still included in the disaggregate models. The small negative elasticity of the consumer inflation differential indicated that although a higher differential imposes a small impedance effect on the volume of air cargo transported, the low elasticity implies a small sensitivity to the differential. The explanatory power of the model is fairly good, with an R 2 of 0.6691. However, although two thirds of the variation in the total volume of exports by air is explained, there are other unidentified explanatory variables which account for approximately one third of the variation. 7.5 Results of the Disaggregate Models The five models used in the disaggregate model were estimated individually, and initially included all explanatory variables. It is interesting to note the different models had different significant variables in the final equation: all models included both population and income, but the inclusion of distance, tariffs and consumer inflation differential differed. This highlights some of the commodity specific characteristics referred to by Alcady. 7.5.1 Confidential Commodities: Eij The model for confidential commodities was the only model that included all of the original variables, although distance and tariffs were only significant at the 10% level. As discussed earlier, confidential commodities provide no detail as to the actual types of commodities included. Thus, it is difficult to analyse the consumer behaviour associated with these commodities. The signs of all the coefficients were consistent with a priori beliefs. The population coefficient was a little lower than in the aggregate model, but still elastic enough to suggest a strong relationship between the size of the population and demand for Confidential Commodities. Income elasticity is highly elastic in this model, greater than unity. This implies a great sensitivity of air export volumes to changes in income. Distance again provided the expected positive sign, with a very elastic coefficient of 1.55. This implies the demand for these commodities is very sensitive to the distance: the further the country from Vancouver, the higher the volumes of confidential commodities consumed. Despite the large elasticity associated with the distance variable, it is only significant at the 10% level, indicating that distance is not a strongly significant variable. © Sheelah Anne Turner 40 The influence of tariffs is very close to that seen in the aggregate model: a relatively low elasticity, with the expected negative sign. Finally, consumer inflation differential was a very significant variable in the model, in sharp contrast to the insignificance evident in the aggregate model. The explanatory power of the model is slightly stronger than the aggregate model, with an R of 0.6833. 7.5.2 Machinery & Instruments: E2j The final Machinery & Instruments model only contained three of the original variables, but the included variables were all significant at the 0.01 level. A l l the variables displayed the expected signs in the final model. In the initial model, however, consumer inflation had a marginally positive coefficient. Due to the high p-value of 0.94, the variable was removed. The population elasticity was almost unitary, similar to the aggregate model. This implies a proportional relationship between the volume of machinery and instruments exported by air, and the size of the population. The income elasticity is lower than that in the aggregate model. Lastly, the tariff coefficient is similar to the aggregate model, with a relatively more inelastic behaviour. Consequently, a large decrease in tariffs is associated with only a small increase in demand for machinery and instruments. Distance and consumer inflation were excluded from the final model. It appears that the volume of machinery and instruments exported is not influenced by the distance of the receiving country from Vancouver, nor by the inflation differential between the countries. The explanatory power of the Machinery & Instrument model is higher than in the aggregate model. The R 2 is 0.7193, indicating that the 3 variables discussed above account for over 70% of the variation in the volume of machinery and instrument exports. Although distance and consumer inflation differential are not significant explanatory variables, there may exist other commodity specific variables that would explain some of the remaining variation in Machinery & Instrument export volumes. 7.5.3 Perishable Products: ESJ The Perishable Products model excluded the distance variable in its final form, and included consumer inflation differential at only a 0.10 significance level. The signs of the coefficients in the final model all agreed with prior beliefs. In the initial mode, however, distance displayed a negative coefficient. This high p-value of 0.23 resulted in it being removed from the model. The population coefficient was highly significant, but the elasticity was much lower than seen in all previous models. One explanation is that countries with larger populations tend to rely more on self-provided perishable products. The income coefficient was likewise lower than the aggregate model: although highly significant, the elasticity was only 0.5822. This lower elasticity, however, is consistent with the relatively inelastic demand for essential economic goods. © Sheelah Anne Turner 41 The impact of tariffs is more pronounced in the Perishable Products model than was evident in any previous models. It is still inelastic, indicating changes in the tariff rates has a very small impact on the volumes of perishables that are received by air. A disturbing element to the model is the very low explanatory power - an R of 0.4846. It seems that there should be numerous other commodity specific variables which can explain the variation in export volumes of perishables. These may relate to weather conditions, or percentage of land under cultivation. 7.5.4 Mineral, Metal & Wood: E4j The Mineral, Metal & Wood model also included only three of the five variables, with tariff and consumer inflation differential being insignificant. A l l included variables displayed signs consistent with a priori beliefs. Population and income continue to display the very high significant levels. Similar to the Perishable Products model, the population variable in this model shows a much lower coefficient value than in other previous models. This implies that increases in population are associated with much smaller percentage increases in demand for minerals, metals and wood from air cargo. It seems that these commodities are obtained by other means: by marine shipping, or self produced. The income coefficient in this model is similar to the Confidential Commodities model. It is more elastic, at 0.7468. Distance is the other significant variable in this model, and is highly elastic at 1.55. The explanatory power of this model is very low, with an R 2 of only 0.3075. This is not a very good model, and there certainly must exist other variables which could better explain the demand for minerals, metals and wood transported by air. 7.5.5 Other: £5, Rather than continue estimating models with weaker and weaker explanatory powers, the remaining commodities were included together in the final model: Other. This included a very diverse range of commodities, from works of art, to textiles, to arms and ammunition. Not all of the initial variables were significant in the Other model: distance and the consumer inflation differential were insignificant. However, all variables displayed the expected signs. Population continued to be a very significant variable (at 1%) but the coefficient is markedly lower than in the aggregate model. There is a less elastic response in Other volumes to changes in population. The income coefficient is fairly similar in magnitude to many of the other models. The value of 0.7334 indicates a relatively elastic response to changes in income. Tariffs are the last significant variable included in the Other model. The value was similar to previous models, at an inelastic level of -0.2590. Although significant in the model, large changes in the tariff rate are only associated with small changes in volumes exported. © Sheelah Anne Turner 42 The Other model has a slightly higher explanatory power than the previous two models, with an R 2 of 0.6001. This may be in part due to the aggregation over smaller commodity groups. There are still other variables which could be included in the model to explain the remaining variation in Other export volumes. 7.6 Empirical Results Comparison The six models examined have given an array of values for the variable coefficients. Further, different combinations of variables are found to be significant in different models. The range of values of variables in the various models indicates varying levels of stability. T A B L E 7.6: C O E F F I C I E N T S O F G R A V I T Y M O D E L E S T I M A T I O N " r , • .",..-( ••-<.-<i-~ . v.': t' Disaggregate Range: Min 0.6948 0.5822 1.5540 -0.4043 -3.0965 0.3075 Max 0.9998 1.1106 1.5541 -0.2590 -0.3944 0.7193 Aggregate Value 0.9208 0.8501 1.3364 -0.3162 - 0.6691 Models Included in: Confidential Commodities • •/ • • • Machinery & Instruments • S • Perishable Products V V • Minerals, Metal & Wood • • Other V 7.6.1 Population: Pj The coefficients of population were highly significant in all models, with p=0.0000 in most models. This indicates the high explanatory power of population in the gravity model and the valuable force of attraction population provides. The range of elasticities can be seen in table 7.6. The range is wide, indicating a less stable influence of the variables on different commodity groups. 7.6.2 Income: Ij Similar to population, income was included in all models at a very high significance level (p=0.0000 in most models). Income is thus another important variable in the gravity model, serving as a strong force of attraction. The range of elasticities is very wide (as seen in table 7.6), indicating a less stable influence of income on the demand for different commodity groups. 7.6.3 Distance: Dj The distance variable was only included in three of the six models estimated. This is interesting in that it indicates that only certain commodities consider the distance between Vancouver and the receiving country. In the models where distance was included, it displayed a very narrow, very elastic range (table 7.6). This implies that distance is stable in its effect on demand, but only for the commodities where it is significant. © Sheelah Anne Turner 43 Most importantly, though, is the sign of the coefficient. Although intuitively contradictory, the positive sign supports findings of research in passenger models that air transport distance is positively correlated with traffic volume. This is important in the understanding of the possible destinations for air cargo, in that further destinations may provide higher volumes in the future. 