Cross-country variation in economic preferences and the asset composition of international investment positions

This is a self-archived version of an original article. This version may differ from the original in pagination and typographic details. Author(s): Title: Year: Version: Copyright: Rights: Rights url: Please cite the original version: CC BY 4.0 https://creativecommons.org/licenses/by/4.0/ Cross-country variation in economic preferences and the asset composition of international investment positions © 2024 The Author(s). Published by Elsevier Ltd. Published version Nieminen, Mika; Kuziemska-Pawlak, Kamila Nieminen, M., & Kuziemska-Pawlak, K. (2024). Cross-country variation in economic preferences and the asset composition of international investment positions. Journal of International Money and Finance, 146, Article 103130. https://doi.org/10.1016/j.jimonfin.2024.103130 2024 Journal of International Money and Finance 146 (2024) 103130 Available online 2 July 2024 0261-5606/© 2024 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/). Cross-country variation in economic preferences and the asset composition of international investment positions Mika Nieminen a , * , Kamila Kuziemska-Pawlak b a School of Business and Economics, University of Jyv¨ askyl¨ a, Mattilanniemi 2, Building Agora, P.O. Box 35 FI-40014, Jyv¨ askyl¨ a, Finland b University of Lodz, Narutowicza 68, 90-136 Lodz, Poland ARTICLE INFO JEL classification: E71 F21 F32 Keywords: Time Preference Risk Aversion Negative Reciprocity International Investment Position Equity Investment Foreign Direct Investment ABSTRACT A stylized fact of international capital markets is that advanced countries tend to be long and developing countries short in risky assets (i.e., portfolio equity and foreign direct investment (FDI)). In other words, residents of advanced countries hold a larger stock of portfolio equity abroad than residents of developing countries, and firms in advanced countries have more foreign subsidiaries than firms in developing countries. This paper is the first to utilize a large-scale international survey on economic preferences to propose a behavioral explanation for the heterogeneity in the asset composition of international investment positions. We provide robust empirical evidence that countries with a high time preference (i.e., patience) or a high risk preference (i.e., risk-taking) tend to have a positive net international investment position and a positive net risky position. In addition, we show that countries with a high degree of negative reciprocity (e.g., willingness to punish for unfair action) tend to have a positive net FDI position. Overall, our findings suggest that preferences are important determinants of cross-country variation in net foreign asset positions. 1. Introduction The international financial landscape is characterized by the growth of cross-border gross asset and liability positions and large differences in the asset structure of these positions (Gourinchas and Rey, 2014). If the seven largest advanced economies, the G7, are compared with the four major emerging economies, BRIC, there seems to be a clear division between the two country groups. The sum of cross-border assets and liabilities has increased much more in the G7 countries than in the BRIC countries (see Figure A1 in the Appendix). Furthermore, residents in the G7 countries have heavily invested in portfolio equity abroad and foreign direct investment (FDI), which can be considered risky assets, whereas residents in the BRIC countries have preferred safer assets such as foreign government bonds. As a result, residents in the G7 countries hold more risky assets abroad than foreigners in the G7 countries (i.e., a positive net risky position) and residents in the BRIC countries own less risky assets abroad than foreigners in the BRIC countries (i.e., a negative net risky position) (see Figure A2 in the Appendix). The positive relationship between risk preference and net risky position has been modeled in the finance literature, for example, by Dumas (1989) and in the economics literature, for example, by Gourinchas and Rey (2022) and Stepanchuk and Tsyrennikov (2015). However, possibly due to the lack of a global dataset on economic preferences, the relationship between economic preferences and net positions of different asset classes has not been empirically tested * Corresponding author. E-mail addresses: [email protected] (M. Nieminen), [email protected] (K. Kuziemska-Pawlak). Contents lists available at ScienceDirect Journal of International Money and Finance journal homepage: www.elsevier.com/locate/jimf https://doi.org/10.1016/j.jimonfin.2024.103130 Journal of International Money and Finance 146 (2024) 103130 2 before our paper. We utilize the Global Preferences Survey 1 (GPS) introduced by Falk et al. (2018) and provide robust empirical evidence that cross-country variation in economic preferences explains a large share of the heterogeneity in the asset composition of international investment positions. In other words, our analysis suggests that the division in the asset structure of gross positions between the G7 and the BRIC countries arises from the differences in their preferences as the inhabitants of the G7 countries have, on average, a higher degree of patience and risk-taking than those in the BRIC countries. 