Abstract
This paper analyses the Household Budget Surveys prepared by the Turkish Statistical Institute to reveal the empirical importance of precautionary saving in Turkey. The most difficult aspect of the empirical analysis is the approximation of labour income risk as a proxy variable for future labour income uncertainty. Individual disposable income is interacted alternately with the probability of being unemployed and with the probability of job loss in the next period to generate the labour income risk variables. The econometric results support the precautionary saving hypothesis and labour income risk emerges as one of the main determinants of household saving decisions. Moreover, households implement alternative strategies to smooth out their income streams such as holding a second job and increasing the number of income earners in the family. However, it is evident that they are still vulnerable against labour income risk, which underlines the need for an effective and efficient social security system.
Similar content being viewed by others
Explore related subjects
Discover the latest articles, news and stories from top researchers in related subjects.Avoid common mistakes on your manuscript.
1 Introduction
The aim of this empirical research paper is to analyse the impact of labour income risk on household saving decisions in Turkey. Although the analysis of household consumption and saving behaviour is arguably one of the most interesting topics in economic theory, the empirical literature is far from satisfactory. Specifically, there is a significant gap in the literature from a micro-economic point of view within the context of developing countries. Therefore, this paper will analyse the Household Budget Surveys prepared by the Turkish Statistical Institute (TURKSTAT) to reveal the empirical importance of precautionary saving in Turkey.
The discussion of the nature and significance of precautionary saving is not only a theoretical issue, but it is also important for public policy formation, especially in the design of health care and unemployment insurance schemes (Attanasio and Weber 2010). There is a rich empirical literature on precautionary saving in developed countries; i.e. Lusardi (1997, 1998), Kazarosian (1997), Carroll and Samwick (1997, 1998), Guariglia (2001), Guariglia and Rossi (2002), Carroll et al. (2003), Fuchs-Schündeln and Schündeln (2005) and Benito (2006). Although there are significant contributions such as Kochar (1999), Guariglia and Kim (2003, 2004) and Chamon and Prasad (2010), empirical research on precautionary saving for developing countries has to date been rather limited.
The relationship between household saving and labour income risk is analysed using the TURKSTAT Household Budget Surveys. The most difficult aspect of the empirical analysis is the approximation of labour income risk. Individual disposable income is interacted alternately with the probability of being unemployed and the probability of job loss risk in the next period to generate two separate proxy variables for labour income risk. The econometric results show that there is a positive and significant relationship between household saving and permanent income. Labour income risk emerges as one of the main determinants of household saving decisions. In addition, durable goods purchases are also negatively affected by labour income risk, but the sensitivity of durable goods purchases to labour income risk is lower than the rest of household consumption expenditures. Thus, the empirical analysis supports the precautionary saving hypothesis.
Moreover, households implement alternative strategies to smooth out their income streams to smooth out their consumption patterns such as holding a second job and increasing the number of income earners in the family. In particular, the presence of multiple income earners in the family raises both household income and saving level. However, households are still vulnerable to labour income risk. At the same time, it is observed that families without social security coverage save significantly less than the rest of society. Therefore, the empirical analysis underlines the need for an effective and efficient social security system in Turkey.
Generally, the labour market of Turkey is not considered as flexible. The situation of civil servants is often cited to support this argument. However, private sector employees experience certain difficulties. Private sector employees constitute 81.2% of wage and salary earners according to the TURKSTAT Household Budget Surveys. The majority of the labour force receives the minimum wage in the private sector and it is estimated that half of the labour force works in the unregistered economy. Thus, they are unable to benefit from provisions of the social security system such as health care and pensions. Moreover, union membership is limited among private sector employees. Furthermore, it is difficult to qualify for unemployment insurance and unemployment insurance payments are only half of the minimum wage.
Upon the analysis of the TURKSTAT Household Labour Surveys, Tansel (1992) observed that second job holding is quite common, especially among farmers in rural regions in Turkey. Metin-Özcan et al. (2003) discovered a positive relationship between private savings and inflation variability using macroeconomic figures for the Turkish economy and they considered that as empirical evidence in support of the precautionary saving hypothesis. Akbay et al. (2007) analysed household food consumption patterns using the TURKSTAT Household Budget Survey for 2003 and they concluded that regional and seasonal differences are as important as price and income dynamics in order to understand households’ food consumption patterns in Turkey. Moreover, Van Rijckeghem and Üçer (2009) showed that household saving is inversely related to health expenditures risk using the TURKSTAT Household Budget Surveys. However, to my knowledge, the impact of labour income risk on household saving decisions has not previously been investigated for the Turkish economy using micro-economic data, which is more appropriate for the analysis of household behaviour under risk and uncertainty.
