The effect of self-control problems on different types of savings:
Evidence from Dutch households
Mareille Meijering
1Master Thesis, MSc Economics & MSc Finance
Thesis supervisor: C. Laureti
June 2018, University of Groningen
Keywords: Household savings, self-control, income, illiquidity JEL classification: D14, D15, G41
1 S2525038; m.meijering.1@student.rug.nl
2
1. INTRODUCTION
Economic theory usually assumes that people behave rationally. According to the life-cycle model, rational consumers maximize their expected life-time utility given their life-time income. To optimize life-time utility, individuals have to smooth consumption over their life time. In order for this to be possible, people have to save during periods of high income (working life) to be able to cope with unexpected negative income shocks or anticipated shocks such as retirement. In addition, saving is also necessary from a macroeconomic point of view, considering that without savings, investment will not be possible and therefore no capital accumulation can occur. It turns out, however, that on average people tend to save too little. One reason for this is that the determination of the ‘optimal’ savings rate is a difficult task (Thaler and Benartzi, 2004). Determining the optimal savings rate is already a hard task for economists, making it nearly impossible for individuals who have only little or zero understanding about the economics behind their savings decision and the life-cycle hypothesis. Another reason for low savings rates is that people simply do not have the self-control to lower current consumption in order to save for future consumption (Thaler and Shefrin, 1981).
3
This thesis complements the existing literature by using real-life survey data. Most papers, including Carvalho et al. (2016), use experiments in order to test the relationships between self-control and savings, and self-control and income. These outcomes may depend on the set-up and environment of the experiments, using survey data gets around this problem. Furthermore, the extensive dataset of the DNB Household Survey that I will use, is never before used to explore the relationship between self-control and savings.
The DNB Household Survey contains data of over 2000 households that participated in the CentERpanel. The survey collects the data by means of annual questionnaires, which date back to 1993. For this thesis, the 25th wave is used which consists of data that is obtained between April 2017 and October 2017. The questionnaires were filled out by all individuals above the age of 16. The data includes information on: income, savings, risk and time preferences, demographic factors (such as age and gender) and behavioral aspects of the participants. The different types of savings that will be used are liquid savings, illiquid savings, investment savings (which include both liquid and illiquid shares, stocks and bonds). OLS regressions are used to analyze these types of savings and the effect of the level of self-control of the participants on these levels of savings. A set of self-control variables is added to the model which includes individual characteristics as well as risk and time preferences and some other behavioral aspects, such as mental accounting and precautionary saving. Furthermore, to determine the relationship between income and the ability to exercise self-control, a regression is carried out where a measure of self-control depends on income and a vector of control variables.
4
This thesis continues as follows: section 2 reviews the existing literature. Then, section 3 explains the data, continued by section 4 which describes the methodology. In section 5 the results are discussed, followed by the conclusions in section 6.
2. LITERATURE
This section reviews the related literature. It starts with the neoclassical view on savings with the life-cycle hypothesis and the permanent income hypothesis. Then, it continues with literature on behavioral biases and commitment devices, and the relation between income and these behavioral biases. Lastly, related literature that also uses the DNB Household Survey is reviewed.
2.1 A neoclassical view on savings
Decades ago the foundations for the life cycle hypothesis were already laid down with Modigliani and Blumberg’s (1954) life-cycle theory of saving and the permanent income hypothesis of Friedman (1957). Traditionally, in the life-cycle hypothesis individuals or households would maximize their expected life-time utility restricted by their life-time income. People are assumed to prefer smoothing their consumption over their lives. At younger ages, individuals will have to borrow money in order to be able to consume at the level they prefer. When these individuals start working, a period of saving starts. This period of saving accommodates the future drop in income at retirement. When reaching the retirement age, the level of income drops and people will start to dissave again and consume their previously build up savings. Figure 1 gives a schematic approach to the basic life-cycle hypothesis.
5
Figure 1. Schematic approach basic life-cycle hypothesis
A second reason for saving, aside from consumption smoothing, is the precautionary savings motive. Individuals save when their future income is uncertain, they accumulate wealth to be able to cope with negative income shocks in the future (Mazzocco, 2004). Savings should therefore increase with the level of absolute risk aversion of the individuals. Numerous articles have been published about risk aversion which show that more than half of the individuals exhibit risk aversion (see for example Eckel and Grossman, 2002; Holt and Laury, 2002) and should therefore according to Mazzocco (2004) have increasing savings with regards to their level of risk aversion. Besides being, on average, risk averse, people also tend to be impatient. People prefer having the benefits right now and delaying the costs to the future. It is usually assumed that people discount the future exponentially. That is, people who are more impatient will have higher discount rates and will attach more weight to the present compared to the future (O’Donoghue and Rabin, 1999). This has an effect on the amount of savings these individuals have. People who attach more weight to the future, are more likely to save sufficient amounts to be able to consume the preferred amounts in the future. Therefore, these individual time preferences have an influence on the amount of savings people have. Lawrance (1991) found that time preferences differ between income classes, low-income individuals have on average higher discount rates, making them more impatient.
