Endnotes

¹See Chen and Ravallion (2004) for more in-depth discussion of poverty measures and trends in global poverty.

²Winters (2002) identified seven linkages between reform and poverty: Changes in i) consumer prices and availability of goods; ii) factor prices and quantities employed; iii) taxes and transfers influenced by shifts in government revenue; iv) the terms of trade and other external shocks; v) investment and innovation that affect the long-run growth path; vi) remittances; and vii) short-run risk and adjustment costs. While the primary reform scenario considered by Winters is trade reform, the same pathways exist for changes in transportation costs.

³This study also briefly reviews some of the earlier CGE studies of India, Bangladesh, and Nepal.

⁴Equivalent variation is the monetary value of the increment in income that would have to be given to (or taken away from) a household at today's prices to make them as well off today as they would be under the proposed policy change.

⁵The SAM is an account of all of the flows between economic agents at a given point in time.

⁶A newer SAM from Saluja and Yadav (2006) has a base year of 2003–2004, 73 productive sectors and 10 household categories, defined by expenditure level. We may update our model to include information from this SAM at a later date.

⁷The procedure we used was first to split the factor income proportions across skilled and unskilled labor using the aggregate level of factor use in GTAP7 and the allocation of labor to agricultural/non-agricultural activities. Once this mapping was complete, we were able to construct household incomes consistent with the GTAP7 data. These generally matched the proportions in the original data quite closely. We then matched the consumption categories to GTAP categories, and used the overall GTAP consumption proportions to split the individual household proportions where necessary. Finally, we used the RAS method (Bacharach 1970) to ensure that the household consumption shares were consistent with the household incomes and total expenditures in GTAP7. This process was undertaken for each country in the model.

⁸This is probably an upper bound of the impact for the first year, assuming that at least some costs are fixed. On the other hand, there are several reasons to expect the estimates of this type of model to be conservative, as discussed further below.

⁹The implicit adjustment time frame in this type of simulation is roughly 10–12 years.

¹⁰We have followed Gilbert and Wahl (2003) and used a uniform margin, here a default standard deviation of 7.5% of the mean value from GTAP7, implying that almost all variation will occur within 25% of the mean. The results give us a measure of the underlying sensitivity of the outcomes.

¹¹This general technique is valid for any type of model structure and the computational complexity does not increase with the number of parameters that are allowed to vary. It is, however, computationally expensive. Variance reduction techniques can therefore be usefully be applied here. The two techniques that we used are to run alternative simulations using common random numbers, and to adopt antithetic variates in the sampling. The former ensures that the same pseudo-random numbers are drawn for alternative simulations, and therefore that alternatives can be compared without the risk of a skewed draw. Antithetic variates use the mean of symmetric draws from the underlying distribution as the estimator for mean predictions. Since most of the variables of interest vary monotonically with the elasticities that we treated as random variables, this technique dramatically reduced the standard errors in our preliminary tests (i.e., improved the accuracy of the mean estimates). In simulations with 500 draws, the standard errors were roughly halved relative to 1000 fully random draws, a reduction that would require quadrupling the number of iterations under fully random draws. That is, simulations using 500 antithetic variates generate a level of accuracy equivalent to that of 4000 fully random draws.

¹²Roughly, a result can be considered robust to the assumed underlying parametric uncertainty if it retains the same sign within two standard deviations of the mean.

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

Comment(s)

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