Determinants of household savings
An international cross-country analysis to detect the determinants of household savings
Cajsa Fredriksson
Examensarbete - Civilekonomprogrammet Master of Science in Business and Economics
D-Thesis
Term: Spring 2020 Supervisor: Klaas Staal
II Acknowledgements
I would like to acknowledge the help from my peers that has been participating in all the online seminars and has been giving helpful advice and input into the work of this paper. But also, a great thanks to Jesper Huric Larsen, Henrik Jaldell, Karl-Markus Modén and my supervisor Klaas Staal for sharing your expertise and giving me a good guidance in the process of writing my first larger thesis.
Cajsa Fredriksson 2020-06-09 cajsa_fredriksson@hotmail.com
III Abstract
The purpose of this paper is to look into the determinants of household savings in an international cross-section. The focus is on the effects from social security, old-age dependency, participation rate and change in unemployment, among other variables as an addition to the disequilibrium saving hypotheses, which is the base theory for the savings function. The fixed-effect least square dummy variable method is used on panel data of 14 OECD countries over the time-span 2000 to 2018.
The determinants that has a significant effect on household saving in the empirical result is unanticipated income; a positive sign supports the permanent-income hypothesis and the disequilibrium saving hypothesis. This means that individuals tend to save the transitory income. The next significant variable is the lagged savings rate, which indicates inactivity in the savings behavior. The change in the unemployment rate is also significant and the positive sign supports the uncertainty hypothesis, indicating that individuals tend to save for
precautionary reasons. The last significant variable was social security and it had a negative effect on household savings; which is supported by the life-cycle hypothesis, and can indicate a wealth substitution effect or general confidence in the social security system.
Keywords: Determinants of household savings, household savings, cross-country, social security, participation rate, old-age dependency rate, panel data, OECD.
IV Table of contents:
1. Introduction ... 1
2. Theory ... 5
2.1 Life-cycle model ... 5
2.2 Wealth substitution effect ... 5
2.3 Permanent-Income hypotheses ... 6
2.4 Precautionary saving ... 7
2.5 Bequest motives ... 7
2.6 Ricardian equivalence ... 7
2.7 Psychological factors ... 8
2.8 Financial literacy ... 9
2.9 Capital accumulation and economic growth ... 9
3. Literature review ... 11
4. Stylized facts on some regressors, demographic factors & replacement rate etc. ... 14
5. Data ... 20
5.1 Household savings rate ... 20
5.2 Unanticipated inflation ... 20
5.3 Unanticipated income ... 21
5.4 Interest rate ... 21
5.5 Social security ... 22
5.6 Dependency rate ... 22
5.7 Unemployment rate ... 23
5.8 Participation rate ... 23
5.9 Descriptive statistics ... 24
6. Methodology... 25
7. Empirical results ... 28
8. Discussion ... 32
9. Robustness checks ... 34
9.1 Heteroscedasticity robust standard errors ... 34
9.2 Other determinants and transformations ... 35
9.3 Test for cross-sectional dependence ... 35
9.4 Panel unit root ... 36
10. Conclusion ... 38
11. Summary ... 41
Appendix ... 42
Appendix 1: Data description ... 42
V
Appendix 2: Hausman test of endogeneity ... 43
Appendix 3: The Durbin-Watson h-statistics calculations ... 43
Appendix 4: Breusch-Pagan-Godfrey test ... 44
Appendix 5: Graphical test of heteroscedasticity ... 46
Appendix 6: Pesaran CD test for cross-sectional dependence ... 47
Appendix 7: Transformations and other determinants ... 48
Appendix 8: Test of normality ... 50
12. References ... 51
List of figures: FIGURE 1: Household savings during 2000-2018 ... 14
FIGURE 2: Social security as percentage of GDP during 2000-2018 ... 15
FIGURE 3: Old-age dependency during 2000-2018 ... 15
FIGURE 4: Participation rate of people over 65, during 2000-2018 ... 16
FIGURE 5: Net replacement rate as a percentage of pre-retirement earnings 2018. ... 17
FIGURE 6: Aggregate replacement ratio for pensions, 2010-2018 ... 18
FIGURE 7: Life expectancy after 65 for the OECD sample; 2000-2017... 19
FIGURE 8: Test for cross-sectional dependence ... 36
FIGURE 9: Scatterplot of residuals and predicted values, OLS [1]. ... 46
FIGURE 10: Scatterplot of residuals and predicted values, OLS [2]. ... 47
FIGURE 11: Scatterplot of residuals and predicted values, OLS [3]. ... 47
FIGURE 12: Scatterplot of residuals and predicted values, 2SLS [4]. ... 47
FIGURE 13: Normality plot and boxplot ... 50
List of tables: TABLE 1: Table over the descriptive statistics ... 24
TABLE 2: Table of estimations ... 28
TABLE 3: Table of signs; theory, previous studies and own estimations... 32
TABLE 4: Estimations with heteroscedasticity-robust standard errors ... 34
TABLE 5: Test of unit root in panel data ... 37
TABLE 6: Table of data description ... 42
TABLE 7: Hausman test of endogeneity ... 43
TABLE 8: Breusch-pagan and Koenker test, OLS [1] ... 45
TABLE 9: Breusch-pagan and Koenker test, OLS [2] ... 45
TABLE 10: Breusch-pagan and Koenker test, OLS [3] PR ... 46
TABLE 11: Breusch-pagan and Koenker test, 2SLS ... 46
TABLE 12: Test results and p-value Pesaran CD ... 48
TABLE 13: CPI as a measure of inflation. ... 48
TABLE 14: OLS regression of [1], [2] and [4] with added public budget balance ... 49
VI Glossary of terms
Organization for Economic Co-operation and Development (OECD)
The organization for economic co-operation and development is an organization that has a goal to create evidence-based solutions to different economic and societal challenges. OECD has 37 member countries and cooperates with different organizations to create policies or collect data for different analysis (OECD 2020).
Pay-as-you-go (PAYGO)
Pay-as-you-go is a pension system that funds current retirees based on payments from current workers, it is also known as an unfunded plan. After the great depression, there was an idea that savings was greatly diminished for a lot of people and the thought was that they deserve to have a better support as they grow older. The alternative fully founded system was also criticized by politician, that thought that the accumulated money would not be managed properly and could go to other things than pensions and therefore PAYGO was introduced.
Social security today is often a partially founded system, with some of the taxes used and some surplus is collected in a trust fund. The PAYGO financing is therefore an important component (Rosen and Gayer 2014:226-227).