6
of savings without being aware of it. Clark et al. (2006) found that participants in a seminar about financial education changed their goals regarding their retirement savings. Moreover, they found that women were more likely to alter their retirement goals after the seminar than men.
Naturally, there is quite some diversity among people when it comes to their income levels. Some people are better able to save part of their income than others. However, everyone is used to a standard of living that matches his/her income level. For retirement, roughly 70% of this income is necessary to maintain the same standard of living (Ackert and Deaves, 2016). This implies that in absolute terms, rich individuals should save more than poor individuals to maintain the same standard of living when reaching retirement.
The abovementioned variables are investigated using OLS regressions, they are mostly used as control variables. I will check how these time and risk preferences, age, education levels, income levels, and precautionary savings motives influence the savings levels.
2.2 Self-control and commitment devices
Another reason that is put forward to explain why people tend to save too little is that they simply do not have the self-control to lower present consumption to be able to consume more in the future (Thaler and Shefrin, 1981). To make the life-cycle hypothesis more behaviorally realistic, Thaler and Shefrin (1988) add behavioral characteristics related to self-control, mental accounting and framing to arrive at the ‘Behavioral Life-Cycle Hypothesis’. An important characteristic of this model is the distinction between the ‘planner’ and the ‘doer’ within each individual. The planner has to impose self-control rules for the short-sighted doer. This can for example be done with commitment devices such as mortgages and illiquid assets that make it costly to reverse the decision to invest in it or withdraw money. Such devices are therefore helpful to force people to save more.
7
functions is wrong in many cases. A variety of studies has shown that people exhibit time-inconsistent preferences (for example: Laibson, 1997, O’Donoghue and Rabin, 1999). In fact, these people have hyperbolic discount functions rather than exponential ones. This hyperbolic discounting is characterized by high discount rates in the short run compared to lower discount rates for longer time spans. More weight is attached to the present-self. People who exhibit present-biased preferences know that they have to save more, yet, prefer to do it tomorrow instead of today. They lack self-control to decrease current consumption and start saving.
Repaying a mortgage, defined-benefit pensions and social security all do not require a lot of willpower. And as it turns out, people who make use of these devices are doing a rather decent job in saving an adequate amount for retirement (Thaler and Benartzi, 2004). These channels can be regarded as commitment devices. Sophisticated agents, people who are aware of their self-control problems, prefer commitment devices since this restricts their future choices and forces them to start saving (Beshears et al., 2015). Another sort of commitment device can be the co-holding of high cost credit card debts and low yield savings accounts. This ‘co-holding puzzle’ is analyzed by Gathergood and Weber (2014). They use UK survey data to investigate the existence of credit card debts and liquid savings and find that co-holding is related to impulsive spending behavior. People therefore make use of it as a tool to manage their self-control problems. Moreover, it is found that co-holders of credit card debt and liquid savings were mostly higher (financially) educated people with above average levels of income. Hence, different people may decide to save through other savings channels. Therefore, I expect to find a negative relationship between savings and self-control problems. Moreover, I expect to find that people with higher self-control problems will not have significantly lower illiquid savings. These expectations are based on the abundant literature on the subject. However, these studies are mostly based upon theoretical or econometric frameworks, or experimental data. On the contrary, I will use real-life survey data to complement the existing literature. The advantage of survey data is that the decisions of the participants are not influenced by the set-up of the experiment.
2.3 Income and self-control
8
restricted) and an after payday group (not financially restricted) and found that the before payday group acted more present-biased. This before payday group could be seen as a group of ‘poor’ individuals, implying that poor individuals are behaving more present-biased than the non-poor. Related to these self-control problems are temptation goods. Temptation goods provide instantaneous utility for the individual that consumes them right away, but does not provide utility when an individual foresees that the good will be consumed in the future. If you are more prone to temptations, this can be interpreted as an equivalent to low self-control. Banerjee and Mullainathan (2010) showed in their model that there is a declining impact of temptations when income levels increase. This would imply that temptations have a larger impact on poor individuals, which could result in being less able to save enough.
2.4 DNB Household Survey
Numerous studies use the DNB Household survey. Since the survey is very extensive, the subjects vary greatly. A health-related paper is written about the relationship between health and financial strain (Prentice et al., 2017). Whereas Bucciol and Miniaci (2018) investigate if individuals’ risk propensities change over time, which is a completely different subject. Furthermore, abundant literature examines retirement planning using the DNB Household Survey (for example: van Rooij et al., 2012, Alessie et al., 2006). Even though the literature using this dataset is extensive, the relationship between self-control problems and savings is never examined before. Where Nyhus and Webley (2001) and Nyhus (2002) use the survey data to investigate the psychological determinants of savings, they merely focus on personality traits, leaving the influence of self-control problems aside. This thesis therefore contributes to the existing literature by exploring a different aspect of this rich dataset. The next section describes the dataset.
