Docstoc

Forecasting Demand for Australian Passports

Document Sample
Forecasting Demand for Australian Passports Powered By Docstoc
					Forecasting Demand for
Australian Passports

                      Dr. Roselyne Joyeux
                     Department of Economics
                       Macquarie University

                     Dr. George Milunovich
                     Department of Economics
                       Macquarie University



   R. Joyeux and G. Milunovich – Forecasting Australian Passports
Prepared for the 28th Annual International Symposium on Forecasting
Outline of presentation

   Objectives
   Data
   Methods
   Forecast Evaluation Methodology
   Recommendation

       R. Joyeux and G. Milunovich – Forecasting Australian Passports
    Prepared for the 28th Annual International Symposium on Forecasting
         Objective
        Forecast demand for Australian passports: the number of
         passport applications lodged with the Department of Foreign
         Affairs
           1.     Aggregated Demand for Adult and Senior Citizens - 10 years & 5 years
           2.     Demand for Children’s (Minors) Passports – 5 year passports

        Forecasting horizons:
    1.      Short-term forecasts (forecast horizon < 1 yr.)
    2.      Medium to Long-term forecasts (> 1 yr.)

        Outcomes of this project currently assist agency planning and
         budgeting.



                   R. Joyeux and G. Milunovich – Forecasting Australian Passports
                Prepared for the 28th Annual International Symposium on Forecasting
     Data
   Stochastic trends and other patterns in the number of passport
    applications make them difficult to forecast.


         Monthly Frequency                                  Quarterly Frequency




            R. Joyeux and G. Milunovich – Forecasting Australian Passports
         Prepared for the 28th Annual International Symposium on Forecasting
    Literature
   Forecasting passport demand:
       In the US: BearingPoint
       In Canada: Passports Canada

   Related Literature: Forecasting Tourism Demand
       Andrew, Cranage and Lee (1991); Carey (1991); Lim (1997);
        Morley (1993); Witt and Witt (1995), Wong (1997);

       Methods: Time-Series, Regression and ANN analyses

       Relevant variables: income, travelling cost, relative prices,
        exchange rates, population

            R. Joyeux and G. Milunovich – Forecasting Australian Passports
         Prepared for the 28th Annual International Symposium on Forecasting
Forecasting Models
    Univariate models
    1.   ARIMA
    2.   ARIMAX - dynamic regression models

    Multivariate models
    1.   Vector Error Correction Models (VECM) models
              no exogenous variables
    2.   Vector Error Correction Models (VECM) models
             explanatory variables specified as exogenous

    Apply the models to:
        Levels
        Log Levels


       R. Joyeux and G. Milunovich – Forecasting Australian Passports
    Prepared for the 28th Annual International Symposium on Forecasting
    Identification/Model Selection
    Methodology
   Univariate models: Box-Jenkins (1976)

   Multivariate models – General to
    Specific (GETS) Hendry and Richard
    (1990)
       methodology that identifies a number of
        statistically significant explanatory
        variables from a given pool of potential
        candidates.
           R. Joyeux and G. Milunovich – Forecasting Australian Passports
        Prepared for the 28th Annual International Symposium on Forecasting
         Data
   Dataset includes three different types of variables:
        economic
        demographic
        chronological: “event” dummy variables

   Time Period: January 1987 – June 2007
        Monthly data: 247 observations
        Quarterly data: 83 observations

   Trade-off between the:
        amount of information - monthly frequency
        number of explanatory variables - quarterly frequency


                R. Joyeux and G. Milunovich – Forecasting Australian Passports
             Prepared for the 28th Annual International Symposium on Forecasting
  Data: Monthly Frequency

Variable                                                       Available for Period
All Ordinaries Stock Market Index                               1987:M1 - 2007:M6
Spending                                                        1987:M1 - 2007:M4
Commodity Price Index (COM)                                     1987:M1 - 2007:M5
USD/AUD Exchange Rate                                           1987:M1 - 2007:M6
NZD/AUD Exchange Rate                                           1987:M1 - 2007:M6
Trade Weighted Index                                            1987:M1 - 2007:M6
Interest Rate                                                   1987:M1 - 2007:M6
Australian Population (population)                              1987:M1 - 2007:M6
Number of Passport Inquiries to the Call Centre – Calls         2000:M7 - 2007:M6
Demand for Passports – Adults                                   1987:M1 - 2007:M6
Demand for Passports – Minors                                   1987:M1 - 2007:M6
Demand for Passports - Seniors                                  1987:M1 - 2007:M6