7.6.4 Tariffs: Tj Only one model - Mineral, Metal & Wood - did not include tariffs in the final model. In the other five models, tariffs displayed different levels of significance: from p-0.0194 to p=0.0680 depending on the commodity. The range of elasticities of tariffs is very narrow, and rather inelastic. This very narrow range is testimony to the stability of tariffs in influencing the demand for commodities. It was interesting, however, to note the inelastic nature of exports to tariffs. This implies that large decreases in the tariffs would only result in small increases in export volumes. 7.6.5 Consumer Price Inflation Differential: Cj Consumer inflation differential was only significant in two of the five disaggregate models, and not in the aggregate models. While it was very significant in the Machinery & Instruments model (p — 0.003), it was less significant in the Perishable Products model (p=0.060). The two values were quite different, again indicating the instability of consumer inflation. It is also not very strong in providing an overall understanding of demand. 7.6.6 Explanatory Power: R2 The aggregate model displayed a fairly strong explanatory power. The disaggregate models, however, had a very wide range in the R2 values, as can be seen in table 6.6. Although the first two disaggregate models had better fitting models than the aggregate model (with higher R2 values), the remaining three models had a lower explanatory power. Thus, although the included variables perform well in some models, the other models would need to include more explanatory variables in order to perform better. 7.7 Discussion of Objective Findings The results have been discussed extensively above. The discussion concludes with a discussion of the objectives of the research. This research had two main objectives: (1) to examine the impacts of tariffs in the context of gravity models; (2) to examine whether aggregation over commodities tends to eliminate commodity specific characteristics. In the analysis of tariffs, it is apparent that tariffs have a significant influence on both the aggregate and the disaggregate models. Although the coefficients indicate a relatively inelastic response of air cargo volumes to changes in the tariff rate, the variable is still significant as an impedance factor in the gravity model. Moreover, the small range of the tariff coefficient indicates stability of the variable in influencing the demand for various © Sheelah Anne Turner 44 commodities. The only concern is that tariff was not significant in one model, Metals Minerals & Wood. For a truly stable variable, it should be significant in all models. In analysing the types of commodities in Metals Minerals & Wood, it seems that these goods may not utilise air transport as the primary means of transport. Thus, only a small portion of the total Metals Minerals & Wood destined for the receiving country are sent via air. The demand for air transportation of these goods may indicate they are more essential in nature, and hence do not consider tariffs as a relevant barrier. It is important to note that the gravity model performs according to international trade theory: that tariffs do indeed provide a barrier to international trade. In deciding whether aggregation eliminates some commodity specific characteristics, there are a few aspects to consider. Firstly, the disaggregate models have different significant explanatory variables in the final model. This indicates that different models react to different included variables, and certainly to specific variables which are not included. Secondly, the wide range of variable coefficients in the disaggregate models indicates the instability of the variable in response to the commodity being exported. The various commodities clearly display different sensitivities to changes in the explanatory variables. These provide evidence of commodity specific characteristics emerging in the disaggregate models. The benefit of the aggregate model is exactly that the commodity specific characteristics are eliminated. The model fits better than most of the disaggregate models, and the coefficients provide a measure of an "average" effect of each of the variables. 7.8 Criticisms of the Model Despite the widespread and growing use of gravity models for analysis of transportation, it is not without criticism. Not always detrimental, the criticisms provide direction for further research, and hopefully the development of better models in the future. One of the recurrent criticisms of the model is that the model is not based on the theoretical foundation of standard microeconomic models of consumer behaviour (Oum). The model is based almost entirely on the idea that the traffic flow is directly related to the size of the two areas under consideration and inversely related to impedance factors -which represent friction, inconvenience and cost of transportation. Another significant problem is the measurement of the variables. The "mass" variable is conventionally explained as population. However, other variables could be used which would provide better "mass" properties. These could be total economic activity, investment in infrastructure, value added in manufacturing. Similarly for the "impedance" factors: different measures may provide better estimates in the model. Other variables such as travel time (including loading and unloading) and travel cost could be used. © Sheelah Anne Turner 45 A basic issue in the model relates to weights being applied to the masses. (Isard 506). Isard argues that different kinds of people may contribute differently to the economy, and the population should be weighted accordingly. However, this raises the question of what weights would be applicable. Furthermore, there is a lack of theory to explain the values estimated for the exponents (Isard 515). Initial gravity models assumed values for the exponents, based on the assumptions of impacts of population and distance on the dependant variable. Later models (including this research) estimated the exponents. Finally, the air cargo forecasting model does not take into account the derived nature of the air cargo industry. Since the majority of international export cargo from Y V R is transported in passenger aircraft, cargo is restricted to the routes and flights that are available to passengers. Cargo may thus be restricted by the available capacity in the passenger flights. This is not adequately dealt with in this forecasting model, since the data may be truncated. © Sheelah Anne Turner 46 8. F O R E C A S T I N G USING G R A V I T Y M O D E L S Understanding the demand for air transportation of cargo provides insight into the determining factors and significant contributors. However, the real benefit of these models is derived from harnessing these factors and using them to forecast air cargo flows in the future. Understanding the possible future volumes assists in the strategic planning of infrastructure and other service provision. 8.1 Forecasts using Aggregate and Disaggregate Models Having analysed the aggregate and disaggregate models, both models will be used to forecast the data. As has been discussed earlier, the aggregate model eliminates commodity specific characteristics. Thus, for more accurate forecasts, it is expected that the disaggregate models together will better predict the individual commodities, and hence provide a more accurate forecast. The base period was taken as 1998. First, the estimated model was used to predict total exports for 1998. Then exports for 1999 were forecast. This year was chosen as the aggregate data was available for 1999, and thus the predictions of the model could be compared. Exports for each commodity group were unavailable for 1999. There has not been much literature on the accuracy of the forecasts obtained from the gravity model. For this reason, there has been some apprehension in forecasting of international exports. The desire is to forecast However, since the model is in terms of logs, there is some bias when taking exponentials of the estimates. In particular, although j J it is not true that j J There is an element of bias that is introduced, which should be adjusted for in the forecasts. Specifically, the expected bias of each observation is defined as E(Bias) = E(e^ - £. ) = E{Pj - einEj ) © Sheelah Anne Turner 47 The expansion of eu is as follows: °° 11' e"=Y-For simplicity, the second order approximation is used. 2 e = 1 + H -I For a simple model, with a single explanatory variable, the equation can be expressed follows: Yi = KXU-, Taking natural logarithms: In?;. = A" + alnJr , +e'. Without loss of generality, this can be expressed as: Wt = a + pvt + si Now, E{Bias) = E{e*-< -ew>) = E(e*+M - ea+pVi+£t) By using the second order approximation for e", this can be expressed as: = E[\ + a + pV; + i ( a + pV.f - {1 + a + pvt + s. + ^(cx + pVi + erf}] By expanding and taking expectations of the terms, this simplifies to = ^ Var{a) + Vfiovia, + ~ Var(p) -1 Var(et) Thus, after forecasting, the adjusted forecast is obtained as follows: bias factor - naive prediction - corrected prediction © Sheelah Anne Turner Thus, corrected prediction = naive prediction - bias factor This methodology can be followed for models with the necessary number of explanatory variables, and the corresponding bias factor can be calculated. Using the equations, the exports for 1998 are forecast using both the aggregate and disaggregate models. Three scenarios are used: i . Naive (biased unadjusted) forecast which uses the biased constant and does not adjust for the bias factor; i i . Adjusted forecast which uses the biased constant and adjusts for the bias factor; i i i . Unbiased forecast which uses the unbiased constant term and adjusts for the bias factor. The aggregate results of the three methods are shown in the table below. The percentage of actual exports is shown in parentheses. Full forecasts are shown in Appendix 3. T A B L E 8.1: G R A V I T Y M O D E L PREDICTIONS -1998 * i . A G G R E G A T E 38,840 22,998 (59.2%) 23,045 (59.3%) 47,118 (121.3%) DISAGGREGATE 38,840 21,065 (54.2%) 21,342 (54.9%) 64,260 (165.4%) Model 1 16,686 10,768 (64.5%) 10,842 (65.0%) 41,871 (250.9%) Model 2 10,381 7,580 (73.0%) 7,611 (73.3%) 13,627 (131.3%) Model 3 7,377 1,151 (15.6%) 1,228 (16.6%) 4,549 (61.7%) Model 4 2,699 667 (24.7%) 726 (27.2%) 2,467 (91.4%) Model 5 1,697 899 (53.0%) 935 (55.1%) 1,746 (102.9%) From the table above it is clear that the gravity models estimated in this research do not perform very well as a forecasting tool. Even after adjusting for the bias factor and the bias in the constant term, the