Time Preferences, Culture, and Household Debt Maturity Choice

Working Paper Series National Centre of Competence in Research Financial Valuation and Risk Management Working Paper No. 674 Time Preferences, Culture, and Household Debt Maturity Choice Wolfgang Breuer Thorsten Hens Astrid Juliane Salzmann Mei Wang First version: January 2011 Current version: January 2011 This research has been carried out within the NCCR FINRISK project on Behavioural and Evolutionary Finance

Time Preferences, Culture, and Household Debt Maturity Choice Wolfgang Breuer*, Thorsten Hens #, Astrid Juliane Salzmann +, and Mei Wang January 2011 Abstract. Our paper explores the role of time preferences on household debt maturity choice. We find that in countries where people are more patient in the long term, planning horizons in household debt portfolios are significantly longer, as the optimal maturity of loans is considerably higher. The relationship is derived in a theoretical approach and a numerical analysis, and supported by an extensive empirical analysis with 31 countries. Controlling for various economic, institutional, and demographic variables as well as risk preferences of a country, we show that the long term discount factor δ is a sound predictor for the planning horizon of households borrowing decisions. We further investigate on the foundations of time preferences and explore the role of culture in this context. We show that culture is not only a reasonable predictor for planning horizons, but also a sound predictor for time preferences as well. The findings complete the picture about the mechanisms between culture, time preferences, and planning horizon, and highlight the importance of culture for financial decisions. JEL classification code: D90, F30, Z10 Keywords: borrowing, culture, household finance, planning horizon, time preferences *Prof. Dr. Wolfgang Breuer, RWTH Aachen University, Lehrstuhl für Betriebswirtschaftslehre, insb. Betriebliche Finanzwirtschaft, Templergraben 64, 52056 Aachen, Germany. email: wolfgang.breuer@bfw.rwth aachen.de. # Prof. Dr. Thorsten Hens, Swiss Finance Institute and Department of Banking and Finance, University of Zurich, Plattenstrasse 32, 8032 Zurich, Switzerland. email: thorsten.hens@bf.uzh.ch and Norwegian School of Economics and Business Administration, Helleveien 30, N 5045, Bergen, Norway. + Dr. Astrid Juliane Salzmann, RWTH Aachen University, Lehrstuhl für Betriebswirtschaftslehre, insb. Betriebliche Finanzwirtschaft, Templergraben 64, 52056 Aachen, Germany. email: astrid.salzmann@bfw.rwth aachen.de. Prof. Dr. Mei Wang, WHU, Otto Beisheim School of Management, Burgplatz 2, D 56179 Vallendar. email: meiwang28@gmail.com. Acknowledgment: Financial support by the University Research Priority Programme Finance and Financial Markets of the University of Zurich and the national center of competence in research Financial Valuation and Risk Management is gratefully acknowledged. 1

1 Introduction Researchers have long been aware that people can perceive time in various ways, and that these perceptions affect their behavior. Perception of time is an elementary construct in the building of mindsets, and many other viewpoints will be biased in one direction or the other depending on the person s perception of time (Graham, 1981). This paper examines the link between perception of time and household debt maturity choice. We find that in countries where people are more patient in the long term, planning horizons in household debt portfolios are significantly longer as the optimal maturity of loans is considerably higher. We also elaborate on the foundations of time preferences and show that household debt portfolio choice is rooted in culture, affecting planning horizons through time preferences. During the last decade, asset allocation reemerged at the forefront of financial research (Guiso et al., 2002). In part, this research is due to the multitude of financial innovations offered by the financial services sector and the increased importance of retirement savings because of population aging. It is well documented that both demographic variables (such as age, gender, family size, and education) and economic factors (such as employment, wealth, and private business risk) are important determinants of households portfolio decisions (Campbell, 2006). However, over the last few years, an expanding literature has investigated less traditional explanations of household portfolio choices. Renneboog and Spaenjers (2010) explore the role of religion on household finance and identify thrift, risk preferences, responsibility, social capital, and planning horizon as main channels through which religion affects household financial decision making. Hong et al. (2004) provide evidence that sociability fosters stock market participation. Guiso et al. (2008) show that trust has a significant influence on the level of stockholdings. Puri and Robinson (2007) observe that optimism is related to stock investment and saving behavior. Brown et al. (2008) establish a causal relation between community effects and stock market participation. Bilias et al. (2008) identify the willingness to take 2

