Always Look On The Bright Side Of Life? Investigating The Impact Of Optimism Upon Household Saving Behaviour

Always Look On The Bright Side Of Life? Investigating The Impact Of

Optimism Upon Household Saving Behaviour

University of Groningen

Faculty of Economics and Business

Master Thesis Economics

Name: Joseph Tichband

Student ID number: S3453898

Student email:

j.tichband@student.rug.nl

Date Paper: 12

th

June 2019

2

Contents

1. Introduction ... 3

2. Literature Review ... 4

2.1 Household savings behaviour ... 4

2.2 Optimism ... 5

3. Data ... 8

3.1 Data Selection ... 8

3.2 Construction of key variables ... 11

3.2.1 Household Saving Behavior ... 11

3.2.2 Optimism ... 12

... 17

3.2.3 Control Variables ... 17

4. Methodology ... 18

4.1 Dichotomous Dependent Variable Models ... 18

4.2 Continuous Dependent Variable Models ... 20

5. Results ... 20

5.1 Preference to Save ... 20

5.2 Preference to save in future ... 22

5.3 Total Amount Saved In a Given Year ... 23

5.4 Total Balance of Saving & Deposit Accounts ... 27

5. Conclusion ... 30

3

1. Introduction

The determinants of household financial behaviour have long been a hotly investigated topic within the field of economics. Following work by Katona (1968), emphasizing the impact psychological factors play upon consumer behaviour, increased research attention has turned to potential psychological determinants. It is hoped including these psychological variables within models will improve their accuracy with regards to explaining the patterns displayed by empirical data upon household financial behaviour. Although various financial behaviours exist, household saving behaviour has been of particular interest to academics (Harris et al., 2002) and policy makers alike. Household savings behaviour makes up the central focus of this paper.

The motivations behind an individual’s decision to save are extensive and long established (Browning & Lusardi, 1996)1_{. Savings allow an individual to maintain their marginal utility from consumption} throughout the course of their entire life. Importantly this allows for individuals to maintain their marginal utility from consumption when faced with reductions in income such as retirement. It is this consumption smoothing aspect of savings that lays the foundation for the Life-Cycle hypothesis of savings behaviour (Modigliani & Brumberg, 1954). Without an adequate stock of savings an individual is unable to smooth their consumption over their lifetime; facing reduced marginal utility of consumption post retirement. Savings also act as a buffer for any unplanned expenditures a household may face such as medical bills. This contributes to the precautionary motive of saving along with income uncertainty (Guiso et al., 1992). Without a suitable stock of savings, a household may be required to increase its indebtedness when experiencing unplanned expenditures, reducing future consumption capability of the household.

Individuals are becoming far more independently responsible for their stock of savings in the lead up to retirement (Poterba et al., 2007). Coupled with ever-increasing life expectancies stresses the importance for policymakers of ensuring households develop a suitable stock of savings. Recent empirics however, point towards household saving levels being far below that of an optimal level in order to sustain current consumption behaviour post retirement. In the United Kingdom over 11 million people have insufficient savings for their retirement (Balasuriya et al., 2014). Insufficient saving underlies the importance of understanding the determinants behind household saving behaviour in order for policy makers to ensure household saving can be raised in the future. This paper is of importance as it investigates whether or not holding optimistic views impacts upon household saving behaviour.

4 Since their inception, models of household savings behaviour have heavily stressed socio-demographic variables as key determinants. These models, however are based upon strong assumptions of rationality and perfect information on the part of consumers. These assumptions are identified as inaccurate in many areas of household financial behaviour (Barberis and Thaler, 2003) Thus in recent decades studies upon household financial behaviour has turned to the relatively new field of behavioural economics. Household savings behaviour has been no exception. It is upon these psychological factors and their impact upon household saving behaviour this research is focused, in particular optimism. This study therefore aims to investigate the following research question: Whether

optimism has an impact upon household financial behaviour?

Optimism, defined as an individual exhibiting positive expectations about the future (Scheier et al., 1994) has had a mixed impact upon household saving behaviour. Early theories suggest that individuals whom are more optimistic about future economic conditions are less likely to save (Katona, 1975). In contrast, recent findings pinpoint to the opposite relationship holding true. Puri & Robinson (2007) for example, find that in a data sample for individuals in the United States optimism, particularly moderate optimism, results in increased saving. This result was not recreated for individuals in Australia however, Harris et al., (2002) found that due to increased precautionary motive of saving, it is in fact pessimism that increases household saving. Therefore there still remains no concrete consensus on the impact optimism has upon household saving behaviour.

This paper contributes to the literature through investigating the impact of optimism upon household saving behaviour in a population sample not utilised before. Further this study utilizes two differing measurements of optimism, the increasingly popular life expectancy miscalibration measure (Puri and Robinson, 2007) along with economic optimism (Harris et al., 2002).

In order to answer the question of whether optimism impacts household savings behaviour this paper investigates the impact of optimism on savings behaviour of a representative population group of individuals in the Netherlands. The remainder of the paper is structured as follows. Section 2 provides a review of prior literature upon the research area. Section 3 introduces the data utilized in this study, along with the construction of both the dependent and independent variables of importance. Section 4 provides information upon the methodology undertaken in the analysis. Results from the analysis are provided in section 5 and a conclusion given in section 6.

2. Literature Review

5 Initial models of household saving behaviour took the form of the Permanent Income Hypothesis (PIH) (Friedman, 1957) and the Life-Cycle (LC) hypothesis (Modigliani & Brumberg, 1954). The LC hypothesis which has become the standard for modelling household savings over time (Yuh & Hanna, 2010). According to the model, individuals will decide upon a consumption path in order to maximize expected lifetime utility and hence will seek to keep the marginal utility from consumption at a constant throughout their lifetime (Yuh & Hanna, 2010). In order to smooth consumption like this individuals will borrow when incomes are low to suffice consumption and will save when incomes are high, ultimately resulting in dissaving when retirement is reached (Browning & Lusardi, 1996). Although the result of this is that consumption is not linked to current income, the model relies upon the expectation that individuals will choose a consumption and hence saving path based upon lifetime income.

The central tenet of household savings behaviour thus focused on within this LC model and further extensions is the variance of future income (Yuh & Hanna, 2010). Through assuming individuals seek to optimise lifetime utility given lifetime income, these models solely stress the ability an individual possesses in order to save as determining their savings behaviour. Seminal work by Katona (1968), however, discusses the concept that consumer savings behaviour is a function comprising not only of individuals’ ability to save but also their willingness to (Gerhard et al., 2018). It is this willingness to save of individuals which is impacted by psychological factors. A well-known example in the field of behavioural economics is the presence of “hyperbolic discounting” (Laibson, 1997). Individuals who exhibit “hyperbolic discounting” place a large discount rate on tomorrow relative to today, followed by sensible discount rates being applied to future time periods (Deaton, 2005). This formalizes the concept of procrastination (Deaton, 2005). This is useful for modelling household saving behaviour as procrastination is witnessed in the real world with individuals waiting too long to start saving for retirement (Deaton, 2005).

Overtime, modern theories of household savings behaviour have thus incorporated psychological factors. For example the Behavioural Life Cycle (BLC) hypothesis as developed by Shefrin and Thaler (1988). The BLC hypothesis introduces the concepts of self-control, mental accounting and framing effects in order to predict actual household saving behaviour (Shefrin & Thaler, 1988). Through introducing psychological elements like these, modern savings models are thus able to ease the restrictive and unrealistic assumptions of perfect information and foresight of consumers and instead allow for real human behaviour not simply rational behaviour. Therefore models like the BLC hypothesis are better equipped to explain real world data i.e insufficient retirement savings.

6 Being optimistic is shown to benefit an individual’s mental health. Carver and Gaines (1987) find that women who displayed optimism during the third trimester of pregnancy were less likely to develop depressive symptoms in the three weeks post-partum than their pessimistic counterparts. Optimism also improves physical well-being. Optimistic individuals are further along with their recovery 6 months post coronary artery bypass surgery than pessimistic individuals (Scheier & Carver, 1992).

