Assets and Poverty Status Dynamics in 5 Main Regions in Indonesia

Comparing household expenditure and national poverty line, about 24.78% of households in Indonesia experienced poverty (expenditure below the poverty line) at least once within 14-year period. By utilizing the Ordered Logit Model, this study examines the determinants of household poverty status and analyses the relative effect of different household assets and characteristics on their poverty status. Employing three waves of Indonesia Family Life Survey (IFLS) consisting of household level data from the year of 2000, 2007 and 2014 and categorized households into five main regions based on their location. This study finds that assets (building, vehicle, jewellery, and savings) play important role in determining poverty status of households in Indonesia. Some demographic and socio-economic variables are confirmed to be statistically significant to poverty status in Indonesia. However, the determinants of poverty status vary within regions.


Introduction
As one of the developing countries, Indonesia has made enormous efforts both through government and NGO with social assistance programs, community-driven development programs and many other programs to eliminate poverty. However, many people are still vulnerable to poverty and even non-poor groups of the society face the possibility of becoming poor in the near future. Indonesia enjoys increasing growth as well as a decreased poverty rate but the rate of decline is getting slower and slower as we can see in Table 1. The fact that poverty is still exist and the rate of decline is slowing down lead some to challenge the efforts that have been formulated and executed as ineffective or not on target. To answer that challenge, many studies have been conducted but most research on poverty in Indonesia still focus on a static poverty analysis that analyzes the proportion of people being poor in a given single point of time (Afandi, Wahyuni, & Sriyana, 2017;Hariadi, 2009). However, when we discuss policy related to the poverty problem, we need to consider that there is a lag between policy implementation and the emergence of the results, such as the impact of subsidy on education or return to assets on poverty. These studies, that only investigate the factors of poverty in a static or short term period study, fail to account for that lag. Balisacan & Fuwa (2007) mentions that the main cause of poverty or economic mobility is a low level of assets endowment, low return to assets and the inability to cope with negative income shocks. Naschold (2012) introduces the concept he calls "stock of assets" to explain how a household's economic wellbeing can be analyzed as that which, accumulated over a period of time, makes it possible for the household to move out of poverty through income gain using those assets. The next question would seem to be, to what extent assets endowment could help the poor? Could higher assets endowment assure that vulnerable people will not be poor in the future or even be better-off? Therefore, the objectives of this study are: 1) to examine the determinants of household poverty status divided into categories such as chronic poor, transient poor and never poor; 2) to understand to what extent household assets and other household characteristics measured in the initial period affect the dynamics of household poverty status in the following periods.

Research Method
This study employ the last three panel survey of Indonesia Family Life Survey (IFLS) data, a longitudinal socioeconomic and health survey conducted by The RAND Corporation: IFLS3, IFLS4 and IFL5, selecting the household which were re-interviewed for all three waves making a total of 9,229 households. The advantage of using the IFLS datasets to analyze poverty dynamics is due to this high re-contact rate. This survey enables us to track the same households and follow their welfare dynamics over time, even though a household may have migrated to another region.
The analysis of poverty dynamics starts from defining poverty. This study defines poverty in each given year as expenditure of a household below the poverty line. Data of expenditure was gained from IFLS 2000, 2007 and 2014 datasets by aggregating the data on food expenditure and non-food expenditure calculated monthly, in real terms. To get per capita expenditure, the monthly expenditure was divided by household size.
The poverty line used in this study is gained from the rural-urban specific per capita poverty line published by Statistics Indonesia for year 2000, 2007 and 2014. Statistics Indonesia defines this poverty line on the basis of the calculation of essentials food and non-food estimates combined. To estimate minimum food needs, the rupiah value to fulfill 2,100 kcal/day energy intakes monthly per capita is used and for the non-food, the rupiah value of basic monthly per capita needs including housing, clothing, schooling, transportation and other basic needs are used. (2011, pp. 6). Next, this study compares the per capita expenditures to the official poverty line, to obtain the data of "poor" and "non-poor" household in each year.
After determining who the poor are, the severity levels of their poverty are classified in terms of length. This classification is based upon the foundational concept of spells of poverty. Using this approach, we construct the poverty status as the explained variable with categorization into the following spells: (1) Chronic poor, if a household is poor three times; (2) Usually poor, if a household is poor only two times; (3) Occasionally poor, if a household is poor once; and (4) Never poor, if a household never falls into poverty. withthesespecifications: Using an ordered logit model, these poverty status categories were regressed on the factors that determine them. The ordered logit model was utilized for the reason that the poverty status categories have an order of preference where one status is better and preferred over the others. The order of poverty status is "never poor" > "occasionally poor" > "usually poor" > "chronic poor". The most preferred poverty status is "never poor" and the least preferred poverty status among them is "chronic poor". The ordered logit model for this study is specified as the following: The explanatory variables that were utilized in this study consist of variables from the initial year (2000) and their subsequent changes. These initial variables characterize a set of certain conditions assigned to each household and used to explore whether these characteristics determine the poverty status. Additionally, this study used the variables of change identified as the years between 2000 and 2007, and between 2007 and 2014 to represent how a change in a characteristic (for example, change in location from rural to urban or urbanization) affects a household's poverty status in the following periods. The descriptions of each explanatory variable along with their expected values are presented in the Table 3.