3. DATA
9
age, gender, risk and time preferences, education, income levels, savings levels and behavioral aspects. Since behavioral aspects are quite stable over time (see for example Eigsti et al., 2006), running a panel data model is not necessary and data of one year is therefore sufficient for the analysis on the relationship between self-control and savings and the relationship between self-control and income.
3.1 Savings and income levels
To investigate the effect of self-control on different types of savings, the savings are categorized into three groups, namely; liquid savings, illiquid savings and investment savings. Liquid savings include the balances of checking accounts, deposit/savings accounts, deposit books and money lent to family. Illiquid savings include savings certificates, the amounts of annuity/pension insurance and the total amount of endowment insurance. For the investment savings, the total amounts invested in mutual funds, shares/stocks, bonds and/or mortgage bonds are taken together. Net income of the participants is calculated by adding all sources of income and wage-replacing transfers, then mortgage interest payments and taxes are subtracted. This calculation is already done by the DNB Household Survey.
3.2 Individual characteristics
The dataset consists of information on the individual characteristics of the participants, which includes the year of birth of all the participants. The age is calculated, using the year of birth, at the time the participants answer the questionnaire, thus, in 2017. Moreover, the gender is reported as one for males and two for females. This is coded into a dummy variable where one means that the participant is male and zero means that the participant is female. Furthermore, there are questions regarding the number of kids that are present in the household of the participant, whether there is a partner present in the household, and whether the participant does the financial administration in the household. Lastly, there is a variable on whether the participant smokes, which is coded into a dummy where one means that the participants smokes and zero means that the participant does not smoke.
10
special education, pre-vocational education or other sorts of education. Participants who completed pre-university education or senior vocational training are placed into the category medium education level. Lastly, high education includes participants who completed university education or vocational colleges.
The questionnaire includes six statements related to risk preferences where the subjects indicate to what extent they agree with the statements on a scale of 1 to 7. Some statements are asked in the light of risk loving behavior while others focus on risk aversion. According to Coppola (2014), self-assessment questions on risk preferences seem to have high predictive power on the actual risk-taking behavior of the subjects. Therefore, these self-assessment questions on the willingness to take risk from the CentERpanel can be used as a measure of risk preferences of the subjects. Since the questions are asked differently, all answers are coded as such that 1 means that people are risk averse and 7 being risk loving. The average reply is taken out of these six questions to measure the risk preferences of the subject. Resulting in a number between 1 and 7 to specify the risk preferences. Similarly, the participants answered some statements about the future, whether they focus on the short run or on the long run. All the answers are coded as such that 1 implies participants have mainly a short run focus and 7 implies they have a long run focus. Again, the average of these six questions is calculated and used as a proxy for time preferences of the participants.
3.3 Behavioral characteristics
11
immediately’ are seen as the most impulsive ones, and therefore the ones with the highest self-control problems. Thus, participants with a score of 7, are very impulsive and thus, have little self-control. The variable on impulsiveness can be used for robustness checks since this variable is similar to the self-control variable.
Two other interesting variables in the dataset are the variables about mental accounting and the precautionary savings motive. Mental accounting is a trick that people might use as some sort of commitment device. They put money in separate jars or bank accounts and can only withdraw money for certain goals, for example a bank account with only money for vacation, or an account with money for clothes, etc. A dummy variable is included in the analysis where one means that people use mental accounting tricks and zero means that they do not. Furthermore, there is a question related to saving motives that states whether they save for unexpected expenses, also called the precautionary savings motive. The participants indicated whether they find this very unimportant or very important on a scale from one to seven.
12
Table 1: Descriptive statistics
N Mean Std.dev Min Max
Individual characteristics:
Age 1,336 55.9371 16.0891 20 92
Male 1,336 0.5756 0.4944 0 1
Number of kids 1,336 0.4843 0.9231 0 6
Partner present 1,336 0.7186 0.4499 0 1
Does financial administration 1,336 0.7635 0.4251 0 1
Smoker 1,336 0.1467 0.3539 0 1 Education levels: Low education 1,336 0.2590 0.4382 0 1 Medium education 1,336 0.3346 0.4720 0 1 High education 1,336 0.4064 0.4914 0 1 Risk preferences 1,336 2.8004 1.0179 1 6.5 Time preferences 1,336 4.0697 0.8987 1 6.83 Net income 1,336 27,318.86 17,433.31 0 169,527.80 Behavioral factors: Mental accounting 1,336 0.3945 0.4889 0 1 Precautionary savings 1,336 0.8263 0.3790 0 1 Self-control 1,336 2.7717 1.4635 1 7 Impulsiveness 1,336 5.2515 1.1928 1 7 Savings levels: Total savings 1,241 51,020.65 155,815.30 0 3,245,889 Liquid savings 1,241 31,316.98 70,345.17 0 1,047,071 Illiquid savings 1,241 8,107.11 97,328.10 0 2,993,434 Investment savings 1,241 11,596.56 60,352.75 0 1,008,500
4. METHODOLOGY
Two different models are used in this thesis. The first ordinary least squares (OLS) model investigates the relationship between self-control problems and different types of savings. The second OLS model describes the relationship between self-control problems and income.