forecasts do not provide satisfactory predictions. The naive and adjusted forecasts both underperform significantly, while the unbiased forecast overpredicts. This may indeed be a characteristic of logarithmic models in general. © Sheelah Anne Turner 49 Furthermore, the aggregate model slightly outperforms the disaggregate models (-21.3% vs. -65.4%). Viewed in conjunction with previous findings, the elimination of commodity specific characteristics in the aggregate model seemed to provide a better overall estimate. This is not consistent with prior beliefs. Using data for 1999, the exports are forecast for 1999. The results are shown in table 7.2, with full forecasts available in Appendix 3. TABLE 8.2: GRAVITY MODEL PREDICTIONS -1999 A G G R E G A T E 54,133 22,998 (24.5%) 23,045 (42.6%) 48,145 (88.9%) DISAGGREGATE 54,133 18,150 (33.5%) 18,429 (34.0%) 48,003 (88.7%) Model 1 6,749 6,824 26,244 Model 2 7,730 7,761 13,898 Model 3 2,078 2,155 3,575 Model 4 678 738 2,509 Model 5 915 951 1,777 Forecasting for 1999 performed surprisingly better than for 1998. The discrepancy between the aggregate and disaggregate forecasts has decreased (88.9% vs. 88.7%). This seems to indicate that the commodity specific factors are not that instrumental in forecasting exports. The strength of the forecasts has improved (from 121.3% in 1998 to 88.9% in 1999). Thus, it seems that the biased gravity model estimated has not performed well when compared to actual volumes in the forecast period. The adjustment for the bias factor also does not seem to provide a significant improvement in the forecasts. The unbiased model (correction of the constant term) seems to perform much better. The dramatic improvement in the forecasting should be viewed cautiously. The improvement is more a result of increases in actual volumes, which rose sharply to more closely match the forecast, rather than the forecast adjusting to forecast the actual volumes more accurately. It appears that there are other explanatory variables which should be included to more accurately capture short term changes in export volumes. © Sheelah Anne Turner 50 8.2 Problems in Forecasting Demand It is apparent in the above discussion that the forecasts are problematic. The most severe problem is the adjustment for bias that is necessary. The adjustment factor that was used in this research did not seem to adjust the forecasts sufficiently, and the adjusted forecasts were still inaccurate. The unbiased forecasts (adjusting the constant term) performed significantly better, but there is still room for improvement. Further, the model does not seem to be capturing the changes in the environment affecting the volumes of air cargo exported. The most obvious shortcoming is that the variables included in the model are stable over time, or only changing at a very slow rate. This makes the gravity model unsuitable for short-term forecasting, and apparently more suitable for long-term forecasting. Thus, explanatory variables should be included which have more fluctuations in the short term. However, in long-term forecasting the assumption of a stable model structure and constant parameter estimates seems unreasonable. It is more likely that the model structure will also change over time, than the elasticities of variables will remain unaltered over time. Ultimately, as Quandt indicates, the long run the model should incorporate changes in the exponents over time. Model specification also affects the accuracy of forecasts. The inclusion of behavioural characteristics, such as consumer tastes and lifestyle changes, has been overlooked. It can not be denied that these factors do change over the long-term, and would alter the demand for commodities. This in turn would alter the demand for air transportation. Attention also should be given to the interdependence of the demand and supply for air transportation. It may be necessary to model this relationship by a system of simultaneous equations. Forecasting error can have significant impacts, and there is concern over whether it is better to overestimate or underestimate. Both occurrences can have dramatic impacts, particularly where infrastructure investments are concerned. The preferable case may depend ultimately on the symmetry of the costs of the error. A case of over-estimation may result in the development of a half-empty airport; under-estimation may result in an airport with insufficient capacity which could lead carriers to seek other airports. © Sheelah Anne Turner 51 9. F U T U R E IMPACTS O N A I R C A R G O Air cargo has shown remarkable growth in the past years, both within Canada and globally. With continuing changes in the global economy, rapid expansion of the Internet and its capabilities, and developments within the airline industry regulation, this growth could potentially continue, given the right circumstances. Unfortunately, not all global influences are positive, and the recent terrorist attack in New York is likely to have a dampening effect on the growth in air cargo. This section will go into further detail on some of these aspects. 9.1 Canada's International Air Policy The Canadian government, as with governments of other countries, has a regulatory policy governing the operation of air transportation services between Canada and other countries. Any airline that would like to provide international air services between Canada and another country needs to obtain the approval of both governments prior to beginning scheduled flights. For these scheduled air services, the approval takes the form of a "bilateral trade agreement" between Canada and the applicable country. These agreements include numerous provisions such as the cities to be served, number of carriers that can fly the routes, and the flight frequency. Canada currently has 70 such bilateral agreements. The current international air policy was introduced in December 1994 and completed in March 1995. The only change since March 1995 was made in December 1999. The current international air policy is under discussion and review. The new policy is intended to take into account a wider spectrum of Canadian stakeholders, including air carriers, airports, communities, travellers and the trade and tourism sectors. However, this may be challenging, given the potential conflicts of interest that may arise between different stakeholders. The following possible changes to the current policy could increase the volumes of air cargo in Canada: i . Canada's negotiating position with its major bilateral trading partners is one of searching for more open agreements. This could result in an increased number of cities to be served within the foreign country, and an increase in the flight frequency to the country. This potential increase in the market size within the bilateral trading country could result in an increase in air cargo transported. i i . Air relations between countries have historically tended to focus on passengers services, and, as such, most bilateral agreements to date have not distinguished © Sheelah Anne Turner 52 between passenger and all-cargo services. As all-cargo services continue to grow in importance, it appears important to distinguish between passenger and all-cargo services. Further, there has been an interest from some stakeholders to negotiate open all-cargo agreements in an attempt to encourage foreign carriers to serve Canadian markets. The introduction of additional freighters would provide new capacity for products which can only be carried by freighters. i i i . Fifth freedom rights provide for a carrier to transport goods and passengers between two countries outside of the home country. By allowing carriers fifth freedom rights, a foreign carrier could combine different markets on a single flight. This is particularly advantageous in cases where a market is inaccessible on a non-stop flight, but may be accessible i f fifth freedom rights are allowed in an intermediary country. For example, an Asian carrier could combine Canada and South American in a single flight. In the past Canadian carriers have opposed the granting of fifth freedom rights, as it is viewed as diverting traffic from Canadian carriers. However, fifth freedom rights could make some routes more competitive and provide additional air cargo capacity, and for this reason more liberal fifth freedom rights are being investigated. iv. Canadian bilateral agreements typically contain various provisions regarding the ownership and control of foreign carriers serving Canada. The standard provision requires that carriers should be "substantially owned and effectively controlled" by nationals of the countries designating them. The primary reason for this stipulation is to have some element of control over who receives the economic benefits from the service. If this becomes a more negotiable point, it is likely that additional foreign carriers may have access to the Canadian markets, bringing with them additional cargo. The revisions to Canadian International Air Policy could have major impacts on air cargo. With a move towards more liberal policies and the consequential opening of markets, additional markets would become more accessible, and additional flights would provide additional air cargo capacity. These revisions together could provide additional air cargo for Canada. 9.2 The Internet The growth of the Internet and e-commerce is likely to have the most significant impact on air cargo. Due to the continuous rapid expansion of the Internet and its capabilities, predicting the actual effect is challenging at best. There has been an unprecedented increase of Internet transactions, particularly in the area of business-to-business (B2B). This growth is expected to continue in the future, as is a smaller growth in the smaller business-to-customer (B2C) sector. The growth will increase the demand from both businesses and customers for time-definite delivery of ordered goods. For businesses, the Internet provides a greater transparency in markets. © Sheelah Anne Turner 53 This enables them to dramatically reduce the cost of inventories with a concurrent increase in frequency of procurement. Internet consumers are, however, sensitive to price and this may favour the use of ground transportation within North America. As intercontinental Internet transactions become more common, it can be expected that there will be an increase in air cargo. Document transmission remains an uncertain area when trying to understand its effect on air cargo. It is expected that there will be some movement away from paper transmission of documents towards Internet transmission. However, there are still some obstacles to be overcome, including security issues and dependability factors. Until these issues are fully resolved, it is not expected that significant document volumes will be diverted from air express to the Internet in the immediate future. 