The precautionary motive for saving may be even more important for households in developing countries. Lack of a satisfactory social security system, which will meet the needs of society, might raise the amount of precautionary saving. However, empirical research for developing countries is scarce mainly due to the lack of appropriate micro-economic data sets (Deaton 1997). The TURKSTAT Household Budget Surveys present a good opportunity to examine the empirical importance of precautionary saving in Turkey. The case of Turkey is remarkable since she possesses a sizeable and growing economy coupled with a young and increasing population. Therefore, the empirical analysis of the relationship between household saving decisions and labour income risk using micro-economic data from a developing country is the main contribution of this paper to the existing literature on household consumption and saving behaviour.
The outline of the paper is as follows: Section II presents a formal interpretation of the Permanent Income Theory and the precautionary saving hypothesis. Then, section III shows the approximation of labour income risk following a descriptive analysis of the TURKSTAT Household Budget Surveys. Section IV presents the econometric results. Finally, Section V concludes this empirical paper with a brief summary of empirical findings.
2 Theoretical background
2.1 A formal interpretation of the permanent income theory
The main idea behind the Permanent Income Theory is the fact that an individual’s lifetime consumption cannot be greater than his/her lifetime resources (Friedman 1957). It is assumed that there is a rational and risk-averse individual, who is a representative economic agent for the rest of society. The only source of utility is consumption and the individual aims to maximise utility from consumption with respect to the budget constraint, which represents the total lifetime resources of the individual. As a result, saving is defined simply as the difference between current income and consumption. It is assumed that consumption adopts a steady pattern throughout the individual’s life following permanent income, but saving is quite volatile in this period. In addition, it is expected that transitory income changes will be reflected in saving, which increases its volatility.
According to this interpretation of the Permanent Income Theory, the ultimate purpose of saving is future consumption. Campbell (1987) suggests that it is reasonable to evaluate this definition of saving as “saving for a rainy day”. An individual raises the amount of saving if his/her future income prospects are bleak or uncertain. This interpretation allows the establishment of a direct link between saving decisions in the current period and future income prospects. Thus, saving will serve as a good predictor of expected income changes. In this framework, following Deaton (1992), it is possible to define consumption as the present value (PV) of wealth and the expected lifetime income (Eq. 1):
In this equation, c t is the real consumption, y t is the real labour income, A t is the real value of financial assets, r is the real interest rate, which is constant, and finally Ω t is the information available to the individuals at time t, upon which their expectations are based.
The Eq. 2 is substituted into the Eq. 1 to express the “saving for a rainy day” concept formally. Saving at time t (s t ) is the present value (PV) of all future expected falls in income, as shown in Eq. 3 below. In this equation, the symbol ∆ indicates that real labour income (y t ) is first differenced with respect to the previous period.
At this point, it is important to highlight that the information at time t Ω t is only available to the individual. Therefore, the information matrix of the individual Ω t needs to be replaced by the information available to the researcher H t . The researcher has only limited information compared to the individual, H t ⊆ Ω t . Consequently, the Eq. 4 becomes a formal expression with observable variables, which is appropriate for an empirical analysis.
The main idea behind this interpretation of the Permanent Income Theory is that an individual will increase his/her saving in the current period if he/she anticipates that his/her future income will be lower than its lifetime average. This is certainly the case for many households in rural regions of developing countries. Since their agricultural revenues are dependent on favourable weather conditions, rural households are able to forecast their agricultural income level accurately by considering seasonal weather changes of previous periods (Deaton 1997). Hence, they adjust their saving level according to the available information.
2.2 The precautionary saving hypothesis under labour income risk
The precautionary saving hypothesis proposes that households are forced to postpone their consumption expenditures and raise their saving level to ensure their well-being under risk and uncertainty. The postponement of consumption expenditures and the rise in the amount of saving will allow households to accumulate financial assets. The main reason individuals opt for financial wealth is the fact that it can be used almost instantaneously in times of need due to its liquid nature. Hence, the existence of financial wealth guarantees the well-being of the family. In this respect, precautionary saving is defined as the amount of financial wealth that households keep to safeguard themselves against future labour income uncertainty.
There is a significant theoretical and empirical difference between the saving for a rainy day concept and the precautionary saving hypothesis. Saving is expected to be a good predictor of future income changes, i.e. an individual will increase his/her saving level in the current period, if he/she expects that income will fall in the next periods. However, the precautionary motive for saving will only emerge if there is uncertainty about future labour income prospects, such as a spell of unemployment.
Although the precautionary saving hypothesis is widely accepted from a theoretical point of view, previous empirical research indicates that the share of precautionary saving in household saving is small and quantitatively unimportant (Browning and Lusardi 1996; Attanasio and Weber 2010). One plausible explanation of this dilemma could be the failure to control for the self-selection of more risk-averse individuals to perform less risky occupations in the empirical analysis (Fuchs-Schündeln and Schündeln 2005). Another crucial aspect of the discussion on precautionary saving is that there are different types and sources of risk and uncertainty in the economy. Moreover, it is suggested that the complexity of the development of proxy measures for uncertainty contributes to the underestimation of the empirical importance of precautionary saving. Households are not only concerned with the possibility of losing their jobs, but they are also concerned about health care issues due to the size of out-of-pocket health expenditures. Therefore, it is essential to establish an alternative approach to understanding the empirical importance of precautionary saving. A feasible option is to analyse the impact of each type of income risk on household saving decisions separately.
