El proceso de pruebas en RUP (Proceso Unificado de desarrollo)

ACRE as a regional supply model is comparable to another supply model which is also applied for agricultural policy analysis: RAUMIS (Regionalisiertes Agrar- und UMweltinformatIonsSystem für die Bundesrepublik Deutschland, engl. regional agricultural- and environmental information system for Germany). Since 1993 the modelling system RAUMIS is implemented by the Federal Ministry of Food, Agriculture and Consumer Protection (BMELV) and the Johann Heinrich von Thuenen Institut (vTI), Braunschweig.

Since 1997 RAUMIS is also implemented by the Research Society for Agricultural Policy and Agricultural Sociology (FAA), Bonn (ILR 2010). RAUMIS is a regional agricultural and environmental information system that is based on a positive mathematical programming approach with a non-linear objective function. The model is used to simulate the impacts of policy measures on agricultural production and environment (Henrichsmeyer et al. 1996; vTI 2008). RAUMIS is part of the vTI model framework¹⁷ which supports policy decision making of the BMELV by prospective quantitative policy scenario analysis. The model framework is used to carry out investigation of developments and policy impacts at different scales: at world and EU markets as well as at sector, regional and farm scale. The scenario simulations are focussed on impacts of trade, agricultural and environmental policy, as well as selected regional policies (Offermann et al. 2010; vTI 2008).

Treatment of the economic equilibrium

RAUMIS represents agricultural production in Germany and is a supply model. Thus, the economic equilibrium is treated by simulating the supply side while the market is represented by exogenous parameters. Agricultural production in RAUMIS is represented for Germany by 31 crop production activities which can be produced by 48 production intensities out of which 2 are intensities for grassland. Animal production is represented by 16 activities fed by 224 feeding alternatives (Cypris 2000: 48, vTI 2008).

17 For a description of the vTI model frame work see Appendix 2.3.

Modelling technique and model structure

Modelling technique

The model simulating agricultural production in RAUMIS is a non linear programming model based on the PMP approach, which maximizes agricultural total gross margin by optimizing the extension of agricultural production activities. In the following a simple PMP model for crop production is described according to Howitt (1995) by the equations Eq. 2.3.1 to Eq.

2.3.4. These equations describe a PMP model calibrated according to the cost sided approach which explains the decrease in marginal gross margin by the increase of marginal production costs. For details see Howitt (1995), Umstaetter (1999) and Röhm and Dabbert (2003).

Eq. 2.3.1 is the total gross marginal (TGM) function, which is the objective function of the LP model. TGM is maximized by the LP model subject to Eq. 2.3.2 to Eq. 2.3.4. Xi is the optimized extension of the production activities i, and Xˆ_i is the extension of the production activity i observed in the calibration situation. The index i represents the crop production activity.

) (

max f X where

∑

^{+ −}

=

=

i

i i i

i

i y p SUB c

X TGM

X

f( ) ( *[ * ])

Eq. 2.3.1 with

i: total crop activity i (e.g., wheat, rye or grassland) X : simulated acreage of crop production activity i [ha] i

Xˆi: observed acreage of crop production activity (statistic) i [ha]

y : crop yields of crop activity i [dt hai ^-1] p : price for crop activity i [EUR dti ^-1]

SUB : subsidies for crop activity i [EUR hai ^-1]

c : variable costs for production of crop activity i [EUR hai ^-1]

subject to

∑

∑

^≤

i i i

i X

X ) ( ˆ )

( Eq. 2.3.2

Eq. 2.3.2 limits the resource land and produces the dual valueλ_land, which is used to calculate the shadow price of the marginal crop activity.

) 1 ( ˆ *

ε1

+

≤ i

i X

X _{Eq. 2.3.3}

The constraint on the amount of crop activity is represented by Eq. 2.3.3. The total crop activity restriction produces the dual value λ_i^.