more than average risk, long investment horizon, and financial alertness as essential for determining stockholding levels. Haliassos and Bertraut (1995) find that inertia arising from cultural influences discourages stockholding. This overview displays that current knowledge of this field is rather incomplete and suggests that more research is clearly needed. Portfolio choices are among the most important economic decisions which people face; however, as they occur relatively infrequently, there is little data available on which to take a decision. These are the decisions that are most ripe for being influenced by psychological factors like attitudes and emotions (Puri and Robinson, 2007). We incorporate the wellknown phenomena of time discounting in our analysis on financial decision making. Focusing on liabilities of households we present an intertemporal optimization model to derive a relationship between the maturity of loans and subjective discount rates. We extend our theoretical findings by a numerical analysis and derive a testable hypothesis. Combining data on time preferences and household debt portfolios from 31 countries, we then examine empirically whether the decisions households make can be reconciled with their preferences. We find that the long term discount factor δ is more effective for the prediction of the household planning horizon than the present bias β. Furthermore, we show that national culture is a powerful factor for explaining time preferences and debt maturity decisions, and that the Schwartz cultural model is more capable for the analysis than the Hofstede model. The balance of the paper is structured as follows. In Section 2, we document the distribution of planning horizons in household debt portfolios. Section 3 describes the dataset. Section 4 presents the theoretical background and the numerical analysis and derives our main research hypothesis. In Section 5, we present an empirical analysis of the impact of time preferences on the planning horizon of households. Section 6 discusses the relevance of culture for the planning horizon of household debt portfolios. Section 7 compares the results with evidence on the firm level and on the individual level. A summary section concludes the paper. 3

2 Planning horizons in household portfolios Household portfolios consist of different types of assets and liabilities which might vary in their maturities considerably. Short term assets or liabilities typically comprise assets or loans respectively with an original maturity of one or two years or less. Long term assets or liabilities typically comprise assets or loans respectively with an original maturity of more than one or two years. Table 1 gives us a general idea of the relation of short term to long term assets and liabilities. [Insert Table 1 about here.] Short term assets can hardly be found among household portfolios worldwide. As a rule, the amount of long term assets is about 20 times as high as the amount of short term assets. The United Kingdom, where the ratio of short term to long term assets amounts to just 7.56 %, is already the frontrunner of short term assets in our whole database. Short term loans, on the other hand, can be found much more frequently in household portfolios and its use varies considerably. In Argentina, we have the highest amount of short term loans, with a ratio of short term to long term loans of 100.52 %, so that the distribution of short term and long term loans is indeed well balanced here. In Germany and Norway, we find the lowest shares of short term loans of about 6 % only. Because of the little relevance that short term assets have in household portfolios worldwide we restrict our analysis to the maturity of liabilities. The key variable we look at in the following is the ratio of long term loans versus short term loans. We call this ratio the (borrowing) planning horizon and observe a wide range of variation in this variable among countries. The smallest ratio of long term to short term loans can be found in Hong Kong, where it is almost 0, which means that we find hardly any long term loans here. The largest 4

ratio of long term to short term loans can be found in Japan with 23.8, which is equivalent to a ratio of short term to long term loans of about 4 %. [Insert Figure 1 about here.] 3 Data description In his survey of the literature on household finance, Campbell (2006) lists a number of characteristics an ideal data set for research on household finance should possess: the set of households should be representative of the total population, it should disaggregate wealth into a number of distinguished categories and asset classes, and it should feature a high level of accuracy. Such a dataset for computing households planning horizons as defined in the previous section is provided by the Economist Intelligence Unit (EIU) World Data panel of 2009, which yields unique information for studying these issues. The EIU World Data (2009) is a panel data set that follows households over time. This allows us to average each data over 2003 to 2007, which yields more reliable statistics for our focal variable. Data on time preferences is from the International test of risk attitudes (INTRA) survey carried out among economics students in 45 countries. A total of 5,912 university students participated in the survey. Each participant was asked to fill in a questionnaire that included several questions on decision making, cultural attitudes, and some information about his or her personal background (Wang et al., 2009). In the following, we restrict ourselves to the analysis of 31 countries for which we have got data on household finance and planning horizons. To measure the implicit discount rate, participants were asked to give the amount of a delayed payment which makes them indifferent with an immediate payment. The two questions are: 5

Please consider the following alternatives A. a payment of $100 now B. a payment of $ X in one year from now X has to be at least $, such that B is as attractive as A. (one year matching question) Please consider the following alternatives A. a payment of $100 now B. a payment of $ X in 10 years from now X has to be at least $, such that B is as attractive as A. (ten year matching question) In order to infer discount rates from intertemporal decisions, the quasi hyperbolic discounting model of the form is employed. In this context, z t is a given individual s exogenous consumption at time t from t = 0 to t =. Time preferences are described by parameters β and δ. When 0 < β < 1 and 0 < δ < 1, people appear to be more patient in the long run and less patient for the immediate future. We rely on the quasi hyperbolic discounting model which assumes a declining discount rate between this period and the next, but a constant discount rate thereafter. This approach has often been discussed in the context of irrationality. In particular, β refers to the degree of present bias, a larger value of β implies less present bias. δ is called the long term discount factor. An elaboration of different measures of time discounting can be found in Wang et al. (2009). When we assume a linear utility function v( ), the two matching questions about time discounting can be represented as follows, where F 1year and F 10year denote the responses from the respective matching questions. 100 = β δ F 1year, (2a) 100 = β δ 10 F 10year. (2b) Proxies for time preferences can then be calculated as follows: 6