An alternative formulation of household consumption and saving behaviour, which incorporates the precautionary saving hypothesis along with the Permanent Income Theory, can be presented formally as shown in Eq. 5. The virtue of this alternative formulation is that it is suitable for the empirical verification of the theory under risk and uncertainty. This alternative formulation of the saving function depends on Campbell’s (1987) “saving for a rainy day” interpretation of the Permanent Income Theory. For instance, Guariglia (2001) and Guariglia and Kim (2003, 2004) followed a similar approach to estimate the empirical importance of precautionary saving.
The dependent variable (S) of this equation is household saving. There are two essential explanatory variables on the right hand side of Eq. 5. The first variable is the estimation of the permanent component of household head’s income (Y P) and the second one is the approximation of the household head’s labour income risk (U). The social and demographic characteristics of households (Z) are incorporated into the econometric investigation process as control variables.Footnote 1
3 Empirical analysis
3.1 TURKSTAT Household Budget Surveys
I analyse seven consecutive waves of the TURKSTAT Household Budget Surveys from 2003 to 2009. The TURKSTAT Household Budget Surveys are actually repeated cross-sectional surveys, which do not have a panel dimension. However, the surveys provide information about family types, economic indicators, social and demographic characteristics both at individual and household level. Moreover, the surveys provide detailed data about the sub-items of household consumption expenditures based on the United Nations (UN) Classification of Individual Consumption According to Purpose (COICOP) as well as income distribution at household level distinguishing between rural regions and urban regions.Footnote 2 Among the sub-items of household consumption expenditures, there is the housing and rent category, which includes household expenditures on actual and imputed rentals, maintenance of the dwelling and utilities such as water, electricity and gas. However, this category does not measure house purchases. Therefore, it is not possible to approximate housing wealth as a flow variable for empirical analysis as in Fuchs-Schündeln and Schündeln (2005).
Traditionally, family is the most important aspect of social life, which makes it the focus of empirical research on household consumption and saving behaviour as well. Therefore, household saving is the dependent variable in the econometric investigation process. Two different definitions of household saving are analysed in this empirical paper.Footnote 3 The first definition of household saving (SAVI) is merely the difference between household disposable income and consumption expenditures. The second definition of household saving (SAVII) is the difference between household disposable income and consumption expenditures; however, in this case, household saving includes expenditures on durable goods from consumption since durable goods are generally considered as part of household saving in the economics literature.Footnote 4 It is calculated that 56.9% of total households have positive savings with respect to the first definition of household saving (SAVI) in the pooled sample, but this ratio rises to 62.8%, when the second definition of household saving (SAVII) is analysed (Table 1). Moreover, the household saving ratio increases from 8.7 to 17% for the pooled sample, if household expenditures on durable goods are included in household saving rather than in consumption.Footnote 5
According to the classification of the TURKSTAT Household Budget Surveys, a family member who plays a greater role than the rest of the members in at least one important issue is selected as the household head. Bringing income into the family is not the main criteria in the selection of the household head. The household head may be male or female though over 90% of them are actually male. He/she does not have to be the highest income earner in the family, but he/she is responsible for managing household income and consumption expenditures. Household head characteristics have a strong influence over household saving preferences. The level of household saving increases with the education level of the household head. Household saving level is higher, if the household head has compulsory health insurance coverage. More importantly, there is a positive relationship between the age of the household head and the level of household saving (Table 1).Footnote 6
3.2 The approximation of labour income risk
The main focus of the econometric investigation process is labour income risk, which is associated with future labour income uncertainty in the economy. The development of a proxy variable for uncertainty, which separates anticipated income changes from unexpected negative income shocks such as a spell of unemployment, is essential for the analysis of the precautionary saving hypothesis.Footnote 7 A suitable proxy variable for uncertainty to capture its implications for household saving decisions can be based on the prediction of unemployment risk. Labour income risk is generated by interacting unemployment risk of the individual with the variance of income following Lusardi (1998), Guariglia (2001) and Guariglia and Kim (2004), as shown in Eq. 6.