13
decision, which makes these savings channels illiquid. Investment savings can be either liquid or illiquid, depending on for example the type of bonds and whether the stocks are listed on an exchange. This information was not available and therefore a different category for these combined liquid and illiquid investment savings will be used. Equation 1 shows the OLS regression that will be performed.
ln(𝑠𝑎𝑣𝑖𝑛𝑔𝑠𝑖) = 𝛼0+ 𝛼1∗ 𝑠𝑒𝑙𝑓𝑐𝑜𝑛𝑡𝑟𝑜𝑙𝑖 + 𝛼2∗ 𝑿𝒊+ 𝜀𝑖 (1)
The dependent variable savingsi is the level of savings of the participant, as explained
above, divided into the categories liquid savings, illiquid savings, investment savings.
selfcontroli measures the ability of the participants to control their expenses, thus, their ability
to exercise self-control on a scale of one to seven, where one implies that people have low self-control problems and seven corresponds to high self-control problems. Xi is a vector of control variables including individual characteristics: age, gender, income level, education level, number of kids, whether there is a partner present in the household, whether the participant does the financial administration, risk- and time preferences, and two dummy variables which measure whether the participants uses mental accounting tricks and has a precautionary savings motive. 𝜀𝑖 is the error term. Since there is data available of multiple individuals within the same household, this potentially introduces some source of heteroskedasticity. Using clustered standard errors may therefore be a potential solution. Moreover, to try to convert the distribution into a normal distribution, logarithms are used for the levels of income and all types of savings.2
The second OLS model runs a regression to investigate how the level of income of the individuals is associated with their ability to exercise self-control. Equation 2 shows this OLS regression.
𝑠𝑒𝑙𝑓𝑐𝑜𝑛𝑡𝑟𝑜𝑙𝑖 = 𝛽0+ 𝛽1∗ 𝑛𝑒𝑡𝑖𝑛𝑐𝑜𝑚𝑒𝑖 + 𝛽2∗ 𝒁𝒊+ 𝜀𝑖 (2)
The dependent variable selfcontroli measures the individual’s inability to exercise
self-control, again, on a one to seven scale where the higher the value of this variable, the higher the self-control problems of the individual. This self-control variable is assumed to depend on
14
the level of income, the variable netincomei. And 𝒁𝒊 is a vector of control factors including
the abovementioned individual characteristics (age, gender, education, number of kids, partner present in household, whether the individual does the financial administration, risk and time preferences and, a dummy for a precautionary savings motive). Furthermore, a dummy is included on whether the participant makes use of mental accounting tricks. This is an interesting variable to look at, since mental accounting can be regarded as a soft form of commitment device. Therefore, it is appealing to see whether this variable is related to self-control problems. Moreover, a dummy variable which states whether the individual is a smoker or not is included. Smokers are expected to have stronger self-control problems (Adams, 2009). And lastly, 𝜀𝑖 is the error term.
5. RESULTS
This section provides the results of the OLS regressions. First, the OLS model on savings and self-control is discussed, then the regression on self-control and income.
5.1 Savings and self-control
The OLS model regarding the relationship between the different types of savings and self-control potentially suffers from heteroskedasticity, this is because of the possible existence of correlation in the errors within households. Therefore, a Breusch-Pagan test for heteroskedasticity is performed. The null-hypothesis of homoskedastic errors is rejected3 with a p-value of zero, therefore clustered (robust) standard errors are applied to the model to correct for the heteroskedasticity problem. Table 2 shows the regression results of the relationship between self-control and savings.