The perturbation coefficientε₁, in equation Eq. 2.3.3 is a small positive number. This coefficient enlarges the restrictions of the observed amounts of the activities Xˆ_i by a small value, which allows the LP model to produce dual values for each crop activity. Nevertheless, the number of constraints exceeds the number of variables by one, which is why one total crop activity constraint produces the dual value of zero. The dual value of zero for the least profitable total crop activity requires a special method of calibration for this so called marginal crop (or marginal activity). The calibration of the marginal activity requires inter alia the shadow price for land λ_land (for details cf. Röhm and Dabbert, 2003; Röhm, 2001;

Umstätter, 1999).

Eq. 2.3.4 describes the classical version of the objective PMP function for crop activities. For a better overview, yield, price, subsidies, and cost terms in Eq. 2.3.4 are replaced with the equation: GM_i = y_i*p_i +SUB_i −c_i

∑



































 − +

=

i i

i i

i i

i X

GM X X

X

TGM * λ * 1 ˆ _{Eq. 2.3.4}

The positive mathematical programming model in RAUMIS is only a part of the complete information system RAUMIS. The mathematical formulation of the model is described in detail in Cypris (2000).

Model structure

RAUMIS is built in a modular way, which makes it possible to use single modules separately from the others. The modular structure allows for dealing with the huge information requirements and the parallel working with the model by scientists from different institutions.

The results of the single modules are exchanged in between the modules and used as input data for other modules (Cypris 2000: 7). RAUMIS consists of four different modules, which are briefly described below.

The module "Grunddatensammlung" (engl. basic data base) contains the original data of the ex-post period. The “Konsistenzrahmen-Modell” (engl. consistency framework model) provides the ranges for the model data used to define the base year. The ranges are derived from data of official statistics. A check of consistency of the base year data is provided for the agricultural production activity extension, input, output and for the monetary data. The

"Entscheidungsmodul fuer die Basisjahre" (engl. decision module for the basic years) contains the calibration LP model. It calculates dual values for scarce input factors. The check for deviation between the calculated dual values and statistics provides information about the

validity of the model. The calibration parameters for the non linear programming model are calculated here (Cypris 2000: 7, 9).

In submodules the exogenous parameters, calculated by other models, as well as parameters representing progression and trends for the simulation period are implemented. Furthermore, a submodule calculates the optimal special intensity for crop activities, derived from price relations. This submodule determines the extension of production intensities (Cypris 2000:

10).

The "Entscheidungsmodul fuer das Zieljahr" (engl. decision module for the simulation period) contains the non linear programming model which simulates via a process analytical approach the agricultural production in the simulation scenarios. The "Modul zur Loesungsaufbereitung (engl. module for the processing of results) is a framework where the simulation results are discussed with policy decision makers. This module provides the feedback loop from the experts back to the model. Via this module calibration parameters, exogenous and trend parameters can be corrected to aim at simulation results which are consistent with the experts' knowledge (Cypris 2000: 11).

The structure of RAUMIS and the included consistent checks allow for consistency of agricultural and environmental results with official statistics. A coupling with different types of economic and natural models is also possible (Kreins et al. 2010, Offermann et al. 2010, Gömann et al. 2009, vTI 2008).

Aggregation of results

RAUMIS simulates agricultural production at NUTS3 level for Germany. City counties are statistically defined as NUTS3 because of their high population density. Their agricultural area and agricultural production is relatively small. Thus most of the city counties are aggregated with neighboured NUTS3 counties with larger UAA and of higher importance of agricultural production. The simulation results can be analysed for the German 326 single (or aggregated) NUTS3 counties, can be aggregated to 38 administrative districts, different river catchment areas or 76 agricultural regions. The borders of administrative districts and the agricultural regions are congruent with the borders of the Federal States and thus the regional analysis within political borders is possible (Cypris 2000: 31).

Temporary dimension

RAUMIS is a comparative static model, calculating from a base year to a simulation year.