Results and Discussion
The estimated coefficient signs in the ordered logit models give the same indications compared to the linear regression results in terms of direction. However, in the logit model the analysis additionally shows the marginal effect (dy/dx) of how changes in explanatory variables affect the probability of a household to be chronic poor, usually poor, occasionally poor and never poor. Table 6 and 7 shows the marginal effects (dy/dx) of changes in the probability of households being poor, usually poor, occasionally poor and never poor in response to a change in the explanatory variables, while setting all the explanatory variables at their mean values. As shown in Table 6, the probability of households in Indonesia to be chronic poor, usually poor, occasionally poor and never poor are 0.64%, 5.26%, 24.3% and 69.78%, respectively. In the Java and Bali region, the probability of households to be chronic poor is 0.66%; usually poor is 4.98%; occasionally poor is 22.19%; and never poor is 72.16%. However, in the remaining regions, the probability of households outside Java and Bali collectively to be chronic poor is lower compared to Java and Bali (0.52%). When we set the explanatory variables at their average value, generally, the probability over 14 years of a household in Java and Bali region to be either always poor or never poor is higher than it would be for households outside Java and Bali.
The third model (national model or full sample) shows that all demographic and socio-economic variables are statistically significant in determining the poverty status except marital status and location. In Model 1, the significant demographic and socio-economic variables vary across the regions.
In terms of asset holdings, Model 3 confirms that building, vehicle, jewelry, and savings are statistically significant in determining poverty status in Indonesia as a whole. However, Model 2 shows that for the regions examined separately, house, building, jewelry and saving are statistically significant as poverty status determinants in Java and Bali. In regions outside Java and Bali (collectively), land, poultry, vehicle, jewelry and savings are statistically significant. Moreover, Model 1 confirms that the significant factors for poverty status in Sumatera are only jewelry and savings; in West Nusantara they are land, vehicle and savings; in Kalimantan they are house, building and jewelry, while in Sulawesi the significant factors are only house and savings. These findings illustrate how different assets influence a household's welfare in different way.

108
Economic Journal of Emerging Markets, 9(1) April 2017, 104-113 Referring to Table 7, setting savings at its mean value, a 100% change in savings is associated with a 0.45% increased probability of a household in Indonesia identifying as never poor. However, a 100% change in savings is associated to increase the probability of a household in Indonesia identifying as chronic poor, usually poor, and occasionally poor by 0.014%, 0.11%, and 0.33% respectively. A 100% change in value of jewellery, leads to an increased probability of the household in Indonesia to become never poor by 0.32%. However, 100% increase in value of jewellery decreases the probability of a household in Indonesia becoming chronic poor, usually poor, and occasionally poor by 0.0098%, 0.07%, and 0.24% respectively.   This finding confirms the study by Jalan & Ravallion (2000) that finds chronic and transient poverty to be reduced greatly by physical capital. Savings and jewellery are rarely employed as a proxy of assets, with many studies using land ownership/value instead (Jalan & Ravallion, 1998;McCulloch & Baulch, 2000;Haddad & Ahmed, 2003;Woolard & Klasen, 2005;Dartanto & Nurcholis, 2014). Savings and jewellery are found to significantly impact poverty status since they are more liquid or can more easily to be turned into cash to provide additional funds in the face of some negative shock whereas selling assets like land or house takes more time. Therefore, encouraging the poor to accumulate assets such as gold and bank account savings could be used as a preventive strategy for facing shocks and to lift them out of poverty as well.     15e-05*** -0.000767*** -0.00264*** 0.00348* ** -5.88e-05*** -0.000450*** -0.00142*** 0.00193*** ln_jewelr y -7.12e-05** -0.000506*** -0.00160*** 0.00218*** -0.0001 04*** -0.000979*** -0.00336*** 0.00445* ** -9.81e-05*** -0.000752*** -0.00237*** 0.00322*** ln_saving -0.000125 *** -0.000888*** -0.00281*** 0.00382*** -0.0001 44*** -0.00135*** -0.00465*** 0.00614* ** -0.0 00137*** -0.00105** * -0.00332*** 0.  In general, this study supports the results of the previous studies related to the determinants of poverty. Assets, such as building, vehicle, jewellery and savings are important in determining poverty status of households in Indonesia. In addition, demographic and socio-economic variables such as household head's gender, age and education, member's education, household size, nutrition, work sector, employment sector and access to electricity are confirmed to be statistically significant to poverty status in Indonesia. Additionally, government assistance and access to credit also affect the poverty status of households in Indonesia. Changes in household size, employment sector (from formal to informal) and gaining credit is significant in determining poverty status during 2000-2007 and finally, during 2007-2014, a change in household size, marital status (divorce) and work sector led to an increase in the probability of households to be poor.

Conclusion
Using the three last waves of the Indonesia Family Life Survey (IFLS) consisting of household level data from the years 2000, 2007 and 2014, this study identified the poverty status dynamics of households in Indonesia and their determinants. This study used ordered logit model to examine the determinants for poverty status of households in Indonesia (chronic poor, usually poor, occasionally poor and never poor) by grouping the households into five regions (Sumatera, Java and Bali, Kalimantan, Sulawesi and West Nusa Tenggara) as well as into two subnational grouping (Java and Bali and outside Java Bali) and, finally, by a full sample analysis.
The results show that the determinants of poverty status vary from region to region. One of the most interesting findings is that unlike previous studies which found land to be the common indicator of household assets, it is only significant in the West Nusa Tenggara region. Another interesting finding is that the probability of households either to be chronic poor or never poor is higher in Java and Bali, while the probability of transient poverty is found to be higher outside Java and Bali. Finally, this study finds that there is no indication of chronic poverty in the Kalimantan region.
In general, the study finds that assets play an important role in determining poverty status of households in Indonesia. However, the only assets variables that are statistically significant in determining poverty status are building, vehicle, jewelry and savings. Besides assets, demographic and socio-economic variables such as household head's gender, age and education, member's education, household size, nutrition, work sector, employment sector and access to electricity are confirmed to be statistically significant to poverty status in Indonesia. Additionally, the positive shocks variables, such as government assistance and access to credit also affect the poverty status of households in Indonesia.