However, optimistic individuals are likely to suffer from an optimistic bias, whereby they overestimate the probability of favorable outcomes occurring and/or underestimate the probability of negative outcomes occurring (Puri & Robinson, 2007). This logic is applied when judging their own risk of a negative outcome to that of other individuals (Helweg-Larsen & Sheppard, 2001). Individuals suffering from an optimistic bias believe that they will not be subject to misfortune, but others in fact will fall victim (Weinstein, 1980). Individuals holding miscalibrated beliefs about personal risk have been found for a multitude of negative outcomes such as automobile accidents, victims of crime and suffering from disease (Weinstein, 1980). This highlights the danger of optimism, whereby individuals may fail to take the necessary precautions in order to achieve a desired outcome. With regards to household saving behaviour this dangerous side of optimism may result in individuals under saving throughout their lifetime believing that they will not be the ones who are subject to insufficient funds in retirement but instead the statistics highlighting this phenomena simply apply to others.

Within the field of economics itself, the prevailing view is that optimism and its proxies result in suboptimal decisions undertaken by individuals (Puri & Robinson, 2007). Optimism has been found to induce suboptimal business entry (Camerer and Lovallo, 1999), irrational behaviour on the part of entrepreneurs (Coelho et al., 2004) and individual equity ownership (Odean, 1998). Optimism is oft viewed through the lens of related psychological concepts; also believed to result in suboptimal behaviour. In particular, optimism is related to overconfidence (Puri & Robinson, 2007). Overconfidence has been shown to lead to lower investment returns for men when compared with women and excessive trading (Barber & Odean, 2001). With a plethora of evidence in which optimism is shown to induce sub optimal outcomes for individuals, this begs the question of why in fact individuals would hold systematically miscalibrated beliefs about future events. Brunnermeier and Parker (2005) theories that as utility contains an anticipatory component. Therefore when the ex-ante gain to overall utility, due to increased expected future utility as a result of optimistic beliefs, outweighs the ex-post cost of negative outcomes to which said optimistic beliefs may lead to, holding optimistic beliefs is justified (Brunnermier & Parker, 2005).

7 and intuitive part of the brain. System 2 on the other hand is slower, more analytical and thus deliberate. Unlike system 1 which is active at all times and leads to fast responses of the brain, system 2 requires logical thinking and reason is able to dominate. System 1 to a large extent is far more influential in relation to decision making and steers system 2. It is this limbic system 1 that is susceptible to optimism, resulting in the individual facing all problems such as whether to save or not with an optimistic bias (Brunnermier & Parker, 2005).

Katona (1968) identifies due to the dynamic impact of optimism that when an individual is optimistic they will save less in order to demand more today knowing they will have the income in the future to smooth consumption. This is supported somewhat by the finding of Webley and Nyhus (2001) that debtors are more optimistic about future income. This anticipated impact relates to the LC hypothesis as expectations about increased future income will lead to less saving as consumption is smoothed in the current period. This leads to the first hypothesis:

H1: Optimism reduces the preference for saving

Further, optimistic individuals will have underweighted the probability of negative outcomes occurring during their lifetime; rendering the need for precautionary savings to be far lower. This leads on to the second hypothesis:

H2: Optimism reduces the amount saved

However the empirics in the literature have thus far not been able to reach a consensus upon the impact of optimism upon household savings behaviour. Puri & Robinson (2007) in their study upon optimism and economic choices of individuals in the United States develop a novel measure of optimism. An individual whose self-assessed life expectancy2 _{is greater than what actuarial tables} would predict are determined to be optimistic. Through OLS regressions, controlling for determinants of savings behaviour such as age and gender, the authors identify that optimism induces greater saving. This relationship however differs with the degree of optimism. Extreme optimists3 exhibit lower savings behaviour than mild optimists. Harris et al., (2002) undertake a similar survey based investigation upon individuals in Australia. Unlike Puri & Robinson (2007) the authors focus on a measure of optimism that is linked to economic situation as opposed to life expectancy. This composite measure was calculated as an average of responses to survey questions upon future family finances and the development of the Australian economy over the next 1 and 5 years. Through estimating an ordered probit model, including controls such as a measure of income and homeownership, the

2 _{Determined through a survey provided to individuals}

8 authors find that being optimistic reduces the probability of an individual being in a higher savings group.

3. Data

This section provides an overview of the data utilized in this study. Followed by a description of the variable construction.

3.1 Data Selection

In this paper use is made of data of the DNB Household Survey as provided CentERdata. The DNB Household Survey (DHS) offers a multitude of data upon differing socioeconomic and psychological characteristics. The survey has been conducted annually since 1993 and is distributed to just over 2,000 households, representing the population of the country as a whole (Bucciol and Veronesi, 2014; Webley and Nyhus, 2006). Only individuals aged 16 and above are entitled to participate. The survey consists of six questionnaires: general information upon the household, household & work, accommodation & mortgages, health & income, assets & liabilities and economic & psychological concepts4_.

The data utilized in this paper is comprised of questions included in three of the questionnaires: general information upon the household, health & income and economic & psychological concepts. In addition to this data from the aggregated assets data provided is also used to construct one of the four dependent variables. The general information questionnaire provides information such as age, gender, education, position in household etc. The questionnaire upon health & income provides information such as subjective and objective health measures, along with a subjective survival probability. A second source of data required is the Human Mortality Database. This provides mortality data in the form of both period life tables and cohort life tables for many countries. The period life table data for the Netherlands is utilized in order to construct actuarial survival probabilities and ultimately the main variable of interest.

This paper uses data from 2004 to 2016. Pre 2000 surveys are not comparable with those completed post 2000 in terms of sample design, and over represent wealthy households (Bucciol and Veronesi, 2014). Further, the surveys completed between 2000 and 2004 suffer from a significant amount of missing data upon crucial questions required to construct key variables. Thus these years are dropped. Individuals whom did not participate in all of the three questionnaires required for the analysis were also dropped from the data set. A large number of individuals did not provide answers to the economic

9 & psychological concepts questionnaire. As a result of this the dataset is comprised of 20,216 observations. The summary statistics are provided in Table 1.

In order to undertake the analysis, further restrictions are then placed upon this whole data set. Importantly only individuals identified as the financial administrator of their household are included. The financial administrator of the household is determined by answering yes to the question “Are you the person who is most involved with the financial administration of the household? By financial administration, we mean making the payments for rent or mortgage, taking out loans, taking care of tax declarations, etc.” Therefore these individuals are the ones most likely to have the knowledge upon saving decisions. This individual is representative of the household as a whole for the proceeding analysis. In addition to this due to the nature of the optimism variables constructed, only individual’s aged 70 years or below are included during the analysis. Ultimately this results in 13,337 observations. Table 2 provides the summary statistics for the individuals included in the proceeding analysis.

Variable Mean Std. dev. Median Min Max Observations Preference To Save .7102729 .4536478 1 0 1 18690 Preference To Save In The Future .8179281 .3859143 1 0 1 18273

Total Saved Over Last 12 Months (€) 5296.49 8027.776 3250 -75000 75000 13234 Total Savings in Deposit Accounts (€) 16755.23 59222.18 3000 0 3700000 19220 General Optimism .5701083 .1696902 .5889701 0 1 19766 Economic Optimism .3773559 .2387325 .2666667 0 1 19041 Household Size 2.621636 1.298849 2 1 8 20216 Number Children .814256 1.110816 0 0 6 20216 Time Horizon 2.439563 1.144748 2 1 5 19599 Age 49.04759 13.69126 50 16 70 20216 Retired .161852 .3683241 0 0 1 20216 Employed .5958152 .4907457 1 0 1 20216 Female .4708647 .4991628 0 0 1 20216 High Education .3830135 .4861336 0 0 1 20216 Medium Education .573704 .4945501 1 0 1 20216 Low Education .0423922 .2014872 0 0 1 20216 Financial Literacy .2582608 .437689 0 0 1 20216 Household Net Income 39113.26 19873.97 31000 5000 75000 17977 Risk Aversion 31.55395 6.16212 32 6 42 18238

10 Looking at Table 2, just over 70% of respondents report a preference to save over the past 12 months. Over 80% report a preference to save over the next year. The average value saved over the last year is €5269.195 _{and the average total held in an individual’s seven most important savings & deposit} accounts is €20032.26. Both independent variables have been standardized between 0 and 1. 44% of respondents are female, just under 62% are employed and just under 18% are retired. The average age of a respondent is 50 and the youngest is 18. The median household size is 2 individuals in total and on average respondents have below 1 child present in the household.