Several calibrated base years are available, with 2003 and 2007 being the most recent ones.

The current data base is already updated to the year 2007. The simulation period between base year and simulation year is 10 to 20 years. Thus, with a base year of 2007 the simulation year of 2027 is possible. The shortest simulation period is one year.

Validation of RAUMIS by ex-post analysis

In order to validate the RAUMIS model and to get information on the forecasting quality an ex-post validation has been done (Cypris 2000: 133). Within an ex-post validation results from simulation are compared with statistical data representing the reality. The deviation is a quality measure, indicating how good the model simulation matches reality.¹⁸ Cypris (2000) did the ex-post validation for RAUMIS for two different simulation periods, each of 8 years length. As measure he used the “Mittleren absoluten prozentualen Fehler” (MAPF) (the mean absolute percentage deviation). The MAPF is calculated according to the following equation:

MAPF ˆ *100

∑

* ⁻

=

i i

i i

i X

X

w X Eq. 2.3.5

with

i: production activity

wi: weighting of production activity according to the income value X : simulated extent of production activity i i

Xˆi: observed extent of production activity (statistic) i

Table 2.4-4 presents the MAPF for two simulation periods for the agricultural sector, for single products and for the NUTS3 counties. The MAPF for the sector and the products are evaluated by the benchmarks of prognosis quality which is represented by the forecasting error. According to Hazell and Norton (1986: 271) a forecasting error (here MAPF) between 10% and 13% is 'acceptable', a MAPF of 14% is acceptable but has to be improved, while a MAPF of 15% and greater are not acceptable.

The prognosis quality of RAUMIS for the complete agricultural sector is evaluated with a MAPF of 14% as acceptable however with needs for improvements. The MAPF for the crop production are all not acceptable, while the MAPF for animal production is acceptable (for pork and poultry in the period from 1983 to 1991) and with less then 10% can be even regarded as 'good' (Hazell and Norton 1989: 271). Nevertheless the regional results of the ex-post prognosis of RAUMIS for 1987 and 1991 are not convincing. The mean value of the MAPF of the NUTS3 counties (MAPFcounty) calculated for the simulation periods from 1979

18 For a more detailed description of the forecasting error see Subsection 2.3.4.

to 1987 and 1983 to 1991are 20% and 24%, while only 50% and 30% respectively of the NUTS3 counties show a MAPF_countysmaller than 20%.

Thus the forecasting quality at sector scale is acceptable while it is not acceptable at regional NUTS3 level (Cypris 2000: v). However, it is important to keep in mind that RAUMIS is used as a simulation model not as a forecast model. A sensitivity analysis has shown that the model reactions of RAUMIS are plausible with respect to regional adaptation of agricultural production (Cypris 2000: 158). Thus, regional policy impact analysis with RAUMIS is possible also without matching a good forecasting quality. It should be noted that the results of the post validation described by Cypris (2000) are more than 10 years old and no ex-post analysis has been published using the updated version of RAUMIS.

Table 2.3-1: Forecasting errors of crop and animal production in RAUMIS.

MAPF Prognosis quality according to Hazell and Norton, 1986: 271 Simulation period

from 1979 to 1987 from 1983 to 1991

% %

MAPFsector 14 14 acceptable but to be improved

MAPF_products Crop production 24 24 not acceptable

Cash crop production 25 24 not acceptable

Fodder crop production 21 21 not acceptable

Animal production 11 10 acceptable

Cattle and sheep 10 12 acceptable

Pork and poultry 12 7 acceptable

% of number of NUTS3 counties

MAPFcounty Mean value of all MAPFcounty 20% 24% not acceptable

counties with MAPF < 20% 50% 30% not acceptable

Source: Cypris (2000: 143)

In document Plan de Pruebas de Modulo para la Unidad de Compatibilizacion, Integracion y Desarrollo de Productos Informaticos para la Defensa (página 36-42)