Methodology

For this project we custom-built a CGE model of South Asia, with sub-economy models for key countries in the region, programmed using the general algebraic modeling system (GAMS) system. This section outlines key characteristics of the model structure and experimental design. The model is a multi-regional competitive CGE covering India, Bangladesh, Sri Lanka, Nepal, and Pakistan, as well as an incompletely modeled rest of the world (ROW) region. Overall, the structure of the model that we built for this study is a regional CGE similar in many respects to GTAP and other global models. Therefore we will keep our description brief.

4.1 Model

The model identified 16 production sectors. Each sector produced a joint product for domestic and foreign markets, with the allocation between the two based on a constant elasticity of transformation (CET) function. The production functions were nested constant elasticity of substitution (CES) functions with intermediate goods used in fixed proportions and all primary factors in variable proportions having a common elasticity. Intermediate inputs were composites of imported goods and domestic production, with variable proportions specified independently by industry. In this version of the model, the domestic transport sector was included in the services category, implying that changes in the output structure of all transportation services was in fixed proportions.

Competitive conditions hold, so firms paid market prices for all inputs, and made zero (economic) profit. Primary endowments were fixed, and could be treated as specific or mobile. The dataset contained five primary factors. In the default medium-run closure, we treated all factors except natural resources as mobile across economic activities.

The model identified several consumption agents: the government, investment, and multiple consumer households. The number of consumer households varied by region depending on available data, with between four and 19 categories in the various regions. Final consumption of each household was modeled using Stone-Geary utility functions, which generate linear expenditure systems (LES) characterizing demand for each household category. Changes in household welfare were measured by equivalent variation (EV).⁴ The parameters of the functions varied by household to capture differences in consumption patterns. The quantity of government consumption and investment was held constant in the default closure. All agents consumed composites of imported goods and domestic production, with variable proportions specified independently by agent (sometimes called the SALTER specification). On the income side, factors were owned in varying proportions by the households, and we maintained fixed proportions in household savings, taxation, and government transfers.

The exportable produced by domestic firms was allocated over destination regions using a second-level CET function, hence the aggregate exportable was a composite of exports to the various regions (the elasticity of both CET functions was set such that export destinations are very close to being perfectly substitutable, with elasticities of 20 and 40 at the lower and upper levels, respectively). Similarly, on the import side, the imports of each country were a CES composite of regional imports (i.e., a second-level Armington function). In contrast to the first level, this function was common across all agents in the domestic economy. Demand for regional exports was derived from the Armington import structure for all regions that were explicitly modeled. For regions that were not explicitly modeled, here the ROW region, we reduced the computational complexity of the model by using constant elasticity of demand (CED) functions to represent demand responses. The prices of imports from the ROW region were fixed.

An international transportation sector accounted for the difference between the free on board (FOB) price of exports and the cost, insurance, and freight (CIF) price of imports. Transportation margins varied by commodity on all international routes. Unlike in the GTAP model, because of our focus on a single, relatively small (in global terms) region, we fixed the price of international transportation services.

The price normalization and closure rules were similar to those used in many single-country models. The current account balance was fixed and the nominal exchange rate was allowed to vary to maintain balance within each country. The numeraire in each country was the consumer price index. We also had to define a numeraire region for which the nominal exchange rate was fixed, which in this model was the ROW region.

The model included a full range of distortions in the form of taxes and subsidies on economic activities at all levels to ensure that the second-best implications of the policy scenarios were adequately accounted for.

4.2 Data

The CGE model required appropriate data in the form of a SAM for each country, trade flow matrices, and estimates of the model parameters and their distributions.⁵ These were compiled from various sources, and were reconciled prior to model implementation.

The base data on trade, production, aggregate consumption, and employment were extracted from the GTAP7 (pre-release) database, which was made available by the ADBI for this project. GTAP7 has a base year of 2004. Information on sources of household income (ownership of primary factors and transfers/taxes) and variation in consumption patterns across households were obtained from Pradhan and Sahoo (2006) for India, Fontana and Wobst (2001) for Bangladesh, Naranpanawa (2005) for Sri Lanka, Roland-Holst (2008) for Pakistan, and Acharya (2007) for Nepal.⁶ The household categories used in the model are listed in Table 2 [ PDF 19.3KB | 1 page ]. The information in each study was aggregated/disaggregated and rebalanced where necessary to match the dimensions of our model and to be consistent with the aggregate GTAP7 household consumption data.⁷

Model elasticity parameters were obtained from the existing estimates in GTAP7. Armington elasticities have recently been estimated by Hertel et al. (2007). Base substitution elasticities in production were also obtained from GTAP7.

4.3 Experimental Design

The model is quite general in purpose, and can in principle be useful to examine a variety of developments in South Asia. For now, the shock magnitudes chosen to represent the effect of transport infrastructural developments are based on the original RETA (ADB 2007). This suggests a 20% reduction in transportation and processing time, which we assume is directly reflected in transportation margins.⁸ Because the investment is in land networks, we used only the land transport component of international trade margins (see Table 3 [ PDF 23.5KB | 1 page ] for summary data from GTAP7). Transport margins and the primary mode of transport vary by product and route, and this information was taken into account when calculating the shock values.

We considered two alternative versions of the impact of the reduction in international transportation costs. In the first (Scenario 1), we assumed that the impact was only on the SASEC member economies in the model (i.e., India, Nepal, and Bangladesh). In Scenario 2, we assumed that the measures would also lower transportation costs of intra-regional trade for the other South Asian economies in the region (i.e., Pakistan and Sri Lanka). Again, the cuts were based only on the land transport costs, and were also adjusted downward to reflect the fact that only a part of the route is modernized under the proposal (i.e., the intra-SASEC component).

The simulations were run as comparative statics, so the results should be interpreted as representing how the economic system would have appeared in the base year had the proposed changes been implemented and the economic system given sufficient time to adjust to the new equilibrium. As noted above, the factor market closure allowed all factors except natural resources to be mobile across economic activities, implying that the simulation is medium run in nature.⁹

Sensitivity analysis was implemented within the simulations using an unconditional approach adopted in Gilbert and Wahl (2003). This approach improves the policy value of the simulations by highlighting results that are unlikely to be robust, and by providing an estimate of the range of potential outcomes rather than a point estimate. To undertake the analysis, key parameters (the trade elasticities) are treated as normally and independently distributed random variables.¹⁰ Each simulation was run as a Monte-Carlo experiment, with a series of pseudo-random parameter values chosen from the underlying distributions. With a large number of iterations (we used 500) of the simulation we could approximate the mean predictions of the variables of interest, along with indicators of their susceptibility to parametric uncertainty (the standard deviations), and the accuracy of the simulation procedure (the standard errors).¹¹

Download this Paper [ PDF 210.1KB| 26 pages ].

Comment(s)

There are [0] comment(s) for this entry. Post a comment.