           R. Joyeux and G. Milunovich – Forecasting Australian Passports
        Prepared for the 28th Annual International Symposium on Forecasting
Data: Explanatory Variables
Monthly Frequency




     R. Joyeux and G. Milunovich – Forecasting Australian Passports
  Prepared for the 28th Annual International Symposium on Forecasting
Data: Quarterly Frequency
                  Variable                     Available for Period
 Net Migration (migration)                      1987:Q1 - 2006:Q4
 Spending                                       1987:Q1 - 2007:Q1
 Nominal Disposable Income (NDI)                1987:Q1 - 2007:Q1
 Real Disposable Income (RDI)                   1987:Q1 - 2007:Q1
 Consumer Price Index (CPI)                     1987:Q1 - 2007:Q1
 Real Gross Domestic Product (RGDP)             1987:Q1 - 2007:Q1
 Nominal Gross Domestic Product (NGDP1)         1987:Q1 - 2007:Q1
 Real GDP per capita (RGDP_PC)                  1987:Q1 - 2007:Q1
 GDP per capita (GDP_PC)                        1987:Q1 - 2007:Q1
 Population                                     1987:Q1 - 2007:Q1
 Passport Demand - Adults                       1987:Q1 - 2007:Q2
 Passport Demand - Minors                       1987:Q1 - 2007:Q2
 Passport Demand - Seniors                      1987:Q1 - 2007:Q2
 House Price Level (house)                      1987:Q1 - 2007:Q1
 Travel Prices                                  1987:Q1 - 2006:Q4
 Earnings per capita                            1987:Q1 - 2006:Q4
 All Ordinaries Stock Market Index (allords)    1987:Q1 - 2007:Q2
 USD/AUD Exchange Rate                          1987:Q1 - 2007:Q1
 TWI                                            1987:Q1 - 2007:Q1
 Interest Rates                                 1987:Q1 - 2007:Q1
 Cost of Passports for Adults                   1987:Q1 - 2007:Q2
 Cost of Passports for Minors                   1987:Q1 - 2007:Q2
Data: Explanatory Variables




     R. Joyeux and G. Milunovich – Forecasting Australian Passports
  Prepared for the 28th Annual International Symposium on Forecasting
Other Explanatory Variables
1.    28th of August 1986 adult passport validity
      switched from 5 years to 10 years.
2.    September 2000: Sydney Olympics;
3.    September 2001: Terrorist attacks;
4.    October 12th 2002: Bali bombings;
5.    April 2003: SARS epidemic;
6.    December 2004: South East Asian Tsunami;
7.    July 2005: London bombings;
8.    August 2005: New Orleans floods in the U.S.
        R. Joyeux and G. Milunovich – Forecasting Australian Passports
     Prepared for the 28th Annual International Symposium on Forecasting
Selected Variables – Joint Demand for
Adult and Senior Passports
   Monthly Data:
        TWI
        All Ordinaries Share Price Index
        1991 - 1996 Dummy variable
        1991:M2 Dummy Variable – beginning of the 1st Gulf War
        1997:M4 Dummy Variable – East Asian Financial Crisis
        2003:M4 Dummy Variable – SARS epidemic, 2nd Gulf War
        Seasonal Effects

   Quarterly Data
        TWI
        GDP
        1991 – 1996 dummy variable
        Seasonal Effects

           R. Joyeux and G. Milunovich – Forecasting Australian Passports
        Prepared for the 28th Annual International Symposium on Forecasting
Selected Variables – Demand for
Children’s Passports
   Monthly Data:
          TWI
          All Ordinaries Share Price Index
          Population
          1991:M10 Dummy Variable
          2003:M4 Dummy Variable – SARS, 2nd Gulf War
          Seasonal Effects

   Quarterly Data:
          TWI
          All Ordinaries – Share Price Index
          Spending
          House Prices
          Travel Costs
          Seasonal Effects