Table2

Summary Statistics of variables utilized in regressions for all individuals self-identified as financial administrators of the household.

Variable Mean Std. dev. Median Min Max Observations Preference To Save .7030793 .4569189 1 0 1 12990 Preference To Save In

The Future

.8131266 .3898252 1 0 1 12768

Total Saved Over Last 12 Months (€) 5269.187 8142.204 3250 -75000 75000 9316 Total Savings in Deposit Accounts (€) 20032.26 51904.78 6000 0 2871271 12582 General Optimism .5656878 .1719583 .5863117 .0067388 1 13002 Economic Optimism .3773902 .236926 .2666667 0 1 12946 Household Size 2.368299 1.268916 2 1 8 13337 Number Children .6657419 1.030499 0 0 6 13337 Time Preference 2.443752 1.131676 2 1 5 13005 Age 50.41554 12.61013 51 18 70 13337 Retired .1777761 .3823385 0 0 1 13337 Employed .6151308 .4865826 1 0 1 13337 Female .4418535 .4966261 0 0 1 13337 High Education .4279073 .4947939 0 0 1 13337 Medium Education .5420259 .4982494 1 0 1 13337 Low Education .029242 .1684903 0 0 1 13337 Financial Literacy .3067406 .4611581 0 0 1 13337 Household Net Income 38450.68 19828.12 31000 5000 75000 12540

11

Risk Aversion 31.4212 6.236384 32 8 42 12158

Body Mass Index 26.25916 6.594491 25.35154 3.501278 247.3958 13305

Smoker .2228387 .4161666 0 0 1 13337 GP Consultations 2.813601 5.278468 2 0 214 13337 LIDH .2660268 .441895 0 0 1 13337 Improved Economic Situation .2500609 .4330655 0 0 1 12309 Happy .8192287 .3848435 1 0 1 12939 Good Health .7787358 .4151135 1 0 1 13337

3.2 Construction of key variables

3.2.1 Household Saving Behavior

The dependent variables of household saving behavior is represented by four different proxies; (i) the willingness to save in a given year, (ii) the desire to save in the upcoming year, (iii) the total amount saved in the previous 12 months, and (iv) the total amount in savings deposit accounts.

First, in order to measure the willingness to save in a given year a dummy variable is constructed. This dummy takes the value 1 if the respondent answered yes to the question asking whether the individual put money aside in the past 12 months and a value of 0 for a response of no6_{. This is in line with that} utilized in previous studies (Bucciol and Veronesi, 2014; Webley and Nyhus, 2012).

Second, following the literature (Webley and Nyhus, 2012) the desire to save in the future is represented through the answer to the question “Is your household planning to put money aside in the

next 12 months?” The answers provided are in an ordinal scale whereby a higher value represents a

stronger desire to save in the future7_{. From this a dummy is constructed; taking the value 1 if the} respondents answered yes, certainly or yes, perhaps. The dummy takes the value 0 if the answer provided was either probably not or certainly not.

The amount saved in the previous 12 months is derived from a closed question with 7 different categorical answers. In order to create a continuous variable the amount saved is equal to the central value of each answer range. For the extreme range of €75,000+ the threshold value of 75,000 is used. For individuals who answered no to the willingness to save question, the value provided by their answer represents a negative savings amount. During the analysis the natural logarithm of this savings amount is utilized, thus all those individuals with a negative savings value are not included8_.

6 _{A response of “Dont Know” were coded as missing values} 7 _{Exact answers are provided in Table A1}

12 Finally the total amount in savings and deposit accounts is provided by the DHS data set itself in the information upon aggregated assets. The variable provides an aggregate of the balance upon an individual’s seven most important deposit and savings accounts in Euros. The natural logarithm of the total amount in savings and deposit accounts were included in the analysis in order to reduce the impact of outliers.

3.2.2 Optimism

In order to measure dispositional optimism, this study adopts two different methods utilized in previous literature. The main focus is upon general optimism (Angelini and Cavapozzi, 2017; Puri and Robinson, 2007) supplemented with investigating economic optimism (Harris et al., 2002).

General Optimism

Following the work of Puri & Robinson (2007) and Angelini & Cavapozzi (2017), dispositional optimism is measured by the discrepancy between self-assessed survival probability and actuarial survival probabilities. This is shown by equation 1.

𝑂𝑝𝑡𝑖𝑚𝑖𝑠𝑚_𝑖 = 𝐸_𝑠(𝑆𝑢𝑟𝑣𝑖𝑣𝑖𝑛𝑔 𝑡𝑜 80)_𝑖− 𝐸_𝐴(𝑆𝑢𝑟𝑣𝑖𝑣𝑖𝑛𝑔 𝑡𝑜 80)_𝑖 (1)

In order to gauge an individual’s self-assessed survival probability 𝐸𝑠(𝑆𝑢𝑟𝑣𝑖𝑣𝑖𝑛𝑔 𝑡𝑜 80)𝑖, the answer to the question “How likely is it that you will attain at least the age of 80?” is used9_{. The answers} provided are within the range 0-10. The answers are then rescaled to the interval 0 to 1 e.g. an answer of 0.3 represents a self-assed probability of 30%. Following the work of Peracchi & Perotti (2010), the actuarial survival probability 𝐸𝐴(𝑆𝑢𝑟𝑣𝑖𝑣𝑖𝑛𝑔 𝑡𝑜 80)𝑖 is constructed using data provided in the Human Mortality Database10 _{for the Netherlands. It is calculated based upon a vector of characteristics: the} individuals’ gender, their birth year and year of the DHS questionnaire.

13 In line with Angelini & Cavapozzi (2017) the variable is standardized to lie in the range 0 and 1. This is carried out in order to define clear benchmarks for the most pessimistic and most optimistic individuals (Angelini & Cavapozzi, 2017). Higher levels of this indicator therefore represent a higher level of optimism. The cumulative distribution function of this general optimism indicator is represented by

figure 1.

The discrepancy between actuarial and subjective survival probabilities may be a result of individuals possessing more accurate information regarding their longevity than demographers (Perozek, 2008), e.g. upon their life-style and family genetic disease history. Therefore, life expectancy miscalibration may be unrelated to an individual’s optimism level. Consequently Puri & Robinson (2007) conduct a variety of validity checks upon whether or not life-expectancy miscalibration captures optimism. They show that the measure correlates with positive expectations upon future economic conditions; both general macroeconomic and individual income growth. Along with this, their measure of life-expectancy miscalibration also correlates with a commonly utilized psychometric test of optimism – the Life Orientation Test (LOT) as introduced by Scheier and carver (1985). This correlation between life-expectancy miscalibration and LOT scores is also confirmed by Angelini & Cavapozzi (2017). Angelini & cavapozzi (2017) further validate life-expectancy miscalibration as a measure of optimism. They show a strong correlation with individuals believing their standard of living will improve in 5 years and more likely to rate their self-assessed health as at least good.

Following this, several tests are carried out to validate the measure of general optimism within the sample presented in this paper. Life-expectancy miscalibration is compared with outcomes related to holding a general positive attitude towards life. The DHS provides data upon individual’s expectations

14 regarding their household’s economic situation in five years’ time, their self-assessed happiness level and subjective health. In order for overestimating survival probability to reflect higher levels of general optimism, ceteris paribus, a positive correlation is expected between this variable and individuals household economic situation expectations, self-reported happiness and subjective health.

