Appendix 1: Territorial Forecasting Methodology

Introduction

1.          The recommended territorial forecasting approach utilises econometric forecasting techniques to forecast total demand for retail floorspace. The approach adopts three simple economic theories which each use a number of independent variables as indicators of the total demand for retail floorspace. Indicators which show a statistical relationship to the dependent variable - total demand - are used to predict its future value. This appendix presents the forecasting approach in a series of sequential steps.

            Step 1- Adopt Economic Theories

2.          Three simple economic theories are used to identify appropriate independent variables:

• consumer expenditure (demand), i.e. the level of consumer expenditure is the principle influence upon the quantity of retail floorspace. The greater the level of consumer expenditure, the more retail floorspace will be required;

• performance and characteristics of the retail industry (factors of production), i.e. the structure and organisation of the industry in any one year is the principle influence upon the quantity of retail floorspace. The greater the level of activity, the more retail floorspace will be required; and

• performance and indicators of the retail property market (supply), i.e. the supply or demand for retail floorspace in the preceding year, and the rent/price of that floorspace, is the principle influence upon the demand for retail floorspace. The demand is a result of the interplay between supply and price.

             Step 2 - Construct Database

3.          The forecasting approach adopts the use of regression analysis. Variables are divided into:

• dependent variable - the future value is predicted using the independent variable. The dependent variable adopted for model testing is the total demand for private retail floorspace (i.e. total stock minus vacancy). This represents the total occupied IFA of retail premises; and

• independent variable - used to predict the future values of the dependent variable. The independent variables, and their expected signing, are presented in Table A1.1.

Table A1.1: Independent Variables and their Expected Signing with Retail Total Demand

[ Table Summary ]

Economic Theory/Independent Variables	Expected Relationship with Total Demand for Retail Floorspace
Consumer Expenditure (Demand) - Gross Domestic Product, Local Population, Working Population, Visitor Arrivals, Average Labour Earnings, Local Retail Expenditure, Visitor Retail Expenditure	Positive
Retail Industry (Supply) - Retail Employment, Retail Establishments, Retail Sales, Value Added, Average Unit size	Positive
Average employment per unit floor area	Negative
Property Market (Factor of Production) - Lagged total demand of private retail premises, Supply of Private Retail Premises	Positive
Price of Private Retail Premises, Rental of Private Retail Premises, Property Construction Costs	Negative

4.          For both the dependent and independent variables, time series historical data are collected from 1976 where possible. Where statistics back to 1976 are not available, the missing values are estimated. Lagged, log transformed and lagged log transformed values of each variable are also incorporated into the database for model testing.

             Step 3 - Model Testing

5.          The database is subjected to statistical tests in "SPSS for Windows". This generates a number of potential equations. They must satisfy the following standard requirements:

• signs on the independent variables are consistent with the applicable theory;

• the f-test must show that at least one of the independent variables is significant. The probability associated with the F value must be smaller than 0.01;

• the t-value associated with each independent variable is significant. The t-values must be less than 0.05;

• the adjusted R² values indicate the 'goodness of fit' of the model with the dependent variable. The adjusted R² should be greater than 0.90;

• residual analysis should show that the assumptions behind regression analysis have not been violated. No assumptions should be violated;

• a good historical fit for the past 20 years by "backtrending" the forecasts over the past 20 years; and

• the results of the models should indicate realistic and plausible forecasts when compared to the current market sentiment.

6.          The model testing undertaken in 1997 produced the following six forecasting models:

Equation 1:

Total Demand _{year t} = - 5.986 + 4.556 ln( Local Population _{year t-2} ) + 7.845 ln ( Visitor Retail Expenditure _{year t} )

Equation 2:

Total Demand _{year t} = - 8.179 + 6.329 ln( Visitor Retail Expenditure _{year t} ) + 4.947 ln( Working Population _{year t-1} )

Equation 3:

Total Demand _{year t} = 3.710 + 12.647 ( Retail Employment _{year t} ) - 5.451( Average Retail Employment per Unit Floor Area _yeart-1 )

Equation 4:

Total Demand _{year t} = 2.858 + 38.654 ( Total Demand _{year t-1} ) - 3.197 ( Retail Property Price _{year t} )

Equation 5:

Total Demand _{year t} = - 16.632 + 6.190 ln( Visitor Arrivals _{year t} ) + 2.426 ln( Retail Sales _{year t} ) - 2.961 ln( Retail Property Price _{year t} )

Equation 6:

Total Demand _{year t} = - 8.862 + 6.923 ( Visitor Arrivals _{year t} ) + 10.004 ln( Retail Sales _{year t} ) - 5.406 ( Retail Property Price _{year t} )

where ln = natural logarithm

7.          The models can be re-calibrated and new models can be constructed annually utilizing new data for the dependent and independent variables.

Step 4 - Develop Floorspace Forecasts

8.          Forecast data for all the independent variables of the preferred equations are generated. The forecast data are fed into the preferred equations to produce a "range" of floorspace forecasts. The forecasts for the total retail floorspace demand via the six forecasting models utilising data available in year 1997 are outlined in Table A1.2 below:

Table A1.2: Model Forecasting Results

[ Simple Table Format ]

Equation No.	Total Demand for Retail Floorspace (million sq.m. IFA)
	1995	2001	2006	2011
Equation 1	7.6	8.8	9.8	10.8
Equation 2	7.7	8.9	9.9	10.8
Equation 3	7.1	7.3	7.3	7.2
Equation 4	7.5	8.1	8.5	8.9
Equation 5	7.6	8.6	9.3	9.9
Equation 6	7.7	9.6	11.1	12.7

9.          The forecast range produced by the six equations ranges from 7.2 to 12.7 million sq.m. IFA in 2011.

             Step 5 - Develop the Forecast Band

10.        The forecast range may be too large to be practical for planning purposes. The forecast range can be narrowed down into a more manageable forecast "band" by adopting one or more of the following pragmatic approaches:

• review the theories underpinning the equations as one or more of the models may be too similar in nature;

• consider past trends and the ability of the property market to absorb the projected levels of take-up; and

• exclude any outlier equations.

11.        The width of the band also needs to be pragmatically determined. It should widen over time to reflect the increasing uncertainty over time. A band widening to 10% of the lower limit is reasonable and flexible for planning purposes. This can be based on 5% either side of the mid-way point.

12.        The range produced for the 1997 forecasting was considered too large to be practical for planning purposes. The range was narrowed down to the band presented in Table A1.3 below using a combination of the above criteria.

Table A1.3 The Forecasting Band

[ Simple Table Format ]

Year	Total Demand for Retail Floorspace (million sq.m. IFA)
	Lower Band	Mid-Point	Upper Band
1996	7.6	7.7	7.8
2001	8.2	8.4	8.7
2006	8.7	9.1	9.4
2011	9.2	9.7	10.1