The approximation of labour income risk, which is based on the probability of being unemployed in the next period, is more appropriate to reveal the empirical importance of precautionary saving, since unemployment risk is a more relevant concern for working-class individuals rather than the volatility of income.Footnote 8 Thus, the variance of individual disposable income is interacted with unemployment risk to create the labour income risk variable. The individual has zero income with the probability (p) and with the probability (1 − p) the individual gains his/her disposable income (I). The sum of the two possibilities will be the individual’s expected income in the next period. The subscript (i) indicates that the model is estimated using individual observations.Footnote 9
Unemployment is defined as the situation, when an individual is not working, but actively seeking a job during the survey month. The probability of being unemployed is predicted from a probit model. The dummy variable for being unemployed, which takes the value of one if the individual is unemployed and zero otherwise, is regressed on gender, age-group, age-group squared, health insurance coverage and education level. It is thought that health insurance coverage is an important factor affecting individuals’ labour force participation choices, since most individuals obtain health insurance coverage as well as social security coverage through their employment contracts. Moreover, time dummy variables for survey years and a dummy variable for the rural regions are also included in the probit model. There are 112,205 individuals participating in the labour market and 9,666 of them are unemployed individuals, constituting 8.61% of the labour force in the pooled sample, which is consistent with national unemployment rates.
The sample set is restricted to household heads, who are of working age—between 15 and 64—and who participate in the labour market voluntarily. Unpaid family workers are excluded from the sample, since they pre-dominantly work in family farms in agricultural production in the rural regions. Thus, these individuals have significant differences in their labour market choices compared to the working-class individuals. Moreover, retired individuals are excluded from the sample set, since their perception of unemployment risk and income loss would be significantly different than young and active individuals in the labour market. As a result, the probit model is estimated for 47,884 household heads, who satisfy these criteria from the pooled sample of the TURKSTAT Household Budget Surveys between 2003 and 2009 (Table 2).Footnote 10
Moreover, it is possible that individuals pool the risk of being unemployed and losing their labour income by living with their family in the same way as they share their income and consumption in the family. It is necessary to take this issue into consideration in the approximation of labour income risk. The probit model for the probability of being unemployed is estimated for household heads using dummy variables at individual level. The robust standard errors, which are estimated using cluster option for household units, are reported in Table 2 for the probit model.
Unemployment risk is higher for young household heads, but it decreases with age as expected and it is observed that women suffer from a higher level of unemployment risk compared to men. Household heads, who have compulsory health insurance coverage, have a lower unemployment risk. It is thought that an employment opportunity in the registered economy provides both job security and compulsory health insurance coverage. Moreover, university graduates have a lower unemployment risk compared to other education categories except for vocational school (Table 2).Footnote 11 The probability of being unemployed in the next period is predicted from the probit model and utilised in the approximation of the first labour income risk variable (LIRI). The predicted probability of being unemployed in the next period is interacted with the square of the household head’s disposable income, as shown in Eq. 6.
Job loss is defined as the situation, when the individual was employed in the previous year, but he/she is not working and actively searching for a job in the survey month. In other words, the individual lost his/her job recently. It is observed that there are 2,806 individuals, who lost their jobs in the survey year and constitute 2.5% of the labour force in the pooled sample. The probability of an individual losing his/her job is estimated using a pooled probit model for 47,884 household heads, who are between the ages of 15 and 64. Unpaid family workers and retired individuals are excluded from the sample as before. The econometric results from the probit model for job loss risk are similar to the econometric results from the probit model for unemployment risk (Table 3).Footnote 12 The predicted probability of job loss from the probit model is interacted with the square of the household head’s disposable income to generate the second labour income risk (LIRII), as shown in Eq. 6.
4 Econometric results
Households may underreport their disposable income for various reasons, which is observed in many developing country examples, while household consumption expenditures are collected in a detailed manner. Thus, household saving is underestimated, since it is calculated as the residual between household disposable income and consumption expenditures (Deaton 1997). Van Rijckeghem and Üçer (2009) argue that household saving levels are questionably negative for several observations. In addition to that, Yükseler and Türkan (2008) claim that the TURKSTAT Household Budget Surveys fail to observe individual interest and rent income accurately. For this reason, the lowest and highest 1% quintiles of household saving variables (SAVI and SAVII) are trimmed out to remove potential outliers from the sample. Household units that are composed of individuals, who live together and families, whose household head is an unpaid family worker are excluded from the sample. Finally, the econometric regressions are performed for families, whose household head’s age is between 20 and 64.
Permanent income is estimated for household heads, who have a positive income level and who are between 20 and 64 years old, using a Heckman two-step selection model (Heckman 1979).Footnote 13 Household head’s disposable income is regressed on age-group, age-group squared and the dummy variables for gender, education, employment sector, job-status, occupation, social security coverage and employment sector (King and Dicks-Mireaux 1982; Kazarosian 1997). The fitted values from the Heckman two-step selection model are saved and used as the permanent component of current income (Guariglia 2001; Guariglia and Kim 2004).