9.3 Electronic Data Interchange The increased use and development of electronic data interchange (EDI) will also impact air cargo volumes. EDI should allow improved inventory handling by businesses, and thus drive businesses towards "just-in-time" inventory control. By managing inventory on a "just-in-time" basis, minimal inventory is held by businesses, and better integration of data transfer, ordering and payment systems into one management system can occur. As a result, there may be a more frequent demand for goods, with a smoother ordering and payment system. Electronic Data Interchange is also used by Customs, although currently in a very limited capacity. "Wheels up" clearance involves the electronic submission of cargo air waybills from the aircraft to Customs at the destination airport for processing while the aircraft is still en route. Customs "wheels up" clearance has a great impact on the processing of air cargo once it arrives at its destination airport. With the wider use of the "wheels up" clearance process, the customs clearance process is simplified significantly, and is hence faster. This shorted processing time of air cargo reduces potential delays at airports, and can make air transportation a more attractive transportation option for shorter distance shipments. 9.4 Electronics Industry Growth One of the major commodities transported world wide by air is electronics. The electronics industry has seen substantial growth in the past few years, and this growth is expected to continue in the future. Researchers in the supply chain of electronics have identified changes in the electronics industry that may have impacts on the need for transportation of electronic components. In particular, the coming years are expected to see product lifecycles becoming shorter and shorter: products will be reaching obsolescence much faster and wil l be requiring replacement more frequently. Mass customisation of products, similar to the way that Dell operates, is expected to become © Sheelah Anne Turner 54 more common. As a result, companies are more inclined to perform more frequent, smaller shipments, than larger less frequent shipments. In addition, the low cost barrier to entry is bringing heavy competition from numerous smaller start-up companies. A trend towards manufacturing and logistics outsourcing is driving significant growth in the electronics manufacturing service and third party logistics providers. Outsourcing requires additional transportation for components prior to assembly, followed by transportation of the final product to its market. These factors all translate into increases in air transportation of electronic components, and hence total air freight. 9.5 Impact of September 11th 2001 Terrorist Attack Unfortunately, despite the potentially positive growth influences discussed above, the impact of the terrorism attack on the World Trade Centre in New York is having a very dramatic negative effect on the economy and the airline industry, including air cargo. The overall economic impact of the attack is spreading much further than the airline industry. There has been a sharp drop in consumer spending, and a loss in consumer confidence. This is predicted to translate into an economic contraction in the US particularly, and worldwide potentially, in both the third and fourth quarters of 2001. The close relationship between air cargo and economic growth could result in a decline in air cargo volumes over the same period. The heightened awareness of airline security issues will impact both passengers and cargo. Within countries, as well as across borders, the increased security precautions could raise business costs and lower productivity. Further, it is expected that aircraft will no longer be able to turn around in under 45 minutes, as has been possible in the past. This will lengthen the processing time of air cargo, and may make other modes of transport more attractive for short haul carriage. Further, these slower transfer times may drive airlines towards larger aircraft and lower flight frequency, reducing cargo carrying capacity. The impact of the terrorist attack on the airline industry itself will be devastating in both the short-term and long-term. The industry has lost in excess of $650 million as a result of closed air space in the US alone. The total is expected to rise to $2 billion for the third quarter. Analysts expected further losses of $3 billion for the fourth quarter. Canada and European countries have also experienced significant losses. Within the US, the airline industry has shed more than 100,000 jobs. Internationally, the job losses have been on a smaller scale. A l l this does not bode well for economic growth in the short term. One of the most significant effects on the airline industry has been the increased cost of insurance. Airline underwriters lowered the coverage limit from $700 million to $50 million for damages to third parties in the event of attacks on airliners. Most airports and aircraft leasing contracts stipulate minimum coverage of $750 million to $1 billion. As airlines have been hard pressed to cover the additional insurance costs, many have grounded their aircraft, or sought government assistance to provide emergency insurance © Sheelah Anne Turner 55 coverage. This has resulted in increased insurance costs for airlines, and reduced cargo carrying capacity from grounded aircraft. The additional costs incurred by the airlines, including cargo airlines, can be expected to filter through to the end users in the form of higher prices. Northwest Cargo was the first cargo airline to alter its pricing policies to compensate for additional aircraft hull insurance and new security procedures. In the US, cargo airlines will receive approximately 10% of the $5 billion financial aid package offered by the US government, which should compensate them for some of the additional costs. Despite the financial aid packages and government assistance with insurance, many airlines still face the threat of bankruptcy. As a result, many airlines have been cutting routes, reducing flight frequency on certain routes, and reducing their fleet size. Fleet size reduction has been either through grounding of aircraft, or the sale of aircraft. Airlines facing these problems include Swissair, Varig, Continental and Virgin Atlantic. This will reduce the worldwide carrying capacity of cargo both as belly hold in passenger aircraft, and in freighters. © Sheelah Anne Turner 56 i o . D I R E C T I O N S F O R F U R T H E R R E S E A R C H This research is part of a limited literature into the application of gravity models to air cargo. The main limiting factor in this is the lack of accurate data. However, should data become available in the future, there is a wealth of opportunities for future research. 10.1 Time Series The form of the "ideal" data set would be pooled data. Pooled data would include the volume of exports and imports per country by commodity by volume (in tonnes) for a range of years. This would allow analysis of the changes of trading partners over the years, as well as changes in the commodities consumed. Further, changes in consumer tastes could be identified, and possibly included in future models. By using pooled data, exponent adjustments discussed by Quandt could be included. The model suggested is as follows: p«i(0pa 2U) j<*i(t) jat(i) E = K"^0 ij r )« 5 (')/-" <*6 (') 7-«?(0 As a first approximation, the functions aft) may either be expressed as linear or quadratic functions of time. 10.2 Other Variables There are variables in the model which could be altered in favour of better measures. Although it is not expected that the improved measures would dramatically change the results, more accurate measures may introduce more stability in the models. As mentioned earlier, a measure of per capita income instead of per capita GDP would give an indication of the purchasing power of the consumers, after the effects of taxes. This should also be measured in terms of PPP and not in terms of a fixed currency. The distance between cities should be measured in terms of actual flying distance, and not the straight-line distance. Other variables that could be included as impedance factors are the cost of transport, and the time for travel. Research exists on the testing of various travel related variables as impedance factors, but there is inconclusive evidence as to which of distance, cost or time is the best measure. As has been discussed earlier, service level also impacts the demand for transportation. Variables such as number of flights per day and reliability of service can capture an element of service quality, and could be included in further models. With the inclusion of time series data through the use of pooled data, currency exchange rates could be included in future models. The changes in currency value over time influences the volume of trade between countries. This could be not investigated in a one period model, but should be considered in a multi-period model. © Sheelah Anne Turner 57 Another important aspect to measure is the structure of the economy. An economy that is more focussed in primary industries will demand different commodities than an economy that is more focussed in secondary and tertiary industries. Furthermore, the secondary and tertiary industries dominant in the economy will alter the nature of commodities demanded. Although percentage urbanisation was used to attempt to capture this, it did not perform well, and another measure should be sought. 10.3 Transformations on the Data Log-linear models require all data to have positive values. In the current context, this was not possible for all variables. Crude data transformations were used to ensure all data was positive, but there may be better ways to achieve this. Either truncated distributions should be used, or better transformation should be used in the future. 10.4 Inclusion of other Origin Points The current research used a single origin point. As such, all explanatory variables relating to the origin city were absorbed in the constant term, and no specific analysis relating to the origin could be included. Thus, no generation factors could be included in the models. By incorporating numerous origin points, generation factors could be included in the model. Factors such as population and income at the origin could also influence the volume of exports. 10.5 Air Cargo as a Derived Industry As has been mentioned earlier, air cargo in Canada is a derived industry, since it relies on passenger flights. Thus, the truncated nature of the data could play a vital role in the construction of the model. In order to understand whether the data is truncated, it may be necessary to undertake a capacity study of the passenger flight network, and determine whether there is excess demand or excess supply. In the case of excess demand, the data may be truncated and modelling should take this into account. 