 So the measurement of long-run household wealth is extremely important for us as researchers but it is difficult to do and what people have tried to do so far is to construct asset indices based on durable goods ownership from household surveys and although a range of different indices exist the underlying theory connected with the different indices are somewhat informal and this is this is problematic because it doesn't help us decide how to choose between the different indices when they give us different results in terms of identifying which households are poor, the levels of poverty and how poverty has changed over time. So in my work I basically propose an alternative asset index which is based directly on economic theory. I estimate my index using data from Malawi and show how it compares to some of the existing indices and then I argue that it offers an improvement because it's structurally estimated based on demand for durable goods and is therefore able to incorporate information on prices, household demographics and location. So the model is fairly simple, households are going to be maximizing utility subject to their budget constraint and they're going to be choosing over some consumption good as well as a set of discrete assets and what they will do ultimately is compare the utility that they would get from the different bundles that are available to them and end up purchasing whichever bundle gives them the most utility. So this is illustrated in this figure here and this is the case where there's two discrete bundles so you can imagine having a television and a radio and in that case there's four potential bundles you can have. The first is the bundle with neither the television or the radio, the second is the bundle with the radio but not the television, the third is the bundle with the television but not the radio and the fourth is the bundle with both and each of these bundles has a utility function, an indirect utility function associated with it and households will just compare the different functions and choose which is optimal for them. So we see that at very low levels of income it's optimal to purchase neither good at very high levels it's optimal to purchase both of the goods and so on and so forth. So from this model what we're interested in is estimating the marginal utility associated with each of the discrete goods and I can do so by writing down the likelihood function of seeing the data that we observe and then going ahead and using maximum likelihood estimation methods. One additional assumption that I make is that we'd like to incorporate things like housing quality which don't often have prices associated with them so I assume a link between marginal utility and price and that that is increasing in price. So I go ahead and estimate this using two rounds of data from the Malawi integrated household surveys using seven durable goods as well as three types of housing quality and what I find is in the bold column for the utility index which is what I term my proposed index. We find that the marginal utility is broadly increasing in price with the exception of CD players which don't seem to provide much of a utility boost and then in terms of housing quality we see that the marginal utilities are relatively low. What I show in the latter two columns is what we get from some of the existing indices. So these are factor analysis which is very similar to principal components analysis and this is based on statistical estimation of weights and often people argue it doesn't really have that much meaning and then the last column shows the inverse frequency index which basically weights the different goods depending on the inverse of the population ownership level and the main thing that we see is that the weights that we get are very different depending on the method and again I argue that my method offers an improvement because it's based on economic theory. In terms of the different weights how much do they matter for the conclusions that we get and I argue that they matter a lot so this first bar graph shows what happens when we're trying to classify households into different wealth deciles and what we see is that the decile that you get depends very much on the index that you use and for instance using the factor analysis index over 40 percent of the sample is classified in a wealth decile that differs from my benchmark by two or more groups. Again when we're thinking about poverty headcounts it really matters the index that you use and this shows the poverty headcount using a 10 percent relative poverty line from 2004 and what we see is that the level of poverty really matters depending on the index that you use particularly for the equal weights index which is just the asset count and the inverse frequency index and these show relatively high levels of poverty mainly because the indices create somewhat lumpy classifications so you can't distinguish between large groups at the bottom of the distribution. Again when we think about how poverty has changed we see that it again this matters depending on the index that you choose and we see that the first three indices show that poverty has increased slightly while using factor analysis you see that the factor analysis index shows that poverty has decreased over time so in light of the substantial discrepancies between the different indices I argue that we should really be using economic theory to choose between the indices and my method is just one way that we can do this I argue that it offers some improvement over the existing indices because it's structurally estimated and again therefore is able to incorporate price household demographics and location because of that the interpretation of the weights is more meaningful it generates smoother distributions of household income and ultimately can be used for policy simulations.