Dummy variable
Different attributes as something belonging to one country or not; as in this paper, could be quantified with a value of 1 if it has the attribute or 0 if not. The type of variables that has this binary character are called dummy variables and is used in statistics to classify data into different exclusive attributes (Gujarati & Porter 2009:277-278).
Heteroscedasticity
An assumption that is important for a classic linear model is that the error terms are
homoscedastic; meaning that they have the same variance. When the error terms do not have the same variance then there is heteroscedasticity. The problem of heteroscedasticity is also common in cross-sectional data (Gujarati & Porter 2009:365-366).
Endogeneity
A regressor could seem to be an explanatory variable, but could also be an endogenous variable and then there is an endogeneity problem. The problem is then that the endogenous explanatory variable is usually correlated with the error term. As a consequence, the OLS becomes inconsistent. (Gujarati & Porter 2009:673)
1 1. Introduction
This paper looks into the determinants of household savings in an international cross-section and the different variables that can change the behavior in regards to household savings. A lot of countries will experience an increase in the older population looking at the old-age
dependency ratio and forecasts. Therefore, social security and savings have been a highly discussed topic in many countries. The reason could be the concern for the ability to maintain the retirement standard or for possible improvements.
According to the life-cycle hypotheses, the midlife is the time when the individual saves the most for the future, because the income is at its highest. The income is either consumed or saved and when consumption goes up savings decreases. However, the commonly used life- cycle hypothesis has been criticized in overestimating the consumption during the old ages, due to not including bequest motives and the precautionary savings that may occur (Mankiw 2009). The theory has also been criticized based on behavioral deviations and the discussion of rationality in the population. The model is often used as a base and then studies add variables to detect determinants that are not limited to demographic factors. The motive to save is subjective and can be hard to identify, but the primary determinants of savings are important to understand because it could indicate problems and uncertainties among the population that is important to understand for policymakers.
Knowledge about the population demographic and economic dependency is also important for savings, policymakers and for the general population, due to information about the possible social security income and how retirement could change. A young individual might start to save more and earlier for the future if the demographic is changing and the dependency is increasing. Changes in the retirement age and the change in old-age level of 65, could make young individuals prone to change their savings behavior to retire earlier or to increase the future income. The discussion of the old-age dependency and variables such as participation rate among different demographic groups is important since a high participation rate after 65 indicates that the older population tends to be more active and the dependency could decrease;
even though the dependency ratio itself could be high.
Household savings is an interesting and important subject to research because saving is closely related to physical capital and economic growth. Saving rates can change investment rates; because if savings decrease, there could be less funds for investments and different projects, which can decrease investment in physical capital and that can decrease economic growth (Weil 2009: 388). Treating the savings rate as an endogenous variable, the
2 government can see it as a tool to increase the national income. One way of increasing the savings rate have in history been to fix old-age pension plans. In the eighties in Chile they introduced mandatory pension savings, which led to an increased private savings rate by 17 percent by the beginning of 1990 and a lot of other Latin-American countries followed with similar policies after the Chilean success. Singapore implemented a pro-saving policy in the 1950 requiring workers to allocate part of their wage in a fund that can be used to finance retirement, but also to purchase housing and financing medical costs, leading to really high savings rate in the country. Public persuasion; with infomercials, was used in Japan after the second world war to increase savings among the citizens (Weil 2009: 72).
One debated determinant of household savings is social security and its effects. The result from previous studies on social security and the effects on savings has been ambiguous.
Feldstein (1974) argues that social security depresses household savings and Koskela and Virén (1983) have argued that social security does not have a clear effect. According to Ang (2009), the expected pension benefits have a positive influence on household savings in India, but the opposite effect of expected pension benefits was presented for China. De Freitas &
Martins (2014) emphasized that the health care system needs to be one of the main determinants of savings and analyzed the effects of health, life expectancy and pension programs on household savings.
The purpose of this paper is to look into the determinants of household savings in an
international cross-section. The focus is on the effects from social security and the variables for old-age dependency, participation rate, change in the unemployment rate, among other variables as an addition to the disequilibrium saving hypotheses. The disequilibrium saving hypotheses is the base theory used for the savings function in this paper. The methodology is based on the study of Koskela and Virén (1983) and the time-frame is from 2000 to 2018 on 14 OECD countries. All the countries in the cross-country data have different social security programs and different institutional effects that could affect the savings rate and therefore is the least-square dummy variable method used; also suggested by the Hausman test, and each country has an individual intercept.
The variables in the regressions have expected signs by different theories, for example, the life-cycle model expects that the old-age dependency should be negative and emphasizes the effects from demographic variables on household savings. Social security and the effect on household savings have been mixed; expected negative by the life-cycle model, but proven to be negative, positive and sometimes not having an effect at all by empirics. The model used
3 includes lots of variables and different hypothesis are constructed based on theories and previous studies and the predicted sign are tested using ordinary least squares (OLS) and two- stage-least squares (2SLS). It is interesting to see if the different hypothesis holds for new data from 2000-2018 and if the results from Koskela and Virén (1983) are robust.
The empirical results indicate that the life-cycle predictions for the variables of old-age dependency and social security had the expected sign, the social security variable is
significant in the first OLS regression and in the 2SLS regression, but the old-age dependency was not. Social security could, therefore, have a negative effect on household saving
according to the results. The results could then be due to the effect of wealth substitution being greater than the retirement effect, as the retirement effect was not supported in the results. Legislative action has been introduced to create disincentives for earlier retirement in some OECD countries and people may therefore work longer. High replacement rate as a combo with good social security could also lead to a reduction in savings, or that individuals have a good confidence in the social security system and therefore the savings decreases.
The uncertainty hypothesis holds according to empirical results, with a significant positive sign from the proxy for income uncertainty. The inertia in the saving behavior is also evident;
the lagged savings rate is significant and positive and were high in all OLS estimations.
Inertia is also been evident in other studies; Koskela and Virén (1983), Aizenman et al (2019) and De Freitas and Martins (2014). The unanticipated income is also significant and positive in all OLS regressions as expected by the disequilibrium saving hypothesis, so the hypothesis holds partly according to own estimations. The participation rate is positive, but insignificant in two of the OLS regressions. The variable was expected negative by Koskela and Virén (1983) and could identify potential retirement effect, but is inconclusive according to empirics in the results.
The limitations in this paper are mostly restricted by the data availability, and the amount of countries used is the maximum amount of countries that has values for the variables during the time-span. The countries in the sample are mostly European countries except for Canada and the USA and it is a sample that contains developed countries only. The results are therefore not applicable to developing countries or countries that are far different than the OECD-sample.