3a (3b) The median value of β across all 31 countries under consideration is 0.65 (mean = 0.63, standard deviation = 0.30). The median value of δ is 0.84 (mean = 0.83, standard deviation = 0.04). Note that the variation in the present bias β is much higher than the variation in the long term discount factor δ. The responses to the two questions are highly correlated ( = 0.78), however, the present bias parameter β and the long term discount factor δ are only moderately correlated ( = 0.25), indicating that the two components may correspond to different psychological constructs. Figure 2 plots the parameter estimates of β and δ for each country under consideration. [Insert Figure 2 about here.] 4 Theoretical background There is little empirical or theoretical work on how time preferences affect the planning horizons of households. We consider an individual with an objective function according to the quasi hyperbolic discounting model of (1) that takes a loan of amount L at t = 0 with maturity M. r M is the gross interest rate paid per period for loans with maturity M, such that periodical interest payments of r M L are made at times t = 1,, M. In addition, at time t = M the loan is redeemed. We assume that the vol 7

ume L of the loan as well as consumption levels z 0, z 1,, z T before borrowing are given. Summarizing, this leads to the following objective which is to be maximized with respect to the choice of M: Apparently, because of the exogeneity of z 0, L, and β, the present bias β is irrelevant for the optimal value of M. Moreover, in order to simplify the problem and to focus on the impact of δ for the determination of M, we assume u to be linear. As a consequence, consumption values z t for all t and the loan volume L do not affect the optimal solution for M any longer. Thus our objective function reduces to which has to maximized with respect to M. Note that. We thus have: For a flat term structure of interest rates, i.e. r M = r for all M, the sign of is simply identical to that of 1 δ (1+r) because of ln /(1 ) < 0 for 0 < < 1. When switching from maturity M to M+1, the individual saves $1 at time M, but has to pay an additional dollar amount of 1+r at time M+1 which is subjectively discounted by the long term discount factor. For rather impatient individuals, i.e. individuals with small values of, this maturity prolongation will be advantageous. In fact, the sign of 1 δ (1+r) is identical to that 1/r δ /(1 δ). δ /(1 δ) is the subjective net present value (or cumulated utility) of a consumption stream of $1 from t = 1 to, while 1/r is the objective net present value for a perpetuity of $1. If the individual thus discounts future payments more than the capital market does (that is, if the individual is sufficiently impatient ), we get a border solution M. The opposite (M 0) is true for 1/r < δ /(1 δ). For a flat term structure of interest rates, there are thus only border solutions possible which is intuitively appealing because of u being linear. When 8

looking at the term structures of interest rates as of the beginning of 2007 for the 31 countries depicted in Figure 2, there are 17 of them for which the Bloomberg financial database offers data for at least 10 different maturities. For all of them (and for all maturities) 1/r M > δ /(1 δ) holds. Because of the empirical evidence, we rely on this inequality in the following. From (6), we obtain the following first order condition for the optimal value of M: Apparently, for an inner solution, i.e. 0 < δ M < 1, we need r M > 0 and thus an at least locally increasing yield curve. We are mainly interested in the relationship between the optimal value of M* and preference parameter δ. However, even for interest rates r M that are linearly increasing in maturity M, it is not possible to derive an unambiguous sign of M*/ δ. In order to understand this result, once again we have to take a look at (6) which reveals that there are two different marginal utility effects at work when maturity M is varied. The first term in brackets on the right hand side of (6) is identical to and thus describes the negative utility effect of higher periodic interest payments when M is increased due to a normal yield curve. Apparently, this negative utility effect of increased periodic interest payments becomes weaker for an individual s smaller patience level δ, as higher periodic interest payments now hurt less. The second term describes the consequences of a marginal increase in maturity M under the assumption of a flat term structure of interest rates. It consists of two components. The difference in brackets characterizes the consequences of switching from M to M+1 while the factor ln /(1 ) corrects for the fact that M is indeed only marginally increased. In what follows, we examine somewhat more thoroughly the term in brackets. As already pointed out, simply is (for L = 1) the 9

net value of an increase in maturity M to M+1 as evaluated at the original time M for constant interest rates. Once again, smaller patience levels make higher maturities more attractive. However, in order to compute the net effect as seen from t = 0, this difference has also to be discounted by. For smaller patience levels, the net effect of maturity prolongation is discounted more severely and thus ceteris paribus loses relevance as seen from time t = 0. The consequences of varying values of on the strength of this second marginal utility effect are thus unclear. In particular, this second positive utility effect of maturity prolongation may also become weaker with smaller levels of. Therefore, it is an empirical issue whether smaller values of the long term discount factor lead to higher or smaller optimal maturities M. Against the background of our theoretical considerations, we therefore now turn to a numerical analysis of the decision problem according to (5) based on actual market data. As mentioned above, we obtain historical yield curves from Bloomberg financial database for 17 countries. To alleviate the effects of the financial crisis, rates are per 01/01/2007. We linearly interpolate yield data for missing maturities but assume constant values before the first and after the last data point reported. As the collected zero bond rates i M differ slightly from the r M in our theoretical model we calculate the correct values using the relationship. For all maturities M = 0 to M = 100 we compute our objective of M, r M and δ. The optimal maturity M* is reached where with corresponding values is maximal. Figure 3 reports the observed relationship [Insert Figure 3 about here.] 10