2 In his 2012 Ely Lecture, Obstfeld (2012) noted that the growth of gross asset and liability positions poses a danger to financial stability if the composition of the asset and liability side differs. 3 Illiquid assets cannot pay short-term debts. According to Gourinchas and Rey (2007), the asymmetric composition of the US external balance sheet (i.e., high return on risky assets and low-yielding safe liabilities) has provided a sizeable excess return for it. Even though the composition of external balance sheets might have these important implications, only a few papers have examined the determinants of external positions. When Lane and Milesi-Ferretti (2001) consider country’s size, GDP per capita and trade openness, they find that countries with a high GDP per capita invest heavily in FDI, and countries with a high degree of openness receive a lot of FDI. Consequently, countries with a higher GDP per capita tend to have a positive net FDI position, while there is a negative correlation between trade openness and net FDI position. Mendoza et al. (2009) show that if countries differ in contract enforceability, a country with the highest degree of enforcement of financial contracts will accumulate a net positive position in risky productive assets but a much larger negative position in safe bonds. This is due to the ability of this country to insure away idiosyncratic risk. Accordingly, Vermeulen and de Haan (2014) investigate the relationship between a country’s domestic financial development and the composition of its net foreign asset position. They find that financial development leads to higher net equity (i.e., net risky position), to a lower net debt position, and overall to a lower net foreign asset position. In our paper, we show that economic preferences are more important determinants of net risky positions than financial development is. 4 Our main interest in terms of economic preferences are time preference and risk preference. Time preference is one of the most fundamental concepts in economics, and according to the Metzler diagram, an agent with a higher time preference factor becomes a net creditor while an agent with a lower time preference factor becomes a net debtor (see, e.g., Obstfeld and Rogoff, 1996, pp. 31-34; Metzler, 1960). As already noted, the relationship between risk preference and net risky position has been theoretically modeled both in finance and economics (see, e.g., Dumas, 1989; Gourinchas and Rey, 2022; Stepanchuk and Tsyrennikov, 2015). Yet, in addition to time and risk preference, we test all other preference measures which are included in the Global Preferences Survey (GPS): negative reciprocity, positive reciprocity, altruism and trust. We find that negative reciprocity (i.e., the willingness to punish unfair behavior even at one’s own cost) has explanatory power on net FDI positions. Investment is classified as direct investment (FDI) when it means having control or a significant degree of influence on the management of an enterprise in another economy (IMF, 2009, p. 100). According to the empirical literature on FDI determinants, contract enforcement increases FDI attractiveness (see, e.g., Bailey, 2018; Contractor et al., 2020). In the behavioral economics literature, (negative) reciprocity is seen as a contract enforcement device (see, e. g., Fehr et al., 1997; Falk et al., 2018). However, we are not aware of a theoretical model that would connect negative reciprocity to the net FDI positions. The empirical literature on the determinants of net international investment positions is scarce. This is, to some extent, surprising because, as Obstfeld (2012) states, the main reason why current accounts matter is that, over the long run, the accumulated current account tracks the net foreign asset position. While introducing an updated version of the External Wealth of Nations database, Lane and Milesi-Ferretti (2018) provide a descriptive analysis of the cross-border holdings of financial assets and liabilities. Our estimates of external assets and liabilities are taken from the External Wealth of Nations database. Turrini and Zeugner (2019) adopt the concept of net international investment position norms, obtained as cumulated current account norms, to estimate country-specific benchmarks. They consider a very large set of explanatory variables but not economic preferences. Nieminen (2022) examines the relationship between time preference and net foreign asset position, providing empirical evidence for a positive association between the two. Yet, he considers neither the asset composition of international investment positions nor other preferences. According to the current account literature (see, e.g., Chinn and Prasad, 2003; Chinn