This paper is structured into the following 12 parts; starting with the theory that the model is based on and it presents potential effects and determinant of household savings. The third part
4 is the literature review that is closely related to the theory and goes through the previous empirical and theoretical studies on determinants of household savings and previous empirical results on social security and household savings is also presented. The fourth part presents the stylized facts for some regressors or explanatory variables, with time-series figures. The fifth part explains all the data leading up to the sixth part which is the methodology. The seventh part is the empirical results and the eighth part is the discussion which is followed by a robustness check. The tenth part is the conclusion connected with the eleventh part which is a summary and then there is the appendix in which referenced data, figures and tables are presented and references is the last twelfth part.
5 2. Theory
2.1 Life-cycle model
A consumption function that is most commonly used when looking into the determinates of household savings is Modiglianis (1986) life-cycle model. According to Modigliani, saving allows consumers to move income from times when income is high, to times when income is low. Income varies in a person’s life due to example childhood and retirement. To smooth out consumption, young people tend to save and elderly tends to dissave. (Mankiw 2009). A person with diminishing marginal utility is by theory also likely to prefer a consumption that is smooth over time and saving helps to achieve that preference. (Rosen and Gayer 2014:236).
Problems with the life-cycle model is that empirics have found that the elderly does not dissave as much as life-cycle model predicts and the chief explanations are precautionary savings that arises from uncertainty and bequest to children. (Mankiw 2009) Explanatory variables from this theory is primarily demographic variables such as old-age dependency ratio, participation ratio, age or life expectancy. If the dependency ratio is high, then it is going to have a negative effect on savings and the life-cycle hypothesis also suggest a positive effect of income growth (Ang 2009).
The life-cycle model was used in Feldstein (1974) and spurred on the view that the pay-as- you-go (PAYGO) social security program depressed private savings. Due to the fact that social security is paid out to older individuals and leads to a substitute for pension savings during the working years, so social security is expected to have a negative sign on household savings according to the life-cycle model (Barro & Mcdonald 1979).
2.2 Wealth substitution effect
As previously explained social security can change the allocation of savings during a lifetime and one theory is the wealth substitution effect. If there is a belief that social security will guarantee a good income as a person grows older or when they are in need of security, then that could crowd out private savings. If for example the payments of social security are seen as a decrease in present consumption, but added to the future consumption. Then looking at the lifetime allocation as inspired from the life-cycle model; under an intemporal budget constraint, the person needs to save less to attain the optimal point, when social security increases. Which leads to less savings than before the increased social security (Rosen and Gayer 2014:236-238).
6 The specific effect of social security crowding out private savings is however, under the assumption of diminishing marginal utility; meaning the indifference curves are convex to the origin. Regarding saving and social security there could be other utilities that effects the saving decision, for example, anticipated utility. The anticipated utility is based on the expected utility and if the social security is believed to yield a good future income, then the endowment of utility could be changed. If the positive endowment is greater than the negative endowment and if the benefits is not enough, the utility could still increase, because of the anticipation, but then decrease due to realization of low retirement income (Wilkinson and Klaes 2012:92).
The PAYGO-system does not lead to direct capital accumulation, it could even lead to a reduction of saving, if the public saving is not compensating the decreased private saving, benefits are paid out to the current beneficiaries and not to the person that is currently a contributor and savings decreases (Rosen and Gayer 2014:236). The replacement rate could indicate the efficiency of the pension system and if it is high together with good social security, then the wealth substitution effect could be strong.
2.3 Permanent-Income hypotheses
The permanent income hypotheses from Friedman (1957) suggest that consumption should not depend on current income solely, an individual experience random and temporary changes to their income from year to year, which is a contrast to the life-cycle model. Income has two components; permanent income and transitory income, 𝑌 = 𝑌𝑃+ 𝑌𝑇. Different income has different persistence; for example, could education give a higher permanent income and a fortunate situation could give higher transitory income. Friedman said that consumption depends most on permanent income; 𝐶 = 𝛼𝑌𝑝, 𝛼 is a constant of consumed permanent income. Because consumers use saving and borrowing in response to transitory income, the consumption is proportional to permanent income and then people save rather than consume their transitory income. Consumption is, therefore, also affected by individual expectations (Mankiw 2009).
Mankiw (2009) also mentioned that for this hypothesis to be true, the consumption should be unpredictable over time and the variable of consumption should follow a random walk.
Changes in the savings rate should also follow a random walk, as it is expected to change with transitory income and those deviations from permanent income is unexpected. The
7 unexpected increase in income should therefore have a positive relationship to household savings.
2.4 Precautionary saving
Uncertainty about future income will reduce current consumption and increase current savings and is known as precautionary saving, also known in the literature as the uncertainty
hypothesis. One proxy of income uncertainty that is used in this paper is the change in unemployment rate. There are also other types of variables that could show the effect of precautionary savings, but only one variable is in the model. Precautionary saving could also be that an individual’s saves to take on a risk and could show the saving behavior of a risk- averse individual. Carroll and Samwick (1995) investigated how much savings that is due to some households having greater income uncertainty than others. If household behave
according to the buffer-stock model, then they could derive good measures of uncertainty and find the precautionary saving response. They found strong evidence for precautionary saving as a response to income uncertainty in the future and if all households had the same
uncertainty; constructed from a simulation focused on the lowest uncertainty group, they explained that 39 to 46 % of savings was due to uncertainty and 32 to 50% was due to uncertainty in a later paper (Carroll and Samwick 1995, 1998).
2.5 Bequest motives
Bequest could be testaments and other transfers to children or to a recipient belonging to the next generation and can diminish the dissaving that is proposed by the basic life-cycle model.
The social security decreases savings by allocating capital from the taxpaying children to the parents and then they could bequest it back to the child. (Rosen & Gayer 2014) There could be a lot of reasons for intergenerational transfers, one could be that the children have a different income risk profile or it could just be altruistic motives behind it. Scervini and Trucchi (2019) looked at the intergenerational precautionary saving and the income risk of children to see if altruistic behavior; due to the income risk of their children, effected savings.
They then found that the income risk of children effected precautionary savings.
2.6 Ricardian equivalence
According to the theory of the Ricardian equivalence, the governments choice to borrow and create deficit in the budget, could lead to increased savings. If the individuals in society understand that deficit could lead to increased taxes in the future, they could increase the intergenerational transfers, so that their descendants could have the same consumption level as they had and savings is increasing. The ability, however, that the elderly will foresee future
8 taxes is a quite controversial statement. The assumption by the model is that increased budget deficit will lead to an increase in household savings, but the empirical evidence has been mixed (Rosen & Gayer 2014:461). A later explained study by De Freitas & Martins (2014);
included the public budget balance and the variable had a negative sign in all estimations by different models; as expected by the Ricardian equivalence theory.