The numerical analysis with real world data clearly exhibits a positive relationship between the longterm discount factor δ and the optimal maturity choice of loans (ρ = 0.74, p = 0.0007). Individuals who exhibit higher values of δ prefer liabilities with longer maturities. The relation can also be confirmed by a linear regression which estimates M* = 392.1 δ 267.3. Both coefficients and the F statistic for the overall fit are highly significant, and R 2 amounts to 0.56 in this simple regression model. Our computations rely on nominal interest rates. However, referring to inflation adjusted, i.e., real interest rates would only be necessary if varying inflation rates over time were to be taken into account (i.e. if inflation rates in future years vary considerably from those that prevailed when performing the INTRA survey). Because of a lack of data regarding inflation rates for the very far future, we refrain from such an approach. Nevertheless, we control for cross country differences in inflation rates in our regressions in the next sections. From our theoretical model and our numerical analysis we derive a positive relationship between the long term discount factor and the maturity of household loans. Individuals who are more patient in the long run tend to choose debt portfolios with higher maturities. Thus, one should expect that more patient households should choose loans with higher maturities resulting in larger ratios of long term to short term loans. Households which are more patient in the long run have longer planning horizons. Our findings can therefore be summarized as follows. Hypothesis 1: Regarding loans, countries with more patient individuals, i.e. higher long term discount factors, have longer planning horizons, whereas countries with less patient individuals have shorter planning horizons. The present bias does not affect planning horizons. 5 Empirical analysis 5.1 Methodology To empirically capture the relationship between the planning horizon of households, measured by the ratio of long term versus short term loans, and time preferences, represented by the present 11

bias parameter β and the long term discount factor δ, we estimate country level ordinary least squares regression models. Our basic regression model is of the form PlanHor = const + a β + b δ + Σ c i Control i + ε. Control i is a set of control variables representing the economic, institutional, and demographic development and risk preferences of a country. A description of these control variables can be found in Table 2. [Insert Table 2 about here.] In addition, we also include measures of risk preferences as control variables, in order to check the relationship of time preferences with risk preferences. These measures were derived from hypothetical lottery questions in a different section of the INTRA survey. According to basic prospect theory, these parameters are risk aversion towards gains, risk seeking towards losses, probability bias, and loss aversion (see, for example, Tversky and Kahneman, 1981). Table 3 illustrates the main summary statistics and cross correlations of our data. We standardize all variables before we estimate the regression models so that coefficients can be compared directly. [Insert Table 3 about here.] 5.2 Results The regression results for our basic model can be found in Table 4. The dependent variable is the planning horizon of households. The independent variables are time preferences of households, 12

represented by the present bias parameter β and the long term discount factor δ. The table reports the ordinary least squares estimates. In this as well as the subsequent regressions, we control for a number of variables. As pointed out previously, we include various economic, institutional, and demographic variables as well as risk preferences which might influence the planning horizon of households. Unfortunately data availability for some of our control variables is very limited, which does not allow us to control simultaneously for all of them. We include the control variables separately and in groups to obtain sample sizes of at least 20. [Insert Table 4 about here.] The first column reports the estimates of the basic specification, where we insert both variables for time preferences. While the present bias parameter β turns out to have little predictive power, the effect of the long term discount factor δ is positive and highly significant. In countries where we have higher values of the long term discount factor δ we have higher ratios of long term to shortterm loans. These findings turn out to remain stable throughout all other specifications where we include control variables and are perfectly in line with or theoretical considerations. Thus we infer that people who are more patient in the long term tend to have higher planning horizons. To account for the impact of outliers, we repeat our calculations by removing these. We remove all data which depart from the mean of the data by more than three times the standard deviation. All of our results remain robust so we abstain from reporting the estimates here. We check the robustness of our results by using an alternative specification of our dependent variable. We replace our dependent variable long term versus short term loans by a dummy which is equal to one if a country has a ratio of long term to short term loans above the median value of all countries, and zero otherwise. We model the probability of a longer planning horizon as a function 13

of time preferences and the controls that we have included before using a logistic regression model. The results can be found in Table 5. [Insert Table 5 about here.] We find that the logistic regression models have less explanatory power than the corresponding linear regression models, which is only logical, as our dummy variable for the planning horizon of households is less precise than the original variable. Nevertheless, we can confirm our general finding that countries which have higher values of the long term discount factor δ tend to have a longer planning horizon. Even though the long term discount factor δ loses its significance slightly in two of the regression models, its impact is positive and significant in the remaining models. 6 Discussion 6.1 Cultural explanations There has been an increasing awareness of the necessity to understand culture in order to better understand behavior. Although the importance of culture at the economic level has been established, it is still unclear in how far culture drives people s financial decisions (Hens and Wang, 2007). In order to investigate the effect of culture on planning horizons, we repeat the above analysis using direct measures of culture as main explanatory variables. Though culture is a very vague concept, many scientists developed different frameworks to describe cultural paradigms. Basically, such a framework consists of several cultural dimensions which can be used to categorize countries. One of the most influential approaches to characterize cultures has been developed by the Dutch sociologist Geert Hofstede during his cross country research on organ 14