10.6 Bias Estimation The current method of bias estimation and adjustment could be improved upon. Although the theoretical approach behind the adjustment seemed good, the adjustment factors calculated did not seem to provide significant adjustments, and the forecasts were not improved much by the adjustment factor. This aspect of the forecasting requires substantial attention. 10.7 Non-Linear Estimation Although theory indicates that the estimation of the log-linear model is equivalent to that of the log-linear model, it may not indeed be the case. For this reason, future research could be pursued in the area of estimation of a non-linear gravity model. © Sheelah Anne Turner 58 i i . C O N C L U S I O N The undertaking in this research was to investigate the application of the gravity model to air cargo and forecast future international export volumes. Previous applications of the gravity model have focussed primarily on passenger transport and interurban travel. There has been very little research in the area of air cargo. Before analysing air cargo, it is important to understand how different factors impact the choice of transportation mode. The underlying price of the transported goods, as well as the cost of transportation, impact the choice of transportation mode. Furthermore, shippers will frequently trade off the cost of transport against the time taken to transport to the goods. The additional services provided by transport carriers may favour one transport mode over another. These and other factors are considered when selecting the transportation mode. Vancouver International Airport (YVR) was the focus of the research, with international air cargo exports providing the data for air cargo flows. Y V R has shown substantial growth in air cargo volumes since 1994, and processes a wide range of commodities annually. The primary trade partners are Asian, receiving more than 60% of international air cargo exports from Y V R . It was also evident that the growth in air cargo is closely related to economic growth. In order for Y V R to provide the necessary facilities and infrastructure in the future to assist in the transportation of cargo, they need to understand the potential growth of air cargo. The gravity models were estimated both on an aggregate and a disaggregate level (by commodity type). There were two main objectives in the analysis of the model performance: (1) to examine the effects of tariffs in the context of gravity models, and (2) to examine whether aggregation eliminated some commodity specific characteristics in the disaggregate models. The results indicated that tariffs performed well as an impedance factor, with high levels of significance. This is consistent with international trade theory. Further, the aggregation of export data across commodities seemed to eliminate some of the commodity specific characteristics in the disaggregate models. The preference for one over the other rests on which provides more accurate forecasts. Forecasting of air cargo is essential, and may be used in support of strategic planning and investment. The forecasting ability of the estimated gravity model was tested by predicting 1998 (the year of fitting) and forecasting for 1999. Overall, the forecasts did not perform well. After adjusting for expected bias, the forecasts were not significantly improved beyond the naive forecasts. The unbiased forecasts were an improvement, but still do not account for all the exports. Further, it is apparent that the model does not fit the data well, and that there are explanatory variables which should be included to improve model fit and forecasting ability. The aggregate model performed better than the disaggregate model, although the prior belief was that the commodity specific disaggregate models would capture changes better. © Sheelah Anne Turner 59 It is evident in both the fitting of the model and the forecasting that there are still many problems that require attention. The inclusion of additional explanatory variables should assist both the fit of the model, as well as the accuracy of forecasts. The future of air cargo is dependant on many factors. The characteristics of both air transportation and competitor transport modes may result in switching from one mode to another. In addition, the changing nature of international business, and the manner in which business is conducted, could see further changes in the entire logistics industry as consumers' demand for time-definite delivery of goods. In the short-term, the impact of the economic downturn and terrorist attack in New York and Washington needs to be worked through before recovery of the air transport industry can be expected. However, once recovery has occurred, it is expected that the air transport industry will again fulfil a vital role in international logistics, and show strong growth again. © Sheelah Anne Turner 60 R E F E R E N C E S Airbus Global Market Forecast 2000, Airbus, 29 March 2001 < http://www2.airbus.com/media/market.asp>. Armbuster, William. "Airline bailout includes insurance." AirCargo World Online, 25 September 2001 <http://www.aircargoworld.com/Break_news.htm>. Armbuster, William. "Continental: 12,000 jobs to go." AirCargo World Online, 25 September 2001 <http://wrww.aircargoworld.com/Break_news.htm>. Armbuster, William. "Northwest hikes cargo charges." AirCargo World Online, 25 September 2001 <http://www.aircargoworld.com/Break_news.htm>. Alcady, Roger E. "Aggregation and Gravity Models: Some Empirical Evidence", Journal of Regional Science Vol 7. No. 1. (1967): 61-82. —. "The Demand for Air Travel", Studies in Travel Demand, Ed. Ronald E. Miller. Mathematica, Princeton, (1965): 49-97. Ashall, Mark. "Air cargo growth strong at about 10% in 2000", 25 September 2001, <http://www.tdctrade.com/sj ippers/13/01 annual/annu 10.htm> Ashley, David J., Paul Hanson and J Veldhuis. " A policy-sensitive traffic forecasting model for Schipol Airport", Journal of Air Transport Management Vol . 2. No. 2. (1995): 89-97. Barnard, Bruce. " E U averts air insurance crisis." AirCargo World Online, 25 September 2001 <http://www.aircargoworld.com/Break_news.htm>. Barnard, Bruce. "Swissair cuts 3,000 jobs." AirCargo World Online, 25 September 2001 <http://www.aircargoworld.com/Break_news.htm>. Baumol, W.J. and H.D. Vinod. "An Inventory Theoretical Model of Freight Transport Demand", Management Science, Vol . 16. No. 7, March, (1970): 413-421. Biederman, David. "Worsening Crisis", Traffic World, September 24 , h 2001. <http://www.trafficworld.com/news/headlines/air.html>. Boeing: World Air Cargo Forecast 2000/2001, 29 October 2001 <https://www.boeing.com/commercial/cargo/cargo01.html>. Button, Kenneth, J., Transport Economics, 2 n d Edition, Cambridge, University Press, 1993. Folley, Patrick and Mark Heuchen. " IATA Industry Outlook", Paper Presented in Budapest, 25 September 2001. 29 October 2001 < http://www.iata.org/pdf/AirTransportOutlook2001 .pdf> Goldberger, Arthur S., "The Interpretation and Estimation of Cobb-Douglas Functions", Econometrica, Vol 36, Issue 3/4 Jul - Oct (1968): 464-472. "How big a blow?" The Economist September 22 n d 2001: 29. © Sheelah Anne Turner 61 Howrey, Philip E., "On the Choice of Forecasting Models for Air Travel", Journal of Regional Science, Vol . 9. No. 2 (1969): 215-224. Isard, Walter. Methods of Regional Analysis: an Introduction to Regional Science, Cambridge: MIT Press, 1960. Jackson, Paul. "Dawn of the 21 s t Century", delivered September 2000, 29 October 2001 <http://www.triangle.eu.com/press_releases/pr46/speech%20Washington.htm> Kasilingam, R.G., "Air cargo revenue management: Characteristics and complexities", European Journal of Operational Research, 96 (1997): 36-44. Makridakis, Spyros, Steven C. Cartwright and Rob J. Hyndman, Forecasting Methods and Applications, United States of America: John Wiley & Sons, Inc., 1998. McArthur, Keith. "Airlines' losses seen totalling $2-billion", The Globe and Mail,[Toronto] Thursday, October 18, 2001: B5 McFadden, Daniel L. , "The Theory and Practice of Disaggregate Demand Forecasting for Various Modes of Urban Transportation" reprinted in Transport Economics, ed. Tae Oum et al, Amsterdam: OPA 1997, 51-79. Myers, Raymond H. Classical and Modern Regression with Applications, Second Edition, Belmont, California: Duxbury Press, 1986. Ogier, Thierry. "Varig cuts staff, fleet." AirCargo World Online, 25 September 2001 <http://www.aircargoworld.com/Break_news.htm>. Oum, Tae Hoon, Demand for Freight Transportation with a Special Emphasis on Mode Choice in Canada, Vancouver: Centre for Transportation Studies, 1980. Peskin, Henry M . "Some Problems in Forecasting Transportation Demand", Studies in Travel Demand, Ed. Ronald E. Miller. Mathematica, Princeton: (1965): 22-32. Pogge, Richard W. Lecture 18: The Apple and the Moon: Newtonian Gravity, Ohio State University, 29 October 2001 <http ://www. astronomy .ohio-state. edu/~pogge/Ast 161 /Unit4/gravity .html>. Quandt, Richard E. "Some Perspectives on Gravity Models", Studies in Travel Demand, Ed. Ronald E. Miller. Mathematica, Princeton: (1965): 33-46. —. and William J. Baumol, "The Demand for Abstract Transport Modes: Theory and Measurement", Journal of regional Science, Vol . 6, No. 2, (1966): 13-26. —. and Kan Hua Young, "Cross-sectional Travel Demand Models: Estimates and Tests", Journal of Regional Science, Vol . 9. No. 2, (1969): 201-214. Sen, Ashish and Tony E. Smith. Gravity Models of Spatial Interaction Behaviour, Berlin: Springer, 1995. Shawdon, Christopher. "The impact of the internet in air cargo", Syntegra, November 1998, <www.ccx.com>. Sletmo, Gunnar Kristoffer, "Demand for Air Cargo: An Econometric Approach", Unpublished PhD thesis, Columbia University, 1971. © Sheelah Anne Turner 62 Taneja, Nawal K. , Airline Traffic Forecasting: A Regression Analysis Approach, Lexington: D.C. Heath and Company, 1978. Transport Canada, Canada's Policy For International Scheduled Air Services - Issues For Discussion, 29 October 2001 <http://vvvvw.tc.gc.ca/pol/en/airpolicy/IntlAirPolicy/PolicyScheduledAirServices/IA PRDocumentintro%20.htm#Introduction>. Tretheway, Michael W. and Tae H. Oum, Airline Economics: Foundations for Strategy and Policy; Vancouver: Centre for Transportation Studies, 1992. "Unchartered airspace", The Economist, September 22 n d 2001: 51-52. Vedantham, Anu and Michael Oppenheimer, "Long-term scenario for aviation: Demand and emissions of C 0 2 and N O / ' , Energy Power, Vol . 