and Ito, 2007; Gruber and Kamin, 2007; Ca’ Zorzi et al., 2012; Coutinho et al., 2022), macroeconomic, demographic, and institutional factors are the most important medium-term determinants of current accounts. As the current account balance measures net capital flow, net foreign asset positions and current account balances are closely related in the long run. Our set of control variables is derived from the current account literature, which is much larger than the literature on net international investment positions. We use Bayesian Model Averaging (BMA) as a model selection method for finding the most important determinants of total net foreign asset position and net positions of different asset classes. The BMA method has been used in the current account literature (see Ca’ Zorzi et al., 2012; Coutinho et al., 2022). Among the economic preferences that we include in our analysis, time preference has been studied the most in economics. Based on previous empirical studies, patience is positively correlated with macro-level outcomes such as GDP per capita and saving rates (Sunde 1 The Global Preferences Survey is the first global dataset on economic preferences that is representative at the country level. 2 Based on the Global Preferences Survey, the median of patience is 0.54 and the median of risk-taking is −0.03 for the G7 countries. In contrast, the median of patience is −0.09 and the median of risk-taking is −0.26 for the BRIC countries (measured as differences to the world mean in the standard deviation of the respective preference measure). 3 Cat˜ ao and Milesi-Ferretti (2014) find that the ratio of net foreign liabilities to GDP is a significant crisis predictor, and this is mainly due to the net position in debt instruments. In other words, from the recipient’s perspective, debt is riskier than equity or FDI. 4 Stepanchuk and Tsyrennikov (2015) point out that as both an increase in risk aversion and an increase in the variance of income induce precautionary savings, the assumption of differences in risk aversion and the assumption of differences in financial development are two sides of the same coin. M. Nieminen and K. Kuziemska-Pawlak Journal of International Money and Finance 146 (2024) 103130 3 et al., 2022), as well as with current account balances and total net foreign asset positions (Nieminen, 2022). These macro-level findings are consistent with experimental individual-level studies that show that patience is positively related to saving (Sutter et al., 2013; Falk et al., 2018) and negatively related to indebtedness (Meier and Sprenger, 2010). Korff and Steffen (2022) consider a wider set of preferences and provide evidence that if countries differ in negative reciprocity, they tend to trade less with each other. Our contribution derives from combining data on external assets and liabilities (the External Wealth of Nations database) with a dataset on economic preferences (the Global Preferences Survey, GPS) and testing the hypotheses on the relationships between the components of international investment position and time, as well as risk preferences that have previously been formulated but not yet tested in the economics literature. Overall, in our empirical analysis, we find that 1) patience is a very important determinant of total net foreign asset positions, net risky positions, and net FDI positions with a positive coefficient; 2) risk-taking is an important determinant of net risky positions with a positive coefficient; and 3) negative reciprocity is an important determinant of net FDI positions with a positive coefficient. The rest of the paper is structured as follows. Section 2 outlines a simple theoretical framework and testable hypotheses. Data and methodology are described in Section 3. Section 4 presents the empirical results and several robustness checks. Section 5 offers a concluding discussion. 2. Theoretical framework and testable hypotheses In this Section, we describe a simple theoretical framework from which the relationship between time preference and net foreign asset position as well as the relationship between risk preference and net risky position can be derived. In all variants, we assume that the world consists of two countries, A (“advanced”) and E (“emerging”), which are populated by representative agents. 2.1. Cross-country variation in time preference and net foreign asset position Agents in country A have a higher time preference factor than agents in country E. Consequently, country A has a lower autarky interest rate than country E. There is only one security, which is risk-free. As countries A and E become financially integrated, A will become a net lender and E a net borrower due to the difference in autarky interest rates. Hence, we expect the following: Hypothesis 1. (There is a positive relationship between time preference and net foreign asset position) The positive relationship between time preference and total net foreign asset position has been illustrated in international macroeconomics by the Metzler diagram (see, e.g., Obstfeld and Rogoff, 1996, pp. 31-34; Metzler, 1960). 