Both definitions of household saving (SAVI and SAVII) are regressed on the labour income risk variables (LIRI and LIRII) using pooled OLS regressions and the econometric results are presented in Tables 4 and 5. The econometric results of the pooled OLS regressions are similar for both definitions of household saving. The pooled OLS regressions on household saving, which includes expenditures on durable goods (SAVII), have a higher explanatory power compared to the first definition of household saving (SAVI). It is observed that the regression coefficient of permanent income is positive and statistically significant at 1% significance level. The econometric results indicate that families, whose household head does not have social security coverage, have a lower saving level than the rest of the society contrary to initial expectations. It is thought that the lack of social security coverage reduces household saving, especially among lower income groups. Moreover, the regression coefficients of the labour income risk variables (LIRI and LIRII) have the expected positive signs and they are also statistically significant at 1% significance level (Tables 4, 5). Therefore, the econometric results are in favour of the precautionary saving hypothesis.Footnote 14
According to the econometric results, when the elasticity of uncertainty is calculated at the sample means, ceteris paribus, it is observed that a 10% rise in the first approximation of labour income risk (LIRI) leads to an increase of 3.3% in the first definition of household saving (SAVI) and an increase of 2.2% in the second definition of household saving (SAVII). In addition to that a 10% rise in the second approximation of labour income risk (LIRII) leads to an increase of 3.5% in the first definition of household saving (SAVI) and an increase of 2.3% in the second definition of household saving (SAVII). It is possible to interpret the increase in household saving due to a rise in labour income risk as precautionary saving.Footnote 15 Moreover, these percentages indicate that households postpone their consumption and increase their saving amount against labour income risk considerably, since a 10% increase in the labour income risk is actually a modest rise.Footnote 16 Thus, the econometric results are similar to the findings of Guariglia (2001) and Guariglia and Kim (2003, 2004).Footnote 17
Household expenditures on durable goods are negatively affected by labour income risk as stated by Carroll et al. (2003) and Benito (2006). The estimated increases in the second definition of household saving (SAVII) due to a 10% rise in the labour income risk variables (LIRI and LIRII) are lower than the increases in the first definition of household saving (SAVI) in the pooled OLS regressions, contrary to the findings of Carroll et al. (2003). At the same time, households can also be discouraged from making house purchases due to labour income risk in Turkey.Footnote 18 House purchases require a long term commitment, which households may refrain from making under labour income risk. The presence of labour income risk might also act as an artificial borrowing constraint, which prohibits households from obtaining credit from financial institutions for property purchases as pointed out by Zeldes (1989) previously.
Moreover, it is thought that households implement alternative strategies to protect themselves against labour income risk, if it is not possible to accumulate savings for precautionary purposes. They might try to increase their income sources, which would diminish the negative effects of the household head’s labour income risk. Tansel (1992) and Kochar (1999) demonstrated that farmers shift their labour from agriculture to the related sectors of the economy to smooth out their income streams in order to smooth out their consumption expenditures. Moreover, Guariglia and Kim (2004) observed that households use moonlighting as a self-insurance mechanism against future labour income uncertainty in Russia. Nevertheless, it is observed that the dummy variable for the additional employment of the household head is not statistically significant in the pooled OLS regressions.Footnote 19 The dummy variable for the presence of multiple income earners in the household is statistically significant in all of the pooled OLS regressions. Moreover, the interaction terms of the dummy variables and the labour income risk variables (LIRI and LIRII) are not statistically significant in the pooled OLS regressions (Tables 4, 5).Footnote 20
The presence of multiple income earners in the family makes a positive contribution to household saving, even if it does not reduce the statistical significance and the size of the household head’s labour income risk. There is a significant difference between the predicted probability of being unemployed and job loss in the next period of household heads and family members. Moreover, disposable income levels of household heads are significantly higher than those of the rest of family members. Therefore, the contribution of family members to household income and saving level is not strong enough to diminish household heads’ labour income risk and to reduce the need for precautionary saving.Footnote 21
Household size continuously decreases both in urban and rural regions over the years in Turkey. Despite this fact, traditional and extended families constitute 18% all households even in 2009. The labour force participation rate of children, who are at working age, is actually higher than spouses. Household heads, who have a second job constitute approximately 8% of total household heads, while in 40% of families there is more than one income earner in the pooled sample. It is thought that young family members work and support family income provided that they can find a job. Thus, it will be a positive contribution to implement a comprehensive public policy to decrease the high unemployment ratio among young university graduates in Turkey.
Unemployment insurance usage is not widespread, since qualifying conditions are very strict.Footnote 22 Only individuals who worked in the registered economy and paid their social security premiums regularly for a sufficient period of time can apply for unemployment benefits. In addition to that, such workers have a lower level of unemployment risk compared to individuals who work in the unregistered economy without formal employment contracts. As a result, the percentage of unemployed individuals who benefit from unemployment insurance is quite low. Therefore, households remain vulnerable to labour income risk, which underlines the need for an effective and efficient social security system.
