Estimating Consistent and Complete Trade Matrices and Supply Utilization Accounts

Funding

Food and Agriculture Organization of the United Nations, Rome, Italy

Duration

February 2004 – May 2004 (Study #1)

October 2004 – February 2005 (Study #2)

December 2005, April 2006 (GAMS training)

Objectives

In its FAOSTAT2 project, the FAO is reconsidering technical solutions, organizational aspects and the underlying methodology of its statistical services. In this context the problem arises to consolidate both Supply Utilization Accounts and trade matrices. The overall aim was to develop and test a methodology, the related algorithm and software solution which makes best use of the available data and the inherent restrictions in hard- and software. Consolidation includes

• complete and closed balances for each product in the SUAs,
• trade flows which are bilaterally consistent and correspond to total imports and exports in the SUAs.
• Technical relations within product trees are imposed as well, but have become less important in the last contract (due to vertical integration of most product trees to primary product equivalents).

Methodology and results of study 1

Initially the approach started from the SUAs with the following two step procedure:

• In the first step supply utilization accounts were consolidated, i.e. gaps closed and eventual imbalances removed. Technically, the estimation process was broken down to individual countries, but all products in one product tree and all years were processed simultaneously.
  • Firstly, simple trend regressions were performed to backcast or forecast the series in case of missing values. A Hodrick-Prescott filter was used to smooth these estimates and to interpolate gaps within a series.
  • The non consolidated series were inputs to a consolidation step based on the Highest Posterior Density estimator presented by Heckelei, Mittelhammer, Britz 2005 at the 89th EAAE Seminar.
• These consolidated SUA results gave starting values for a second step, where SUAs for all countries and products in a product tree are estimated together with trade matrices, year by year.

The main results of this approach were the following:

• The prototype for the wheat tree demonstrated technical feasibility and reasonable results
• The methodology refrained to a large extent from subjective ad hoc decisions which were replaced by rules casted into program statements.
• The methodology had a sound basis in econometric theory and exploited the informational content of the interrelated time series to a large extent.
• But solution time was unsatisfactory in view of applicability to all products,
• and the possibility that errors may spread in step 2 from one SUA to other countries was troublesome.

Methodology and results of study 2

To solve the last two problems the sequence of steps was reversed in the second study:

• In the first step trade matrices were consolidated:
  • First, missing notifications on bilateral trade flows were completed with regressions and the Hodrick-Prescott filter as above.
  • The completed series of notifications from exporter and importer were aggregated using reciprocal standard deviations as weights.
• The consolidated trade matrices gave total exports and imports as input values to the second step which estimated time series of complete and consistent SUAs based on these trade data, a number of other hard data (production etc.), and a set of coefficients defining 'reasonable' relationships among SUA elements.

The main results of the revised approach were:

• The procedure was successfully tested on all food products and a number of agricultural inputs.
• The approch preserved the objective, rule based character of study 1.
• Because the SUA estimation could not feed back to the trade matrix results there was some loss of informational efficiency,
• but this also isolated the different SUA estimations from each other, avoiding disemination of errors.
• Solution time per product improved considerably but is still high if applied to 200 countries and about 140 food items.

The approach might free resources currently used to set up SUAs and shift it to validation and quality control. On the other hand, the experiences accumulated by FAO staff is still disregarded to a large extent. Currently the procedure is being implemented, evaluated, tested, and refined within FAO.

Issues of GAMS training

The General Algebraic Modeling System (GAMS), a high-level modelling system, was starting to be widely used by the technical Divisions at FAO as a modelling tool. Subsequently, the staff required training on the GAMS application and use in agricultural modelling as well as in policy analysis. The new FAOSTAT System is also using GAMS for estimating missing observations, calibrating and validating data using maximum entropy techniques. It was therefore important that FAO staff is trained in GAMS, its applications and use in the FAOSTAT System.

Additional information:

• Britz, W., Witzke, H.P., Heckelei, T. (2004): Estimating Trade Matrices and Supply Utilization Accounts using a Bayesian Estimator, Eurocare Final Report to FAO (Study 1).
• Britz, W., Witzke, H.P., Heckelei, T. (2005): Estimating Consistent Trade Matrices and Supply Utilization Accounts for FAOSTAT2, Eurocare Final Report to FAO (Study 2).
• Witzke, H.P., Britz, W. (2005): Consolidating trade flows and market balances globally using a Highest Posteriori Density estimator, Paper on the 8th Annual Conference on Global Economic Analysis , June 9 - 11, 2005, Luebeck, Germany.

---

Last updated: Tuesday, 16-Sep-2008 10:36:46 CEST