Aggregated Agricultural Statistics: problems of modelling with real data

1
Aggregated Agricultural Statistics problems of
modelling with real data

• Ignacio Pérez, University Bonn

2
Table of Contents

• Introduction a modelling approach
• Statistical Sources
• The Problem
• The CoCo Solution
• Data flows in the CAPRI Model
• Conclusions

3
Introduction a modelling approach

• The CAPRI Modelling was born as an initiative to
  simulate economic effects of policy changes in
  the European agricultural sector
• It was and is still maintained by a coordinating
  team (IAP, University of Bonn) and a network of
  partners
• It is a mathematical equilibrium model,
  comparative-static, spatial and includes
  endogenous price feedback
• Its main parts are a supply, a market and a young
  animal trade modules
• And needs to be fed with data

4
Statistical Sources

• Initially, the CAPRI data base built upon three
  major data sources
• The SPEL-EU data base from EUROSTAT
• consistent national frame for farm and market
  balances, activity levels (hectares of crops and
  animal herd sizes)
• unit value prices
• Economic Accounts for Agriculture.
• The REGIO domain of EUROSTAT main source for
  regional activity levels and physical outputs
• Additional data supplied by
• national experts of the project (to fill gaps or
  correct errors in REGIO and to supply data not
  comprised in it)
• other institutions (FAOSTAT, EU, ), mainly data
  on policy instruments

5
The Problem

• But as the SPEL-EU data base was growing over the
  years, more or more rules to create the best
  possible corrections and estimates for missing
  values were introduced.
• Users could not determine which data were
  original, trend estimates, calculated ones, etc
• Moreover, EUROSTAT kept introducing updates in
  its time series.
• And in 1999 decided to stop the SPEL exercise and
  initiate a new project called AGRIS (Agricultural
  Information System).
• so we decided to go for an own solution and
  create the CoCo Data Base

6
The CoCo Solution

• CoCo stands for completeness and consistency
• The basic idea was
• To reproduce the accounting principles underlying
  the consistency framework of SPEL (wherever
  appropriate)
• To use GAMS, common programming language in CAPRI
  (and many other models)
• To storage meta-information (data points from
  statistics or trend estimates)
• To find a solution less dependant on threshold
  values and individual decisions about functional
  forms and corrections (reduction of update costs)

7
The CoCo Solution

• General approach minimising normalised least
  squares under constraints
• (differences between corrected and given data
  and differences between trend forecasts and
  given or fitted data)

8
The CoCo Solution

• Definition of upper and lower bounds (estimation
  corridor)

9
The CoCo Solution

• Data estimation hanging on bounds
  (infeasibilities)

10
The CoCo Solution

• Data estimation reaching the optimum (relaxing of
  bounds)

11
Data Flows in CAPRI
12
Conclusions

• CoCo is actually digested by the CAPRI Model
  without major necessary updates.
• The estimation process ensures that
  infeasibilities are avoided in most instances,
  thus reducing the control cost for a new data
  update.
• The risk that original data are corrected with
  inconsistencies is close to zero.
• The sweeping tail problem of trend estimates is
  up to a certain extent reduced by introducing
  bounds.
• and in any case original data are kept in the
  system for comparison purposes (transparent
  incremental procedure)

13
Thanks for your attention !!!