5 Conclusion
This paper analyses the impact of labour income risk on household saving decisions in Turkey using the TURKSTAT Household Budget Surveys from 2003 to 2009. The most difficult aspect of the empirical analysis is the approximation of labour income risk. Individual disposable income is interacted alternately with the probability of being unemployed and the probability of job loss risk in the next period to generate two separate proxy variables for labour income risk. The econometric results show that there is a positive and significant relationship between household saving and permanent income. Moreover, labour income risk emerges as one of the main determinants of household saving decisions. The estimated increases in household saving level due to labour income risk are sizeable in comparison to current trends in household saving ratios. Thus, the empirical analysis supports the precautionary saving hypothesis.
Moreover, households implement alternative strategies such as holding a second job and raising the number of income earners in the family. Households try to smooth out their income streams to smooth out their consumption patterns, if they are unable raise their saving level further for precautionary purposes. Household head’s additional employment and the presence of multiple income earners in the family contribute to household income and saving level. However, households still remain vulnerable to labour income risk, which underlines the need for a satisfactory social security system. Finally, the positive relationship between permanent income and household saving indicates that a growth policy, which stimulates investment and employment, will contribute to the rise of household saving level in Turkey.
Notes
The introduction of permanent income and the social and demographic characteristics into the econometric investigation process aims to capture life-cycle effects such as saving for the retirement period. However, the precautionary motive for saving is independent of life-cycle motives and it emerges only if there is uncertainty about future income prospects. Therefore, the approximation of the household head’s labour income risk enters the household saving equation as an independent variable.
A settlement unit like a village or town is defined as an urban region, if the total population of the place is equal to or >20,000. If the population is <20,000, it is considered to be a rural region. However, this definition of a rural region does not take into account economic sectors such as the role of the agricultural sector or tourism revenues in the local economy. Therefore, social and economic characteristics of rural regions might differ significantly between the west and east of the country.
Consumption expenditures are available only at household level and as monthly figures in the TURKSTAT Household Budget Surveys. Monthly consumption expenditures are multiplied by twelve to reach an annual estimate of household consumption expenditures with the assumption that household consumption follows a steady pattern throughout the year. Moreover, in the surveys individual disposable income is available both monthly and annually, but household disposable income is available only annually. Thus, household saving is calculated as the difference between annual household disposable income and annualised household consumption expenditures.
Durable goods, which are considered as part of household saving, are home appliances, medical equipment, consumer electronics, new and second-hand automobile purchases and jewellery and watches for personal consumption. The second definition of household saving (SAVII) includes durable goods, but it does not contain housing wealth.
TURKSTAT collects individual and household disposable income figures for the twelve-month period prior to the survey month, but not for the calendar year due to the design of the survey questionnaires. For instance, if a household participates in the Household Budget Survey in September 2008, then the annual household disposable income will refer to the twelve-month period between September 2007 and September 2008. However, monthly inflation rates are quite high and there are significant differences in the inflation rates of geographical regions in Turkey. TURKSTAT has included a regional and monthly inflation variable in the Household Budget Surveys since 2003. Household disposable income and household consumption are inflated to the year-end (December) prices of the corresponding survey year by multiplying with this inflation index. Annual household disposable income and household consumption expenditures are divided by year-end consumer price indices for each survey year and all economic variables including household saving figures are analysed in 2003 TL prices.
The TURKSTAT Household Budget Surveys do not provide information about individuals’ ages. Instead, the surveys specify the age intervals of the individuals. The empirical analysis can only be realised with respect to individuals’ age groups, which increase in 5 year intervals. For that reason, it is not possible to create a pseudo-panel data set using birth cohorts, since it is not possible to determine the individuals’ birth years exactly. Remaining age groups, which are not shown in Table 1, are children between 0 and 5, 6 and 14, 15 and 19 besides individuals, who are 65 or older.
According to Browning and Lusardi (1996), a potential uncertainty measure must be an observable variable, but an exogenous one to the individual’s decisions and behaviour. Finally, a potential uncertainty measure must be variable across the population to account for the heterogeneity in society.
Carroll et al. (2003, p. 586) points out that “… a tenured college professor who, by choice, teaches or consults every other summer may have more variable annual income than a factory worker, but does not face the uncertainty of being laid off during a recession.”
The common approach in the previous literature is to approximate labour income risk by using the individuals’ subjective evaluation of unemployment risk. However, the TURKSTAT Household Budget Surveys do not have a question about the individuals’ subjective evaluation of the probability of being unemployed in the next period, which restricts the scope of the empirical analysis. At the same time, it is observed that the objective probability of being unemployed in the next period is also used to generate labour income risk. For instance, Guariglia and Kim (2004) used the objective probability of unemployment risk to develop a second approximation of labour income risk to check the robustness of their econometric results.
A shortcoming of the TURKSTAT Household Budget Surveys is that no information is provided about the income and employment prospects of the individual, if he/she is already unemployed. Hence, it is possible to observe the employment sector and job status of the individual, only if he/she is currently employed. Moreover, it is not possible to learn whether an individual has social security coverage or not, if he/she is unemployed.