26 No. 8. (1998): 625-641. Waters, W.G., Extracts from "Transport Demands and Forecasting", Chapter IV of manuscript, University of British Columbia. World Bank, "2001 World Development Indicators", CD-Rom, 2001 World Customs Organisation, Harmonized System, <http://www.wcoomd.Org//ie/en/topics_issues/harmonizedsystem/hsconve2.html#Ge neral information^ © Sheelah Anne Turner 63 A P P E N D I X I : D A T A TABLE A l . l : AGGREGATE AND DISAGGREGATE EXPORT DATA (IN TONNES) -' 'IIII'''-^ '. BSB9& IllliilyP iJHlillill BsllIBS • BI1BB Japan 10,049 3,296 3,309 3,039 83 322 Hong Kong 8,015 4,533 600 1,890 780 212 Germany 2,606 690 1,197 563 88 68 Australia 1,944 935 521 21 381 86 United Kingdom 1,784 683 405 360 162 174 China 1,615 736 236 616 0 27 Singapore 1,548 763 640 45 64 36 Netherlands 1,216 504 226 352 90 44 France 824 432 134 120 111 27 Thailand 773 498 123 3 144 5 Belgium 724 195 310 40 40 139 South Korea 658 388 183 12 33 42 Switzerland 623 259 250 94 2 18 Italy 598 209 240 35 86 28 Spain 593 253 44 2 206 88 South Africa 471 350 101 0 15 5 Malaysia 462 190 115 21 7 129 Sweden 445 228 185 2 9 21 Norway 425 260 124 33 4 4 New Zealand 314 236 60 0 9 9 Brazil 251 156 71 0 1 23 India 236 100 121 0 0 15 Austria 195 114 61 14 3 3 Finland 195 78 108 0 2 7 Egypt 194 83 103 0 4 4 Turkey 165 0 67 0 89 9 Denmark 161 115 11 9 2 24 Mexico 157 81 64 0 7 5 Sri Lanka 139 0 0 0 139 0 Chile 137 82 31 2 22 0 Ireland 125 58 19 36 10 2 Poland 108 0 30 0 0 78 Philippines 107 0 104 0 3 0 Czech Republic 103 59 24 7 11 2 Mozambique 98 98 0 0 0 0 Russia 91 0 80 0 4 7 Indonesia 76 3 73 0 0 0 © Sheelah Anne Turner 64 •III fSsBflll I S M S Pakistan 63 0 62 0 0 1 Saudi Arabia 61 4 57 0 0 0 Bangladesh 51 0 43 0 8 0 Lithuania 51 0 51 0 0 0 Venezuela 47 0 47 0 0 0 Greece 40 4 30 1 0 5 Nigeria 40 0 0 0 40 0 Colombia 32 2 20 0 10 0 Tunisia 29 0 29 0 0 0 Jamaica 28 0 0 28 0 0 Trinidad and Tobago 28 0 1 26 0 1 Romania 28 0 10 4 14 0 Tanzania 22 0 13 0 1 8 Peru 21 2 8 0 9 2 Zimbabwe 20 0 20 0 0 0 Morocco 17 0 3 0 0 14 Portugal 10 0 7 0 0 3 Luxembourg 10 9 0 0 1 0 Algeria 8 0 5 0 3 0 Uruguay 2 0 2 0 0 0 Botswana 2 0 0 0 2 0 El Salvador 2 0 2 0 0 0 Latvia 2 0 0 2 0 0 Ecuador 1 0 1 0 0 0 Total for modelling 38,840 16,686 10,381 7,377 2,699 1,697 Residual* 4,629 1,714 1,283 1,355 142 135 Total 43,469 18,400 11,664 8,732 2,841 1,832 * residual includes all countries not included in the model © Sheelah Anne Turner TABLE A1.2: EXPLANATORY VARIABLE DATA ^^^^^^^^^^^^^^^^^^^ B S l l i l p i ' 1 ' IBSIS^ Japan 126.4 42,285 7,573 4.8 -0.34 Hong Kong 6.6 21,801 10,279 0.0 1.85 Germany 82.0 31,285 8,078 3.5 -0.06 Australia 18.8 22,821 13,638 5.7 -0.13 United Kingdom 59.3 20,718 7,604 3.5 2.43 China 1,242.2 725 8,796 16.8 -1.83 Singapore 3.9 25,297 12,837 0.0 -1.25 Netherlands 15.7 29,293 7,722 3.5 1.00 France 58.4 28,243 7,946 3.5 -0.31 Thailand 59.8 2,629 11,819 21.6 7.08 Belgium 10.2 29,016 7,842 3.5 -0.03 South Korea 46.4 11,022 8,179 8.7 6.53 Switzerland 7.1 44,988 8,343 0.0 -0.88 Italy 57.6 19,911 9,023 3.5 0.97 Spain 39.4 16,391 8,436 3.5 0.85 South Africa 41.4 3,922 16,431 8.5 5.90 Malaysia 22.2 4,380 12,788 7.1 4.28 Sweden 8.9 28,796 7,456 3.5 -1.12 Norway 4.4 37,053 7,203 2.9 1.27 New Zealand 3.8 16,564 11,336 2.8 0.30 Brazil 165.8 4,501 11,200 13.6 2.21 India 979.7 430 11,118 32.2 12.24 Austria 8.1 30,962 8,525 3.5 -0.08 Finland 5.2 29,257 7,536 3.5 0.41 Egypt 61.5 1,144 10,867 20.5 3.19 Turkey 63.4 3,175 9,636 8.2 83.65 Denmark 5.3 36,864 7,661 3.5 0.86 Mexico 95.3 3,540 3,934 10.1 14.94 Sri Lanka 18.8 789 13,391 20.1 8.38 Chile 14.8 5,247 10,523 10.0 4.12 Ireland 3.7 23,154 7,178 3.5 1.44 Poland 38.7 3,396 8,224 13.1 10.74 Philippines 72.9 1,124 10,563 10.0 8.73 Czech Republic 10.3 5,129 8,277 6.8 9.69 Mozambique 17.0 188 16,680 16.9 -0.43 Russia 146.8 2,134 8,229 13.9 26.68 © Sheelah Anne Turner ^^^^^^^^^^^^^^^^^^^^^^ -' if • i r Indonesia 203.7 975 13,340 10.9 56.66 Pakistan 131.6 500 11,729 46.5 5.24 Saudi Arabia 19.7 6,866 11,993 12.6 -1.59 Bangladesh 125.6 350 11,366 22.0 7.30 Lithuania 3.7 2,055 8,114 3.9 4.09 Venezuela 23.2 3,531 6,701 12.6 34.80 Greece 10.5 12,269 9,801 3.5 3.77 Nigeria 120.8 254 11,929 21.8 9.33 Colombia 40.8 2,404 6,752 11.8 19.37 Tunisia 9.3 2,279 9,418 18.4 2.14 Jamaica 2.6 1,712 5,393 17.9 7.64 Trinidad and Tobago 1.3 4,651 7,083 18.4 4.63 Romania 22.5 1,309 9,209 13.1 58.11 Tanzania 32.1 184 15,018 21.0 11.81 Peru 24.8 2,354 8,160 13.0 6.26 Zimbabwe 11.7 715 15,822 22.2 30.83 Morocco 27.8 1,392 8,848 22.1 1.92 Portugal 10.0 11,976 8,311 3.5 1.79 Luxembourg 0.4 49,620 8,028 3.5 -0.03 Algeria 29.5 1,542 9,117 25.0 1.32 Uruguay 3.3 6,461 11,414 4.6 9.82 Botswana 1.6 3,611 16,182 8.5 5.67 El Salvador 6.0 1,727 5,014 6.7 1.56 Latvia 2.4 2,335 7,851 5.6 3.65 Ecuador 12.2 1,560 7,053 12.9 35.11 © Sheelah Anne Turner A P P E N D I X 2: R E G R E S S I O N R E S U L T S Aggregate ModekTotal Exports Regression Equation Section Independent Regression Standard T-Value Prob Decision Power Variable Coefficient Error (Ho: B=0) Level (10%) (10%) Intercept -16.82293 6.055529 -2.7781 0.007426 Reject Ho 0.864218 Lpopulation 0.9208361 0.1128238 8.1617 0.000000 Reject Ho 1.000000 Lincome 0.8501331 0.1434657 5.9257 0.000000 Reject Ho 0.999987 Ldistance 1.336442 0.5916226 2.2589 0.027799 Reject Ho 0.721337 Ltariff -0.3392972 0.1244191 -2.7271 0.008517 Reject Ho 0.852921 R-Squared 0.669067 Regression Coefficient Section Independent Regression Standard Lower Upper Variable Intercept Lpopulation Lincome Ldistance Ltariff T-Critical Coefficient -16.82293 0.9208361 0.8501331 1.336442 -0.3392972 1.672522 Error 6.055529 0.1128238 0.1434657 0.5916226 0.1244191 -26.95094 0.7321357 0.6101836 0.3469396 -0.5473909 -6.694925 1.109536 1.090083 2.325943 -0.1312035 Standardized 90% C L . 90% C.L. Coefficient 0.0000 0.6951 0.6438 0.1894 -0.2742 Analysis of Variance Section Sum of Mean Prob Power Source DF Squares Square F-Ratio Level (10%) Intercept 1 1364.922 1364.922 Model 4 167.2981 41.82454 28.3046 0.000000 1.000000 Error 56 82.74875 1.477656 Total(Adjusted) 60 250.0469 4.167448 Root Mean Square Error Mean of Dependent Coefficient of Variation Sum |Press Residualsj 1.215589 4.730304 0.256979 61.30065 R-Squared Adj R-Squared Press Value Press R-Squared 0.6691 0.6454 102.6971 0.5893 Normality Tests Section Assumption Value Probability Decision(10%) Skewness -1.2573 0.208653 Accepted Kurtosis 0.4990 0.617802 Accepted Omnibus 1.8297 0.400573 Accepted Multicollinearity Section Eigenvalues of Centered Correlations No. 1 2 3 4 Eigenvalue 2.037217 0.956257 0.687455 0.319072 Incremental Percent 50.93 23.91 17.19 7.98 Cumulative Percent 50.93 74.84 92.02 100.00 Condition Number 1.00 2.13 2.96 6.38 © Sheelah Anne Turner 68 All Condition Numbers less than 100. Multicollinearity is NOT a problem. Plots Section Normal Probability Plot of Residuals of Lexp Residuals vs Predicted J2 -0.5 -1.5 0.0 1.5 Expected Normals Residuals vs Lpopulation -2.3 o tfo o° « OCBP >^ o o o o o c? 2.5 5.0 7.5 Predicted Residuals vs Lincome °o ° o § o « o °o 0.5 3.0 5.5 Lpopulation Residuals vs Ldistance o ° 0 o ° «b° $> o °To 5.0 6.5 8.0 9.5 Lincome Residuals vs Ltariff CO ! -0.5 co CD cc -2.3 -4 .0-8 o o ^ ° °"° o co oo ° ° o o o ^ o o « ° o ° o ° P3 oo oo 8.5 9.0 Ldistance -1.0 Ltariff cBCoo ^ D °8 8 o CP ^ O O o 1.5 © Sheelah Anne Turner Disaggregate Model 1: Confidential Commodities Regression Equation Section Independent Regression Standard T-Value Prob Decision Power Variable Coefficient Error (Ho: B=0) Level (10%) (10%) Intercept -21.58878 8.454693 -2.5535 0.013468 Reject Ho 0.809771 Lpopulation 0.8770252 0.1535141 5.7130 0.000000 Reject Ho 0.999968 Lincome 1.110614 0.2084918 5.3269 0.000002 Reject Ho 0.999850 Ldistance 1.554051 0.8100709 1.9184 0.060255 Reject Ho 0.598860 Ltariff -0.3161702 0.1698519 -1.8614 0.068028 Reject Ho 0.577012 Linflation diff -0.6387044 0.2062693 -3.0965 0.003080 Reject Ho 0.921197 R-Squared 0.683338 Regression Coefficient Section Independent Regression Standard Lower Upper Standardized Variable Coefficient Error 90% C.L. 90% C.L. Coefficient Intercept -21.58878 8.454693 -35.73377 -7.443794 0.0000 Lpopulation 0.8770252 0.1535141 0.6201909 1.13386 0.4820 Lincome 1.110614 0.2084918 0.7618 1.459428 0.6123 Ldistance 1.554051 0.8100709 0.1987749 2.909327 0.1604 Ltariff -0.3161702 0.1698519 -0.6003382 -3.200215E-02 -0.1860 Linflation diff -0.6387044 0.2062693 -0.9837999 -0.2936088 -0.2727 T-Critical 1.673034 Analysis of Variance Sect ion Sum of Mean Prob Power Source DF Squares Square F-Ratio Level (10%) Intercept 1 543.6776 543.6776 Model 5 322.3651 64.47303 23.7374 0.000000 1.000000 Error 55 149.3852 2.716094 Total(Adjusted) 60 471.7503 7.862505 Root Mean Square Error 1.648058 R-Squared 0.6833 Mean of Dependent 2.985423 Adj R-Squared 0.6546 Coefficient of Variation 0.552035 Press Value 192.6827 Sum |Press Residuals| 84.21948 Press R-Squared 0.5916 Normality Tests Section Assumption Value Probability Decision(10%) Skewness 0.2037 0.838566 Accepted Kurtosis 0.2100 0.833691 Accepted Omnibus 0.0856 0.958106 Accepted Multicollinearity Section Eigenvalues of Centered Correlations Incremental Cumulative Condition No. Eigenvalue Percent Percent Number 1 2.299673 45.99 45.99 1.00 2 1.023070 20.46 66.45 2.25 3 0.833934 16.68 83.13 2.76 4 0.548442 10.97 94.10 4.19 5 0.294882 5.90 100.00 7.80 All Condition Numbers less than 100. Multicollinearity is NOT a problem. © Sheelah Anne Turner Plots Section Normal Probability Plot of Residuals of Lexpl 4.0-, 4.0 0.04 -4.0 4.0 -4.0 -1.5 0.0 1.5 Expected Normals Residuals vs Lpopulation O CD O C O ° o o„ o l o o o o „ 0.5 3.0 5.5 Lpopulation Residuals vs Ldistance O o CD R ° O 8.0 8.5 9.0 Ldistance Residuals vs Predicted cr 8a> o -2.0 0.5 3.0 5.5 Predicted Residuals vs Lincome 6.5 8.0 9.5 Lincome Residuals vs Ltariff -4.0 CP o o -6.0 -3.5 -1.0 1.5 Ltariff 4.0 © Sheelah Anne Turner Residuals vs Linflation diff o o o ° ° Co 0.0 2.0 4.0 Linflation diff © Sheelah Anne Turner 72 Disaggregate Model: Machinery & Instruments Regression Equation Section Independent Regression Standard T-Value Prob Variable Intercept Lpopulation Lincome Ltariff R-Squared Coefficient -5.196062 0.9997922 0.7617486 -0.3664224 0.719278 Error 1.22586 0.1004883 0.1184249 0.108595 (Ho:B=0) Level -4.2387 0.000083 9.9493 6.4323 -3.3742 0.000000 0.000000 0.001337 Decision (10%) Reject Ho Reject Ho Reject Ho Reject Ho Power (10%) 0.994502 1.000000 0.999999 0.954381 Regression Coefficient Section Independent Regression Standard Lower Upper Standardized Variable Intercept Lpopulation Lincome Ltariff T-Critical Coefficient Error -5.196062 0.9997922 0.7617486 -0.3664224 1.672029 1.22586 0.1004883 0.1184249 0.108595 90% C.L. 90% C L . Coefficient -7.245735 0.8317729 0.5637388 -0.5479963 -3.146388 1.167812 0.9597584 -0.1848484 0.0000 0.7732 0.5910 -0.3034 Analysis of Variance Section Source Intercept Model Error DF 1 3 57 Total(Adjusted) 60 Sum of Squares 785.896 171.3506 66.87513 238.2257 Mean