            R. Joyeux and G. Milunovich – Forecasting Australian Passports
         Prepared for the 28th Annual International Symposium on Forecasting
Comparison of Forecasting
Models
   We use 3 measures of forecast
    accuracy:

    
              1 T
              n t T n
                            
        Bias =  Vt 1  Vt ,t 1
                          ˆ                  
                             1 T
       Mean Absolute Error =  Vt 1  Vt ,t 1
                                         ˆ
                             n t T n
                                 1 T
        Root Mean Squared Error =  Vt 1  Vt ,t 1                         
                                                                                  2
                                            ˆ
                                 n t T n
       R. Joyeux and G. Milunovich – Forecasting Australian Passports
    Prepared for the 28th Annual International Symposium on Forecasting
Forecast Horizon
   Models specified on the following sub-sample:
       1987:M1 – 2004:M6 or 1987:Q1 – 2004:Q2


   Models Evaluated on:
       2004:M7 –2007:M6 or 2004:Q3 –2007:Q2
       2005:M7 –2007:M6 or 2005:Q3 –2007:Q2

       36 months or 12 quarters
       24 months or 8 quarters
        R. Joyeux and G. Milunovich – Forecasting Australian Passports
     Prepared for the 28th Annual International Symposium on Forecasting
Findings: Joint Demand for Adult
and Senior Passports
Monthly data
   Short run Forecasting - up to one year
          ARIMA models outperform the other models.

       Long run forecasting:
          from 1 to 3 years: VECM in log form with exogenous variables,
           VECM in log form with only endogenous variables, ARIMA and
           log ARIMA all perform well.

          from 2 to 3 years: VECM in log terms with exogenous variables
           performs best

Quarterly data
       the quarterly models are outperformed by the aggregate forecasts
        obtained from the monthly models.
       R. Joyeux and G. Milunovich – Forecasting Australian Passports
    Prepared for the 28th Annual International Symposium on Forecasting
Findings: Demand for Children’s
Passports
 Monthly data
 Short run forecasting up to one year:
       ARIMA. ARIMAX, VECM log endogenous and VECM log exogenous
        all perform well.
   Long run forecasting:
       from 1 to 3 years: VECM in log form with exogenous variables,
        VECM in log form with only endogenous variables, ARIMA and log
        ARIMA all perform well.

       from 2 to 3 years: ARIMAX and log ARIMAX perform best, followed
        by the ARIMA and log ARIMA models.

Quarterly data
       the quarterly models are outperformed by the aggregate forecasts obtained
        from the monthly models.

        R. Joyeux and G. Milunovich – Forecasting Australian Passports
     Prepared for the 28th Annual International Symposium on Forecasting
Conclusions
   We construct a number of univariate and multivariate models with
    about the same degree of forecasting accuracy.

   Since multivariate models require inputs of certain macroeconomic
    variables, which are difficult to forecast (e.g. exchange rate), our final
    key recommendations are:

        Use the univariate ARIMA models for short to medium (i.e. 1-2 years) term
         forecasting.

        For longer term forecasts multivariate VECM models (without exogenous
         variables) should be preferred.

   Additionally, models constructed for monthly data outperform those
    formulated for quarterly data for the same forecasting horizons.


        R. Joyeux and G. Milunovich – Forecasting Australian Passports
     Prepared for the 28th Annual International Symposium on Forecasting
Examples of forecasts
                              Log Monthly Adults+Seniors
                                    ARIMA model
                                 Forecasts from 2004:7




     R. Joyeux and G. Milunovich – Forecasting Australian Passports
  Prepared for the 28th Annual International Symposium on Forecasting
                                                                         VECM – No Exogenous Monthly Log Adults+Seniors
ARIMAX - Monthly Log Adults+Seniors - Forecasts from 2004:7                          Forecasts from 2004:7




                      VECM (with exogenous) - Monthly Log Adults+Seniors - Forecasts from 2004:7




                     R. Joyeux and G. Milunovich – Forecasting Australian Passports
                  Prepared for the 28th Annual International Symposium on Forecasting

				
DOCUMENT INFO
Shared By:
Categories:
Tags:
Stats:
views:9
posted:12/3/2011
language:English
pages:22