DHS respondents are asked “How do you think the economic situation of your household will be in five

years’ time in comparison to the current situation?” They are then able to choose from five answers

ranging from much worse to much better. The median response being the situation will be (about) the

same. From this a dummy variable is created, taking the value of 1 if the individual believes the

economic situation will be better or much better, and 0 otherwise. As we are interested in the direction and significance of the correlation, a simple linear probability model with robust standard errors is estimated of this outcome on the measure of general optimism and all variables included in the main analysis. The results are displayed in column 1 of Table 3. The coefficient upon the general optimism indicator is both positive and statistically significant at the 1% level. Ceteris paribus, the higher the general optimism, the higher the probability an individual will believe the economic situation of their household will be better in five years’ time as compared to now. An extremely optimistic individual is 26 percentage points more likely to believe their situation will improve as compared to an extremely pessimistic individual. Thus, this result suggests a strong correlation in the expected direction between life-expectancy miscalibration and individuals holding positive views about their future economic situation.

In line with Angelini and Cavapozzi (2017) the correlation between general optimism and self-assessed health is investigated. 77.87% of individuals in the sample categories their health in general as good or excellent rather than fair, not so good or poor. Within this regression, more objective measures of health are controlled for. The individuals Body Mass Index is calculated from their height and weight measurements11_{. A dummy variable Smoker equals 1 if the individual smokes cigarettes at all. A} dummy LIDH is created which is equal to 1 if the respondent suffers from a long illness, disorder, handicap, or if they suffer from the consequences of an accident. Lastly the number of consultations an individual has had with their general practitioner (GP) over the last year is controlled for12_{. The} results show that the correlation between general optimism and self-reported health is positive and statistically significant when objective health indicators and socioeconomic status are controlled for (see column 2). Holding everything else constant, an extremely optimistic individual is 46 percentage points more likely to rate their health as at least good compared to an extremely pessimistic individual.

11 _{The equation for calculating the BMI is provided by the CDC, for more information see}

https://www.cdc.gov/healthyweight/assessing/bmi/index.html

15 This supports the results of Angelini and Cavapozzi (2017). However, these results should be interpreted with caution due to the presence of reverse causality between subjective health and life expectancy miscalibration (Angelini & Cavapozzi, 2017).

Finally, the correlation between the general optimism indicator and self-reported happiness is investigated. In the sample 81.92% of individuals consider themselves as either a happy or very happy person, as opposed to indifferent, unhappy or very unhappy. As shown in column 3, when controlling for objective health measures and socioeconomic status, higher general optimism makes individuals more likely to consider themselves as generally happy at least. Looking at the statistically significant coefficient, extremely optimistic individuals are 33 percentage points more likely than extremely pessimistic individuals to consider themselves at least happy generally.

Table 3 Validation of the general optimism indicator.

This table presents the linear regression models estimated by OLS showing the effect of optimism upon differing outcomes associated with holding a general positive outlook on life. Standard errors are presented in brackets and the significance levels of 10%, 5% and 1% are denoted by *, ** and *** respectively.

Improved Economic Situation

Happy

Good Health

General Optimism

0.255***

0.333***

0.464***

(0.0230)

(0.0229)

(0.0225)

Household Size

-0.0170**

0.161***

0.0362***

(0.00843)

(0.00871)

(0.00730)

Number Children

0.00740

-0.151***

-0.0349***

(0.0105)

(0.0100)

(0.00885)

Time Preference

-0.0191***

0.00500

-0.00123

(0.00336)

(0.00313)

(0.00300)

Age

-0.0134***

-0.000383

-0.000262

(0.000422)

(0.000361)

(0.000346)

Retired

0.000580

0.0873***

0.0761***

(0.0118)

(0.0136)

(0.0138)

Employed

-0.0301***

0.0667***

0.0987***

(0.0112)

(0.0107)

(0.0106)

Female

-0.0376***

0.0933***

0.0866***

(0.00873)

(0.00807)

(0.00771)

High Education

0.0262

0.0596**

-0.00722

(0.0235)

(0.0237)

(0.0221)

Medium Education

-0.0224

0.0408*

-0.0361

(0.0229)

(0.0235)

(0.0220)

Financial Literacy

0.0504***

0.0632***

0.0380***

(0.00858)

(0.00687)

(0.00696)

Household Net Income

-3.92e-07*

1.13e-06***

5.49e-07***

(2.14e-07)

(1.84e-07)

(1.78e-07)

Risk Aversion

-0.00347***

0.00147***

-0.000870

(0.000637)

(0.000557)

(0.000549)

16

(0.000492)

(0.000581)

(0.000543)

Smoker

0.0264***

-0.0340***

-0.0304***

(0.00940)

(0.00885)

(0.00816)

GP Consultations

-0.000880

-0.00223**

-0.0114***

(0.000761)

(0.00110)

(0.00178)

LIDH

0.000801

-0.0423***

-0.317***

(0.00883)

(0.00908)

(0.0110)

Constant

1.008***

0.115***

0.523***

(0.0459)

(0.0440)

(0.0424)

Observations

11,111

11,318

11,632

Economic Optimism

A composite measure to reflect economic optimism is created. DHS respondents are asked what they expect to be both the lowest and highest total net yearly income their household may realize in the next 12 months. From this they are asked to provide a probability in percent that the net yearly income of their household will be less than €𝑋 in the next 12 months. The value 𝑋 is constructed by the equation shown below:

𝑋 = [𝐿𝑜𝑤𝑒𝑠𝑡 +(𝐻𝑖𝑔ℎ𝑒𝑠𝑡 − 𝐿𝑜𝑤𝑒𝑠𝑡) × 8

10 ]

From this, the probability the individual expects their household net income to be above 𝑋 is calculated13_{, giving a score for income expectation. This score is divided by 10 in order to standardize} it on a scale 0 to 1014_{. This expected income score is then combined with a value reflecting an} individual’s longer term view upon their household financial situation. This value is extracted from each individuals answer “How do you think the economic situation of your household will be in five years’ time in comparison to the current situation?” The answers equate to the following scores: much worse = 1, worse = 2, the same = 3, better = 4 and much better = 5. The mean score for the economic optimism composite score is 5.66. This compound score is standardized to lie between 0 and 1 in order to identify benchmarks for most optimistic and pessimistic individuals (Angelini & Cavapozzi, 2017). A higher score in this indicator represents an individual being more optimistic with regards to their household’s future financial situation. The cumulative distribution function of this economic optimism indicator is displayed in figure 2.

13 _{Via undertaking 100 − 𝑝(𝑋).}

17 Fig. 2. Cumulative distribution function of the economic optimism indicator

3.2.3 Control Variables

Several control variables are included in the analysis covering socioeconomic, psychological and health characteristics. The respondent’s age at the beginning of the year in which the survey is conducted is included15_{. The household size reflects the number of individuals living in the house of the respondent.} The number of children in the household is also controlled for, as having dependents like children is likely to reduce the ability to save. A binary variable indicting whether the respondent is female (1) or not (0) is included, as gender has been shown to impact savings decision (Bucciol and Veronesi, 2014). The highest level of education an individual has achieved is also controlled for. The dummy variable High Education takes the value of 1 when the individual has completed university or vocational college and 0 otherwise. Medium Education takes the value of 1 if the highest education the individual has completed is pre-vocational education or senior vocational training or pre-university education and 0 otherwise. Household net income is calculated from a 7 answer categorical question where the middle value of each range is used. The dummy variable employed takes the value of 1 if the primary occupation of the individual is employed on a contractual basis/works in own business or

18 employed and 0 otherwise. The dummy variable retired takes the value 1 if the individual’s primary occupation is that of retired and takes the value 0 otherwise.

