Abstract

Abstract In this paper we analyze the impact of the expectations about future labor income on the saving behavior of German households. We measure expectations on an individual basis instead of generalized risk measures as it is common in existing studies. We use a unique panel data set on household savings. We find that that a higher unemployment expectation significantly decreases savings. However, we are not able to identify a significant relationship between general future income expectations and savings. Moreover, we find that good health expectations increase savings in Western Germany, but decrease the savings in Eastern Germany. Introduction According to the intertemporal income smoothing motive of savings, people want to smooth their income and consumption over their lifetime. Therefore expectations about the future income influence the saving decisions of individuals and households The expectation about the future income and the variation in the expectation over time are the 2 most important factors for the saving decision. This relation can only be studied by using panel data in which we can observe households over time. Using a panel data set we examine the impact of income and unemployment expectation on the saving behavior of German households [SAVE]. We are unable to identify a clear influence of a higher income expectation on savings. We only find weak evidence for a negative effect of a higher unemployment expectation on savings. Moreover, we find a positive effect of health expectation on savings in Western Germany but a negative effect in Eastern Germany. The reminder of this paper is organized as follows. In the subsequent section 2 we will give a brief introduction of the theoretical literature on household saving decisions. Section 3 describes the data set and section 4 the estimation strategy. The empirical evidence is presented in section 5. In section 6 we summarize the results and discuss the findings. Theoretical Discussion This section summarizes the results of Schlesinger Income Expectation Hypothesis (IEH) If a distribution of future income first-order stochastic dominates another distribution of future income , the optimal level of saving under is always at least as high as under . In other words, a lower future (mean) income expectation should increase the saving. This theoretical proposition will be examined empirically in the following using the German SAVE data set. Since the health of an individual may also affect saving behavior, we also consider the impact of the health expectation on savings.. There is only a very small strand of literature [e.g. [see Engen and Gruber 2 The term is related to the temporal horizon of the investment, meaning how long it takes to divest the capital. 6 The future health expectation is calculated indirectly. The households were asked if they expect to live longer (3), equal (2) or shorter (1) than comparable persons of their cohort. If a person chose "longer" we can assume that this person has a better health expectation than others. We thus create a dummy variable which is one if the member of the household expects a good future health situation (3) and zero of not (when the answers given are either 1 or 2). All information is household based. Thus we calculate household expectation and situation variables from the data on the household members. In the cases of unemployment and income expectation the individual share of the household expectations is equal to the income share of each member of the household. We do so because the impact of the individual expectation/situation on the household savings decision should be more important if the member of the household has a bigger share of the household income. For the health expectation we calculate a household dummy variable similar to the health expectation of the household members. This dummy variable becomes one if the household's overall expectation is larger than two. Thereby we also weighted the individual health expectation of the members of the household by income. Empirical Strategy This section presents the empirical estimation strategy. The theory presented by Eeckhoudt We add several interaction terms, because it is possible that the health situation as well as the health expectation of the householders affect the impact of the income expectation on savings. For example, if an individual has a good health expectation for the future he may reduce his savings because he expects a lower income risk. The same is true for the health situation. A person in poor health may save less money even if he has a bad income expectation (higher first-degree risk). Unfortunately the formulation of a hypothesis about the health expectation is not straightforward, because opposing effects are possible and to our knowledge there is no 3 The results from a pooled regression estimation are similar to the results obtained using panel regression methods 4 The saving rate is calculated as the share of the household's savings on the net income per year. Furthermore using the savings rate eases the interpretation of the estimation results. 5 We also run fixed effect estimations. In this case the savings rate is explained by an additional fixed effect term . 6 The panel includes the individual expectations of the householders but only the overall savings of the household. The individual expectation may affect the individual savings. Because of the household saving information we calculate income weighted household expectations. The individual share on the household's expectations is equal to the individual income share on the overall household income. 7 The dummy variable for a bad health expectation is calculated analogously to the good health expectation dummy. The coefficient of this variable becomes insignificant in the great majority of specifications (44 of 48) but the sign confirms mostly the good health expectation dummy variable. 8 matching theory. 8 Moreover there is only a very small strand of literature on the effect of health expectations on savings in micro-based data. 9 Hence we will present the empirical results and provide an approach for future research. Finally it is also possible, that the income expectations affect savings with different investment periods differently. For example building loan contract may do not reduced because (short-term) income reduction (e.g. unemployment), whereas savings account may decrease. Hence we estimate equation Mutual causality may pose a problem in our regressions since the savings affect the future income expectation. Because of that we also use the expectation about the future employment situation, more precisely the expected probability of becoming unemployed, as an exogenous alternative for the income expectation. We do so because the information about the household's future income expectation in the data set does not differentiate between labour and capital income. Higher savings may increase the future capital income and therefore the income expectation. Thus econometric endogeneity cannot be excluded for sure. To avoid the problem of mutual causality we use also the unemployment expectation to ensure our results. We do so because a higher probability of unemployment is equivalent to a decreasing income expectation [Engen and Gruber 8 The findings of recent studies indicate that wealth may be associated with better self-rated health 9 We estimate the empirical model with focus on the future employment and income expectation of the households for Eastern and Western Germany separately. The households and especially, the income structures differ between both parts of Germany. The female labour market participation and therefore, their share of the household's net income, is much higher in Eastern Germany 10 We also estimate different specifications of equation 11 A second problem is the possibility of collinearity between the exogenous variables, especially between the different expectation and situation variables. An increasing unemployment risk may induce discomfort and therefore a decreasing health expectation. Causality in the other direction is also possible. If individuals expect a bad health situation in the future they may also expect an increasing probability of losing their job/earn less money. 12 Because of this correlation problem we estimate equation 13 10 Regression for Germany with a dummy variable for Eastern Germany show significant differences (the dummy variable is mainly negative and significant). 11 In critical specifications (Wooldridge-test < 0.05) models with an autocorrelated error term were used. These estimation methods confirm the results of the RE and FE specifications. 12 The pair wise correlation of these variables is between 0.4 and 0.7 and significant. 13 The results of this specification mainly confirm the former specification but more variables become insignificant and the explanatory power of the estimation decreases, because a lot of information is (many variables are) not considered. 10 The last specification of equation Empirical Results In this section, we examine if we can find evidence for the IEH from the model by Eeckhoudt and Schlesinger (2008). Does an increasing labor income expectation (decreasing unemployment expectations) lead to increasing savings? At first we present the RE and FE results for the reported (estimated) real savings of the households in Eastern and Western Germany. As mentioned above dis-saving and debt are excluded in this set-up and the value is only an estimation of the householders. We can divide the savings into all savings and savings 14 Furthermore we do a GMM estimation [see 11 for future income (life insurance and building loan contract = long term savings).