| As one of the three greatest economic difficulties, poverty has been concerned by governments and the social public. Governments made their anti-poverty policies mainly on the basis of current poverty assessment. But today is poor doesn't equal to tomorrow is poor, currently poor households may escape poverty quickly or continue to be poor in the future, being stuck in a "poverty trap". Currently non-poor households may fall into poverty due to serious adverse shocks. A household's observed poverty level or status-defined in terms of the household's observed income or consumption-can only measure the household's current well-being statically, without considering future well-being and risks about future well-being. Anti-poverty policies accordingly can only be mending at best, they are not forward-looking and couldn't prevent poverty before it occurs.The concept of "Vulnerability" was officially proposed in The World Development Report (WDR) 2000/2001 by the World Bank. After that, this forward-looking viewpoint became a research hotspot of development economics. A large amount of research achievements have been made about how to incorporate risk into poverty assessment framework and how to understand and measure vulnerability. Learning from foreign experiences, constructing vulnerability research framework suitable to Chinese condition, providing reference to the targeting and the designing of forward-looking anti-poverty policies, has both theoretical and practical value.In this paper, the basic research element is a household, the thesis is the measurement, resources and determinants of vulnerability, the research methods includes literature research, descriptive statistics (including statistical charts, statistical figures, mean, standard deviation), non-parametric method(including rank correlation analysis, kernel density estimation, bootstrap, quartile regression) and econometric models(panel data fixes effects models, regression analysis). The empirical study is conducted with Matlab, Excel, Eviews, SPSS, SAS, using CHNS as data resource.The whole paper is arranged in the logical order of measurement, identification, intervention, and is divided into five stages and ten sections.Stage 1 consists of the first three sections, constructing theoretical and data basis of research, On the one hand, we analysis and summarize the literatures, defining the connotation of vulnerability and construct the research framework. On the other hand, according to the basic theory of labor economics, welfare economics and sociology we conduct data processing and matching and prepare as many variables as possible for the following study.Stage 2 is to measure, consisting of the next three sections.Secion 4 provides the method system to measure households' vulnerability, including descriptive and inferential methods. There are seven descriptive indicators:marginal-poor vulnerability, occasionally-poor vulnerability, temporary poverty I (TP1), temporary povertyâ…¡(TP2), negative variability rate, absolute variability rate, average variability rate. There are three inferential indicators: EP0,EP1,EP2. For inferential measurement, we need to distinguish different estimation methods of future income distribution:intertemporal mean and variance of panel data (method 1), regressing mean and intertemporal variance of panel data (method 2), regressing mean and variance of panel data (method 3), regressing mean and variance of cross-sectional data (method 4), nonparametric method (method 5-1/5-2/5-3).In section 5, we use 84 combinations of indicator, distribution estimation method of future income and poverty line(PL), to measure households'vulnerability. We record the measurement results from the prospect of vulnerability and poverty comparison. Due to the differences of angle, scale, and dimension, the results of these measurements are different, but the basic conclusions are the same:Firstly, the poor are more vulnerable. The vulnerability- poverty comparison tables, the vulnerability histograms and the vulnerability occurrence curves all demonstrate this, which implies that it's reliable to forecast poverty with vulnerability. Secondly, vulnerability doesn't equal to poverty. Not all vulnerable households are poor and not all poor households are vulnerable. They have the same parts but are not totally coincident. Thirdly, the income gap and vulnerability gap of urban households are larger than those of the rural households. Fourthly, the rural households are more vulnerable. The urban-rural difference of vulnerability increases with the PL. This pattern exists in population and non-poor households, but in poor households, the pattern is up-side-down, which implies that urban low-income groups face greater risk to be poor.In Section 6, we evaluate the reliability, robustness and accuracy of the measurement combinations, based on the empirical results, thus we choose the best combination of PL, vulnerability line(VL) and estimation method of future income distribution as the research basis of next stage. Firstly, we should choose the 2$ PL. Secondly, it's complicated to choose VL. For low VL, we consider two ways:the average poverty VL and the average vulnerability VL. The descriptive methods only apply to average poverty VL. EP1, EP2 and nonparametric measurement of EP0 apply to the average poverty VL, while parametric measurement of EP0 applies to the average