izational cultures. The Hofstede (2001) cultural theory introduces five cultural dimensions that address basic societal problems. Individualism and Collectivism describes the relationship between the individual and the collectivity that prevails in a given society. Power Distance is the extent to which different societies handle human inequality differently. Masculinity and Femininity refers to the distribution of roles between genders. Uncertainty Avoidance deals with a society s tolerance for uncertainty and ambiguity and refers to its search for truth. Long term Orientation captures the society s time horizon and reflects to what extent it has a dynamic and future oriented mentality. Hofstede (2001) collected data on the cultural dimensions using the Value Survey Module VSM94. Each index is calculated from several questions to which people respond on a scale from 1 to 5. Questions from this module were included in the INTRA survey as well and are therefore directly available for each country of our analysis. However, data are similar to the original Hofstede data. Regarding the Hofstede model, one should expect a significant relationship between the dimension of Long term Orientation and the planning horizon, as the dimension of Long term Orientation is particularly concerned with the time horizon of a society. Hypothesis 2a: The Hofstede cultural dimension of Long term Orientation is positively related to the planning horizon of households in a country. Although Hofstede s framework has been commonly accepted as standard for the description of cultural differences, the Schwartz (1994) cultural theory has caught on among recent research papers. Schwartz s model overcomes some difficulties of Hofstede s approach: It is based on a careful theoretical elaboration and then empirically validated, it offers a more comprehensive set of value dimensions, it uses more recent data, it was obtained from more diverse regions including socialist countries, and it has been tested with two matched samples (Ng et al., 2007). The Schwartz cultural model specifies three bipolar dimensions representing basic problems confronting all societies: Autonomy versus Embeddedness, Egalitarianism versus Hierarchy, and Harmony versus Mastery. An 15

emphasis on the cultural type at one pole of a dimension typically accompanies a de emphasis on the polar type. Autonomy relates to the desirability of individuals independently pursuing their own ideas and draws on values like CREATIVITY, SELF INDULGENCE, FREEDOM, and VARIED LIFE. Embeddedness emphasizes the maintenance of the status quo and relies on values like SOCIAL ORDER, RESPECT FOR TRADITION, RECIPROCATION OF FAVORS, and WISDOM. Egalitarianism relates to sharing basic interests and showing concern for everyone s welfare and bears on values like EQUALITY, SOCIAL JUSTICE, RESPONSIBILITY, and HELPFULNESS. Hierarchy corresponds to the legitimacy on an unequal distribution of power and applies to the values SOCIAL POWER, AUTHORITY, WEALTH, and HUMBLENESS. Harmony points to accepting the world as it is and rests on values like UNITY WITH NATURE, PROTECTING THE ENVIRONMENT, and A WORLD IN PEACE. Mastery supports the idea of getting ahead through active self assertion and refers to values like AMBITION, SUCCESS, and INFLUENCE. These dimensions were constructed out of 45 basic value types (as indicated in the above description). Data describing culture using the Schwartz framework have been collected continuously since 1988 and are currently available for 73 countries from the Israel Social Science Data Center at the Hebrew University of Jerusalem. Using the Schwartz Value Survey, people in each country are asked to judge the 45 basic values according to their importance as Guiding Principles in Her/His Life. The response options range from 7 (of supreme importance) to 1 (opposed to my values). For each country, the values of the different cultural dimensions are then calculated out of these results. Regarding the Schwartz model, one should expect a significant relationship between the dimension of Autonomy versus Embeddedness, which is concerned with the relationship between the person and the group, and planning horizons. The bipolar dimension refers to the desire for independence of a person from the group, with low independence in Embeddedness societies, and high independence in Autonomy societies. Higher independence involves a high level of freedom and ability to make own decisions in life and thus induces higher responsibility for oneself. 16

The Hofstede dimension of Long term Orientation captures the society s time horizon and reflects to what extent it has a dynamic and future oriented mentality. Values associated with Long term Orientation are thrift and perseverance; values associated with Short Term Orientation are respect for tradition, fulfilling social obligations, and protecting one s face. Autonomy relates to the desirability of individuals independently pursuing their own ideas and draws on values like creativity, selfindulgence, freedom, and varied life. Embeddedness emphasizes the maintenance of the status quo and relies on values like social order, respect for tradition, reciprocation of favors, and wisdom. This brief description of dimensions shows that the Hofstede and Schwartz dimensions exhibit certain similarities in their construction, as the basic values they refer to overlap considerably. The Schwartz dimension of Autonomy is related to Long term Orientation according to Hofstede, whereas the Schwartz dimension of Embeddedness is related to Short term orientation according to Hofstede. This reasoning indicates that the dimension of Autonomy versus Embeddedness is related to the time horizon of a society. Countries with a higher emphasis on Embeddedness should have shorter planning horizons, whereas countries with a higher emphasis on Autonomy and thus independence should have longer planning horizons in household debt portfolios. Higher debt maturity simply makes a borrower for a longer time interval independent from the capital market than short debt maturity at least if there are refinancing needs. Hypothesis 2b: The Schwartz cultural dimension of Autonomy versus Embeddedness is positively related to the planning horizon of households in a country. In order to capture the influence of culture on planning horizons, we replace the variables representing time preferences by the cultural dimensions of the Hofstede and Schwartz model respectively in our above regression model. This leads to regression models of the form PlanHor = const + Σ a i CulturalDimension i + Σ c i Control i + ε. For the Hofstede model, CulturalDimension i refers to the five cultural dimensions, Individualism and Collectivism, Power Distance, Masculinity and Femininity, Uncertainty Avoidance, and Long term 17