2.7 Psychological factors
There are many psychological factors that affect the decision to consume or save. One such factor could be the propensity to exercise self-restraint; which is the ability to forecast probable future outcomes of a decision and then have the patience and self-restraint to consider long-term interest before short-terms interest. Another factor could be the uncertainty of human life; meaning that it could be hard to save for a decent future if the individual has a hard time to foresee one. There is also an observed tendency for people to be more economical when living under safe circumstances than under hazardous or unhealthy ones (Wilkinson and Klaes 2012:262).
There is also one factor that has been apparent in society when it comes to saving decisions, which is the urge for instant gratification. Factors of intemporal choice, has been behind the construction of the discounted utility model and that model emphasized the importance of the discount rate, when it comes to the trade-off between valuing the present or the future
(Wilkinson and Klaes (2012:263). A high discount rate for the future, values present satisfaction higher and as mentioned earlier, the factor of instant gratification and the uncertainty of human life, could indicate that an individual has a high discount rate for the future leading to less savings.
However, the discounted utility model assumes a stationary discount rate. Meaning no change of time-preferences over the lifespan, which is evidently not true and has been observed to vary over different ages and is expected by the life-cycle model too. Wilkinson and Klaes (2012:266) even mentioned that time-preferences is inconsistent and the discount rate tends to decline over time. If implicated to the theory of the life-cycle model, that could possibly lead to a higher tendency to save as pension becomes closer in time. Johnson et al (2010:13) explains that procrastination is very evident in the savings behavior for pension and a lot of people tend to save little in advance. There is also possible inertia in the savings behavior;
people who have started to save often proceed, but people who do not save have a low tendency to start in time and consistently continue.
9 2.8 Financial literacy
Financial literacy is how well individuals can understand the economic sphere with pension plans, taxes and the general administration of personal finances. More financial products, loans and saving solutions are available today and navigating in the wide range of supply with different interest rates and cost can be hard. Resulting in individuals solving their financial matters differently with different expenses, due to lack of knowledge. To increase the financial literacy education is the key and can come from either schools and universities or from parental advice. Increased financial literacy has been shown to have a positive effect on the saving behavior based on Lusardi (2008). Lusardi, Michaud, and Mitchell (2015)
investigated how programs to enhance financial literacy could increase savings. They could conclude that the effective programs; those that provided follow-up treatment to the
participants, if their financial education was given to employees, could increase the retirement savings by 10%. As a model, they used the life-cycle hypothesis with endogenous financial literacy to see how the financial education could change economic outcomes, with a focus on workplace educational programs.
Mahdzan &Tabiani (2013) looked at financial literacy and saving in Malaysia in the context of emerging markets and could conclude that financial literacy is an important determinant of private savings. They identified different levels of financial literacy and found that their sample had a basic understanding of interest rate, inflation and percentage calculations, but less understanding of the stock market or the risk-return of assets. Which is a blind spot in educational programs, that should be regarded and improved. However, they explain that the sample could have been biased towards a more literate group in the population. A high percentage of the sample could be MBA-students and therefore are the estimations hard to generalize, but could indicate some preliminary knowledge about financial literacy and the effects on savings.
2.9 Capital accumulation and economic growth
The national savings rate is based on two components; government saving which is the difference between collected taxes and government spending. The other component is private savings, which is household savings and saving by corporations. (Weil 2009: 72) Increasing the savings rate has been a tool for government to increase national income, the Solow model explains that countries with a high savings rate ought to have a higher level of income per capita. The Solow model predicts that countries with high investment rates have higher level of income per capita and the investment rates are related to the savings rate.
10 Weil (2009: 69) explains that every choice of investments corresponds to an act of saving.
Capital accumulation has an opportunity cost; which means that the accumulated money could have been used for consumption or something else. However, the investment rate and the savings rate are not fully related, due to the crossing of national borders with international flows of investments. In Weil (2009: 69) investigation; the international investment flows have an effect, but the investment rates are still significantly dependent on the savings rate.
Household savings can therefore increase accumulation of physical capital and affect
economic growth. Treating the savings rate as an endogenous variable and as changeable has implications on policy changes and how the government relate in regards to the savings rate (Weil 2009: 388). Government can; as a tool, influence the private saving rate by, for example, changing the pension system or create different saving enhancing policies.
11 3. Literature review
When it comes to previous literature on the determinates of household savings, there is a lot of literature with varying methods and data. Le Blanc & Porpiglia et al (2016) look at the saving motives and behavior in fifteen euro-countries, using the Household Finance and Consumption Survey, from the years 2008–11. They found that precautionary savings was the most commonly reported motive in all countries and is followed by the saving for old-age.
Niculescu-Aron and Mihaescu(2012) explains based on their empirical study using fixed- effect dummy variable method and panel data that households in Europe save because of excess income from an increased salary or due to large gains from interests, but this holds only under a period of economic prosperity. European households could also have a tendency to save if influenced by public policy. The variables of interest rate, inflation and percentage of the rural population had an effect on savings according to their panel model.
The result from previous studies on social security and household savings has been
ambiguous. Feldstein (1974) argues that social security depresses household savings and that social security could lead to earlier retirement based on time-series data on a single country;
USA. The study by Feldstein sparked controversy and debate due to the results of the effect of social security on household savings. Leimer and Lesnoy (1982) replicated Feldstein’s
methodology and could conclude that their results indicated no statistical significance of social security having any effect on savings and that the model used had no variable that could indicate a clear retirement effect.
Koskela and Virén (1983) also argues that social security does not have a clear effect.
According to Ang (2009), expected pension benefits could have a positive influence on household savings in India; which could be due to bequest motives, but they had the opposite effect in China. The paper used a model that is based on the extended life-cycle model, comparing China and India. Possible explanations are demographic factors and bequest motives. India has more children per parents than China, which could be due to the old one- child policy that China had. But the effects should be view with caution due to lack of exact measures.
Barro and Mcdonald (1979) did a cross-country comparison with a time-span from 1951 to 1960, to look at the relationship between social security and private savings. They conclude that increased retirement could encourage increased savings due to bequest motives. De Freitas & Martins (2014) analyzed the effects of health, life expectancy, the life lived and the pension systems of different countries and the effects on household savings. They used the
12 life-cycle model but added health care, pension entitlements and public budget deficit among other variables to their empirical models.