Psychological control variables shown to have an impact on savings behavior (Bucciol and Veronesi, 2014) are also included. The variable financial literacy takes the value 1 if the respondent answered

knowledgeable or very knowledgeable to the question “How knowledgeable do you consider yourself with respect to financial matters?” and 0 otherwise. The time horizon variable is ordinal on a scale 1

to 5 reflecting the answers to the question “Which of the periods mentioned below is in your household most important with regard to planning expenditures and savings?” The value increases from 1 for an answer of the next couple of months to 5 for more than 10 years from now. A composite measure for risk aversion is created from seven questions measuring the respondents risk appetite. A higher measure reflects a more risk averse individual.

4. Methodology

The objective of this paper is to investigate what impact optimism has upon household saving behavior. Saving behaviour is measured as (i) the propensity to save over the last 12 months, (ii) the propensity to save in the future, (iii) the amount saved in a given year and (iv) the amount saved in deposit and savings accounts. Due to the nature of these differing household saving behaviors – two binary variables and two continuous variables -different forms of estimators are adopted for suitable analysis.

4.1 Dichotomous Dependent Variable Models

Both the preference to save and preference to save in the future are dichotomous variables. The methodology used to investigate the impact of optimism upon these variables therefore is a regression analysis of a binary outcome model. The dependent variable in this model is a binary outcome response 𝑦𝑖,𝑡. When analyzing an individual’s preference to save this variable takes the value 1 if individual 𝑖 reports they have saved over the previous 12 months in survey year 𝑡, or 0 if not. When analyzing an individual’s preference to save in the future, 𝑦𝑖,𝑡 takes the value 1 if individual 𝑖 answers they intend to save over the next year in survey year 𝑡. The likelihood of an individual having saved or planning to save in survey year 𝑡 is thus measured by an underlying latent variable 𝑦𝑖,𝑡∗ . This likelihood of an individual having saved over the previous 12 months or planning to save in the upcoming year is approximated by the latent variable model 𝑦𝑖,𝑡= 1 [𝑦𝑖,𝑡∗ > 0], t = 2004,2005,…,2016. It is hypothesized 𝑦_𝑖,𝑡∗ _{is a function of the variable of interest; optimism}16_{, and a vector of control variables Χ}

𝑖,𝑡. Further, a linear regression model is specified for the latent response 𝑦𝑖,𝑡∗ . Specifically, the multivariate analysis

19 estimates variations of the random effect model shown in equation (2). A logistic distribution is assumed for the estimators and thus a logit regression is carried out when estimating all forms of the following model:

𝑦

_𝑖,𝑡∗

= 𝛼 + 𝛿𝑂𝑝𝑡𝑖𝑚𝑖𝑠𝑚

_𝑖,𝑡

+ 𝛽Χ

_𝑖,𝑡

+ 𝛽

₂

𝑇

_𝑖,𝑡

+ 𝜐

_𝑖

+ 𝜖

_𝑖,𝑡

(2)

Equation (2) includes as determinants for the preference to save a measure of optimism (𝑂𝑝𝑡𝑖𝑚𝑖𝑠𝑚𝑖,𝑡) and a vector of socioeconomic and psychological characteristics17 _(Χ

𝑖,𝑡). When the life-expectancy miscalibration measure of optimism is utilized, objective health covariates are also included in (Χ𝑖,𝑡)18. Further, it is likely that the state of the national economy may lead households to save more in some years compared to others e.g. in the years immediately post global financial crisis, therefore 𝑇𝑖,𝑡 represents a vector of year dummy variables to capture heterogeneity over time (Bucciol and Veronesi, 2014). In equation (2)

𝜐

𝑖 represents the independent time invariant household-specific error and 𝜖𝑖,𝑡 represents the independent and identically distributed error within and across individual households (Bucciol and Veronesi, 2014). A random effects estimator is preferred to an Ordinary Least Squares (OLS) as the result of Breusch-Pagan Lagrange Multiplier Test (Breusch and Pagan, 1980) is significant, indicating significant difference exists across the individuals.

An issue with estimating the random effects model is the assumption that both varying and time-invariant independent variables are exogenous with respect to the error term (Bucciol and Veronesi, 2014). When this assumption does not hold the estimated coefficients shall be biased stemming from omitted variable bias (Bucciol and Veronesi, 2014). Utilizing a fixed-effect estimator would overcome this issue. A fixed-effect estimator however, would present other potential issues. Households which do not experience a preference to save in a given year or a preference to save in the future will be dropped from the estimator as they would exhibit no variance within the dependent variable. Consequently the results of the study would be conditional upon a household having undertaken, or planning to undertake saving; this is non-desirable. Therefore a random effects model is chosen. As a robustness test of the results from random-effect logistic estimator a random effect GLS model is also estimated. The results of these estimations are provided in the appendix.

17 _{This includes Household size, Number of children, Education level achieved, Whether the individual is retired,} Whether the individual is employed, Household net income, Age, Time horizon, Gender, Financial Literacy and Risk aversion.

20

4.2 Continuous Dependent Variable Models

In order to investigate the impact optimism has upon the total amount saved in a given year and the total amount held in deposit and savings accounts, the following equation is estimated:

𝑆

_𝑖,𝑡

= 𝛼 + 𝛾𝑂𝑝𝑡𝑖𝑚𝑖𝑠𝑚

_𝑖,𝑡

+ 𝛽𝑋

_𝑖,𝑡

+ 𝑇

_𝑖,𝑡

+ 𝜀

_𝑖,𝑡 (3)

Where 𝑆𝑖,𝑡 represents the total amount saved in a given year or the total amount held in deposit and saving accounts; Χ𝑖,𝑡 vector of socioeconomic, psychological and objective health control variables, 𝑇_𝑖,𝑡 a vector of year dummy variables and

𝜀

𝑖,𝑡

is the error term.

Initially Eq. (3) is estimated by OLS. However the undertaking of a Breusch and Pagan Lagrangian multiplier test indicates the presence of significant differences across units. Following this, the results of the Hausman test (p-value 0.000) for each specification indicates correlation between the unique error term and the regressors. A fixed-effect estimator is therefore appropriate and estimated in the following analysis. Eq. (3) is reformulated as follows:

𝑆_𝑖,𝑡

= 𝛼 + 𝛾𝑂𝑝𝑡𝑖𝑚𝑖𝑠𝑚

_𝑖,𝑡

+ 𝛽𝑋

_𝑖,𝑡

+ 𝑇

_𝑖,𝑡

+ 𝑢

_𝑖

+ 𝜔

_𝑖,𝑡 (4)

Here 𝑢𝑖 are individual fixed effects controlling for time-invariant individual characteristics

5. Results

The regression results are presented in four sections, with each presenting the results for a different proxy for saving behavior.

5.1 Preference to Save

Table 4 reports the results from the random effect logistic model with the two distinct measures of optimism estimated. All results are presented in the form of exponentiated coefficents (odds ratios)19 to allow for an easier interpretation of each variables impact upon the probability of a household saving over the previous 12 months. Columns 1 – 4 report the results for general optimism i.e. optimism measured as life expectancy miscalibration. Columns 5 – 7 report the results for economic optimism i.e. a composite score of a households expectations for future income and household economic situation.

Initially looking at the general optimism indicator, the first column displays a parsimonious specification that only controls for basic socioeconomic and demographic characteristics such as age, gender and household net income. The result upon the optimism indicator is statistically significant at

19 _{The odds ratios are expressed as} 𝑝

1−𝑝 where 𝑝 = Pr(𝑦 = 1|𝑋) being the probability a household saved over

21 the 1% level with a positive coefficient greater than unity. This indicates an extremely optimistic individual has odds 2.194 times greater of saving over the previous 12 months than that of an extremely pessimistic individual; thus a greater preference for saving. This optimism coefficient remains statistically significant and greater than unity when other psychological variables are introduced (column 2).

