Improved Mediterranean submodule

1
Improved Mediterranean submodule
2
Objectives

• Methodological work for modelling
• Data base for perennials at the regional level
  for major producing regions.
• Perennial sub-module in GAMS.

Working paper 02 - 07
Sources identified. Almost all data compiled.
Further operations needed
Decision to be taken about the software to
perform estimation and integration into GAMS
3
Methodological considerations

• Problem
• Approaches Based on Time Series Analysis

• Available statistical regional information on
  permanent crops is scarce
• To capture the heterogenity of the production
  capacity
• To capture the lagged decision making process

• State Space Approach Kalman Filter (KF1 and
  KF2)
• Multinomial Logit Model (MLM)

4
Assesment and conclusions

• The KF1 model
• The KF2 model
• .
• The MLM improves the CAPRI approach by

Might be practical for the case of selected
regions with available information, not for the
whole system.
Seems potentially feasible and innovative.
However, still too many parameters to elicit.
Estimation not sure to be ready and assessed
during CAPSTRAT span
Extending the regional information (CAPRI only
used data for one triennial period). Introducing
economic variables at the RHS. Making
simulations possible..
5
Methodological remarks I Multinomial Logit Model

• Purpose to obtain consistent estimated values
  for the shares of different crops in the total
  arable land.
• Shares are dependent on exogenous variables and
  error terms.
• Mathematical tools in order to get equations
  which are linear in parameters.

6
Methodological remarks II State-space approach

• Advantages filling information gaps at regional
  level, separating estimation of the qualitatively
  different planting and removal decisions.
• State-space equations
• y(k) C x(k) eyk
• x(k 1) A x(k) B u(k) exk

• Kalman filter Given currents estimates of the
  state variables x(kk), the Kalman filter
  predicts the state value at the next period k1,
  and then adjust the prediction with the
  measurement information.

7
Model Specifications the MLM approach (I)
First Stage national level. Autorregresive
models Multinomial Logit Model
Olives for oil
Olives
Table olives
Table grapes
Original data
Vineyards
Table wines
Result estimates of the shares of the 8 CAPSTRAT
activities into the broader ones at the national
level
Apples,...
Other wines
Fruits
Citrus
Other fruits
Second Stage regional level Autorregresive
models
Olives
Original data
Vineyards
Result estimates of the broader activities at
the regional level
Fruits
8
Model Specifications the MLM approach (II)

• Third Stage combination of previous calculations

First stage information j and k CAPSTRAT
activities at the national level and the annual
rate of changes for the projection period
? (1 rj)/(1 rk)
Main assumption the growing rate pattern
observed at the national level inside each broad
activity is transferred to the regional level
projected regional ratio (j/k) (initial ratio
j/k) ?
Second stage projections broad activity jk
Final result estimates of the 8 perennial
CAPSTRAT activities at the regional level,
incorporating economic variables in the
projections
9
Model Specifications the State-Space approach (I)
Young trees
Comprehensive and detailed model (see WP 02-07)
Original data acreage of a perennial activity
Productive trees
Exogenous forecasting
Young trees
Projected acreage
Productive trees
STATE VARIABLES SYSTEM FORECASTS
10
Model Specifications the State-Space approach
(II)
Sub-activity 1
Allocation model breakdown of a broad activity
into more detailed ones
Original data acreage of a broad activity
Economic variables
Sub-activity 2
Simulation changes in the economic variables
Exogenous forecasting
Sub-activity 1
Economic variables
Projected acreage
Sub-activity 2
STATE VARIABLES SYSTEM FORECASTS