De Freitas & Martins (2014) also used panel data, but with 22 OECD countries during the period 1970-2009. The conclusions from the paper were that longevity increased savings which are predicted by the life-cycle model and public budget deficit also increase savings which are predicted by the Ricardian equivalence. When the replacement rates are high and the expenditure on health care increases, then that could have a negative effect on household savings. But that is highly due to the consumption patterns of the old, it they change
consumption behavior or receive the ability of buying subsidized health care products, then consumption could go down and savings among the old goes up. It is also emphasized that the health care system needs to be one of the main determinants of savings.
The base in this study is the disequilibrium saving hypotheses, also used in Koskela and Virén (1983) and is based from the paper by Deaton (1977). According to Deaton (1977) the savings function contains of three terms, one term that is derived from the equilibrium consumption function, then one term based on unanticipated income and at last one term about
unanticipated prices and the effect on the savings rate.
The first term is a related to the lagged savings rate and the possible inertia in the savings behavior, the anticipated savings are the previous savings. The second term mirrors the view that the transitory income is saved, the idea is that the consumer cannot react to a stimulus not foreseen. But in future time periods the unanticipated income is going to affect anticipated income and some of the income will be consumed. The third term describes the unexpected inflation effects on savings. If real income is accurately anticipated, unanticipated inflation will cause the savings ratio to rise. But the ratio could fall if the unanticipated prices cause unexpected change in real income. It is explained that there will be an increase in the savings ratio when the accurate observed prices exceed the anticipated prices. Conclusion from Deaton (1977) is that unanticipated inflation and unanticipated income causes involuntary saving.
To understand the disequilibrium saving hypothesis, Deaton (1977) explains that consumers could have quite good information about the prices that they actually buy, but the consumer does not have accurate knowledge of the prices of relative goods in the store. The individual has no possibility to really see relative price changes. The disequilibrium model focusses on the absence of complete price information and the mistakes it causes. In theory by the
13 equilibrium consumption function, the individual purchases will sum up to the total
consumption. But since no individual knows all prices, mistakes are made. In the disequilibrium model, the repeating nature of buying and the lack of complete price information, means that the expected and actual prices can differ not only in the future but also in the present and that is the reasoning behind the hypothesis (Deaton 1977).
Opposed to Feldstein, Koskela and Virén (1983) finds no evidence of social security effecting household savings, and gives three explanations to why social security does not have an effect on household savings. First point is based on the extended life-cycle model and is that the wealth substitution effect and the retirement effect cancel each other out. Second point explained is that the beneficiaries will transfer their gains from an increased social security to next generation that pays more taxes and there is no change in aggregate savings. The last explanation is that an increase in social security benefits, from taxes, decreases consumption of current working people and if the pensioners are liquidity constrained, then they consume all their benefits. The total effect is unchanged aggregate savings. (Koskela and Virén 1983).
14 4. Stylized facts on some regressors, demographic factors & replacement rate etc.
Below is a selection of some graphs of the variables included in the regression and some additional variables that may also affect household savings. Descriptive statistics on all variables are presented in part five about data. Household savings and social security by country and by year are presented below and household savings is quite volatile during the years, but social security indicate a quite similar pattern among the countries in regards to how it changes with time. Sweden has a household savings rate above the other countries and Denmark has had a low savings rate up until 2014. According to a normality test in the appendix (Appendix 8: Test of normality); the dependent variable (s/y), have a possible outlier and it is Denmark’s value of household savings the year of 2000. It can be seen that it is the lowest value in figure 1.
The intercept for the different countries is different and suggest the use of a model with individual intercept. There is a common upward slope around 2009 in the graph presenting social security, but also in the graph of household savings for the countries. This indicates that the financial crisis in 2008 may had an effect on household savings and social security. The uncertainty of income and jobs and the decrease of jobs and income could have stimulated the savings behavior of the households, as an effect from the crisis.
FIGURE 1: Household savings during 2000 -2018
Source: The data is from OECD (2020)
Notes: Constructed time series of the household savings of 14 countries done in SPSS.
15 FIGURE 2: Social security as percentage of GDP during 2000 -2018
Source: The data is from OECD (2020)
Notes: Constructed time series of the social security as percentage of GDP of 14 countries in SPSS.
Below are the demographic variables presented; the old-age dependency ratio (OADR) and participation rate for people over 65. There is an increase for all of the countries when it comes to OADR, Italy has the highest rate and Ireland has the lowest. The participation rate for those above 65 is also increasing for every country in the sample, according to figure 4.
FIGURE 3: Old-age dependency during 2000-2018
Source: The data is from OECD (2020)
Notes: Constructed time series of the old-age dependency of 14 countries done in SPSS.
16 FIGURE 4: Participation rate of people over 65, during 2000 -2018
Source: The data is from OECD (2020)
Notes: Constructed time series of the participation rate >65 of 14 countries done in SPSS.
According to the United Nations, Population Division (2019) the OADR is a poor proxy for the level of economic dependency in a population. OADR does not include the fact that older persons are a heterogeneous group with respect to both economic activity; such as labor force participation, but also, not all persons in the working ages are active in the labor force. The prospective old-age dependency ratio (POADR) is argued to be an alternative way of measuring the dependency in the population, but is a complex indicator for the data set used and for a large data set.
POADR is based on the fact that different countries have different threshold old-age and different amount of remaining years to live. POADR is calculated as the ratio above the age closest to the remaining life expectancy; usually 15 years are used in the calculations, and then it is divided between the working force. The work force is then 15 up to the new threshold age that differs between subjects. The threshold old-age used today is 65, but if looking at figure 4; the participation rate has been increasing and a lot of individuals are not in full pension at 65, which supports the view that OADR does not capture the full economic dependency in the population. The life expectancy after 65; see figure 7, has been increasing for all the 14 OECD countries; during 2000-2017; Hungary is the country with the lowest life expectancy, and also has one of the lowest participation rates.
17 FIGURE 5: Net replacement rate as a percentage of pre -retirement earnings 2018.
Source: The data is from OECD (2020) and the graph is constructed in Excel.
Notes: The indicator is calculated as the net pension entitlements divided by net pre-retirement earnings and the figure presents the net pension replacement rate by men; total data was unavailable.
Changing gender did, however, not yield a much different figure1. Different contributions paid by workers and pensioners, for example, social security and income taxes are also added in this indicator.