Orientation. For the Schwartz model, CulturalDimension i refers to the three bipolar cultural dimensions Autonomy versus Embeddedness, Egalitarianism versus Hierarchy, and Harmony versus Mastery. In consideration of the bipolarity of the cultural dimensions, we collapse each bipolar cultural dimension to a single one by calculating its difference, i.e. Autonomy Embeddedness, Egalitarianism Hierarchy, and Harmony Mastery, as suggested in Schwartz (2004). Control i is the same set of control variables representing the economic, institutional, and demographic development and risk preferences of a country as above. The above Hypothesis 2a cannot be supported. The Hofstede model seems unsuitable to predict the planning horizon of household debt portfolios; in almost all regressions the variable for the Longterm Orientation is insignificant. [Insert Table 6 about here.] Hypothesis 2b can be confirmed. Unlike the Hofstede cultural model, the Schwartz model has a good amount of predictive power for the planning horizon of households. The cultural dimension of Autonomy versus Embeddedness is highly significant throughout all the regression models. [Insert Table 7 about here.] 6.2 The foundations of time preferences The preceding section suggests that culture does have a significant effect on economically relevant beliefs and attitudes. Variation in these preferences directly impacts the financial choices individuals 18

make. Culture is defined as a collective programming of the mind, and deeply rooted in every society. The mental programming is learned from and shared with people from the culture one lives in, and who have gone through the same learning process (Hofstede, 2001). Hence culture has relatively stable and long term effects on how individuals understand the world, think, and make decisions. According to this rationale, culture may not only manifest itself in different patterns of investment behavior, but should be regarded as in important factor for time preferences itself. So far, Section 5.2 has dealt with the direct impact of time preferences on planning horizons and Section 6.1 has focused on the direct relationship between culture and planning horizons. The following section elaborates on the direct impact of culture on time preferences. If such an influence can be confirmed, this would complete the picture of culture, time preferences, and planning horizon, and explain the mechanism how culture works. We use the same proxies for national culture as above, the dimension of Long term Orientation from the Hofstede cultural theory and the dimension of Autonomy versus Embeddedness from the Schwartz cultural theory. Following the rationale from above, we expect that countries with a higher emphasis on the Schwartz cultural dimension of Autonomy and a higher emphasis on the Hofstede cultural dimension of Long term orientation should have higher values of the long term discount factor δ, implying more patience in the long run. Values like perseverance and freedom are important in those countries, increasing the significance of long term planning. Countries with a higher emphasis on the Schwartz cultural dimension of Embeddedness and a lower emphasis on the Hofstede cultural dimension of Long term orientation should have lower values of the long term discount factor δ, implying less patience in the long run. Values like respect for tradition and social order are important in those countries, diminishing the relevance of long term planning. As both the Schwartz and Hofstede cultural dimensions do not relate to short term behavior, we do not expect any significant relationships between the regarded dimensions and the present bias parameter β. 19

Hypothesis 3: Countries with a higher emphasis on the Schwartz cultural dimension of Autonomy and a higher emphasis on the Hofstede cultural dimension of Long term orientation are more patient in the long run. Countries with a higher emphasis on the Schwartz cultural dimension of Embeddedness and a lower emphasis on the Hofstede cultural dimension of Long term orientation are less patient in the long run. There are no differences in short term time preferences. To examine the direct effect of culture on time preferences, we estimate ordinary least squares linear regression models of the form TimePref j = const + Σ a i CulturalDimension i + ε. Time preferences are measured either by the present bias parameter β or the long term discount factor δ. Culture is measured either by the five Hofstede cultural dimensions Individualism and Collectivism, Power Distance, Masculinity and Femininity, Uncertainty Avoidance, and Long term Orientation, or the three bipolar Schwartz cultural dimensions Autonomy versus Embeddedness, Egalitarianism versus Hierarchy, and Harmony versus Mastery. The results can be found in Table 8. [Insert Table 8 about here.] The Hofstede cultural dimension of Long term orientation exhibits no significant influence on time preferences, whereas the Schwartz dimension of Autonomy versus Embeddedness is highly significant. This result is similar to the one reported in the above section, where the Hofstede model had no effect on the planning horizon of households, either. This oddity might be caused by the major drawbacks of the Hofstede model mentioned above. The Schwartz cultural dimension of Autonomy versus Embeddedness seems particularly important for time preferences of households. Countries with a higher emphasis on the cultural dimension of Autonomy tend to have higher long term discount rates δ and thus tend to be more patient in the long run, countries with a higher emphasis on the cultural dimension of Embeddedness tend to have lower long term discount rates δ and thus 20