5 2.2. Cross-country variation in risk preference and net risky position Agents in country A have a lower risk avoidance than agents in country E. Consequently, country A has a lower risk premium than country E. There are two securities that differ in riskiness and expected return; the riskier security has a higher expected return. As countries A and E become financially integrated, A will invest in risky assets in country E due to the difference in risk premium in autarky. Hence, we expect the following: Hypothesis 2. (There is a positive relationship between risk preference and net risky position) The positive relationship between risk preference and net risky position has been modeled in the finance literature, for example, by Dumas (1989) and in the economics literature, for example, by Gourinchas and Rey (2022), Maggiori (2017), and Stepanchuk and Tsyrennikov (2015). 6 3. Data and econometric methodology In this Section, we describe our data and econometric methodology. 3.1. Data Our sample consists of 76 countries in the period 2010–2019. The listing of countries is presented in Table A2 in the Appendix. We include all countries for which we have data on both net positions and economic preferences. For some control variables, there are a 5 If the world consists of two countries with constant but different rates of time preferences, there cannot be a steady state with international mobility of financial capital. To solve this problem, overlapping-generations models and intertemporally nonadditive preferences have been suggested (see, e.g., Buiter, 1981, Fukao and Hamada, 1989; Obstfeld, 1990). 6 The first version of Gourinchas and Rey (2022) is from 2010. Maggiori (2017) assumes that countries differ in financial development. He interprets this difference as the reason why risk preferences might differ. According to Stepanchuk and Tsyrennikov (2015), the assumption that individuals in the US are less risk-averse than elsewhere is another way of modelling the fact that the U.S. has more developed financial markets than the rest of the world. They model the combined effect of dominant currency and difference in risk-aversion. M. Nieminen and K. Kuziemska-Pawlak Journal of International Money and Finance 146 (2024) 103130 4 few missing observations; this typically results in 70 countries in the regression analysis for the period. 7 Our sample period ends in 2019 because it is the most recent year for which we have data on net positions for all 76 countries. The descriptive statistics are provided in Table 1, a detailed description of all the variables and data sources is provided in Table A1 in the Appendix, and the correlation matrix is provided in Table A3 in the Appendix. The possible multicollinearity between explanatory variables is tested by the variance inflation factor (VIF) and reported below the regression tables. 3.1.1. Dependent variables We have three dependent variables: 1) Total net foreign asset position, 2) Net risky position, and 3) Net FDI position. We also tested other asset classes, but these three were chosen for the following reasons: 1) The decomposition of external assets and liabilities into net risky and net safe position is widely used in international macroeconomics (see, e.g., Gourinchas and Rey, 2019); 2) The net FDI position is the main component of the net risky position 8 ; 3) Our empirical analysis in Section 4 confirms the relationships between our dependent variables and economic preferences. 9 Data on net positions are from the External Wealth of Nations database (see Lane and Milesi-Ferretti, 2018). The net positions are stock variables, and they are measured at the end of the particular year. The total net foreign asset (NFA) position (or net international investment position, NIIP) is the difference between an economy’s external financial assets and liabilities: Total net foreign asset position =Totalassets −Total liabilities GDP External assets and liabilities are defined on the basis of residence, not nationality. Net positions are measured as ratios to GDP. Following Gourinchas and Rey (2014), the net risky position is defined as: Net risky position =(Portfolio equity assets −Portfolio equity liabilities) + (FDIassets −FDI liabilities) GDP In other words, riskiness is considered from the investor’s point of view. Portfolio equities are risky due to the high volatility and FDI because of low liquidity. The net FDI position is defined simply as: Net FDI position =FDI assets −FDI liabilities GDP 3.1.2. Main explanatory variables The data on preferences are taken from the Global Preferences Survey (GPS) introduced by Falk et al. (2018). The GPS data were collected within the framework of the 2012 Gallup World Poll. Consequently, the GPS data is time-invariant. By covering 76 countries and more than 80,000 participants worldwide, it is the first global dataset on economic preferences that is representative at the country level. The GPS has data on patience, risk-taking, negative and positive reciprocity, altruism, and trust. We include patience and risktaking in our analysis because patience is a measure of time preference (Hypothesis 1), and risk-taking is a measure of risk preference (Hypothesis 2). However, we also test all other preference measures and find that negative reciprocity has explanatory power on net FDI positions. Negative reciprocity measures the willingness to punish unfair behavior even at one’s own cost. As the economic preferences were measured in 2012 and in our regression analysis, the net positions were from the end of 2019, economic preferences are predetermined. The values of the economic preferences are differences to the world mean in the standard deviation of the respective preference measure. 