CHAPTER 2 FREIGHT FORECASTING MODELSINTRODUCTIONIn this study, forecasts <strong>of</strong> containerised <strong>and</strong> non-containerised trade, cruise passengernumbers, <strong>and</strong> the movement <strong>of</strong> freight, cruise <strong>and</strong> other ships are developed usingeconometric models. A detailed discussion on the models <strong>and</strong> their estimatedparameters is presented in the later sections <strong>of</strong> this chapter.As mentioned in Chapter 1, the forecasts <strong>of</strong> containerised <strong>and</strong> non-containerised trade<strong>and</strong> freight ship movements are developed at the national <strong>and</strong> port level. However, theforecasts <strong>of</strong> cruise passenger numbers <strong>and</strong> the movement <strong>of</strong> cruise <strong>and</strong> other ships couldnot be developed at the port level, mainly due to the lack <strong>of</strong> long time-series port leveldata on cruise passenger numbers <strong>and</strong> the movement <strong>of</strong> cruise <strong>and</strong> other ships. Theforecasts are` developed at the national level only.Since the five main city ports (i.e. Brisbane, Sydney, Melbourne, Adelaide <strong>and</strong>Fremantle) dominate Australia’s containerised trade, accounting for 89.9 per cent <strong>of</strong> thetotal containerised trade, their growth rate forecasts are used to derive forecasts <strong>of</strong>container <strong>and</strong> container ship movements at the national level. This approach has beenused following the predictive-accuracy <strong>of</strong> the port level models relative to the nationallevel models, <strong>and</strong> to maintain consistency in port <strong>and</strong> national level forecasts. Theapproach was not found appropriate to forecast non-containerised trade <strong>and</strong> noncontainership movements at the national level. This is because the five main city portsaccount for only 11.2 per cent <strong>of</strong> the total non-containerised trade <strong>and</strong>, hence, do notdominate Australia’s non-containerised trade. The national level forecasts <strong>of</strong> noncontainerisedtrade <strong>and</strong> the movement <strong>of</strong> non-container ships are developed using thenational level econometric models.In this study, the term ‘all ports’ is used to refer to the total <strong>of</strong> all Australian ports (thatis, national level).Forecasts for other ports are derived by subtracting the total forecasts for the five maincity ports from the national level forecasts.The econometric models have been chosen to forecast container <strong>and</strong> ship movements atthe port level, because they are relatively better than time trend or univariate time-seriesmodels in the sense that the models can accommodate several explanatory variables toanalyse their influence on container <strong>and</strong> ship movements. Since the models are specifiedin a double logarithmic linear functional form, they are easy to estimate, provide5
superior fit <strong>and</strong> the estimated parameters can be directly interpreted as elasticities. Themodels have been widely used in many tourism <strong>and</strong> transport dem<strong>and</strong> forecastingstudies, such as Loeb (1982), Witt <strong>and</strong> Witt (1992), Hamal (1997a, 1997b <strong>and</strong> 2004)<strong>and</strong> BTRE (2002a <strong>and</strong> 2002b).The econometric models include population, income <strong>and</strong> exchange rates as explanatoryvariables. The models were estimated using annual data instead <strong>of</strong> quarterly data, fortwo reasons. Firstly, the models are intended to derive long-run forecasts, not shortrunforecasts. Monthly <strong>and</strong> quarterly data are desirable to develop short-run forecasts.Secondly, the long-run forecasts <strong>of</strong> most macroeconomic variables are available only onan annual basis rather than on a monthly or quarterly basis.In this report, ships operating at the Australian ports are grouped into four categories formodelling purposes. These are: container ships, non-container ships, cruise ships <strong>and</strong>other ships. A container ship is defined as a ship designed to carry containers only. Thecategory <strong>of</strong> non-container ships includes ships carrying refined <strong>and</strong> crude oil, chemicals,gas, bulk dry, livestock <strong>and</strong> general commodities. Cruise ships are defined as thoseships which carry passengers for cruising purposes only. The category <strong>of</strong> other shipsincludes survey ships, navy ships, tugs <strong>and</strong> all other ships which are not included undercontainer, non-container <strong>and</strong> cruise ship categories.CONTAINER AND FREIGHT SHIP FORECASTING MODELSSince the export <strong>and</strong> import <strong>of</strong> full <strong>and</strong> empty containers are linked, a system <strong>of</strong>simultaneous equations underlying the dem<strong>and</strong> <strong>and</strong> supply structure <strong>of</strong> containermovements is likely to present a better set <strong>of</strong> container forecasts. However, such asystem <strong>of</strong> simultaneous equations is found to be complex to estimate, especially whenlong time-series data on supplyside variables are not readily available <strong>and</strong> the length <strong>of</strong>time-series data is not the same for all dem<strong>and</strong>side model variables. In this study, thesingle equation modelling technique which is doable <strong>and</strong> simple to use is applied t<strong>of</strong>orecast container numbers.Although it is more meaningful to forecast shipping activities by commodity, port <strong>of</strong>origin <strong>and</strong> final destination, it could not be done for two reasons. Firstly, long timeseriesdata on exports <strong>and</strong> imports by commodity, port <strong>of</strong> origin <strong>and</strong> final destination arenot readily available for forecasting purposes. Secondly, it is resource intensive tocollect, analyse <strong>and</strong> forecast shipping activities by commodity, port <strong>of</strong> origin <strong>and</strong> finaldestination.Forecasts <strong>of</strong> container <strong>and</strong> freight ship movements are separately developed using theforecasting frameworks presented in Figures 2.1, 2.2 <strong>and</strong> 2.3. Since the volume <strong>of</strong>exports <strong>and</strong> imports directly influences the number <strong>of</strong> container <strong>and</strong> ship movements,econometric models <strong>of</strong> export <strong>and</strong> import volumes are specified to forecast the number<strong>of</strong> container <strong>and</strong> ship movements. The models include export <strong>and</strong> import volumes asdependent variables <strong>and</strong> population, income <strong>and</strong> exchange rates as explanatoryvariables.6