Time dummy variables for survey years are included in the probit model and the omitted year is 2003, since it represents a stable period of the Turkish economy. Only the regression coefficients of the dummy variables for 2008 and 2009 are positive and statistically significant in the probit model, which indicates that household heads’ unemployment risk started to increase in this period compared to 2003 due to the contraction of the economy.
Time dummy variables for survey years are also included in the probit model and the omitted year is 2003 as before, since it is a stable period of the Turkish economy. In this probit model, the regression coefficients of the dummy variables for 2007, 2008 and 2009 are positive and statistically significant, which indicates that household heads’ job loss risk started to increase in this period compared to 2003 due to the contraction of the economy.
The estimation of the permanent component of household heads’ disposable income is discussed in detail in the former working paper version of the paper (Ceritoğlu 2011).
Labour income risk and permanent income are generated variables, which are predicted using auxiliary regressions at the previous stages of the empirical analysis. Therefore, the standard errors of the pooled OLS regressions are estimated using the bootstrap method with 1,000 replications.
I restricted the sample set for household heads, who are 54 and younger and then performed the pooled OLS regressions. The estimated increases in household saving due to labour income risk are higher for the restricted sample compared to the original sample. If the elasticity of uncertainty is calculated at the sample means, it is observed that a 10% rise in the first approximation of labour income risk (LIRI) leads to an increase of 3.8% in the first definition of household saving (SAVI) and an increase of 2.5% in the second definition of household saving (SAVII). In addition to that a 10% rise in the second approximation of labour income risk (LIRII) leads to an increase of 4.2% in the first definition of household saving (SAVI) and an increase of 2.7% in the second definition of household saving (SAVII). Consequently, the precautionary motive for saving is stronger for younger households as expected.
I re-estimated the pooled OLS regressions including interaction terms between the time dummy variables for survey years and the labour income risk variables. The time dummy variables and interaction terms with the first labour income risk (LIRI) are statistically significant in the estimated regression for the first definition of household saving (SAVI). I carried out a joint significance test for the interaction terms. The calculated F-value is 18.80 and the probability value is 0.000. Thus, the null hypothesis that the parameters are jointly equal to zero is rejected. The time dummy variables and interaction terms with the second labour income risk (LIRII) are also statistically significant in the estimated regression for the first definition of household saving (SAVI). I carried out a joint significance test for the interaction terms. The calculated F-value is 19.79 and the probability value is 0.000. Thus, the null hypothesis that the parameters are jointly equal to zero is rejected once again.
I restricted the sample set to households only from urban regions of the country to check the robustness of the econometric results. The econometric results for the restricted sample are very similar to the results presented in the paper. Moreover, I observed that the calculated elasticity of uncertainty and its effect on household saving decisions remains the same, even when the sample set is restricted to households from urban regions of the country. The main difference between urban and rural regions of the country is the situation of unpaid family workers, mostly employed in the agriculture sector. However, these observations have already been removed from the sample set.
Housing investment is not included in the second definition of household saving (SAVII).
It is statistically significant only when the second definition of household saving (SAVII) is regressed on the first definition of the first labour income risk (LIRI) without the interaction terms, which is presented in the fifth column of Table 4.
I introduced the interaction terms for the labour income risk variables alternately as in Guariglia and Kim (2004), but the interaction terms were not statistically significant in the pooled OLS regressions as before.
I also performed joint significance tests for the interaction terms that are included in the pooled OLS regressions. The calculated F-value is 1.26 and the probability value is 0.28 for the OLS regression, which is presented in the third column of Table 4. Thus, the null hypothesis that the parameters are jointly equal to zero cannot be rejected. The calculated F-value is 0.36 and the probability value is 0.70 for the OLS regression, which is presented in the sixth column of Table 4. Thus, the null hypothesis that the parameters are jointly equal to zero cannot be rejected once again. Moreover, the calculated F-value is 0.28 and the probability value is 0.76 for the OLS regression, which is presented in the third column of Table 5. Thus, the null hypothesis that the parameters are jointly equal to zero cannot be rejected. The calculated F-value is 0.15 and the probability value is 0.86 for the OLS regression, which is presented in the sixth column of Table 5. Thus, the null hypothesis that the parameters are jointly equal to zero cannot be rejected once again.
The total number of individuals that benefit from unemployment insurance is only 362, while there are 9,666 unemployed people out of 112,205 individuals in the labour force according to the TURKSTAT Household Budget Surveys.
References
Akbay, C., Boz, İ., & Chern, W. S. (2007). Household food consumption in Turkey. European Review of Agricultural Economics, 34(2), 209–231.
Attanasio, O. P., & Weber, G. (2010). Consumption and saving: Models of inter-temporal allocation and their implications for public policy. Journal of Economic Literature, 48(3), 693–751.