Figure 5 of replacement ratio presents the effectiveness of a pension system when it comes to the retirement income that is going to replace the pre-retirement income. Workers and
pensioners have previously paid in income taxes and social expenses that is going to be allocated to pensioners OECD (2020). However, the pension system is often a partially funded or an unfunded system; meaning that the contribution paid in by current workers goes to current pensioners. The trend in reason years, however, have been to individualize the pension system. In the 1990, Italy and Sweden changed the PAYGO system into a nonfinancial
defined contribution system that have a stronger link between life-time earnings and benefits.
Other changes in countries pension systems is that Spain has increased the number of years that they calculate the reference wage on. Spain, Austria and the USA are currently not calculating the reference wage on the whole career. In the UK, Denmark and Ireland there is an earning-related occupational pension scheme not connected to earlier earnings. Earlier retirement disincentives have also been introduced in some countries, which gives out penalty or give out bonuses to regulate people from retire earlier (OECD 2019). The important
outtake from this section is that the pre-retirement income is not a guaranteed income after pension. The different replacement rates presented indicates different effectivities of the pension system. The OECD average is 58,6 % and the replacement rate at 2018 indicates a current level of effectivity for the specific countries used in the sample.
1 The figure was different for Spain and Hungary, when gender was changed. Hungary had 78,4 % for women and Spain had 83,4 % for women and the OECD average was 57,6% for women.
0 20 40 60 80 100
AUT BEL CAN DNK FIN DEU HUN IRL ITA NOR ESP SWE GBR USA 89,9
66,2 50,7
70,9 64,2
51,9 84,3
35,9 91,8
51,6 83,4
53,4
28,4 49,4
18 FIGURE 6: Aggregate replacement ratio for pensions , 2010-2018
Source: The data comes from Eurostat, specifically EU-SILC and graph is done in SPSS.
Notes: The indicator is calculated as the median gross pensions of the 65 to 74-year old’s relative to median gross earnings of 50 to 59-year old’s, excluding other benefits, such as social security.
Calculations for the indicator from Eurostat differs from OECD and includes only countries in the Euro-area and not the entire OECD sample used in the regression. However, the figure displays the time-series of the replacement rate for some countries in the sample. The
replacement ratio is explanatory of the pension system and its functionality. The replacement ratio could also affect the saving behavior and according to De Freitas & Martins (2014) the ratio is assumed to have a negative effect on household savings and they also obtained a negative effect according to the empirical results.
They also explain that high replacement rates and large public contributions; such as social security has a negative effect on household savings together. Some countries have a high replacement rate according to figure 5 and 6 and if social security is high then the individual’s precautionary savings could decrease if the knowledge about the good coverage is known in the population. The calculations differ, but Italy, Spain, Hungary, Austria and Norway have quite high replacement ratios according to figure 5 and 6. The replacement rate could not be included in the regression, due to lack of data from the indicators from both OECD and Eurostat. The variable could, however, be a determinant and explain a negative effect on household savings from social security; if the efficiency is good.
19 FIGURE 7: Life expectancy after 65 for the OECD sample; 2000-2017.
Source: The data comes from OECD (2020) and the indicator is life expectancy at 65. The figure is constructed in Excel.
Notes: The OECD- countries in the sample follow the same trend with increasing life- expectancy after 65.
20 5. Data
Here is a general explanation of the included variables in the model. The dependent variable is the household savings rate and the possible explanatory variables are; old-age dependency, participation rate over 65, social security, unanticipated inflation, unanticipated income, long- term interest rate, change in unemployment rate and the lagged savings rate. The variables are collected from OECD and some from the system of national accounts 2008 (SNA 2008), SNA 2008 is an update to the 1993 version and include some changes in the calculations of the indicators (System of national accounts 2008).
5.1 Household savings rate
Household savings rate is the primary dependent variable used in the regressions. Private savings according to the national accounts is usually calculated by subtracting consumption from the disposable income. The variable is taken from OECD and the new SNA 2008, there have been adjustments to the calculation of the household savings rate and the disposable income. There was a change in pension entitlements that effected the disposable income and thereby the savings rate. The savings are calculated as the part of the disposable income; after adjustments of pension privileges, that is not being spent on other goods and services. The savings rate could be positive or negative from year to year; if it is positive, then there is unused income. If the sign is negative, then financial or non-financial wealth could have been liquidated (SNA 2008:183).
This variable is the dependent variable in the regression. It is the net percentage of the household disposable income for 2000 to 2018. The lagged household savings rate is also used as a regressor due to the potential persistence in the savings behavior. As mentioned earlier in the section about psychological factors; self-restraint and patience can have possible effect on the savings behavior. The variable of the lagged savings rate is therefore predicted to be positive and indicate inactivity.
5.2 Unanticipated inflation
Unanticipated inflation causes involuntary saving and should affect the savings rate positively according to the disequilibrium saving hypotheses (Deaton 1977). According to the result from Koskela and Virén (1983) and Ang (2009) the inflation variable had a positive sign. The variable used to indicate inflation is the PCE-deflator, also called the personal consumption expenditures price index. There is two common ways of measuring inflation, either the consumer price index (CPI) or the PCE-deflator. The difference between the two measures is first that they are based on different formulas; CPI is based on the Laspeyres index and PCE is
21 based on a Fisher ideal index. Secondly, they are weighted differently and based on different sources, CPI is based on household surveys and PCE is based on business surveys (McCully et al 2007:2).
There are also calculations that are not included in one index, but included in the other one and wise versa. CPI measures the out-of-pocket expenditures and PCE measures goods and services bought by households or non-profits; serving households. One important calculation could be medical expenses that in the CPI includes only those purchases by consumers directly. Then in the PCE, medical expenses are based on the purchases by consumers directly, but also supplemented with purchases on behalf of the consumer by example employer provided health insurance or care paid by government programs, among else.
(McCully et al 2007:12).
The variable is taken from the OECD and is classified as the private final consumption expenditure deflator. For unanticipated inflation the variable is computed to capture both the constant and the static expectations. In the first case the expectations are continuous, the expectations of tomorrow are based on the expectations of today and for static expectations, the expectations of today should be equal to the expectations tomorrow. So, the unanticipated inflation with constant expectations are just the log difference of the PCE-deflator; ∆𝑙𝑜𝑔𝑃𝑖𝑡, and for the static expectations a second difference is added ∆∆𝑙𝑜𝑔𝑃𝑖𝑡.
5.3 Unanticipated income
Unanticipated income also causes involuntary saving according to the disequilibrium saving hypothesis and explained by Deaton (1977). Unanticipated income effects savings positively according to the permanent income hypothesis, that states that a person tends to save the transitory income. The variable used in this method is the household disposable income from OECD, which is the closest variable to real income used in economics. It measures the income of households with wages, self-employed income unincorporated income and other salaries. This is however done after taking account of taxes, net interest and dividends. The variable in this method is in net, which means that depreciation cost has been withdrawn from the income (OECD 2020). The variable is then calculated as ∆𝑙𝑜𝑔𝑌𝑖𝑡under the constant expectations and calculated as ∆𝑙𝑜𝑔𝑌𝑖𝑡under the static expectations.