vulnerability VL. This conclusion questions the routine choice of taking poverty rate as VL. With the idea of non-dimensional disposal We set seven high vulnerability lines(HVL). For descriptive measurements, extreme value HVL has the best performance, suitable to both TP1 and negative variation rate, trimmed 1% extreme value HVL fits absolute variation rate, and trimmed 5% extreme value HVL fits TP2. The HVLs of inferential measurements are more complicated. In general,0.5 HVL fits EP0 best, in accordance with the routine choice of current literatures, while EP1å’ŒEP2 apply to vulnerability extreme value HVL. Thirdly, as for the distribution estimation method, three types of assessment are not the same, we have to make a tradeoff. Parametric methods are better than the nonparametric ones. Among the parametric methods, method 3 is the best. It's worth mentioning that, although method 4 measures vulnerability with cross-sectional data and is naturally in the inferior position, it still has some good performance. Although its reliability is behind other methods, its robustness stays the second, just behind method 3. Fourthly, in terms of measurement indicators, generally, inferential measurements are better than descriptive ones, but descriptive measurements have good performance in some aspects. In reliability evaluation, TP2, TP1, negative variation rate are the second, third and fourth, higher than EP0 and EP1.Occasional poverty vulnerability and average variation rate are ahead of EP0, too. Thus, although descriptive measurements are essentially ex-post measurements and don't have the complete connotation of vulnerability, they provide some viewpoints that inferential measurements can't get to, so they are valuable in the measurement system. The comparison of three inferential measurements are obvious, their order are EP2,EP1,EP0. This conclusion totally coincides with current literatures' theoretical evaluation and our judgment of them in the former analysis.Stage 3 is to identify, corresponding to section 7. Using the best measurement combination, we measure different urban-rural groups'vulnerability by province, education, age, career, dependency ratio. Thus, we identify vulnerable areas and groups, providing reference for the targeting of forward-looking anti-poverty policies. In urban sample, the most vulnerable groups are:Guizhou and Guangxi, close distance group, primary school group and middle school group, non-self-employed group, age group over 65, high dependency ratio group. In rural sample, the most vulnerable groups are:Guizhou and Guangxi, close distance group, illiterate group and primary school group, all-self-employed group, age group over 65, high dependency ratio group.Stage 4 is intervention stage, including section 8 and section 9, the resources and potential determinants of vulnerability are analyzed, providing basis for the designing of pointed and specialized poverty policies.Section 8 decomposes the resources of vulnerability from two angles:vulnerability rate and vulnerability difference, and draw the following conclusions:Firstly, the lower income or the higher variability, the more vulnerable the households are. Secondly, households with the same or similar level of vulnerability may have different resources, some are vulnerable because of low income (LM), the others are vulnerable due to high variability (HV). Thirdly, the decomposition of vulnerability rate shows that, HV and LM vulnerability occupy certain proportions in most groups. The poor households' vulnerability mainly comes from LM, and the non-poor households' vulnerability mainly results from HV. For urban households, the share of HV is greater than that of rural households and the share of LM is lower than that of rural households. Finally, the decomposition of vulnerability difference shows that, HV is the main resource of urban vulnerability, and LM is the main resource of rural vulnerability. Positive variability effect is dominant for the urban high vulnerability groups, while positive level effect is dominant for the rural high vulnerability groups, negative level effect is dominant for both the urban and rural low vulnerability groups.In section 9, we screen the determinants of vulnerability, compare the relative importance of these determinants, and analyze the determinants'impact at different positions of vulnerability distribution. Firstly, the determinants of the urban and the rural households have differences and similarities. The same negative determinants are:number of adults, education, proportion of medical insurance, age and all kinds of assets. The same positive determinants are:household size, proportion of hypertension. The coefficients of whether have skilled member and proportion of self-employed are opposite in urban and rural households, they are negative for the urban and positive for the rural. The BMI of household head and proportion of SOE employees have significant negative impact on the urban and have no significant impact on the rural. Secondly, the importance of the determinants is different. Assets are more important than other determinants for both urban and rural households. It implies that protecting and increasing the vulnerable groups'assets is the basic way to lower their vulnerability. Proportion of medical insurance and age of the household head are the second important determinants for the urban and rural