General Optimism Economic Optimism

(1) (2) (3) (4) (5) (6) (7) General Optimism 2.194*** 2.226*** 1.944*** 2.095*** (0.565) (0.568) (0.499) (0.554) Age 0.976*** 0.971*** 0.972*** 0.968*** 0.978*** 0.973*** 0.969*** (0.00512) (0.00506) (0.00508) (0.00525) (0.00510) (0.00505) (0.00521) Female 1.724*** 1.647*** 1.654*** 1.711*** 1.626*** 1.555*** 1.592*** (0.221) (0.210) (0.210) (0.219) (0.205) (0.195) (0.201) Household Size 1.417*** (0.147) 1.380*** (0.141) 1.340*** 1.243** 1.430*** 1.393*** 1.290** (0.137) (0.128) (0.149) (0.143) (0.134) Number Children 0.569*** 0.582*** 0.592*** 0.629*** 0.560*** 0.572*** 0.608*** (0.0710) (0.0717) (0.0728) (0.0775) (0.0703) (0.0709) (0.0756) High Education 2.229** 2.036** 1.853* 1.672 2.342*** 2.167** 1.940** (0.728) (0.651) (0.591) (0.536) (0.755) (0.684) (0.615) Medium Education 1.815* 1.768* 1.690* 1.642 1.914** 1.888** 1.824* (0.582) (0.555) (0.528) (0.514) (0.605) (0.584) (0.565) Employed 3.005*** 2.934*** 2.799*** 2.569*** 3.137*** 3.058*** 2.807*** (0.368) (0.355) (0.340) (0.315) (0.388) (0.374) (0.347) Retired 2.652*** 2.639*** 2.493*** 2.460*** 2.730*** 2.731*** 2.701*** (0.422) (0.416) (0.393) (0.390) (0.429) (0.425) (0.423) Household Net Income 1.000*** 1.000*** 1.000*** 1.000*** 1.000*** 1.000*** 1.000***

(1.96e-06) (1.95e-06) (1.95e-06) (2.58e-06) (1.95e-06) (1.94e-06) (2.59e-06) Time Horizon 1.326*** 1.321*** 1.321*** 1.329*** 1.326*** (0.0448) (0.0445) (0.0449) (0.0448) (0.0450) Risk Aversion 1.023*** 1.023*** 1.021*** 1.022*** 1.020*** (0.00706) (0.00704) (0.00708) (0.00702) (0.00705) Financial Literacy 1.193* 1.183* 1.172* 1.173* 1.162* (0.108) (0.106) (0.106) (0.105) (0.104) Bmi 0.990* 0.989* (0.00624) (0.00628) Smoker 0.726*** 0.733*** (0.0821) (0.0834) GP Consultations 0.985* 0.986 (0.00886) (0.00889) LIDH 0.826* 0.825** (0.0809) (0.0811) Economic Optimism 1.262* 1.305* 1.131 (0.172) (0.177) (0.180) Constant 0.788 0.266** 0.515 0.511 0.974 0.336** 0.391* (0.393) (0.141) (0.288) (0.291) (0.463) (0.170) (0.202)

22 As previously discussed, a potential criticism against utilizing life expectancy miscalibration as a measure of optimism is that the miscalibration may simply be a product of the individual possessing greater information about health and genetics. Column 3 addresses this issue through the introduction of four objective health measures. These include the individuals BMI score, whether the individual is a smoker, the number of consultations an individual has had with their GP over the past year and an indicator of whether the individual suffers from a long illness/disorder/handicap or the consequences of an accident. The results in column 3 show that optimism still plays a significant role in explaining the preference of an individual to save over the past 12 months. Column 4 introduces year dummies in order to control for time heterogeneity. The statistically significant impact of optimism increasing the odds an individual prefers to save, remains. This result is further confirmed through estimating the preference to save variable via a random effect generalized least squares model with the same specification as column 4 (table A2). Moving from the lowest to the highest level of optimism is associated with an increase in the preference to save of 8.13 percentage points (column 1).

Moving from the lowest to the highest level of economic optimism increases the odds an individual prefers to save by 1.305 times; controlling for both socioeconomic and psychological covariates. This result is significant at the 10% level. However when year fixed effects are controlled for, the positive coefficient losses its statistical significance. This result is also reflected in the random effect GLS model20 _{where the coefficient is positive yet non-significant (table A2). This result may be due to the} state of the economy overall capturing the impact of individual household optimism.

Table 4 also displays that the control variables of financial literacy, time horizon and risk aversion significantly impact saving preference. An increase in all of these indicators induce a higher probability of holding a preference for saving. This is in line with the results found previously by Jappelli and Padula (2013), Fisher and Monalto (2010) and Bucciol and Veronesi (2014).

5.2 Preference to save in future

This section provides the results for the desire to save in the upcoming year as a proxy for the preference to save in the future. Table 5 reports the exponentiated coefficents from the random effect

20 _{The results of this estimation are presented in the appendix.}

Observations 11,629 11,629 11,629 11,629 11,757 11,757 11,757

23 logistic model. Columns 1 – 4 report the results for general optimism, measured as life expectancy miscalibration. Columns 5 – 7 report the results for economic optimism.

Looking at the most parsimonious specification in column 1, moving from the lowest to the highest level of optimism increases the odds an individual desires to save in the future by 2.547 times. This result is statistically significant at the 1% level. When the full set of control variables are introduced, including controlling for year fixed effects (column 4) the coefficient upon the optimism indicator remains significant. The coefficient increases from 2.547 to 2.670. Again, this result is further confirmed through estimating the preference to save in the future variable via a random effect generalized least squares model with the same specification as column 4 (table A3). Here moving from the lowest to highest level of optimism results in an increase in the probability of planning to in the future by 9.15 percentage points. Risk aversion and time horizon remain important determinants of the desire to save in the future. The results show that optimistic individuals are more likely to desire to save in the near future.

If we look at columns 5 to 7 the coefficient upon the economic optimism indicator is greater than unity. However the coefficient is only statistically significant for the full specification (column 7). The result shows that moving from the least to most optimistic views upon the future economic situation of the household increases the odds an individual is planning to save by 1.345 times. This result is confirmed when the model is estimated by a random effect GLS estimator as opposed to the logistic regression (table A3). Here extreme optimism is associated with an increase in the probability of desiring to save in the future of 3.12 percentage points when compared to extreme pessimism.

5.3 Total Amount Saved In a Given Year

Initially Eq. (3); with 𝑆𝑖,𝑡 being the total amount saved in a given year, is estimated by Pooled OLS. These results are displayed in columns 1-4 of Table 6 for general optimism and columns 1-3 of Table 7 for economic optimism. The results for the general optimism indicator are discussed first (Table 6). In all specifications, from the most parsimonious (column 1) to the full specification including year fixed effects (column 4) the coefficient upon the general optimism indicator is negative. Only in the specification with the full set of control variables and no year fixed effects is the coefficient statistically significant. Here moving from the lowest to highest level of optimism is associated with a reduction in the total amount saved over the previous 12 months of 16.1%.

24 a fixed-effect or random-effect model is more appropriate. The null hypothesis of the test, that the preferred model is random-effetcs (Greene, 2008), is rejected. Hence a fixed-effect model is chosen. The regression results from the fixed-effect model are reported in columns (5) to (8) of Table 6. The coefficient upon the general optimism indicator is now positive, suggesting an increase in the level of optimism increases the total value of savings over the previous 12 months. However, in none of the specifications is this coefficient statistically significant. This implies that given an individual has saved in a given year, the level of optimism has no impact upon the total amount saved. Individuals who are employed save a greater amount than those who are not as do those who hold a longer time horizon.

