4.2.1 Data
Data used in this chapter are obtained from the South African Reserve Bank (SARB) from 1990Q1 to 2009Q3. Data are in constant terms, and seasonally adjusted. The analysis makes use of household savings (the rand value) as the dependent variable to explain the relationships that exists with other explanatory variables. Household savings were deflated using the private consumption expenditure (PCE) deflator to obtain constant figures. Household debt was derived using, the published household debt to household disposable income ratio. In this case the mentioned ratio together with rand values for household disposable income was used to calculate the household debt rand values. Thereafter, the PCE deflator was used to derive a constant value. Finally, the prime overdraft rate was used as a proxy for interest rates.
4.2.2 Methodology
The empirical framework involves carrying out cointergartion analysis, as well as analysing the impulse response functions and the variance decomposition. The preliminary step involves applying the unit root tests to all variables to test for their stationarity. The tests used are the Augmented Dickey-Fuller and the Phillips- Perron tests. These tests are performed to avoid spurious regression that could occur when one non-stationary time series is regressed on another non-stationary time series. The first-difference approach is used in instances where variables are found to be non-stationary in order to remove the trend (Bezuidenhout et al, 2008:29). Hafer and Jansen (1991:156) add that time series analysis require data to be stationary, therefore to get stationary results first-difference of a time series is sometimes applied before estimating an economic model. However, Hafer and Jansen (1991:156) states that if variables are found to be non-stationary, the first- differencing approach leads to a loss of information concerning the long-run relationship between variables. The only way an OLS can be performed on non- stationary time series variables is when cointegration is present (Gujarati, 2003:830).
The Johansen test for cointegration will be used to determine if a long run relationship between savings and the explanatory variables exists. According to Bezuidenhout et al (2008:29), time series are said to be co-integrated when a linear combination of two or more non-stationary time series is stationary (i.e the error). The stationary combination of these variables is interpreted as the long-run equilibrium between variables. Co-integration can only exist if variables have the same order of integration (Harris and Sollis, 2003:20). If it is established that there is a long-term relationship between variables then a vector error correction model will be performed. The vector error correction method is a means of reconciling the short-run behaviour of an economic variable with its long-run behaviour (Uddin, 2009:158).
4.2.3 Model
The model is based on the Keynesian savings function following (Fourie, 2001:50). This savings fucttion is specified as follows:
C
Y
S
=
−
Where
S
= savings,Y
= Household disposable income andC
= Household consumption expenditureThe Keynesian savings function states that
S
=Y
+C
. This can also be rearranged to achieveS
=
Y
−C
. This basic savings function will be combined with the classical model theory of interest rates, as well as the life cycle hypothesis for consumption in order to reveal the main determinants of household savings in South Africa. The independent variables include household income, household consumption, household debt and interest rates. The following equation will be estimated:0 1 2 3 4
t t t t t t t
HS =β +βHI +β HC +β HD +β IR +ε
Where:
HS is household savings, HI is household income, HChousehold consumption,
HDis household debt and
IR
refers to interest rates.According to Loayza et al (2000:393) higher savings are positively related to higher income. It is assumed that if income grows then savings will also grow, that is the higher growth in income coupled with a lagged response in consumption, will lead to higher savings (Surrey 1989:148). Therefore this implies that if income grows households will not use the increase for consumption immediately, they will therefore save the portion of the increment. Increasing household expenditure or spending will lead to a decline in savings. Household debt is assumed to have a
negative relationship with household savings, Singh (2004:4) argued that if households are indebted, paying off debts will diminish savings. Household expenditure on food was used as one of the explanatory variables and multicollinearity was detected as a result and therefore was excluded from the model. However, its impact will still be realised in the total consumption by households. Bloemen and Stancanelli (2003:3) argue that among others, food expenditure is less sensitive to income changes, meaning that any means available will be used for food consumption rather than for savings. Table 4.1 will clearly explain each of the relationships.
A study by Surunga and Tachibanaki (1991:354) indicates that some demographic factors such as age have an impact on the savings behaviour of households. This is also substantiated by the life-cycle hypothesis theory. However, this is beyond the scope of this research. Household wealth is another variable assumed to have a positive relationship with savings. According to Laumas and Ram (1982:204) Friedman’s permanent income theory of consumption is evidently a wealth approach to consumption, meaning that permanent consumption depends on consumer wealth. Furthermore the authors stated that in a model, if the rates of return on wealth move procyclical then the coefficient of income would capture the effect of changes in permanent income caused by variations in the return on wealth. In South Africa this was evident in 2006/7 where the residential buildings variable at constant 2005 prices was at 36 million (SARB, 2010:120). This indicator started dropping during 2008 registering 35 million reflecting the impact of the economic downturn. Therefore for the purpose of this research household wealth will be captured by the household income variable.
Table 4.1 Expected signs of coefficients for the variables
Variable Theory intuition
Expected sign
HH
disposable
income
According to the Permanent Income hypothesis (PIH) there is a positive relationship between household income and household savings. Friedman (1957) argued that richer households save more than poor households. This means that household savings should increase as household disposable income increase. An assumption is made that instead of spending money on goods, households will use the money for savings purposes.
Positive (+)
HH
Consumption
The lifecycle hypothesis as stated in (Obwona and Ddumba-Ssentamu, 1998:26) postulates that there is a negative relationship between household consumption expenditure and household savings. This means that if household expenditure increases then savings from households will cease to exist. Therefore, households cannot consume more and save more at the same time, they must sacrifice their current consumption for the future to be left with income for savings.
Negative (-)
HH debt
There is an inverse relationship between household debt and household savings. (Prinsloo 2002:73). Loayza et al (2000:403) argue that credit expansion reduces private savings because households are able to
finance higher purchases at their current income level. This is more evident because if households are in enormous amount of debt, they will need to pay up the debt first, before they can start to save. In this instance, household savings will eventually deteriorate.
Interest rate
According to the classical model theory, Froyen (1983:65) argue that savings were taken to be a positive function of interest rates. There is a positive relationship between interest rates and household savings. The higher the level of interest, the less willing households will be to use significant amount of cash for transaction purposes. Therefore they will ultimately save their money in order to earn a higher return of interest.
Positive (+)