7 Fuel exports has the least observations: 692 out of 760. To ensure consistency, we use only one data source for each variable. For 70 countries out of the 76 included in the Global Preferences Survey, we have data on the net positions in 2019 and at least one observation on all of the control variables listed in Table 1 in the period 2010–2019. 8 The share of FDI assets on risky assets is, on average, 0.775, and the share of FDI liabilities on risky liabilities is 0.829. 9 The distribution of net portfolio equity positions has small variance but high skewness. As a result, it is difficult to link net portfolio equity positions to explanatory variables. The net safe position (reserve assets +debt assets – debt liabilities) was not included because foreign exchange reserves dominate it and hence the determinants are very different from the net risky positions. In our sample, there is a statistically significant concave relationship between patience and net safe position. M. Nieminen and K. Kuziemska-Pawlak Journal of International Money and Finance 146 (2024) 103130 5 3.1.3. Control variables The current account balance measures net capital flow, and in the long-run, total net foreign asset positions and current accounts are closely related. 10 Hence, it is natural that the set of control variables is derived from the current account literature (e.g., Ca’ Zorzi et al., 2012). Our set of control variables covers most of the variables included by Vermeulen and de Haan (2014). The small differences, such as the lack of exchange rate changes, result from the fact that our focus is on deeper determinants. Hence, our empirical methodology is also different. Our set of control variables gauges macroeconomic, demographic, and institutional factors. 11 Due to missing observations, the number of countries decreases if the set of control variables is extended. Empirical results are always conditional on the model used, but we are not aware of any additional control variable that would undermine our empirical results. 12 3.2. Econometric methodology The relationships between the preference measures and the net positions (e.g., Hypotheses 1–2) are tested by estimating the following linear cross-sectional regression model by the OLS estimator: NetPositioni,2019 = α +βPreferencei+xʹ iγ+ ε i,(1) where NetPosition i,2019 is the net position of a particular asset class (as a ratio to GDP) of country i at the end of 2019, α is the intercept, Preference is the measure of economic preferences, x is the vector of control variables, and ε is the residual. The net positions of different asset classes are from the end of 2019 because this is the latest year for which we have data for all of the countries included in the Global Preferences Survey. 13 As a robustness check, we replicate the analysis by taking the net positions Table 1 Descriptive statistics, 76 countries, 2010–2019. Variable Min Max Median Mean St. dev. # Obs. Dependent variables: Total net foreign asset position −1.560 2.348 −0.268 −0.210 0.533 760 Net positions of different asset classes: Net risky position a −1.234 1.362 −0.187 −0.172 0.352 760 Net FDI position −1.236 1.386 −0.168 −0.186 0.297 760 Explanatory variables: Measure of economic preferences: Patience −0.613 1.071 −0.093 −0.003 0.370 76 Risk-taking −0.792 0.971 −0.020 0.013 0.302 76 Negative reciprocity −0.489 0.739 0.004 0.013 0.275 76 Positive reciprocity −1.038 0.570 0.024 −0.034 0.342 76 Altruism −0.940 0.906 −0.097 −0.038 0.343 76 Trust −0.706 0.609 −0.081 −0.022 0.278 76 Control variables: Log GDP per capita 7.208 11.169 9.640 9.585 1.002 750 GDP per capita growth −0.100 0.181 0.022 0.023 0.029 755 Fiscal balance −0.230 0.119 −0.027 −0.031 0.036 755 Fuel exports 0.000 1.000 0.055 0.160 0.246 692 Financial openness −1.924 2.322 1.049 0.723 1.585 730 Old dependency ratio 0.008 0.471 0.117 0.156 0.099 760 Child dependency ratio 0.153 1.001 0.349 0.405 0.201 760 Population growth −2.258 7.687 1.112 1.073 1.169 760 Voice and accountability −1.907 1.690 0.110 0.076 0.934 760 Rule of law −2.323 2.130 −0.074 0.110 0.992 760 Private credit ratio 0.032 1.847 0.496 0.576 0.393 731 See Table A2 in the Appendix for a list of countries. a Net risky position =Net portfolio equity position +net FDI position. (This definition is adopted from Gourinchas and Rey (2014).) 