General Optimism Economic Optimism

(1) (2) (3) (4) (5) (6) (7) General Optimism 2.547*** 2.602*** 2.282*** 2.670*** (0.698) (0.708) (0.626) (0.753) Age 0.959*** 0.956*** 0.956*** 0.952*** 0.961*** 0.957*** 0.952*** (0.00548) (0.00545) (0.00548) (0.00564) (0.00544) (0.00541) (0.00558) Female 1.980*** 1.819*** 1.827*** 1.858*** 1.842*** 1.698*** 1.696*** (0.262) (0.241) (0.242) (0.247) (0.239) (0.220) (0.221) Household Size 1.455*** 1.437*** 1.411*** 1.323** 1.496*** 1.476*** 1.370*** (0.161) (0.158) (0.155) (0.147) (0.165) (0.162) (0.151) Number Children 0.566*** 0.575*** 0.581*** 0.613*** 0.548*** 0.557*** 0.592*** (0.0759) (0.0764) (0.0772) (0.0818) (0.0734) (0.0740) (0.0790) High Education 1.447 1.394 1.284 1.147 1.439 1.392 1.220 (0.486) (0.462) (0.426) (0.383) (0.475) (0.454) (0.400) Medium Education 1.368 1.350 1.295 1.231 1.340 1.327 1.252 (0.451) (0.439) (0.420) (0.401) (0.434) (0.423) (0.401) Employed 3.636*** 3.581*** 3.421*** 3.193*** 3.631*** 3.574*** 3.309*** (0.467) (0.456) (0.439) (0.414) (0.469) (0.459) (0.428) Retired 2.591*** 2.584*** 2.480*** 2.412*** 2.640*** 2.641*** 2.557*** (0.422) (0.417) (0.401) (0.391) (0.424) (0.420) (0.407) Household Net Income 1.000*** 1.000*** 1.000*** 1.000*** 1.000*** 1.000*** 1.000***

(2.17e-06) (2.16e-06) (2.16e-06) (2.83e-06) (2.15e-06) (2.15e-06) (2.84e-06) Time Horizon 1.191*** 1.186*** 1.178*** 1.183*** 1.174*** (0.0438) (0.0435) (0.0435) (0.0433) (0.0432) Risk Aversion 1.028*** 1.028*** 1.028*** 1.027*** 1.026*** (0.00762) (0.00761) (0.00765) (0.00754) (0.00759) Financial Literacy 0.975 0.969 0.939 0.985 0.950 (0.0950) (0.0944) (0.0921) (0.0947) (0.0920) Bmi 0.988* 0.988* (0.00623) (0.00629) Smoker 0.837 0.861 (0.102) (0.105) GP Consultations 0.985* 0.986 (0.00872) (0.00867) LIDH 0.855 0.859 (0.0879) (0.0886)

Table 5 Regression results for the effect of optimism upon the preference to save over the next year.

25 Economic Optimism 1.086 1.109 1.345* (0.161) (0.164) (0.232) Constant 6.271*** 2.225 4.144** 4.050** 9.666*** 3.624** 3.705** (3.316) (1.254) (2.473) (2.472) (4.870) (1.955) (2.046) Observations 11,489 11,489 11,489 11,489 11,636 11,636 11,636 Year Fixed Effects NO NO NO YES NO NO YES

OLS Fixed Effect

(1) (2) (3) (4) (5) (6) (7) (8) General Optimism -0.0699 -0.0703 -0.161*** -0.0688 0.00478 0.0125 0.00523 0.0450 (0.0636) (0.0621) (0.0625) (0.0623) (0.0911) (0.0910) (0.0913) (0.0946) Age 0.00618*** 0.00436*** 0.00492*** 0.00294*** 0.0132*** 0.0116*** 0.0121*** 0.0111*** (0.00105) (0.00104) (0.00104) (0.00103) (0.00331) (0.00332) (0.00337) (0.00415) Female -0.239*** -0.199*** -0.191*** -0.181*** (0.0224) (0.0225) (0.0225) (0.0221) Household Size 0.210*** 0.195*** 0.179*** 0.0952*** 0.143*** 0.141*** 0.140*** 0.120** (0.0226) (0.0221) (0.0221) (0.0222) (0.0487) (0.0487) (0.0487) (0.0486) Number Children -0.284*** -0.273*** -0.266*** -0.184*** -0.161*** -0.159*** -0.159*** -0.144*** (0.0277) (0.0271) (0.0270) (0.0268) (0.0540) (0.0539) (0.0540) (0.0538) High Education 0.425*** 0.355*** 0.332*** 0.270*** -0.0643 -0.0609 -0.0563 -0.0602 (0.0680) (0.0664) (0.0661) (0.0650) (0.189) (0.189) (0.189) (0.188) Medium Education 0.120* 0.0926 0.0907 0.0864 0.0518 0.0553 0.0616 0.0587 (0.0675) (0.0659) (0.0655) (0.0642) (0.183) (0.182) (0.183) (0.182) Employed 0.276*** 0.246*** 0.204*** 0.129*** 0.180*** 0.182*** 0.181*** 0.170*** (0.0309) (0.0302) (0.0304) (0.0302) (0.0524) (0.0523) (0.0524) (0.0523) Retired 0.120*** 0.116*** 0.0770** 0.0445 0.0193 0.0198 0.0195 0.0295 (0.0400) (0.0390) (0.0391) (0.0384) (0.0637) (0.0636) (0.0636) (0.0634) Household Net Income

1.14e-05*** 1.06e-05*** 1.06e-05*** 1.81e-05*** 3.32e-06*** 3.26e-06*** 3.26e-06*** 6.68e-06***

(5.60e-07) (5.49e-07) (5.46e-07) (6.80e-07) (5.99e-07) (6.01e-07) (6.01e-07) (7.92e-07)

Time Horizon 0.150*** 0.145*** 0.137*** 0.0401*** 0.0401*** 0.0435*** (0.00902) (0.00899) (0.00886) (0.0102) (0.0102) (0.0102) Risk Aversion -0.00363** -0.00363** -0.00348** 0.00526** 0.00514** 0.00471* (0.00165) (0.00164) (0.00162) (0.00242) (0.00242) (0.00243) Financial Literacy 0.201*** 0.196*** 0.165*** 0.0457 0.0456 0.0484* (0.0214) (0.0213) (0.0210) (0.0286) (0.0286) (0.0285) Bmi -0.00425*** -0.00438*** -0.00111 -0.00118 (0.00162) (0.00159) (0.00196) (0.00196) Smoker -0.0921*** -0.0764*** 0.0363 0.0387 (0.0243) (0.0239) (0.0504) (0.0502) GP Consultations -0.0180*** -0.0165*** -0.00416 -0.00436 (0.00284) (0.00279) (0.00376) (0.00374) LIDH -0.112*** -0.108*** -0.00967 -0.0127 (0.0245) (0.0241) (0.0330) (0.0329) Constant 6.813*** 6.678*** 6.996*** 6.800*** 7.077*** 6.866*** 6.882*** 6.818*** (0.107) (0.115) (0.124) (0.126) (0.264) (0.276) (0.282) (0.306) Observations 8,063 8,063 8,063 8,063 8,063 8,063 8,063 8,063 Year Fixed Effects NO NO NO YES NO NO NO YES

FE NO NO NO NO YES YES YES YES

Table 6 Regression results for the effect of general optimism upon the natural logarithm of the total amount saved over the previous year.

26 Table 7 reports the results for estimating Eq. (3); with 𝑆𝑖,𝑡 being the total amount saved in a given year, and 𝛾𝑂𝑝𝑡𝑖𝑚𝑖𝑠𝑚𝑖,𝑡 being economic optimism. Columns (1) to (3) show the results when the model is estimated via Pooled OLS. The coefficient upon the optimism indicator is negative across all specifications, and statistically significant in the most parsimonious specification (column 1). Here moving from the lowest to highest level of economic optimism is associated with a 9.11% decrease in the amount saved in a given year. The results of a Bresuch-Pagan Lagrangian multiplier test followed by a Hausman test indicate a fixed-effect model is most appropriate for the estimation. The regression results from the FE model are reported in columns (4) to (6) of Table 7.

