Forecasting Performance of Selected Econometric Models

1

• Forecasting Performance of Selected Econometric
  Models
• By
• Asfaha, TA. and Jooste, A.

2
Outline

• Introduction
• Objective of the study
• Data sources
• Methodology
• Results
• Conclusion

3
I. Introduction

• Prediction is important for efficient marketing
  strategies and effective decision-making (Diebold
  and Mariano,1995).
• The need for market information (projection) has
  grown with the deregulation and liberalisation of
  the agriculture sector.
• Applicability of econometric models also has
  grown.
• Econometric models can provide useful information
  (Gordon and Richard, 1982).

4
Introduction cont

• However, the practice of econometric modelling is
  uneasy task (Garcia Leuthold, 1997).
• Requires understanding of the market, knowledge
  of economic theory and econometric techniques.
• Inadequate market specification
• the choice of inappropriate econometric technique
• Therefore, it is important to build models and
  determine the empirical adequacy, ability to
  forecast and analyze policy implications (Christ,
  1994).

5
2. Objective

• This study examines the forecasting performance
  of SEM, VECM and CDSEM.
• Why these econometric models?
• Both SEM and VECM are commonly applied in
  agricultural economics (Boubaker, 1997 Jose De,
  1999 Meyer and Kirsten, 2005 Poonyth, van Zyl
  and Meyer, 2000 Devadoss and Myers, 1987 Orden
  and Fackler, 1989 In and Mount, 1994 Asfaha and
  Jooste (forthcoming in Agrekon)).
• CDSEM blends SEM and VECM (Hsiao,1997).

6
3. Data sources

• The data used in the study are for the South
  African maize market.
• annually observed for the period 1975-2004.
• were obtained from the Abstract of Agricultural
  Statistics, Reserve Bank of South Africa and the
  U.S. Department of Agriculture.

7
4. Methodology

• The South African maize market model (SAMMM)

8
Methodology cont

• Econometric models used to estimate the SAMMM
• SEM
• Distinguishes between endogenous and exogenous
  variables
• No account for cointegration
• VECM
• Does not distinguish between endogenous and
  exogenous variables
• Accounts for cointegration

9
Methodology cont

• CDSEM
• Blends SEM and VECM
• Distinguishes between endogenous and exogenous
  variables
• Accounts for cointegration
• The RMSE was used to compare the forecasting
  performance of the estimated models.
• RMSE
• where and are the estimated and
  actual value in period respectively, 1,
  2, 3,, is the forecast sample.
• The smaller RMSE, the lower the forecast error.

10
5. Results

• The South African maize market variables are
• integrated and cointegrated.
• Both VECM and CDSEM are used only with variables
• that are know to be integrated and
  cointegrated.
• LinkUnit root and cointegration results.doc
• LinkCoefficients of the estimated models.doc

11
Results contEmpirical adequacy of the
estimated models
Table 2 Autocorrelation, ARCH and
normality test
12
Results cont
Table 3 RMSE of the models
13
Results cont
14
VI. Conclusion

• In general, SEM has a lower forecasting ability
  than the cointegrated models (VECM and CDSEM).
• could be due to the absence of the cointegration
  component.
• The result also reveals that VECM has better
  forecast performance than CDSEM.
• CDSEM imposes restrictions across equations in
  the presence of some of the variables.

15
Conclusion cont..

• Econometric modelling should be preceded by the
  analysis of the statistical properties of the
  variables included.
• This can help a model builder or practitioners to
  choose an appropriate econometric model, e.g. by
  determining whether the theory of cointegration
  should be included or not.

16

• I thank you!