Firstly, when looking at the column of total savings, the positive sign for the coefficient of age, combined with the negative sign for age-squared, confirms the life-cycle hypothesis. When people get older, they start to save more, until they reach a certain age where they save less again. Secondly, males save on average 60% more than females. As already discussed earlier, people are used to a certain standard of living, so when they reach retirement they would like to be able to maintain that standard of living (Ackert and Deaves, 2016). The results are in line with this expectation, since higher income individuals have
15 Table 2: OLS regression on self-control and savings
(1) (2) (3) (4)
VARIABLES Total savings Liquid savings Illiquid savings Investment savings
Age 0.108*** -0.0539** 0.385*** 0.0977** (0.0317) (0.0255) (0.0374) (0.0448) Age2 -0.000712** 0.000633*** -0.00345*** -0.000457 (0.000286) (0.000230) (0.000336) (0.000419) Male 0.600*** 0.466*** 0.330* 0.527** (0.165) (0.168) (0.199) (0.209) Net income 0.387*** 0.357*** 0.153* 0.270*** (0.0950) (0.105) (0.0833) (0.0873) Medium education 0.499** 0.173 0.312 0.826*** (0.208) (0.203) (0.243) (0.254) High education 0.961*** 0.576*** 0.525** 1.085*** (0.201) (0.193) (0.251) (0.263) Number of kids 0.0504 0.0714 0.0111 -0.190* (0.0860) (0.0793) (0.126) (0.104) Partner present 0.0514 0.453*** -0.187 -0.194 (0.163) (0.144) (0.239) (0.253) Does financial administration 1.458*** 1.747*** 0.569*** 0.586*** (0.212) (0.222) (0.207) (0.222) Risk preference 0.323*** -0.0725 0.118 1.278*** (0.0754) (0.0674) (0.0933) (0.101) Time preference 0.281*** 0.309*** 0.384*** 0.342*** (0.0804) (0.0821) (0.110) (0.112) Mental accounting -0.122 -0.118 -0.320* -0.112 (0.149) (0.142) (0.193) (0.202) Precautionary savings motive 0.360* 0.142 0.328 0.609** (0.186) (0.175) (0.239) (0.244) Self-control -0.132*** -0.252*** 0.00953 -0.182*** (0.0509) (0.0510) (0.0629) (0.0649) Constant -3.830*** 3.453*** -12.78*** -10.97*** (1.241) (1.225) (1.374) (1.573) Observations 1,200 1,209 1,241 1,241 R-squared 0.224 0.211 0.098 0.222
Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1
16
Regarding the risk and time preferences, individuals who are risk loving have higher savings. This may seem a bit odd, when taking into account the results of Mazocco (2004). However, the precautionary savings motive variable already includes this effect, by increasing the level of savings significantly when people have a precautionary savings motive. This positive significant effect of risk preferences is presumably largely due to the positive influence of risk-loving individuals on the level of investment savings. Due to the fact that investing in stocks and/or bonds involves more risk, most definitely more risk loving individuals have higher amounts of investment savings compared to risk averse individuals. Since investment savings are included into total savings as well, this can explain why there is a small positive relationship between the risk preferences and total savings. Similarly, the effect of time preferences on the levels of total savings has a significant positive effect. This implies that individuals whose time preferences are more towards the future, instead of the presence, have higher levels of savings. This makes sense, since individuals who attach more weight to the future are more likely to save at this moment in order to be able to consume more in the future. Which is in line with the statements of O’Donoghue and Rabin (1999).
Finally, let us look at the variable of interest, the influence of self-control problems on the level of total savings. The relationship between self-control problems and the level of total savings is significantly negative, implying that people with higher self-control problems have lower total savings compared to people with lower self-control problems.
17
Subsequently, the results for the self-control variable are rather interesting. Where the coefficient of self-control is significantly negative in the regressions on total savings, liquid savings and investment savings, this coefficient turns positive (however, not significant) for the level of illiquid savings. This implies that individuals with high self-control problems have significantly lower total, liquid and investment savings, however, they do not have lower illiquid savings. A possible explanation can be that illiquid savings are some sort of commitment device. Once the money is put into these accounts, it is hard to take your money out of it. Therefore, people with high self-control problems can decide to put their money into illiquid saving accounts and thus forcing themselves to save more, since they cannot easily withdraw the money from these accounts. This is similar to the co-holder puzzle of Gathergood and Weber (2014), since the co-existence of debt and liquid savings is also some sort of commitment device. Additionally, as Thaler and Benartzi (2004) stated, people who make use of low willpower savings methods, such as pension and social security, are doing a rather decent job in saving an adequate amount for retirement. These illiquid savings include pension/annuity insurance and endowment insurance. Once committed to these schemes, it does not require a lot of willpower to build up the amounts put into it.
Taking this altogether, the effect of self-control indeed differs among various types of savings. Where individuals with higher self-control problems are having lower total savings, as well as liquid and investment savings, this difference disappears when looking at illiquid savings. This can probably be explained by the fact that illiquid savings can be seen as some sort of commitment device for people with low self-control.
For robustness, the regression on all categories of savings were also run with the variable measuring impulsiveness instead of the self-control variable. Since impulsiveness is the opposite of self-control, the sign of this variable should be positive regarding the effect on the level of savings. Taking this into account, similar results were obtained.4 Therefore the results are robust and in line with the literature.
