Model specification

The nine independent variables identified in the literature that might influence the savings of poor households are as follows: social grant income; other income (profit or earned income including remittances); income pooling; necessity goods expenditure; normal goods expenditure; luxury goods expenditure; gender of respondents; age of respondents; and household size.

Research has shown that all income acquired from labour market activities, remittances, social grants and borrowed income have a positive impact on households savings (Chakrabarty & Hildenbrand, 2009:649; Esson, 2003:13; Neves et al., 2009:13&17; Paxton, 2009:228; Skidmore, 2001:18; Ting & Kollamparambil, 2015:282). Therefore, the social grant income and other income used in the model of this study are expected to have a positive impact on the savings of the households under consideration, while consumption of necessity, normal and luxury goods have a negative impact as argued by Keynes (1936:20).

According to the literature, households that pool income are presumed to have the ability to save (Burger et al., 2010:3; Case & Deaton, 1998:1340; Esson, 2003:13;

Maltsoglou & Taniguchi, 2004:14; Meyer et al., 2009:43; Woolard, 2002:2; Van der Merwe, 2000:729). Therefore, the expected sign of income pooling as a dummy variable in this study is positive. Since the literature also shows that expenditure has an inverse relationship with saving, the independent variables in this study – necessity, normal and luxury goods expenditure – are expected to have a negative sign.

78 Another independent variable in this study is the gender of household financial handler, which has been identified as an influencing on household savings by Neves et al. (2009:48-49) and Paxton (2009:225). In this study, gender will be regressed as a dummy variable and the coefficient sign of female will be positive in line with Paxton’s (2009:225) study of saving among poor rural households in Mexico on which females are shown to have a positive influence. Similarly, the age of the household financial handler is used as an independent variable in this study as it has been shown in the literature to have a positive correlation with household savings (Beckmann, 2013:9;

Modigliani, 1966:163; Sarantis & Stewart, 2001). The expected sign of age squared in this study is positive and it is expected to account for possible non-linearity in the age of the surveyed households’ financial handlers.

The descriptive analysis of the data indicated that the majority of household financial handlers were between the ages of 28 and 50, with an average age of 43. According to the life-cycle hypothesis, household financial handlers have positive savings during their working years (Modigliani, 1966:163; Ting & Kollamparambil, 2015:282).

Therefore, since the majority of the household financial handlers are within the working age, the expected sign is positive.

Chakrabarty and Hildenbrand (2009:649), Keynes (1936:20) and Skidmore (2001:18) all indicate that household size has an inverse relationship with household savings.

But Van der Merwe (2000:729) argues otherwise, positing that household size, particularly large household size, has a positive relationship with household savings.

Unlike the others however, Van der Merwe (2000:729) merely does not provide empirical results to support his claim. Therefore, in this study, the expected sign of household size in relation to household savings is negative. It is expected that large households, with more than two members, save less compared to small households with one or two members. The descriptive analysis of the study reports that majority of the surveyed households are large. This is shown in Table 5.3 below.

79 Table 5.3: Expected coefficient signs of the variables used in this study

Variable Description

(Measurement)

Expected sign Social grant Social grant is the amount of social

grant income received by a household.

(The social grant income of households was collected by use of a questionnaire and it was measured in ZAR)

+

Other income, including remittances

Other income refers to income received from labour, business, remittances and other income sources. (Other income was measured in ZAR and collected by use of a questionnaire)

+

Pool income Pool income refers to income pooled within a household (pool income was measured in ZAR and collected by use of a questionnaire)

+

Necessity Necessity refers to necessity goods and services consumption of which a household cannot survive without (necessity was measured in ZAR and collected through a questionnaire)

-

Normal Normal refers to normal goods and services consumption of a household, which a household consumes more of when income increases (normal was measured in ZAR and collected through a questionnaire)

-

Luxury Luxury refers to luxury goods and services consumption of a household, which a household consumes more of

-

80 when a household’s income is more

than average income (luxury was measured in ZAR and collected through a questionnaire)

Gender Gender refers to the gender of

household financial handler (Measurement was that either a financial handler is male or female and gender was collected through the use of a questionnaire)

+

Age Age refers to the age of household financial handler (the age of household financial handler was measured in years and collected through the use of a questionnaire)

+

Age^2 Age^2 refers to the squared age of household financial handler (the age^2 was measured in years and age was collected through the use of a questionnaire)

+

Household size Household size refers to number of individuals who have been residing within a household in the past three months (the household size was measured as a number and collected through the use of a questionnaire)

-

Source: Author’s own

Based on the expected coefficient signs of the independent variables, the linear savings function is as follows:

81 𝑠 = 𝛼 + 𝛽₁𝑠𝑦_𝑖 + 𝛽₂𝑦_𝑖+ 𝛽₃𝐷𝑝𝑜𝑜𝑙_𝑖 − 𝛽₄𝑐𝑁_𝑖 − 𝛽₅𝑐𝑁𝑟_𝑖− 𝛽₆𝑐𝐿_𝑖+ 𝛽₇𝐷𝑔𝑒𝑛𝑑𝑒𝑟_𝑖 + 𝐵₈𝐴𝑔𝑒_𝑖 +

𝛽₉𝐴𝑔𝑒_𝑖²− 𝛽₁₀ℎℎ𝑠_𝑖 + +𝐸_𝑖 (4.2)

Empirical results are generally used to determine whether the expected signs were met, whether the independent variables influence the savings behaviour of households with social grant recipients and whether their influence is statistically significant after running a regression model. A series of regression models are applied using the Stata software package.