Square 785.896 57.11685 1.173248 3.970428 F-Ratio Prob Level Power (10%) 48.6827 0.000000 1.000000 Root Mean Square Error 1.083166 Mean of Dependent 3.589365 Coefficient of Variation 0.3017708 Sum |Press Residuals| 57.15238 R-Squared Adj R-Squared Press Value Press R-Squared 0.7193 0.7045 77.56827 0.6744 Normality Tests Section Assumption Value Skewness -0.2185 Kurtosis -0.4352 Omnibus 0.2372 Probability 0.827020 0.663413 0.888182 Decision(10%) Accepted Accepted Accepted Multicollinearity Section Eigenvalues of Centered Correlations Incremental Cumulative Condition No. Eigenvalue Percent Percent Number 1 1.929113 64.30 64.30 1.00 2 0.687630 22.92 87.22 2.81 3 0.383257 12.78 100.00 5.03 All Condition Numbers less than 100. Multicollinearity is NOT a problem. © Sheelah Anne Turner Plots Section Normal Probability Plot of Residuals of Lexp2 3.0-, Residuals vs Predicted -1.5 o.o 1.5 Expected Normals Residuals vs Lpopulation ro •g 0.0 CO 0) tt. -1.5 -3.0 ° o o o o ° o ° ° Oj 0.0 2.0 4.0 6.0 Predicted Residuals vs Lincome o o O o ° o , Q b o OT f cc? o o o * o o „ o o -2.0 0.5 3.0 5. Lpopulation Residuals vs Ltariff CD 1 o.o-l co CD tt -1.5 co o o o o< o oo oo o O) 6.5 8.0 9.5 Lincome 1.5 -6.0 COO " o °8 C P o o o -1.0 Ltariff 1.5 4.0 © Sheelah Anne Turner Disaggregate Model 3: Perishable Products Regression Equation Section Independent Regression Standard T-Value Prob Decision Power Variable Coefficient Error (Ho: B=0) Level (10%) (10%) Intercept -3.348776 2.075079 -1.6138 0.112191 Accept Ho 0.480429 Lpopulation 0.4784038 0.1543967 3.0985 0.003039 Reject Ho 0.921600 Lincome 0.582179 0.1923727 3.0263 0.003736 Reject Ho 0.910620 Ltariff -0.404278 0.1680139 -2.4062 0.019444 Reject Ho 0.767984 Linflationdiff -0.3944186 0.2054465 -1.9198 0.059981 Reject Ho 0.599557 R-Squared 0.484561 Regression Coefficient Section Independent Variable Intercept Lpopulation Lincome Ltariff Linflationdiff T-Critical Regression Coefficient -3.348776 0.4784038 0.582179 -0.404278 Standard Error 2.075079 0.1543967 0.1923727 0.1680139 Lower 90% C.L. -6.819392 0.2201719 0.2604313 -0.685285 Upper 90% C.L. 0.12184 0.7366356 0.9039266 -0.1232711 Standardized Coefficient 0.0000 0.3305 0.4034 -0.2990 -0.3944186 0.2054465 -0.7380325 -5.080468E-02 -0.2116 1.672522 Analysis of Variance Section Source Intercept Model Error Total(Adjusted) DF 1 4 56 60 Sum of Squares 158.3307 144.6831 153.9027 298.5858 Mean Square 158.3307 36.17078 2.748263 4.97643 F-Ratio Prob Level Power (10%) 13.1613 0.000000 0.999997 Root Mean Square Error Mean of Dependent Coefficient of Variation Sum |Press Residuals| Normality Tests Section Assumption Value Skewness 2.2007 Kurtosis 0.2754 Omnibus 4.9191 1.657789 1.611082 1.028991 85.5655 Probability 0.027754 0.783040 0.085473 R-Squared Adj R-Squared Press Value Press R-Squared Decision(10%) Rejected Accepted Rejected 0.4846 0.4477 204.2388 0.3160 Multicollinearity Section Eigenvalues of Centered Correlations Incremental Cumulative Condition No. Eigenvalue Percent Percent Number 1 2.222362 55.56 55.56 1.00 2 0.862866 21.57 77.13 2.58 3 0.549041 13.73 90.86 4.05 4 0.365732 9.14 100.00 6.08 All Condition Numbers less than 100. Multicollinearity is NOT a problem. © Sheelah Anne Turner Plots Section Normal Probability Plot of Residuals of Lexp3 4 . 0 n 4.0 -1.5 0.0 1.5 Expected Normals Residuals vs Lpopulation o o §> o o °?> °8 0 o°°°° % ° 6 b o o o o o 0.5 3.0 5.5 Lpopulation Residuals vs Ltariff 2.0 -4.0 4.0 oo Residuals vs Predicted o o V o 0.0 2.0 4.0 Predicted Residuals vs Lincome " o o o o o 8 o 5.0 6.5 8.0 9.5 Lincome Residuals vs Linflation diff -4.0 -6.0 eg CP ° CD -1.0 Ltariff 1.5 oQ o o °o o 1 8 3 0 0.0 2.0 4.0 Linflation diff © Sheelah Anne Turner Disaggregate Model 4: Metals, Minerals & Wood Regression Equation Section Independent Regression Standard T-Value Prob Variable Intercept Lpopulation Lincome Ldistance R-Squared Coefficient Error -20.01001 0.4674986 0.746758 1.554074 0.307464 7.626818 0.1483685 0.1564027 0.7712751 (Ho: B=0) Level -2.6236 3.1509 4.7746 2.0149 0.011143 0.002593 0.000013 0.048636 Decision (10%) Reject Ho Reject Ho Reject Ho Reject Ho Power (10%) 0.828303 0.928994 0.998937 0.635497 Regression Coefficient Section Independent Regression Standard Lower Upper Standardized Variable Intercept Lpopulation Lincome Ldistance T-Critical Coefficient Error -20.01001 0.4674986 0.746758 1.554074 1.672029 7.626818 0.1483685 0.1564027 0.7712751 90% C L . 90% C.L. Coefficient -32.76228 0.2194222 0.4852482 0.2644793 -7.257755 0.7155751 1.008268 2.843668 0.0000 0.3802 0.6092 0.2373 Analysis of Variance Section Source Intercept Model Error DF 1 3 57 Total(Adjusted) 60 Sum of Squares 221.7576 66.23692 149.1929 215.4299 Mean Square 221.7576 22.07897 2.61742 3.590497 F-Ratio Prob Level Power (10%) 8.4354 0.000099 0.996292 Root Mean Square Error Mean of Dependent Coefficient of Variation Sum |Press Residuals| Normality Tests Section Assumption Value Skewness 1.1033 Kurtosis -0.3080 Omnibus 1.3120 1.617844 1.906665 0.8485205 81.81243 Probability 0.269918 0.758057 0.518910 R-Squared Adj R-Squared Press Value Press R-Squared Decision(10%) Accepted Accepted Accepted 0.3075 0.2710 168.5895 0.2174 Multicollinearity Section Eigenvalues of Centered Correlations Incremental Cumulative Condition No. Eigenvalue Percent Percent Number 1 1.613196 53.77 53.77 1.00 2 0.859518 28.65 82.42 1.88 3 0.527285 17.58 100.00 3.06 All Condition Numbers less than 100. Multicollinearity is NOT a problem. © Sheelah Anne Turner Plots Section Normal Probability Plot of Residuals of Lexp4 4 . 0 . Residuals vs Predicted 2.0 -1.5 0.0 1.5 Expected Normals Residuals vs Lpopulation O o o O o o ° oo <*> ° 8 o o 0 0 o °o 8 °° o° ° cP o -2.0 2.0 0.5 3.0 5:5 Lpopulation Residuals vs Ldistance O o 8 O o o o 0 ° <ob> O O 0 O O o o 0 ° -4.04—, 8.0 8.5 9.0 9.5 Ldistance CD -g 0.0 CO CU oc -2.0 0 0 0 o o -1.0 4.0 •o 0.0 CO 0.5 2.0 3.5 Predicted Residuals vs Lincome o e 5.0 0 o° 0 0 0 0 6.5 8.0 9.5 Lincome © Sheelah Anne Turner Disaggregate Model 5: Other Products Regression Equation Section Independent Regression Standard T-Value Prob Decision Power Variable Coefficient Error (Ho: B=0) Level (10%) (10%) Intercept -5.990654 1.304287 -4.5930 0.000025 Reject Ho 0.998091 Lpopulation 0.6947826 0.1069173 6.4983 0.000000 Reject Ho 0.999999 Lincome 0.7334482 0.1260014 5.8210 0.000000 Reject Ho 0.999980 Ltariff -0.2589814 0.1155426 -2.2414 0.028908 RejectHo 0.715667 R-Squared 0.600116 Regression Coefficient Section Independent Regression Standard Lower Variable Coefficient Error Intercept -5.990654 1.304287 Lpopulation 0.6947826 0.1069173 Lincome 0.7334482 0.1260014 Ltariff -0.2589814 0.1155426 T-Critical 1.672029 Upper 90% C L . 90% C L . -8.17146 0.5160138 0.5227703 -0.452172 -3.809848 0.8735514 0.9441261 -6.579084E-02 Standardized Coefficient 0.0000 0.6027 0.6383 -0.2405 Analysis of Variance Section Sum of Mean Prob Power Source DF Squares Square F-Ratio Level (10%) Intercept 1 204.1601 204.1601 Model 3 113.6138 37.87128 28.5138 0.000000 1.000000 Error 57 75.70583 1.328173 Total(Adjusted) 60 189.3197 3.155328 Root Mean Square Error 1.152464 R-Squared 0.6001 Mean of Dependent 1.82945 Adj R-Squared 0.5791 Coefficient of Variation 0.6299509 Press Value 88.01733 Sum |Press Residuals| 56.63306 Press R-Squared 0.5351 Normality Tests Section Assumption Value Skewness 2.0445 Kurtosis 0.8556 Omnibus 4.9119 Probability 0.040906 0.392222 0.085780 Decision(10%) Rejected Accepted Rejected Multicollinearity Section Eigenvalues of Centered Correlations Incremental Cumulative Condition No. Eigenvalue Percent Percent Number 1 1.929113 64.30 64.30 1.00 2 0.687630 22.92 87.22 2.81 3 0.383257 12.78 100.00 5.03 All Condition Numbers less than 100. Multicollinearity is NOT a problem. © Sheelah Anne Turner Plots Section Normal Probability Plot of Residuals of Lexp5 Residuals vs Predicted © Sheelah Anne Turner A P P E N D I X 3: F O R E C A S T S TABLE A3.1: FORECASTS FOR 1998 1 ^ l l l l l l S U S H I •wr. : t l iD Japan 3,271 3,272 6,848 1,904 1,905 7,402 Hong Kong 1,503 1,503 3,146 452 453 1,758 Germany 2,062 2,063 4,317 1,029 1,031 4,003 Australia 694 694 1,452 395 396 1,537 United Kingdom 994 995 2,081 263 264 1,02! China 676 677 1,415 593 593 2,304 Singapore 1,414 1,414 2,959 1,367 1,368 5,317 Netherlands 401 402 839 158 160 616 France 1,353 1,354 2,832 728 729 2,831 Thailand 169 169 353 19 20 73 Belgium 273 274 571 144 145 560 South Korea 376 376 786 58 59 226 Switzerland 2,251 2,252 4,713 1,728 1,729 6,720 Italy 1,176 1,177 2,462 413 414 1,606 Spain 642 643 1,345 221 222 858 South Africa 360 360 753 52 53 203 Malaysia 169 170 354 28 30 110 Sweden 223 224 468 197 199 767 Norway 147 148 308 61 62 236 New Zealand 120 121 252 56 57 217 Brazil 741 742 1,552 146 147 567 India 382 383 800 18 19 69 Austria 261 261 546 146 148 569 Finland 140 141 293 66 68 258 Egypt 78 78 162 10 11 38 Turkey 221 222 462 6 7 22 Denmark 177 178 371 80 82 312 Mexico 99 100 208 6 7 24 Sri Lanka 25 26 53 2 3 8 Chile 93 94 195 16 18 64 Ireland 79 79 165 28 29 109 Poland 102 103 214 9 11 36 Philippines 110 111 230 8 10 33 Czech Republic 54 55 113 6 7 23 Mozambique 10 10 20 2 3 7 Russia 231 232 484 10 12 40 Indonesia 332 333 696 8 10 32 Pakistan 65 66 136 5 7 21 © Sheelah Anne Turner 81 W' * ' * ' Vj i t l l l i i l l «• - ISSBllililS IlliSSBHIOP Saudi Arabia 168 169 351 184 185 716 Bangladesh 57 57 119 4 5 14 Lithuania 11 12 24 2 3 6 Venezuela 51 52 107 2 4 9 Greece 182 182 380 40 42 157 Nigeria 45 45 93 2 4 9 Colombia 64 65 134 4 5 14 Tunisia 21 22 44 4 5 15 Jamaica 2 3 5 0 1 1 Trinidad and Tobago 4 5 9 1 2 3 Romania 32 33 67 1 2 3 Tanzania 14 14 29 1 2 2 Peru 49 50 103 5 7 21 Zimbabwe 18 19 38 1 2 3 Morocco 33 33 68 5 6 20 Portugal 136 137 286 38 39 148 Luxembourg 23 23 47 16 17 61 Algeria 38 38 79 7 8 27 Uruguay 41 41 85 5 7 20 Botswana 16 17 34 3 4 11 El Salvador 7 7 14 1 2 4 Latvia 7 8 15 1 2 4 Ecuador 15 16 31 1 2 2 Forecast 22,507 22,554 47,118 10,768 10,842 41,871 © Sheelah Anne Turner TABLE A3.2: FORECASTS FOR 1998 '.fXt;;-',^ - • BgsSsj^**.'-.