High food prices and their implications for poverty in Uganda - From demand system estimation to simulation

The repeated occurrence of high food price spells has started intensive research on the impacts of these events on income distribution and poverty, in particular. Initial empirical work attempting to assess the poverty impacts used first-order analysis using household data that differentiates between net buyers and net sellers of food (e.g., Ivanic and Martin, 2008; Wodon et al., 2008; Zezza et al., 2008; Aksoy and Isik-Dikmelik, 2008; and on Uganda: Benson et al., 2008; Simler, 2010). One common feature as well as a major critique of such analyses is that they disregard the second-order impacts of the price changes on both the supply and consumption sides (see, e.g., Aksoy and Hoekman, 2010). The objective of the present study is to quantify the effect of this neglect on the consumption side. More specifically, the aim of this study is twofold. First, to provide an indication whether it is worthwhile to invest in the estimation of a demand system for similar consumption side poverty impact analyses. Second, to provide a sense of the magnitude in the loss of fidelity in using a less flexible instead of a more flexible demand system within computable general equilibrium analyses of poverty impacts. As the basis of this study, a 13-item censored Quadratic Almost Ideal Demand System (QUAIDS) is estimated over data of the Uganda 2005/2006 National Household Survey (UNHS) which covers a representative set of 7426 households. The estimated parameters are used to simulate the impact of the 2007/2008 and 2011 food price spikes on poverty. The shocks are calculated from price time series data for several food commodities and Ugandan market locations which are matched to the households of UNHS. As the Linear Expenditure System (LES) is one of the most popular demand systems used in computable general equilibrium models, a LES is calibrated, for each household separately, to approximate the QUAIDS elasticities in the point of the base data as done in Yu et al. (2003). Then, the non-behavioural first-order poverty impacts as well as the behavioral poverty impacts employing the two demand systems are simulated. The demand system simulations are conducted fixing the base-period utility and calculating the compensating variation based on an algorithm due to Vartia (1983). To date, the estimation of the 13-item censored QUAIDS has been completed and the Vartia algorithm for the simulations has been implemented. We expect that for large price shocks, such as the 2007/2008 and 2011 price spikes, the use of a behavioural demand model will dampen the negative effects on consumption significantly. Nevertheless, it is difficult to conceive what impact the choice between the QUAIDS and the LES will have on the results. The results from this study will: First, provide an indication whether it is worthwhile to invest in the estimation of a demand system for similar consumption side poverty impact analyses. Second, provide a sense of the magnitude in loss of fidelity in using a LES instead of a more flexible demand system like the QUAIDS within CGE model analyses of equivalent price shocks.