Benito, A. (2006). Does job insecurity affect household consumption? Oxford Economic Papers, 58(1), 157–181.
Browning, M., & Lusardi, A. (1996). Household saving: Micro theories and micro facts. Journal of Economic Literature, 34, 1797–1855.
Campbell, J. Y. (1987). Does saving anticipate declining labour income? An alternative test of the permanent income hypothesis. Econometrica, 55(6), 1249–1273.
Carroll, C. D., Dynan, K. E., & Krane, S. D. (2003). Unemployment risk and precautionary wealth: Evidence from households balance sheets. The Review of Economics and Statistics, 85(3), 586–604.
Carroll, C. D., & Samwick, A. A. (1997). The nature of precautionary wealth. Journal of Monetary Economics, 40, 41–71.
Carroll, C. D., & Samwick, A. A. (1998). How important is precautionary saving. The Review of Economics and Statistics, 80, 410–418.
Ceritoğlu, E. (2011). The impact of labour income risk on household saving decisions in Turkey. The Central Bank of the Republic of Turkey (CBRT), working paper no 11/09.
Chamon, M. D., & Prasad, E. S. (2010). Why are saving rates of urban households in China rising? American Economic Journal: Macroeconomics, 2(1), 93–130.
Deaton, A. (1992). Understanding consumption. Oxford: Oxford University Press.
Deaton, A. (1997). The analysis of household surveys: A micro-econometric approach to development policy. Baltimore: John Hopkins University Press for the World Bank.
Friedman, M. (1957). A theory of the consumption function. Princeton: Princeton University Press.
Fuchs-Schündeln, N., & Schündeln, M. (2005). Precautionary savings and self-selection: Evidence from the German reunification experiment. The Quarterly Journal of Economics, 120(3), 1085–1120.
Guariglia, A. (2001). Saving behaviour and earnings uncertainty: Evidence from the British Household Panel Survey. Journal of Population Economics, 14, 619–634.
Guariglia, A., & Kim, B. -Y. (2003). Wage arrears uncertainty and precautionary savings in Russia. Economics of Transition, 11(3), 493–512.
Guariglia, A., & Kim, B. -Y. (2004). Earnings uncertainty, precautionary saving and moonlighting in Russia. Journal of Population Economics, 17(2), 289–310.
Guariglia, A., & Rossi, M. (2002). Consumption, habit formation and precautionary saving: Evidence from the British Household Panel Survey. Oxford Economic Papers, 54(1), 1–19.
Heckman, J. J. (1979). Sample selection bias as a specification error. Econometrica, 47(1), 153–161.
Kazarosian, M. (1997). Precautionary savings—A panel study. The Review of Economics and Statistics, 79(2), 241–247.
King, M. A., & Dicks-Mireaux, L. (1982). Asset holdings and the life-cycle. Economic Journal, 92(366), 247–267.
Kochar, A. (1999). Smoothing consumption by smoothing income: Hours-of-work responses to idiosyncratic agricultural shocks in rural India. The Review of Economics and Statistics, 81(1), 50–61.
Lusardi, A. (1997). Precautionary saving and subjective earnings variance. Economics Letters, 57, 319–326.
Lusardi, A. (1998). On the importance of precautionary saving motive. American Economic Review, 88(2), 449–453.
Metin-Özcan, K., Günay, A., & Ortaç, S. (2003). Determinants of private savings behaviour in Turkey. Applied Economics, 35, 1405–1416.
Tansel, A. (1992). Household saving, income and demographic interactions. METU Studies in Development, 19(1), 91–114.
Turkish Statistical Institute (TURKSTAT). Household Budget Surveys, 2003–2009.
Van Rijckeghem, C., & Üçer, M. (2009). The evolution and determinants of the Turkish private saving rate: What lessons for policy? TÜSİAD—Koç University Economic Research Forum, Research paper, TÜSİAD-T/2008-12/478.
Yükseler, Z., & Türkan, E. (2008). An assessment of households in Turkey: Labour, income, expenditure and poverty (in Turkish). CBRT—TÜSİAD—Koç University Economic Research Forum Joint Research Project, TÜSİAD-T/2008-03/455.
Zeldes, S. P. (1989). Optimal consumption with stochastic income: Deviations from certainty equivalence. The Quarterly Journal of Economics, 104, 275–298.
Acknowledgments
I would like to thank Richard Cornes, Alessandra Guariglia, Sarah Brown, Sarah Bridges, Murat G. Kırdar and Ercan Uygur for their valuable comments and suggestions.
Author information
Authors and Affiliations
Corresponding author
Rights and permissions
About this article
Cite this article
Ceritoğlu, E. The impact of labour income risk on household saving decisions in Turkey. Rev Econ Household 11, 109–129 (2013). https://doi.org/10.1007/s11150-011-9137-2
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s11150-011-9137-2