5.4 Interest rate
The long-run interest rate is used in the method and is taken from OECD for 2000-2018. The long-run interest rate is also a fiscal instrument that can be used to affect savings and is
22 therefore included. There is a lot of previous empirical studies that has included interest rate as a determinant of savings and they looked at the relationship between interest rate and savings.
Aizenman et al (2019) investigated the low interest rate together with certain economic circumstances and the effect of private savings and found that interest rate could affect the savings rate negative if the variables old-age dependency, output volatility; proxy for income uncertainty, and financial development is above a certain threshold. Aizenman et al (2019) also found that when the interest rate is under approximately 1 percent, then greater output volatility would lead to a positive effect on private savings. Old-age dependency and financial development had a negative effect on household savings, but that effect tends to decrease as the interest rate decreases. From the regression result of their baseline model the interest rate had a positive effect, but was statistically insignificant.
5.5 Social security
The effect of social security on household savings have been ambiguous according to previous literature. The indicator used as a social security variable is from OECD and the database social expenditure; aggregated data, and is called social benefits other than social transfers in kind. There are two indicators for social protection, which is transfers in kind and then social transfers that is not in kind. The in-kind transfers are transfers that are connected to a certain good or a service; for example, education. The other one used in this paper is made up by cash transfers in form of pension benefits or non-pension benefits and the last one is either from the government or non-profits institutions; but serving households. The
transfers are used for financial needs to help for example, sickness, unemployment, education or unexpected events, among other needs (OECD 2020). Social security as percentage of GDP is an explanatory variable in all the regressions.
5.6 Dependency rate
There are different measures of dependency in the population, old-age dependency ratio, prospective old-age dependency ratio and the economic old-age dependency ratio, to examine the population demography. (United Nations, Population Division (2019). The old-age
dependency ratio (OADR) is the measure that has been most frequently used when looking at changes in the demographic of old people; usually over the threshold age of 65, in relation to the population of 15-64. However, the measure has its limitations when it comes to the strict definition of dependency, but is the most efficient to be used on the data set in this paper. The variable is from OECD and the database of the labor force statistics.
23 One should however keep in mind that the participation rate and life expectancy among the countries has gone up during recent years and the dependency is not fully represented. The old-age dependency rate has a negative effect on household savings according to Ang (2009), Koskela Virén (1983) and the life-cycle model, that predicts dissaving among the elderly.
5.7 Unemployment rate
This explanatory variable is used as a proxy for income uncertainty and the variable is the difference in the unemployment rate. According to the uncertainty hypotheses the income uncertainty affects household savings positively. The unemployment rate is taken from the OECD labor statistics and is defined as the unemployment rate of the labor force and the labor force is individuals aged 15 to 64. The regular unemployment rate is used lagged in the third participation rate model and then the change in unemployment rate is used in the other three models.
5.8 Participation rate
The participation rate is the rate of the active population that is above the threshold old-age of 65. The variable is taken from the OECD labor force statistics; labor force statistics by sex and age. The labor force participation rate indicator is the percentage in same age group, during the years 2000 – 2018. The participation rate is a regressor in three of the models and is the dependent variable in the third OLS regression. In the model were the participation rate is the regressand then social security is one of the regressors and one effect in literature that could be prominent in the result from this regression is the retirement effect. If social security leads more people to retire, then people tend to save more to finance the increased length (Rosen & Gayer 2014:239). If social security has a negative sign as a regressor to the participation rate, then that negative relationship could indicate that social security leads to early retirement and people tend to save more to finance a longer retirement. This effect could however be linked to the beliefs about future retirement income and if income is not that certain, people may not retire earlier, due to uncertainty coverage of the original believed length of retirement.
24 5.9 Descriptive statistics
The descriptive statistics are presented in table 1. The variables are presented before
transformation that is done to some variables in the regressions. The descriptive explanations presented are observations, mean, standard deviation, skewness, kurtosis, maximum value and minimum value.
TABLE 1: Table over the descriptive statistics
25 6. Methodology
The methodology is based on Koskela and Virén (1983) and the disequilibrium saving hypotheses presented in Deaton (1977); more specific the first three terms are used from the final function and then filled out with other relevant variables. There are also added variables that has an effect by the life-cycle model and other theories. The fixed-effect dummy variable method (FEM) is used on panel data and there is dummies for the countries with individual intercept. By including the individual intercepts, the OLS and 2SLS give consistent values.
OLS and 2SLS can also give inconsistent values, if the assumptions for the methods does not hold, to test some assumptions different statistical test are run. According to the Hausman test, FEM is the appropriate model to be used. The estimated variables presented before the
individual intercept in the results are expected to be the same for all countries; which is a generalization, and for a larger number of cross-section unit it should be valued with caution and should be combined with one-country data for further policy recommendations and investigations into the effects on households savings for a specific country. Cross-country comparison gives more general results about the determinants then country-specific data, but cross-country gives greater variability of experience with the different programs of social security (Gujarati & Porter 2009).
It is important to look out for the possible identification problems; like simultaneity. The Hausman test of endogeneity is used to see if there are endogenous regressors and possible correlation with the error term. There are multiple ways of conducting the Hausman test and the ways used in this method is to regress the participation rate equation with OLS to obtain the residuals and then regress the structural equations on the predicted values of the
participation rate and then add the residuals (Gujarati & Porter 2009: 704) the method is also done with the ordinary participation rate and the added residuals.
The null hypothesis is that there is no simultaneity and if the residuals added to the equation is significant then the null hypothesis can be rejected. Gujarati & Porter (2009: 704) mentioned that Pindyck and Rubinfeld (1990:304) suggested regressing on the ordinary variable of the participation rate and not the estimated value for efficient estimations. The regression is therefore done with one regression with the residuals 𝑣𝑖𝑡and the predicted value 𝑃𝑅̂𝑖𝑡and one with the residuals 𝑣𝑖𝑡and the regular participation rate 𝑃𝑅𝑖𝑡 variable. To solve for possible simultaneity, two-stages least squares (2SLS) is used. There could be possible autocorrelation if variation is low in some variable during a span of years, but can be hard to see when only looking at the subjective graphical display of residuals and predicted values. The Durbin-
26 Watson h-statistic are used and due to the autoregressive form of the structural model; the h- statistic is used instead of the d-statistics. The model has the lagged regressand as a regressor and therefore is the d-statistic given by the regression output not a good indicator of
correlation other than calculating the estimated correlation 𝑝̂ (Gujarati & Porter 2009).
