When measuring poverty, much of the theoretical and empirical work has focused mainly on money-metric measures of poverty. The conventional approach has been the use of a poverty line - often derived from consumption, expenditure or income levels - sufficient to meet primary human needs. However, the money-metric approach to poverty analysis in South Africa is not an appropriate measure, given that the South African environment has a very different outlook, possibly even arriving at a wrong measure of poverty with subsistence farming, where money is not a good measure of poverty. In order to measure poverty accurately in South Africa we need to consider the assets of households and compute an asset-poverty index. Assets are an important indicator of well-being and a measure that is based on assets is likely to capture an important dimension of economic well-being. While there are a few studies that have investigated asset poverty in South Africa, there are serious gaps in the literature. These should be addressed in order to improve policy designed to reduce poverty. Firstly, these studies have mainly relied on cross-sectional data rather than panel data at a national level. The reason for this is mainly the absence of national representative longitudinal data. However, this type of data has become available and is used in this thesis. Secondly, none of these studies have compared results among
subsamples of urban and rural areas. This is very important as the areas are structurally very different, with different characteristics. Thus, it is likely that poverty, asset poverty and the determinants of asset poverty in these areas will differ. Thirdly, previous literature has not investigated the uniqueness of subsistence-farming communities in the measurement of poverty, in which monetary measures have limited application. In these types of economies, monetary measures of poverty are likely to overestimate poverty. Furthermore, the saving behaviour in these communities differs vastly from that of other communities. To address these gaps, the thesis uses a newly-available panel data set named the National Income Dynamics Study (NIDS) observed over the period 2008-2015 in biannual waves to study asset poverty in South Africa. The panel nature of the NIDS data also allows us to overcome common estimation issues of endogeneity. The NIDS contains a comprehensive set of questions relevant to the analysis of asset poverty. However, the NIDS is not without shortcomings. Although it is a national representative data set, it does not capture the uniqueness of remote subsistence-farming communities. Therefore, we decided to contribute to the literature by collecting and analysing data from a subsistence-farming community. The data complements the current data on urban and rural societies in South Africa, as it captures demographic factors and characteristics of poverty and savings, which are unique to subsistence farming communities. The subsistence-farming community used for the purpose of this survey is Hlokozi village in Kwa-Zulu Natal, which mainly relies on subsistence farming. The primary research objective of this thesis was to create an asset index using Principal Component Analysis (PCA) to identify the poor. The asset index was created for the (i) full sample (South Africa), (ii) subsamples based on the urban and rural populations and for (iii) Hlokozi village, an example of a subsistence-farming rural community. The secondary objective was to investigate the determinants of asset poverty within each of these unique geographical areas, using the newly created asset-poverty index as the dependent variable in regressions. We used methodologies
applicable to panel data for the analysis of the NIDS data set (panel random-effects probit and Two-Stage Least Square (IV-2SLS)) and methods applicable to cross-sectional data in the case of Hlokozi village. The estimated results from the random effects probit and the IV-2SLS show that some factors such as household savings, landholding, marital status, level of education and some provincial dummy variables were statistically significant determinants of asset poverty when using the NIDS dataset. With regard to cross-sectional data in the case of Hlokozi village, a probit model was adopted to investigate the determinants of asset poverty. The estimated results further showed that the levels of education, gender, age of the head of the household, marital status had a reducing impact on asset poverty. To address asset poverty, it is important to save. Therefore, the third objective of this thesis was to estimate a households’ savings function using the NIDS data set for the full sample (South Africa) and a binomial probit model for cross-sectional data. Using the fixed-effects and IV2SLS models in the case of the NIDS, the results showed that savings are strongly driven by income earnings, employment status, levels of education, age of the head of the household and the province in which the respondent resides. Using the binomial probit model for cross-sectional data, the results are quite similar to the panel-data
models when using NIDS. The implication in policy terms is that any policy geared towards enabling saving that would facilitate asset accumulation and mitigate the negative influence of asset poverty, should be based on a thorough understanding of households’ savings behaviour, and the factors influencing their saving behaviour.