Table 7 Regression results for the effect of economic optimism upon the natural logarithm of the total amount saved over the previous year.

This table presents the results of the Pooled OLS and fixed effect regressions for the continuous dependent variable of the total amount saved over the previous year. Optimism is measured as a composite score of economic optimism – standardized between 0 and 1. The dependent variable being the natural logarithm of the total amount saved over the previous 12 months by the household. Significance levels of 10%, 5% and 1% are denoted by *, ** and *** respectively.

OLS Fixed Effect

(1) (2) (3) (4) (5) (6) Economic Optimism -0.0911** -0.0552 -0.00849 -0.0501 -0.0453 -0.108** (0.0422) (0.0412) (0.0475) (0.0391) (0.0391) (0.0448) Age 0.00597*** 0.00420*** 0.00209** 0.0116*** 0.0103*** 0.0109*** (0.00105) (0.00104) (0.00103) (0.00335) (0.00336) (0.00406) Female -0.229*** -0.191*** -0.188*** (0.0215) (0.0217) (0.0213) Household Size 0.193*** 0.179*** 0.0908*** 0.131*** 0.128*** 0.105** (0.0224) (0.0220) (0.0221) (0.0487) (0.0486) (0.0485) Number Children -0.266*** -0.255*** -0.171*** -0.150*** -0.148*** -0.130** (0.0275) (0.0270) (0.0268) (0.0540) (0.0539) (0.0538) High Education 0.390*** 0.326*** 0.265*** -0.0694 -0.0672 -0.0732 (0.0669) (0.0654) (0.0642) (0.190) (0.190) (0.189) Medium Education 0.0935 0.0705 0.0708 0.0447 0.0489 0.0439 (0.0663) (0.0648) (0.0635) (0.183) (0.183) (0.182) Employed 0.273*** 0.244*** 0.162*** 0.177*** 0.178*** 0.165*** (0.0309) (0.0303) (0.0300) (0.0530) (0.0529) (0.0528) Retired 0.108*** 0.107*** 0.0740** -0.000104 -0.000622 0.00941 (0.0390) (0.0380) (0.0374) (0.0634) (0.0633) (0.0631) Household Net Income 1.16e-05*** 1.09e-05*** 1.87e-05*** 3.39e-06*** 3.33e-06*** 6.89e-06***

27 In the full specification including year fixed effects (column 6) the coefficient upon the economic optimism indicator is negative and statistically significant at the 5% level. This result shows moving from the lowest to the highest level of economic optimism is associated with a reduction in the total amount saved of 10.8%, given that the individual has saved. This result suggests when an individual is optimistic about the future financial situation of their household, with regards to income expectations and living better off, they will save a significant amount less. This follows the consumption smoothing theory as an individual will seek to consume more now if they believe they will have increased income in the future.

5.4 Total Balance of Saving & Deposit Accounts

Initially Eq. (3); with 𝑆𝑖,𝑡 being the total balance in an individual’s seven most important savings and deposit accounts, is estimated by Pooled OLS. These results are displayed in columns 1-4 of Table 8 for general optimism and columns 1-3 of Table 9 for economic optimism.

Looking at Table 8 in all specifications of the model the coefficient upon the general optimism indicator is negative. This coefficient is statistically significant at the 1% level for all specifications. Looking at the specification with the full set of controls and no year fixed effects (column 3) an increase in the level of optimism is associated with a reduction in the total balance of 9.78%. When year fixed effects are introduced (column 4) this coefficient reduces to 8.81%.

A Breusch and Pagan Lagrangian multiplier test is carried out. The result of which being that a Pooled OLS regression is not appropriate. Following this, a Hausman test is performed in order to examine whether a fixed-effect or random-effect model is more appropriate. The null hypothesis of the test is rejected and thus a fixed-effect model is chosen. The results of which are displayed in columns (5) to (8) of Table 8.

When estimating the model controlling for FE the coefficient upon general optimism remains negative, until year dummies are also included (column 8). However in none of the specifications is this coefficient statistically significant. This implies that moving from the lowest level of optimism to the highest does not lead to a significant impact upon the total balance of an individual’s seven most important deposit & savings accounts. The results imply that even if an individual is optimistic with regards to life expectancy this plays no significant role in determining how much to save in deposit and savings accounts. The implication of this is that the pre-cautionary motive for saving in case of a bad

Year Fixed Effects NO NO YES NO NO YES

28 event in the future does not play a significant role in this sample. If the motive were to be strong optimistic individuals, who underweight the probability of bad outcomes impacting them, would be likely to hold less in their savings accounts.

Table 9 reports the results for estimating Eq. (3); with 𝑆𝑖,𝑡 being the total balance on an individual’s seven most important savings & deposit accounts and 𝛾𝑂𝑝𝑡𝑖𝑚𝑖𝑠𝑚𝑖,𝑡 being economic optimism. Columns (1) to (3) show the results when the model is estimated via Pooled OLS. The coefficient upon the optimism indicator is negative across all specifications, and statistically significant in both the most parsimonious specification (column 1) and full specification without year fixed effects (column 2). In the full specification (column 2) moving from the lowest to highest level of economic optimism is associated with a 2.67% decrease in the total balnce upon the savings & deposit accounts.

The results of a Bresuch-Pagan Lagrangian multiplier test followed by a Hausman test indicate a fixed-effect model is most appropriate for the estimation. The results of the FE model are presented in columns (4) to (6). In all specifications the coefficient upon the economic optimism indicator are positive, yet not statistically significant. The result implies that when determining how much money to hold in savings 7 deposit accounts, an individual’s level of economic optimism plays no significant role.

OLS Fixed Effect

(1) (2) (3) (4) (5) (6) (7) (8) General Optimism -0.738*** -0.734*** -0.978*** -0.881*** -0.134 -0.123 -0.119 0.0129 (0.123) (0.120) (0.121) (0.122) (0.135) (0.135) (0.135) (0.140) Age 0.0359*** 0.0301*** 0.0315*** 0.0284*** 0.0484*** 0.0466*** 0.0478*** 0.0551*** (0.00208) (0.00206) (0.00205) (0.00206) (0.00490) (0.00493) (0.00498) (0.00627) Female -0.186*** -0.208*** -0.219*** -0.221*** (0.0437) (0.0441) (0.0437) (0.0435) Household Size 0.248*** 0.210*** 0.171*** 0.0678 0.128* 0.127* 0.127* 0.0980 (0.0431) (0.0423) (0.0421) (0.0428) (0.0670) (0.0669) (0.0669) (0.0669) Number Children -0.306*** -0.268*** -0.252*** -0.152*** -0.146* -0.145* -0.144* -0.122 (0.0533) (0.0522) (0.0517) (0.0521) (0.0744) (0.0744) (0.0744) (0.0743) High Education 1.224*** 1.091*** 1.018*** 0.943*** -0.103 -0.104 -0.0943 -0.0955 (0.131) (0.128) (0.127) (0.126) (0.295) (0.295) (0.295) (0.294) Medium Education 0.631*** 0.580*** 0.573*** 0.571*** -0.227 -0.227 -0.221 -0.192 (0.130) (0.127) (0.125) (0.125) (0.282) (0.281) (0.281) (0.281) Employed 0.418*** 0.373*** 0.266*** 0.168*** 0.0946 0.0944 0.0914 0.0531 (0.0574) (0.0560) (0.0567) (0.0569) (0.0701) (0.0700) (0.0701) (0.0701) Retired 0.426*** 0.427*** 0.321*** 0.279*** 0.223** 0.226*** 0.226*** 0.208** (0.0757) (0.0739) (0.0738) (0.0734) (0.0865) (0.0864) (0.0864) (0.0863) Household Net Income 8.00e-06*** 7.07e-06*** 6.85e-06*** 1.61e-05***

8.16e-07 7.32e-07 7.39e-07 5.08e-06***

Table 8 Regression results for the effect of general optimism upon the natural logarithm of the total balance of an individual’s seven most important savings and deposit accounts.