tend to be less patient in the long run. Due to lack of data we cannot provide a more formal causal analysis but leave this question for future research. 7 Robustness As our analysis so far is established at the country level, this section provides a comparison at the firm level and individual level to prove the robustness of our results. Firm level results can be obtained from recent literature on national culture and corporate capital structure. Individual level results are derived from cross country surveys on culture, time preferences, and household finance. In recent papers, Chang et al. (2009) and Zheng et al. (2010) analyze the impact of national culture on the choice of corporate debt maturity. Corporate debt maturity is actually a proxy for the planning horizon of firms, indicating whether firms prefer short term or long term debt. Chang et al. (2009) argue that the Hofstede cultural dimension of Uncertainty Avoidance is negatively related to the overall debt maturity in a country, both from a borrower s and lender s perspective, and show substantial cross country evidence to confirm this hypothesis. Zheng et al. (2010) maintain a lender s viewpoint exclusively and claim that countries characterized by high Uncertainty Avoidance, high Collectivism, high Power Distance, and high Masculinity issue debt with shorter maturity, and confirm these findings in an extensive empirical analysis. When turning to the Schwartz cultural dimensions in the robustness section, Zheng et al. (2010) find that the Schwartz cultural dimension of Embeddedness is associated with shorter debt maturity. Referring to the bipolarity of the Schwartz dimensions, this finding yields that the cultural dimension of Embeddedness is associated with a shorter planning horizon, whereas the cultural dimension of Autonomy is related with a longer planning horizon. This result supports our Hypothesis 2 at the firm level. Ideally, one might wish to have individual level data on the planning horizon, time preferences, and culture in order to derive a more integral view on the proposed relationships. To extend our analysis in this vein, we collect cross country survey data correspondingly. A total of 449 economic students 21

participated in the surveys, which were conducted in Germany and Singapore, to ensure sufficient cultural variance in the sample. To capture the planning horizon, participants were asked to judge the question It is important to me that I am allowed to pull back invested money if needed. on a scale from 1 (strongly agree) to 5 (strongly disagree). Certainly, possibilities for pulling back money are more important for individuals with short term debt than for those with long term debt. Time preferences and data on the Hofstede cultural dimensions were derived in a similar manner as described above. Questions on the personal background of the respondents were also included in the survey, replicating the control variables used before. To test our hypotheses on the individual level, we estimate multivariate regression analysis similar to those before. The results are consistent with our country level analyses. We find that the long term discount factor δ has a positive and significant impact on the planning horizon (p = 0.00), which supports our Hypothesis 1, that countries where people are more patient have longer planning horizons. This result holds when controlling for the present bias parameter, risk preferences, other cultural dimensions, sex, nationality, religion, age, housing conditions, educational background, employment, wealth, consumer and savings behavior. However, the Hofstede cultural dimension of Long term Orientation has no significant impact on the planning horizon (p = 0.24) and the long term discount factor (p = 0.27), which is in accordance with our results concerning Hypothesis 2a and 3. Unfortunately, we have no survey data on the Schwartz cultural dimensions of Autonomy and Embeddedness. The detailed empirical results on the individual level can be obtained from the authors upon request. 8 Conclusion Our paper explores the role of time preferences on the planning horizon of households. We focus on household liabilities and develop relationships between the maturity of loans and time preferences in a theoretical approach and a numerical analysis. We derive a testable hypothesis and test our theory in an extensive empirical analysis. Time preferences are measured by the present bias parameter β and the long term discount factor δ derived from the INTRA (2009) study. Planning hori 22

zons of households are measured by the ratio of long term versus short term loans from the EIU (2009) database. Controlling for various economic, institutional, and demographic variables as well as risk preferences of a country, we show that the long term discount factor δ is a sound predictor for the planning horizon of household debt. Household debt portfolios are well reconciled with time preferences of households. We further investigate on the foundations of time preferences and explore the role of culture in this context. Despite its fundamental status, culture has long been neglected in finance research but gained attention in the last few years. We show that culture is not only a reasonable predictor for planning horizons, but also a sound predictor for time preferences as well. This completes the picture about the mechanisms between culture, time preferences and planning horizon. Culture can be established as the essential and constitutional element, it is the driving force for decisions on planning horizons in household debt portfolios. Its influence is exerted via time preferences, as the combination of our empirical analyses shows. [Insert Figure 4 about here.] This paper highlights the importance of culture for financial decisions. As we are provided with a unique dataset that combines data from the cultural level, time preferences and financial outcomes, we are able to derive a number of significant relationships between the variables. Moreover, we are able to explain in detail how culture works, which is a major advancement, compared to existing studies (Breuer and Salzmann, 2010). However, there are still countless open questions on the relationship between culture and investment behavior, and our paper is only a first step towards an extensive field of future research. 23