5.2 Self-control and income
Next, an OLS regression was performed on the relationship between self-control problems and income. Table 3 shows the regression results.
18 Table 3: OLS regression on self-control and income
(1)
VARIABLES OLS self-control
Age -0.0529*** (0.0180) Age2 0.000396** (0.000164) Male -0.0388 (0.0870) Medium education 0.157 (0.108) High education 0.0134 (0.108) Number of kids 0.132*** (0.0478) Partner present -0.147 (0.0982)
Does financial administration -0.245**
(0.100) Smoker 0.234** (0.114) Net income -0.0731** (0.0373) Risk preference 0.0516 (0.0397) Time preference -0.108** (0.0458) Mental accounting 0.212*** (0.0815) Precautionary savings 0.0317 (0.105) Constant 5.461*** (0.630) Observations 1,336 R-squared 0.057
Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1
19
administration within the household shows that the person who does the financial administration has higher control. This could be explained by the existence of self-selection, within the household they decide that the person who has the highest self-control should do the financial administration. When focusing on the variable smoker, which indicates whether someone is a smoker or not, a positive coefficient is observed. This implies that smokers have stronger self-control problems. This is in line with the literature stating that smokers are facing self-control problems because the short-run temptations of smoking are interfering with the long-term goal of quitting, therefore their actions are time inconsistent and causing self-control problems (Adams, 2009). Furthermore, individuals with time preferences on a longer horizon seems to have less problems with self-control. This, indeed, makes sense, since people who are more patient have lower discount rates. They attach more weight to future consumption than impatient individuals would. In order to be able to consume in future periods and obtain the benefits from it, it is necessary to exert self-control and save for future consumption. Thus, these people have higher self-control to be able to obtain the benefits from delayed consumption.
Mental accounting can be seen as some sort of commitment device. By forcing yourself to put money into different account for different purposes, you ‘commit’ yourself to only using the money for that purpose. One would expect that people who already have the self-control to save enough, would not make use of mental accounting. Whereas people with lower self-control might need this extra mental trick to be able to save more. This is exactly what is observed from the regression. The coefficient of mental accounting is significantly positive, which implies that individuals who make use of mental accounting have stronger self-control problems. Lastly, the effect of income on the ability to exercise self-control. Numerous studies investigated this effect (for example Bernheim et al., 2015, Carvalho et al., 2016) and they all found that lower income individuals seems to have more self-contol problems than less financially restricted individuals. When investigating the coefficient of income in the OLS regression, there is a significantly negative effect of income on self-control problems. This implies that lower income individuals seem to have more problems in exercising self-control, which is in line with the literature.
For robustness, the relationship between income and self-control is done by means of ordered probit and order logit models. With only a minor difference in the significance of risk preference5, the results are similar to the OLS regression6. Thus, the result are robust.
5
20
Taking all of the above into account, it can be stated that self-control problems are significantly negatively associated with the level of total savings of the participants. This negative relationship is still present with regards to liquid savings and investment savings. However, this effect disappears when investigating illiquid savings, there is no difference in the levels of illiquid savings between people with low self-control problems and people with high self-control problems. This can be explained by the fact that illiquid savings can be seen as some sort of commitment device, similar to co-holding that was investigated by Gathergood and Weber (2014), once you put your money into these accounts it is harder to get your money out again. Therefore, if you have self-control problems but committed previously to putting your money into illiquid accounts, there is not much you can do about it at a later stage. Moreover, self-control problems appear to be worse for low income individuals, which is in line with the findings of Bernheim et al. (2015) and Carvalho et al. (2016).
6. CONCLUSION
There is substantial evidence that people on average tend to save too little. One reason put forward by Thaler and Benartzi (2004) is that people simply do not have the self-control to lower current consumption in order to increase future consumption. This thesis investigates whether people with higher self-control problems indeed have lower savings. In addition, it is examined whether this effect of self-control on savings differs when looking at different categories of savings. An OLS regression model is executed to investigate these effects. Data from the DNB Household Survey with data of over 2000 is being used in the analysis.
Four OLS regressions are performed to investigate the relationship between self-control problems and the level of different types of savings. The first regression is on total savings, which is then categorized into liquid savings, illiquid savings and investment savings. I found that indeed people with higher self-control problems have lower total savings. When investigating the differences among the savings types, the results suggest that this negative effect of self-control problems on savings is still present when looking at liquid savings and investment savings. Since, in these cases self-control is required to not withdraw money from these accounts. People with higher self-control problems are more inclined to withdraw money from these liquid accounts and therefore have lower levels of liquid savings.
