RDP 2004-01: The Impact of Superannuation on Household Saving Appendix C: Supplementary Results and Robustness

Ellis Connolly and Marion Kohler

March 2004

Table C1: ECM Results for Saving Equations
Sample: 1966/67–2001/02
  Voluntary saving
 		 Voluntary saving
excluding super
Labour incomet−1 0.13**
(0.03)		 0.13**
(0.03)
Compulsory superannuationt−1 −0.38
(0.26)		 −0.31
(0.20)
Voluntary superannuationt−1   −1.30**
(0.18)
Net wealtht−1 0.00
(0.01)		 0.00
(0.01)
Debt to incomet−1 −0.02**
(0.01)		 −0.02**
(0.01)
Δ(Voluntary saving)t−1 0.10
(0.14)		 0.19*
(0.12)
Δ(Labour income)t 0.53***
(0.11)		 0.39***
(0.11)
Δ(Compulsory super)t −1.78**
(0.70)		 −0.93
(0.86)
Δ(Voluntary super)t   −1.47**
(0.38)
Δ(Net wealth)t −0.03**
(0.01)		 −0.01
(0.01)
Δ(Debt to income)t 0.03*
(0.01)		 0.01
(0.01)
Constant −0.68
(0.48)		 −0.53
(0.49)
Speed-of-adjustment −1.30**
(0.20)		 −1.47**
(0.24)
Inline Equation 0.60 0.71

Notes: Numbers in parentheses are standard errors. **, * represent significance at 5 and 10 per cent levels. Standard errors on the long-run variables in the ECMs are calculated using a Bewley Transformation. Standard errors are Newey-West corrected for heteroskedasticity. A negative and significant speed-of-adjustment coefficient is consistent with the presence of cointegration.

Hausman tests for regressor endogeneity in the dynamics do not suggest that there are endogeneity problems that could be corrected using the instruments that we selected. Potentially endogenous variables include the change in labour income and the change in voluntary superannuation. Our instruments include lags of these variables and lagged voluntary super, US output, lagged consumption, lagged government expenditure and the other variables in the ECMs.

We also test for parameter instability using Chow breakpoint tests and recursive regressions, which do not suggest that parameter instability is a serious problem. Note that a possible problem might be posed by the compulsory superannuation variable, which is zero in the first half of the sample. However, the coefficient on this variable does not change if we shorten the time period, as also suggested by the Chow test. The other coefficients, however, can be estimated more efficiently over the longer time period.

Several other variables were introduced into the model to check the robustness of the results to potential omitted variables bias (Table C2). The short-term real interest rate was not significant in regression (c). This is not surprising given that the income and substitution effects of interest rate movements on saving are of opposite signs, potentially producing an ambiguous outcome: Deaton (1992) and Edey and Britten-Jones (1990). A demographic variable (ratio of persons aged 45 to 59 relative to persons aged 15 to 59), was insignificant in regression (d). This result is also not surprising, since Edey and Britten-Jones (1990) could not find evidence that consumption smoothing behaviour is important over individuals' lifetimes. A measure to capture uncertainty, the unemployment rate, was also insignificant in regression (e). This may be because uncertainty is being captured in movements in labour income and wealth.

Table C2: Supplementary ECM Results for Voluntary Saving Equations
Long-run coefficients only, sample: 1966/67–2001/02
  (c) (d) (e)
Labour income 0.15**
(0.05)		 0.11**
(0.04)		 0.14*
(0.08)
Compulsory super −0.33
(0.42)		 −0.40
(0.31)		 −0.34
(0.32)
Net wealth 0.00
(0.01)		 0.00
(0.01)		 0.00
(0.01)
Debt to income −0.02*
(0.01)		 −0.02**
(0.01)		 −0.02*
(0.01)
Ratio of persons aged 45–59
to persons aged 15–59		 0.02
(0.03)		    
Short-term interest rate   0.01
(0.01)		  
Unemployment rate     −0.01
(0.03)
Speed-of-adjustment −1.34**
(0.22)		 −1.15**
(0.22)		 −1.28**
(0.25)
Inline Equation 0.57 0.61 0.57

Notes: Numbers in parentheses are standard errors. **, * represent significance at 5 and 10 per cent levels. Standard errors on the long run variables in the ECMs are calculated using a Bewley Transformation. Standard errors are Newey-West corrected for heteroskedasticity. A negative and significant speed-of-adjustment coefficient is consistent with the presence of cointegration.

We have only dealt with the effect of compulsory superannuation and voluntary superannuation on voluntary saving, without explicitly considering the role of other employer superannuation contributions and superannuation fund earnings. A large proportion of other employer superannuation was provided through unfunded schemes, which are in the process of being phased out. In our model, we would not expect households to offset the portion of compulsory superannuation that merely superseded these unfunded schemes. Past funded employer contributions and superannuation fund earnings can influence current saving through the net wealth variable.[30]

Deaton (1992) argues that it is difficult to use macroeconomic data to test micro theories of consumption and saving due to the aggregation of individuals with heterogenous information and different life spans. Nevertheless, in the absence of appropriate micro data on saving over the period of interest, we believe that the analysis of macro data is an appropriate methodology for measuring the effect of superannuation on aggregate household saving.

Footnote

Unfunded schemes are not included due to data limitations. Since these schemes did not provide employees with vested superannuation accounts, they are less likely to be influencing households' saving behaviour, reducing the potential for omitted variables bias. [30]