References Bilias, Y., Georgarakos, D., Haliassos, M. (2008): Equity Culture and the Distribution of Wealth, SSRN Working Paper. Breuer, W., Salzmann, A. (2010): National Culture and Household Finance, SSRN Working Paper. Brown, J., Ivković, Z., Smith, P., Weisbenner, S. (2008): Neighbors Matter: Causal Community Effects and Stock Market Participation, in: The Journal of Finance, 63, 1509 1531. Campbell, J. (2006): Household Finance, in: The Journal of Finance, 56, 1553 1604. Chang, K., Lee, S.W., Yi, H. C. (2009): Does National Culture Influence the Firm s Choice of Debt Maturity? Working Paper. Guiso, L., Haliassos, M., Japelli, T. (2002): Household Stockholding in Europe: Where Do We Stand and Where Do We Go? SSRN Working Paper. Guiso, L., Sapienza, P., Zingales, L. (2008): Trusting the Stock Market, in: The Journal of Finance, 63, 2557 2600. Graham, R.J. (1981): The Role of Perception of Time in Consumer Research, in: Journal of Consumer Research, 7, 350 401. Haliassos, M., Bertaut, C. (1995): Why Do So Few Hold Stocks?, in: The Economic Journal, 105, 1110 1129. Hens, T., Wang, M. (2007): Does Finance have a Cultural Dimension? FINRISK Working Paper, No. 377. Hofstede, G. (2001): Culture s Consequences: Comparing Values, Behaviors, Institutions, and Organizations across Nations. Hong, H., Kubik, J., Stein, J. (2004): Social Interaction and Stock Market Participation, in: The Journal of Finance, 54, 137 163. La Porta, R., Lopez de Silanes, F., Shleifer, A., Vishny, R. (1999): The Quality of Government, in: Journal of Law, Economics, and Organization, 15, 222 279. Ng, S., Lee, J., Soutar, G. (2007): Are Hofstede's and Schwartz's Value Frameworks Congruent?, in: International Marketing Review, 24, 164 180. Puri, M., Robinson, D. (2007): Optimism and Economic Choice, in: Journal of Financial Economics, 86, 71 99. Renneboog, L., Spaenjers, C. (2010): Religion, Economic Attitudes, and Household Finance, SSRN Working Paper. Schwartz, S. (1994): Beyond Individualism/Collectivism: New Cultural Dimensions of Values, in: Kim, U., Triandis, H., Kagitcibasi, C., Choi, S. C., Yoon, G. (Eds.): Individualism and Collectivism: Theory, Method, and Applications, Sage, Beverly Hills, 85 99. Schwartz, S. (2004): Mapping and Interpreting Cultural Differences around the World. Vinken, H., Soeters, J., Ester, P. (Eds.), Comparing Cultures Dimensions of Culture in a Comparative Perspective, Brill, Leiden, 43 73. Tversky, A., Kahneman, D. (1981): The Framing of Decisions and the Psychology of Choice, in: Science 211, 453 458. Wang, M., Rieger, M., Hens, T. (2010): How Time Preferences Differ: Evidence from 45 Countries, Swiss Finance Institute Research Paper, No. 09 47. 24

Zheng, X., El Ghoul, S., Guedhami, O., Kwok, C.C.Y. (2010): The Influence of National Culture on Corporate Debt Maturity, Working Paper. 25

Country Ratio of short term to long term assets Ratio of short term to long term loans Argentina 4.48 % 105.52 % Canada 4.32 % 38.09 % Germany 0.29 % 5.88 % Italy 1.07 % 14.52 % Norway 3.50 % 5.90 % Russia 4.48 % 36.01 % Taiwan 1.09 % 48.28 % Thailand 4.47 % 36.01 % United Kingdom 7.56 % 19.84 % United States 4.38 % 30.93 % Table 1: Short term vs. long term assets and loans. The table shows the proportion of short term assets or liabilities to long term assets or liabilities. Short term assets or liabilities typically comprise assets or loans respectively with an original maturity of one or two years or less. Long term assets or liabilities typically comprise assets or loans respectively with an original maturity of more than one or two years. 26

Variable Description GDP per head Nominal GDP divided by population. Data is averaged over 2003 to 2007. Source: EIU WorldData (2009) Inflation rate (%) Average inflation rate. Inflation rate is calculated as percentage change in consumer price index. Data is averaged over 2003 to 2007. Source: EIU WorldData (2009) Bank sector The ratio of private sector credit to GDP. Private sector credit measures total loans to development (%) the corporate and household sectors. Data is averaged over 2003 to 2007. Source: EIU WorldData (2009) Stock market The ratio of stock market capitalization to GDP. Stock market capitalization measures development (%) the local stock market capitalization excluding investment funds. Data is averaged over 2003 to 2007. Source: EIU WorldData (2009) State State takes the value 1 if the country is (or previously was) a socialist country, and 0 otherwise. Shareholder rights Aggregation of numerous measures of shareholder rights. Higher values indicate stronger shareholder protection. Source: Financial Development Report (2008) Contract enforcement Dependency ratio (%) Religion Tertiary enrollment Aggregation of measures for the effectiveness of law making bodies, judicial independence, irregular payments in judicial decisions, number of procedures to enforce a contract, time to enforce a contract, cost of enforcing contracts, strength of investor protection, and time to close a business. Higher values indicate stronger contract enforcement. Source: Financial Development Report (2008) The dependency ratio is the sum of the ratio of the population under age 15 to the population ages 15 to 64 and the ratio of the population over age 64 to the population ages 15 to 64. Data is averaged over 2003 to 2007. Source: EIU WorldData (2009) The percentage of the population with Protestant, Catholic or Muslim beliefs. Source: La Porta et al. (1999) Gross tertiary enrollment rate. This variable is the ratio of total enrollment, regardless of age, to the population of the age group that officially corresponds to the level of education shown. Estimates are based on UNESCO s classification of education levels. Tertiary instruction is provided at a university, teachers college, or higher level professional school. Source: Financial Development Report (2008) Table 2: Variables and sources. The table lists the descriptions of control variables and its sources. 27