21
However, this negative effect disappears when investigating illiquid savings channels. This is because once one is committed to putting your money into illiquid accounts, it is harder to withdraw the money again, compared to liquid accounts. Therefore, people with higher self-control problems can use illiquid savings as some sort of commitment device. Often these people know they have self-control problems and therefore save money through illiquid accounts since this makes it harder for them to withdraw the money at a later stage, it forces them to save. All of the above results are robust.
Furthermore, abundant literature found that self-control problems are worse for low income individuals (for example; Bernheim et al., 2015, Carvalho et al., 2016). To examine whether this is indeed the case for the participants of the CentERpanel, an OLS regression is performed on the effect of income on the inability to exercise self-control. It was found that people with lower income indeed have more problems with exercising self-control. Moreover, an interesting aspect of the analysis was that people who make use of mental accounting tricks have stronger self-control problems. This also explains why people with self-control problems do not have significantly lower illiquid savings, since these savings can also be seen as some sort of commitment device similar to mental accounting.
22
References
Ackert, L.F., Deaves, R., 2016. Behavioral finance: Psychology, decision-making, and markets. South-Western Cengage Learning, United States of America.
Adams, J., 2009. The role of time perspective in smoking cessation amongst older English adults. Health Psychology 28, 529-534.
Alessie, R., Hochguertel, S., van Soest, A., 2006. Non-take-up of tax-favored savings plans: Evidence from Dutch employees. Journal of Economic Psychology 27, 483-501. Banerjee, A.V., Mullainathan, S., 2010. The shape of temptation: Implications for the
economic lives of the poor. Unpublished working paper. National Bureau of Economic Research, Cambridge.
Bernheim, B.D., Garrett, D.M., Maki, D.M., 2001. Education and saving: The long-term effects of high school financial curriculum mandates. Journal of Public Economics 80, 435-465.
Bernheim, B.D., Ray, D., Yeltekin, S., 2015. Poverty and self-control. Econometrica 83(5), 1877-1911.
Beshears, J., Choi, J.J., Harris, C., Laibson, D., Madrian, B.C., Sakong, J., 2015. Self control and commitment: Can decreasing the liquidity of savings accounts increase deposits? Unpublished working paper. National Bureau of Economic Research, Cambridge. Bucciol, A., Miniaci, R., 2018. Financial risk propensity business cycles and perceived risk
exposure. Oxford Bulletin of Economics and Statistics 80 (1), 160-183.
Carvalho, L.S., Meier, S., Wang, S.W., 2016. Poverty and economic decision-making: evidence from changes in financial resources at payday. American Economic Review 106(2), 260-284.
Clark, R., d’Ambrosio, M., McDermed, A., Sawant, K., 2006. Retirement plans and saving decisions: The role of information and education. Journal of Pension Economics and Finance 5 (1), 45-67.
23
Eckel, C.C., Grossman, P.J., 2002. Sex differences and statistical stereotyping in attitudes toward financial risk. Evolution and Human Behavior 23 (4), 281-295.
Eigsti, I.M., Zayas, V., Mischel, W., Shoda, Y., Ayduk, O., Dadlani, M.B., Davidson, M.C., Aber, J.L., Casey, B.J., 2006. Predicting cognitive control from preschool to late adolescence and young adulthood. Psychological Science 17 (6), 478-484.
Friedman, M., 1957. A theory of the consumption function. Princeton University Press, New Jersey.
Gathergood, J., Weber, J., 2014. Self-control, financial literacy & the co-holding puzzle. Journal of Economic Behavior and Organization 107, 455-469.
Holt, C.A., Laury, S.K., 2002. Risk aversion and incentive effects. American Economic Review 92 (5), 1644-1655.
Laibson, D., 1997. Golden eggs and hyperbolic discounting. Quarterly Journal of Economics 112 (2), 443–478.
Lawrance, E.C., 1991. Poverty and the rate of time preference: Evidence from panel data. Journal of Political Economy 99 (1), 54-77.
Mazzocco, M., 2004. Saving, risk sharing, and preferences for risk. The American Economic Review 94 (4), 1169-1182.
Modigliani, F., Brumberg, R.H., 1954. Utility analysis and the consumption function: An interpretation of cross-section data. In: Kurihara, K.K., ed., Post-Keynesian Economics, Rutgers University Press, New Brunswick.
Nyhus, E.K., 2002. Psychological determinants of household saving behaviour. Doctoral Dissertation. Norwegian School of Economics and Business Administration, Bergen, Norway.
Nyhus, E.K., Webley, P., 2001. The role of personality in household saving and borrowing behaviour. European Journal of Personality 15, 85-103.
Prentice, C., McKillop, D., French, D., 2017. How financial strain affects health: Evidence from the Dutch National Bank Household Survey. Social Science & Medicine 178, 127-135.
24
Thaler, R.H., Benartzi, S., 2004. Save more tomorrow: Using behavioral economics to increase employee saving. Journal of Political Economy 112(1), 164–187.