households, confirming the rationality of urban-rural medical insurance system reform. It's worth noticing that human capital (number of adults, education, health, skill and technology) and social capital (gifting expenses) are on the third or fourth level of importance. It implies that these determinants haven't played their due roles and have great space to act. Thirdly, the impact of determinants is different at different positions of vulnerability distribution. Anti-poverty policies aiming at the most vulnerable groups should focus on two types of determinants:negative determinants whose impacts increase with the vulnerability quartile (type A) and negative determinants fluctuate but the impact on high quartile are bigger than that on low quartile (type B). For urban households, determinants of type A include assets, self-employed proportion, SOE proportion; determinants of type B include number of adults, skilled members, BMI of household head, east. For rural households, determinants of type A are:various assets, fever cost, gifting expenses, determinants of type B are: number of adults, education, skilled and technical members, BMI of household head.Stage 5 includes section 10, summarizing the conclusion of vulnerability measurement, identification and intervention, providing the corresponding policy suggestions, pointing out future research directions.The innovations of this paper are as follows:Firstly, the uniqueness of vulnerability measurement framework. The measurements are divided into descriptive and inferential methods. For descriptive measurements, we proposed marginal-poor vulnerability, negative variability rate, absolute variability rate, average variability rate for the first time. For inferential methods, current literatures only theoretically discuss rather than analyze empirically EP1 and EP2, we go beyond this, using five estimation methods of future income distribution to measure EPo, EP1, EP2 systematically. So far.no research has conducted so comprehensive measurement. Furthermore, the research framework from measurement to identification, then to intervention is logically strict and structurally novel.Secondly, evaluating measurement methods from different angles. Current vulnerability measurements are varied and diverse. Due to the difference of data used, not only the measurement results but also the performances of measurement methods are not comparable. Evaluation of measurement methods is quite important to anti-poverty practice. After all, it's unrealistic to measure vulnerability with various methods in anti-poverty practice, we need an uniform measurement. But existed literatures lack this kind of research. We try 84 combinations of measurement indicators, PL, estimation methods of future income distribution, to measure households'vulnerability and assess the measurement methods based on the results. Both the evaluation viewpoints and evaluation indicators have certain innovation.Thirdly, the uniqueness of research methods. We use quartile regression to analyze whether the determinants'impact at high quartiles and low quartiles of vulnerability distribution is the same or quite different. It's the first time in vulnerability research area. Appling bootstrap method in vulnerability measurement is the first time in domestic published literatures. Evaluating the robustness of measurement methods with rank correlation is a new attempt.Fourthly, initiative research on selecting VL. In existed literatures, the discussion of VL is limited to EP0, this paper extends it to all measurements. Learning from the idea of setting VL for EP0, we set two low VLs and seven high VLs and choose proper VLs for different measurements. Theoretically speaking, discussion of VL is an useful complement to vulnerability research system. Practically speaking, without setting VL for a measurement, we couldn't judge whether a household is vulnerable, thus we couldn't use this measurement to policy targeting, which limits the application of vulnerability in anti-poverty practice.There are some deficiencies in this paper. Firstly, the content of available data limits the viewpoint and depth of the research. CHNS doesn't provide consumption data, we can only use income data to measure vulnerability and we can't measure vulnerability from the angle of risk exposure. CHNS lack information of risk and risk response, many vulnerability measurement ideas can only be theoretical. Secondly, the length of panel data is limited, our research is still on strong assumptions, the accuracy and reliability of research results is affected. Thirdly, because the sample size.is limited, we only conduct two-level grouping to aggregate the results of measurement and decomposition. So the identified results are too rough and not so specific. Fourthly, there is some subjectivity in the empirical analysis. For instance, most variables used in this paper are not raw data of CHNS, but through matching among different CNNS data files or even processing and recalculating. There is much subjectivity in the data processing, affecting the results to some extent. Finally, this paper is only a preliminary attempt to study China's vulnerability problem on the basis of abroad experiences, especially the applicability of different measurement methods. We didn't propose totally new measurement method and improve the problem of lacking forward-looking.
|
PDF Full Text Request