Heteroscedasticity will be evaluated graphically by looking at the residuals and is
complemented by the Breusch-Pagan-Godfrey test of heteroscedasticity. The first step of the test is to regress the OLS of the four equations to get the ANOVA table and then save the residuals. Then the next step is to obtain the estimated variance that is calculated as the residual’s sums of squares divided by the sample size; which is 265 in this case. Then the squared residuals are divided by the estimated variance to construct a p-variable; not the same as p-value. Then the p is switched with the original dependent variables and regressed on the independent variables. This is conducted four times to all the equations. Then from these four regressions with p as a dependent variable, the explained sums of squares are used in the final test statistic. The test statistic is presented by the Greek letter theta Θ and is the explained sums of squares divided by two (Gujarati and Porter 2009:386). A downloaded macro to the statistical program SPSS by Ahmad Daryanto with the Breusch-pagan and Koenker test is also presented, which use the Lagrange multiplier to yield the test statistics.
The time frame for the data is 2000 to 2018 and all the dummies are included, but there is no intercept to avoid the dummy variable trap. The choice of countries is purely due to data availability under the goal to maximize the time-frame and the amount of countries and also to obtain a balanced panel. An unbalanced panel can create bias that can affect the OLS estimations; OLS assumes unbiased data, and the unit root tests also. The system of national accounts 2008 has been changed a lot and there was a lot of missing values leading to a restricted sample in regards to the time-frame and countries. i indicates countries and t indicates the time, i =1,…,14 and t = 2000,…,2018.
[1] (𝑠
𝑦)
𝑖𝑡= 𝛽1∆𝑙𝑜𝑔𝑌𝑖𝑡+ 𝛽2∆𝑙𝑜𝑔𝑃𝑖𝑡+ 𝛽3(𝑠
𝑦)
𝑡−1+ 𝛽4𝑟𝑖𝑡+ 𝛽5(∆𝑈)𝑖𝑡+ 𝛽6𝑆𝑆𝑖𝑡+ 𝛽7𝑂𝐿𝐷𝑖𝑡+ 𝛽8𝑃𝑅𝑖𝑡+ ∑14 𝑑𝑖𝐷𝑖+ 𝑢𝑖𝑡
𝑖=1
[2] (𝑠
𝑦)
𝑖𝑡= 𝛽1∆∆𝑙𝑜𝑔𝑌𝑖𝑡+ 𝛽2∆∆𝑙𝑜𝑔𝑃𝑖𝑡+ 𝛽3(𝑠
𝑦)
𝑡−1+ 𝛽4𝑟2𝑖𝑡+ 𝛽5(∆𝑈)𝑖𝑡+ 𝛽6𝑆𝑆𝑖𝑡+ 𝛽7𝑂𝐿𝐷𝑖𝑡+ 𝛽8𝑃𝑅𝑖𝑡+ ∑14𝑖=1𝑑𝑖𝐷𝑖+ 𝑢𝑖𝑡
The above equations are done with constant and static expectations of income and inflation and changes of the interest rate. The dependent variable in the structural models are (s/y) t
which is the household savings (net) and (s/y) t-1 is the lagged savings, so there is an
autoregressive form to the equations. ∆logY and is the unanticipated income (net) for constant
27 expectations and ∆∆logY is for static expectations. ∆logP is the unanticipated inflation for constant expectations and ∆∆logp is for static expectations. r is the nominal interest; used in the model with constant expectation. r2 is the nominal interest rate subtracted by the lagged inflation rate and used in the static expectation model. ∆𝑈 is the change in unemployment, the proxy for income uncertainty, SS is social security benefits divided by GDP and OLD is the old-age dependency ratio (OADR) and at last the PR variable, which is the participation rate for people above 65.
[3] 𝑃𝑅𝑡= 𝑎1𝑂𝐿𝐷𝑡+ 𝑎2𝑆𝑆𝑡+ 𝑎3𝑈𝑡−1+ 𝑎4𝑃𝑅𝑡−1+∑14 𝑐𝑗𝐷𝑗
𝑗=1 + 𝑣𝑖𝑡
[4] (𝑠
𝑦)
𝑖𝑡= 𝛽1∆𝑙𝑜𝑔𝑌𝑖𝑡+ 𝛽2∆𝑙𝑜𝑔𝑃𝑖𝑡+ 𝛽3(𝑠
𝑦)
𝑡−1+ 𝛽4𝑟𝑖𝑡+ 𝛽5(∆𝑈)𝑖𝑡+ 𝛽6𝑆𝑆𝑖𝑡+ 𝛽7𝑂𝐿𝐷𝑖𝑡+ 𝛽8𝑃𝑅̂𝑖𝑡+ ∑14𝑖=1𝑑𝑖𝐷𝑖+ 𝑢𝑖𝑡
The variables from the participation rate equation is the OADR again and the social security variable and then the lagged unemployment rate and the lagged participation rate for people above 65. Then dummies for every country is added to both equations and an error term; 𝑢𝑖𝑡 and 𝑣𝑖𝑡 for the PR equation. The 𝑣𝑖𝑡 residuals are saved and used in the Hausman test of endogeneity. To look out for heteroscedasticity with cross-section data, there will be
individual intercept for each country that accounts for institutional and other related effects on the savings rate. Fixed effect least square dummy variable is also known as fixed effect regression model (FEM). The term comes from the fact that the intercept does not vary over time; it is time-invariant, even though it might be different for the countries. To get the fixed- effect to vary among the countries; dummies are used. If dummies are introduced to all the countries as in this paper the intercept has to be removed to not fall into the dummy variable trap. There is also a way to get the intercept to vary over time, namely to introduce time dummies. The model is then known as a two-way fixed-effect model. One problem with the model used is that it takes up a lot of the degrees of freedom, due to a lot of estimates (Gujarati & Porter 2009: 597).
Peseran cross-sectional dependence test is also applied to the four regressions, Peseran (2004), it is an important test to detect interdependencies among countries that has been more common after globalization and can appear in randomly selected household also, for example, due to norms in society. Cross-sectionally dependence is also important to test to apply the first-generation unit root tests for panel data; presented in the appendix, one assumption is namely cross-sectional independence. In the robustness check the panel unit root tests are presented and the one used are; Maddala and Wu (1999), Hadri (2000), Levin et al (2002) and Im et al (2003).