Thaler, R.H., Shefrin, H.M., 1981. An economic theory of self-control. Journal of Political Economy 89(2), 392-406.
Thaler, R.H., Shefrin, H.M., 1988. The behavioral life-cycle hypothesis. Economic Inquiry 26, 609-643.
Appendices
A. Distribution graphs
Graphs A.1 till A.5 show of the distributions of the transformed logarithmic variables, namely, the variables on total savings, liquid savings, illiquid savings, investment savings and net income. As can be seen from the graphs, the distributions on the logarithmic variables now look more closely like a normal distribution, which is shown by the blue line in the graphs.
Graph A.1: Distribution of the logarithm of total savings.
0 .0 5 .1 .1 5 .2 .2 5 D e n s it y 0 5 10 15
25 Graph A.2: Distribution of the logarithm of liquid savings.
Graph A.3: Distribution of the logarithm of illiquid savings.
0 .1 .2 .3 D e n s it y 0 5 10 15
Log of liquid savings
0 .0 5 .1 .1 5 .2 .2 5 D e n s it y 0 5 10 15
26 Graph A.4: Distribution of the logarithm of investment savings.
Graph A.5: Distribution of the logarithm of net income.
Appendix B: Breusch-Pagan heteroskedasticity test
Breusch-Pagan test for heteroskedasticity Chi2 65.85 Probability 0.00 0 .0 5 .1 .1 5 .2 .2 5 D e n s it y 0 5 10 15
Log of investment savings
0 .2 .4 .6 .8 D e n s it y 0 5 10 15
27
Appendix C: Robustness checks
28 Table B.1: OLS regression on the different types of savings and impulsiveness
(1) (2) (3) (4)
VARIABLES Total savings Liquid savings Illiquid savings Investment savings
Age 0.112*** -0.0453* 0.417*** 0.104** (0.0315) (0.0257) (0.0382) (0.0445) Age2 -0.000754*** 0.000548** -0.00373*** -0.000518 (0.000284) (0.000232) (0.000343) (0.000416) Male 0.625*** 0.505*** 0.319 0.556*** (0.164) (0.168) (0.203) (0.207) Net income 0.393*** 0.369*** 0.175** 0.278*** (0.0961) (0.107) (0.0847) (0.0893) Medium education 0.505** 0.177 0.509** 0.833*** (0.207) (0.204) (0.250) (0.255) High education 0.989*** 0.620*** 0.585** 1.121*** (0.201) (0.193) (0.254) (0.265) Number of kids 0.0301 0.0361 -0.00509 -0.220** (0.0857) (0.0779) (0.130) (0.103) Partner present 0.0395 0.446*** -0.199 -0.200 (0.161) (0.145) (0.243) (0.254) Does financial administration 1.486*** 1.803*** 0.553** 0.627*** (0.213) (0.223) (0.215) (0.223) Risk preference 0.339*** -0.0509 0.169* 1.294*** (0.0766) (0.0678) (0.0978) (0.103) Time preference 0.258*** 0.281*** 0.374*** 0.321*** (0.0810) (0.0842) (0.112) (0.112) Mental accounting -0.158 -0.185 -0.263 -0.157 (0.147) (0.142) (0.197) (0.200) Precautionary savings 0.301 0.0522 0.397 0.543** (0.185) (0.171) (0.245) (0.246) Impulsiveness 0.192*** 0.294*** 0.0219 0.216*** (0.0695) (0.0648) (0.0825) (0.0829) Constant -5.258*** 0.983 -14.06*** -12.75*** (1.251) (1.218) (1.378) (1.641) Observations 1,200 1,209 1,241 1,241 R-squared 0.225 0.209 0.106 0.221
29 Table B.2: Ordered probit and logit regression on self-control and income
(1) (2)
VARIABLES Ordered probit Ordered logit
Age -0.0401*** -0.0696*** (0.0131) (0.0221) Age2 0.000298** 0.000514** (0.000119) (0.000201) Male -0.0339 -0.103 (0.0635) (0.108) Medium education 0.120 0.198 (0.0790) (0.137) High education 0.0389 0.0791 (0.0791) (0.136) Number of kids 0.102*** 0.180*** (0.0347) (0.0584) Partner present -0.113 -0.154 (0.0719) (0.124)
Does financial administration -0.197*** -0.308**
(0.0730) (0.125) Smoker 0.188** 0.330** (0.0824) (0.139) Net income -0.0534** -0.0893* (0.0270) (0.0464) Risk preference 0.0486* 0.102** (0.0293) (0.0514) Time preference -0.0721** -0.133** (0.0339) (0.0602) Mental accounting 0.165*** 0.311*** (0.0595) (0.102) Precautionary savings 0.00799 -0.0186 (0.0767) (0.130) Observations 1,336 1,336