-Japan 1,314 1,314 2,362 76 78 302 Hong Kong 398 399 716 110 111 435 Germany 761 761 1,367 56 57 220 Australia 115 115 206 19 20 75 United Kingdom 402 402 722 27 28 107 China 369 369 663 33 34 129 Singapore 264 264 474 179 180 709 Netherlands 139 139 249 20 22 81 France 501 502 901 47 48 186 Thailand 43 44 78 3 4 12 Belgium 89 90 161 20 21 77 South Korea 139 140 251 9 10 35 Switzerland 744 744 1,338 284 285 1,124 Italy 379 379 681 31 32 121 Spain 223 224 402 23 24 91 South Africa 57 58 103 5 6 19 Malaysia 36 36 64 4 6 18 Sweden 78 78 139 25 26 99 Norway 50 50 89 13 15 53 New Zealand 24 24 42 9 10 36 Brazil 214 214 384 11 12 42 India 154 154 277 3 4 11 Austria 75 75 134 18 20 73 Finland 46 46 82 13 14 52 Egypt 24 25 43 2 4 9 Turkey 75 76 136 2 3 8 Denmark 56 56 100 14 15 56 Mexico 114 115 205 5 6 18 Sri Lanka 6 6 10 1 2 3 Chile 24 25 43 4 5 14 Ireland 27 28 49 8 10 33 Poland 41 41 73 3 4 12 Philippines 37 37 66 3 4 10 Czech Republic 19 19 34 3 4 11 Mozambique 2 2 3 1 2 3 Russia 106 107 191 3 4 12 Indonesia 89 89 160 2 3 7 Pakistan 20 21 36 1 3 5 Saudi Arabia 36 37 65 13 14 51 Bangladesh 19 20 35 1 3 5 Lithuania 4 5 7 2 3 6 Venezuela | 26 26 46 2 3 6 © Sheelah Anne Turner IsllliftlBls SI* ''1' J1JI3SI18K1 Greece 48 48 86 8 31 Nigeria 15 15 26 i 2 4 Colombia 34 35 62 2 3 8 Tunisia 6 7 11 2 3 6 Jamaica 1 2 3 1 2 2 Trinidad and Tobago 2 2 3 1 2 3 Romania 11 12 21 1 2 3 Tanzania 3 4 6 0 2 2 Peru 20 20 36 2 4 9 Zimbabwe 3 4 6 0 2 1 Morocco 12 13 22 2 3 8 Portugal 45 45 80 9 10 35 Luxembourg 5 6 9 6 7 22 Algeria 13 14 24 2 3 9 Uruguay 8 9 15 2 3 8 Botswana 2 3 4 1 2 4 El Salvador 5 5 9 2 3 7 Latvia 3 3 5 1 3 5 Ecuador 7 8 13 1 2 3 Forecast 7,580 7,611 13,627 1,151 1,228 4,549 © Sheelah Anne Turner TABLE A3.3: FORECASTS FOR 1998 BSlllIl .... .JSRfjjjgg^J^gg Japan 60 60 221 119 119 231 Hong Kong 15 16 54 47 47 90 Germany 43 44 159 76 77 149 Australia 39 39 143 19 20 37 United Kingdom 25 26 92 45 46 88 China 11 11 39 21 22 41 Singapore 18 19 67 36 36 70 Netherlands 18 19 65 23 24 45 France 33 34 123 56 57 109 Thailand 11 11 39 6 7 12 Belgium 15 16 54 17 18 33 South Korea 15 16 57 19 20 37 Switzerland 19 20 70 83 84 162 Italy 31 32 114 43 44 83 Spain 20 21 75 29 29 56 South Africa 20 21 74 8 9 16 Malaysia 11 12 41 6 7 12 Sweden 13 14 47 15 16 30 Norway 10 12 39 12 13 23 New Zealand 11 12 40 6 7 12 Brazil 23 24 86 21 22 41 India 9 10 34 10 11 20 Austria 16 17 58 15 16 30 Finland 10 11 37 11 11 21 Egypt 5 6 19 4 4 7 Turkey 9 10 34 10 10 19 Denmark 12 14 46 13 13 25 Mexico 3 4 11 13 14 25 Sri Lanka 3 4 11 1 2 2 Chile 8 9 28 5 5 9 Ireland 7 8 25 7 8 14 Poland 6 7 22 6 7 12 Philippines 5 6 19 5 5 9 Czech Republic 4 5 16 4 5 8 Mozambique 1 2 5 0 1 1 Russia 8 9 29 11 12 22 Indonesia 11 12 40 8 9 16 Pakistan 4 5 16 3 3 5 Saudi Arabia 13 14 49 7 7 13 Bangladesh 3 4 12 2 3 5 Lithuania 1 2 5 1 2 2 Venezuela 3 5 13 5 5 9 © Sheelah Anne Turner Ilflllllll i UM IJllSli 981111111 Greece 11 12 41 9 10 18 Nigeria 3 3 10 2 2 4 Colombia 3 4 13 5 6 10 Tunisia 3 4 10 2 2 3 Jamaica 1 2 2 1 1 1 Trinidad and Tobago 1 2 5 1 1 1 Romania 3 4 10 2 3 4 Tanzania 2 2 6 I 1 1 Peru 4 5 13 4 4 7 Zimbabwe 3 4 11 1 1 1 Morocco 3 4 11 2 3 4 Portugal 8 9 30 9 9 17 Luxembourg 5 6 18 3 3 5 Algeria 3 4 13 2 3 5 Uruguay 5 6 19 2 3 5 Botswana 4 5 15 1 1 2 El Salvador 1 2 3 1 2 2 Latvia 1 2 4 1 1 2 Ecuador 2 3 6 2 2 3 Forecast 667 726 2,467 899 935 1,746 © Sheelah Anne Turner TABLE A3.4: FORECASTS FOR 1998 • . • « ^ * t " Japan 3,472 3,477 10,518 Hong Kong 1,022 1,026 3,055 Germany 1,965 1,969 5,898 Australia 587 592 1,999 United Kingdom 761 766 2,030 China 1,026 1,029 3,177 Singapore 1,864 1,868 6,637 Netherlands 358 363 1,056 France 1,366 1,370 4,150 Thailand 82 86 214 Belgium 285 290 886 South Korea 241 245 605 Switzerland 2,859 2,863 9,413 Italy 896 901 2,605 Spain 516 521 1,482 South Africa 142 147 414 Malaysia 85 90 244 Sweden 328 333 1,083 Norway 146 151 440 New Zealand 105 110 347 Brazil 415 419 1,121 India 194 198 410 Austria 270 275 864 Finland 146 151 451 Egypt 45 49 116 Turkey 102 106 218 Denmark 176 180 540 Mexico 141 145 284 Sri Lanka 13 17 35 Chile 56 61 159 Ireland 78 83 230 Poland 65 70 155 Philippines 57 62 137 Czech Republic 36 41 92 Mozambique 6 10 20 Russia 139 143 294 Indonesia 118 122 256 Pakistan 34 38 84 Saudi Arabia 253 257 893 Bangladesh 30 34 70 Lithuania 10 15 27 Venezuela 38 42 83 © Sheelah Anne Turner fists Greece 116 121 332 Nigeria 22 27 52 Colombia 49 53 107 Tunisia 16 21 46 Jamaica 3 8 9 Trinidad and Tobago 5 10 15 Romania 18 23 41 Tanzania 6 11 17 Peru 35 40 86 Zimbabwe 8 12 22 Morocco 25 29 65 Portugal 109 114 311 Luxembourg 34 39 117 Algeria 28 33 77 Uruguay 23 28 67 Botswana 11 15 35 El Salvador 10 14 25 Latvia 7 12 20 Ecuador 12 16 27 Forecast 21,064 21,342 64,260 © Sheelah Anne Turner TABLE A3.5: FORECASTS FOR 1999 f . . . s» 1 If ft mSmm Japan 3,277 3,278 6,861 1,150 1,151 4,470 Hong Kong 1,551 1,551 3,246 1,321 1,322 5,136 Germany 2,089 2,090 4,373 604 605 2,349 Australia 718 719 1,504 207 209 807 United Kingdom 1,012 1,012 2,118 233 234 905 China 716 717 1,500 106 108 414 Singapore 1,487 1,487 3,112 493 493 1,915 Netherlands 413 414 865 103 104 401 France 1,387 1,388 2,903 399 400 1,551 Thailand 175 175 365 32 33 124 Belgium 279 280 584 80 82 312 South Korea 410 411 858 94 95 365 Switzerland 2,283 2,284 4,780 695 696 2,703 Italy 1,190 1,191 2,492 278 279 1,081 Spain 662 663 1,387 138 140 537 South Africa 364 365 762 48 49 186 Malaysia 178 179 372 29 30 112 Sweden 229 230 480 72 73 279 Norway 149 150 313 40 42 157 New Zealand 125 126 261 42 43 163 Brazil 747 748 1,564 91 93 355 India 404 405 846 26 27 101 Austria 265 265 554 86 87 334 Finland 144 145 301 42 44 165 Egypt 82 82 171 9 10 33 Turkey 211 212 442 6 7 24 Denmark 180 181 377 48 50 188 Mexico 102 103 214 6 7 22 Sri Lanka 26 27 55 2 4 9 Chile 93 93 194 14 16 56 Ireland 85 86 179 23 24 88 Poland 106 107 222 10 12 40 Philippines 113 114 236 9 10 34 Czech Republic 54 55 113 9 10 35 Mozambique 10 11 21 1 2 3 Russia 237 238 497 5 6 20 Indonesia 334 334 698 15 16 57 Pakistan 67 68 141 5 6 20 Saudi Arabia 169 169 353 56 57 217 Bangladesh 59 60 124 4 5 14 Lithuania 11 12 23 2 3 7 Venezuela 48 49 100 3 4 10 © Sheelah Anne Turner 8 i Pill Greece 187 188 391 38 40 149 Nigeria 45 46 94 2 4 9 Colombia 62 62 129 4 6 16 Tunisia 22 23 46 3 4 12 Jamaica 2 3 5 0 1 1 Trinidad and Tobago 4 5 9 1 2 3 Romania 31 32 65 1 2 4 Tanzania 14 15 30 1 2 3 Peru 50 51 105 6 7 23 Zimbabwe 18 19 38 1 2 3 Morocco 32 33 68 5 6 18 Portugal 139 140 292 28 30 110 Luxembourg 25 26 53 10 11 39 Algeria 39 39 81 5 6 18 Uruguay 39 40 82 6 7 22 Botswana 17 17 35 2 3 9 El Salvador 7 8 15 1 2 4 Latvia 7 8 15 1 2 4 Ecuador 14 15 29 0 2 2 Forecast 22,998 23,045 48,145 6,749 6,824 26,244 © Sheelah Anne Turner TABLE A3.6: FORECASTS FOR 1999 \ 4 *' — - . - . - T ^ 1111111 * mmms m mm 111111 Japan 1,316 1,317 2,367 56 57 221 Hong Kong 411 411 739 1,386 1,386 841 Germany 770 770 1,384 40 41 158 Australia 119 119 213 13 14 51 United Kingdom 408 409 734 25 26 99 China 389 389 699 11 12 44 Singapore 276 277 497 95 96 375 Netherlands 143 143 256 16 17 62 France 513 513 922 32 34 128 Thailand 45 45 80 4 5 16 Belgium 91 92 164 14 15 54 South Korea 151 151 271 12 13 47 Switzerland 754 755 1,356 162 163 639 Italy 383 383 688 24 25 94 Spain 230 230 413 17 19 68 South Africa 58 58 104 5 6 18 Malaysia 37 38 67 4 6 18 Sweden 79 80 143 13 15 53 Norway 51 51 91 10 12 41 New Zealand 24 25 44 8 9 30 Brazil 216 216 388 8 9 32 India 162 163 292 4 5 14 Austria 76 76 136 13 14 52 Finland 47 47 84 10 11 39 Egypt 25 26 45 2 3 8 Turkey 73 73 131 2 3 9 Denmark 57 57 102 10 12 41 Mexico 117 118 211 4 6 17 Sri Lanka 6 6 10 1 2 4 Chile 24 24 43 3 5 13 Ireland 29 30 53 7 9 29 Poland 42 43 76 3 5 13 Philippines 38 38 68 3 4 10 Czech Republic 19 19 34 3 5 14 Mozambique 2 2 3 0 2 2 Russia 109 109 196 2 3 8 Indonesia 89 90 161 3 4 10 Pakistan 21 21 38 1 3 5 Saudi Arabia 36 37 65 6 7 24 Bangladesh 20 21 36 1 3 5 Lithuania 4 5 7 2 3 7 Venezuela 24 25 44 2 3 7 © Sheelah Anne Turner K i l l Greece 49 50 88 8 9 30 Nigeria 15 15 27 1 2 4 Colombia 33 34 60 2 4 9 Tunisia 7 7 12 1 3 5 Jamaica 1 2 3 1 2 2 Trinidad and Tobago 2 2 3 1 2 3 Romania 11 12 20 1 2 3 Tanzania 3 4 6 0 2 2 Peru 20 21 36 2 4 10 Zimbabwe 3 4 6 0 2 1 Morocco 12 13 22 2 3 7 Portugal 46 46 82 7 9 29 Luxembourg 6 6 11 4 5 17 Algeria 14 14 25 2 3 7 Uruguay 8 9 15 2 4 9 Botswana 2 3 4 1 2 3 El Salvador 5 6 9 2 3 7 Latvia 3 3 5 1 2 5 Ecuador 7 7 12 1 2 2 Forecast 7,730 7,761 13,898 2,078 2,155 3,575 © Sheelah Anne Turner TABLE A3.7: FORECASTS FOR 1999 4 *•. '''"^^^S>T^:-;. 11112 HHPS . f : Japan 60 61 221 119 120 231 Hong Kong 15 16 56 48 48 93 Germany 44 44 161 77 78 150 Australia 40 41 147 20 20 38 United Kingdom 25 26 93 46 46 89 China 11 12 41 22 23 43 Singapore 19 20 70 38 38 73 Netherlands 18 19 67 24 24 46 France 34 35 125 57 58 111 Thailand 11 12 40 6 7 12 Belgium 15 16 55 17 18 34 South Korea 17 18 61 20 21 40 Switzerland 19 20 71 84 85 164 Italy 31 32 116 43 44 84 Spain 21 22 77 29 30 57 South Africa 20 21 74 8 9 16 Malaysia 11 12 42 6 7 12 Sweden 13 14 48 16 16 31 Norway 10 12 39 12 13 23 New Zealand 11 12 41 6 7 12 Brazil 23 24 87 21 22 41 India 10 10 35 11 11 21 Austria 16 17 59 15 16 30 Finland 10 11 38 11 12 21 Egypt 5 6 19 4 4 7 Turkey 9 10 32 9 10 18 Denmark 13 14 47 13 14 25 Mexico 3 4 11 13 14 26 Sri Lanka 3 4 12 1 2 2 Chile 8 9 28 5 5 9 Ireland 7 8 27 8 8 15 Poland 6 7 23 7 7 13 Philippines 5 6 19 5 5 9 Czech Republic 4 5 16 4 5 8 Mozambique 1 2 5 0 1 I Russia 8 9 30 11 12 22 Indonesia 11 12 40 8 9 16 Pakistan 4 5 17 3 3 5 Saudi Arabia 13 14 48 7 7 13 Bangladesh 3 4 12 2 3 5 Lithuania 1 2 5 1 2 2 Venezuela 3 4 12 4 5 8 © Sheelah Anne Turner 5 lflSlJSlt»,'., 1 ;• * slHlliBS BM8ISII11 Greece 11 12 42 9 10 18 Nigeria 3 3 10 2 2 4 Colombia 3 4 12 5 6 10 Tunisia 3 4 11 2 2 3 Jamaica 1 2 2 1 1 1 Trinidad and Tobago 1 2 5 1 1 1 Romania 3 4 10 2 3 4 Tanzania 2 2 6 1 1 1 Peru 4 5 13 4 4 7 Zimbabwe 3 4 11 1 1 1 Morocco 3 4 11 2 3 4 Portugal 8 9 31 9 10 17 Luxembourg 5 7 20 3 3 6 Algeria 3 4 13 3 3 5 Uruguay 5 6 18 2 3 5 Botswana 4 5 15 1 1 2 El Salvador 1 2 3 1 2 3 Latvia 1 2 4 1 1 2 Ecuador 1 2 5 2 2 3 Forecast 678 738 2,509 915 951 1,777 © Sheelah Anne Turner 94 TABLE A3.8: FORECASTS FOR 1999 BS1SII1 Japan 2,701 2,705 7,510 Hong Kong 3,181 3,184 6,865 Germany 1,535 1,539 4,202 Australia 398 403 1,256 United Kingdom 737 742 1,920 China 540 544 1,242 Singapore 920 924 2,931 Netherlands 303 308 832 France 1,035 1,040 2,837 Thailand 98 102 273 Belgium 217 222 619 South Korea 294 298 784 Switzerland 1,714 1,718 4,932 Italy 759 764 2,063 Spain 435 440 1,152 South Africa 139 143 399 Malaysia 88 93 251 Sweden 193 198 553 Norway 124 129 351 New Zealand 91 96 290 Brazil 360 364 902 India 212 216 463 Austria 206 211 611 Finland 120 125 348 Egypt 45 49 114 Turkey 99 103 214 Denmark 141 146 402 Mexico 144 148 287 Sri Lanka 13 18 37 Chile 54 59 149 Ireland 74 79 212 Poland 68 73 164 Philippines 59 64 140 Czech Republic 40 45 107 Mozambique 5 9 15 Russia 135 140 275 Indonesia 126 130 284 Pakistan 35 39 85 Saudi Arabia 118 123 368 Bangladesh 31 35 72 Lithuania 10 15 28 Venezuela 36 41 81 © Sheelah Anne Turner 'SHI i i i i i i i Greece 116 121 327 Nigeria 23 27 53 Colombia 48 53 108 Tunisia 16 21 43 Jamaica 3 8 8 Trinidad and Tobago 5 10 15 Romania 18 22 41 Tanzania 7 11 18 Peru 36 40 89 Zimbabwe 8 12 22 Morocco 24 29 62 Portugal 99 103 269 Luxembourg 28 33 92 Algeria 26 31 67 Uruguay 23 28 68 Botswana 10 15 32 El Salvador 10 14 25 Latvia 7 12 19 Ecuador 11 15 24 Forecast 18,149 18,428 48,003 © Sheelah Anne Turner 

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
https://iiif.library.ubc.ca/presentation/dsp.831.1-0090306/manifest

Comment

Related Items