10 Current account balances account for the bulk of changes in the net international investment position (Turrini and Zeugner, 2019). 11 It is relatively straightforward to identify the most important and the most widely-used macroeconomic and demographic factors from the current account literature. For institutional factors there are more alternatives. We took the institutional factors from the World Governance Indicators (WGI) database which includes six variables on institutional quality. Only Voice and accountability and Rule of law are included as control variables, because 1) due to high pairwise correlations and multicollinearity all of the WGI variables cannot be included together; 2) among the WGI variables Voice and accountability and Rule of law are the strongest determinants of net positions; 3) results on preferences are not sensitive to whether the rest of the WGI variables are included or excluded. 12 In addition to the control variables included in the paper, we tested Public debt, Health expenditure, Trade openness, Privatization of banking sector, Securities market development, Bank entry barriers and Civil liberties. 13 Our version (September 14, 2021) of the External Wealth of Nations database lacks the observation of 2020 for Cameroon, Haiti, Malawi, Venezuela, and Zimbabwe. M. Nieminen and K. Kuziemska-Pawlak Journal of International Money and Finance 146 (2024) 103130 6 from 2015. Following Falk et al., (2018, Tables IX and X), who introduced the GPS, measures of preferences are included one at a time. Control variables are measured as 10-year averages over the period 2010–2019 due to missing observations and possible short-run fluctuations. As a robustness check, we use annual observations instead of multi-year averages. If we consider a linear regression model such as (1), there is uncertainty about which explanatory variables to include on the righthand side of the equation. In order to control for this model uncertainty, we utilize the Bayesian Model Averaging (BMA) method with the uniform model prior and some other reasonable assumptions (see Fernandez et al., 2001). The BMA method has been used in the current account literature (see Ca’ Zorzi et al., 2012; Coutinho et al., 2022). When testing the statistical significance of the explanatory variables, we include five control variables with the highest inclusion probability based on the BMA. This is the maximum number of control variables because we have only 70 observations. However, there are no cases in which a strongly statistically significant control variable would be excluded. 14 In addition, the Schwarz information criterion is used for model selection. There is no particular reason why the relationships between net positions and economic preferences or the relationships between net positions and control variables should be linear. The reason for the assumption of linearity is practical because it is easier to tackle the model uncertainty with this assumption. However, as a robustness check, we do test the quadratic terms of both the preference measures and control variables. 4. Empirical results In this Section, we test the relationships between preference measures and net positions with the data and methodology described in Section 3. 4.1. Variable selection for regression models We utilize the Bayesian Model Averaging (BMA) method to assess the determinants of net positions. The posterior inclusion probabilities of each preference measure are presented in Table 2. The preference measures are included as explanatory variables one at a time. The high inclusion probabilities of patience (from 0.86 to 1.00) mean that it significantly explains the cross-country variation in all three net positions once the potential control variables have been taken into account. In addition to patience and risk-taking, the inclusion probabilities of negative reciprocity are also high. Hence, we also estimate equation (1) for negative reciprocity. Inclusion probability does not give an indication of the sign of the coefficient. That is tested in Sections 4.2–4.4. However, posterior densities for the economic preferences with the highest inclusion probabilities are shown in Figures A3-A5 in the Appendix. The results of posterior inclusion probabilities and posterior densities (Table 2-3 and Figures A3-A5) are based on all possible specifications of the explanatory variables, that is, in this case, 4095 different specifications. The posterior inclusion probabilities of each explanatory variable are presented in Table 3. For patience, risk-taking and negative reciprocity the numbers are the same as in Table 2. Based on the BMA method, patience is the second most important determinant of total net foreign asset positions and the most important determinant of both net risky positions and net FDI positions. Fiscal balance is highly significant for the total net foreign asset position. Risk-taking and the rule of law have a high inclusion probability for the net risky positions. Rule of law is the most widely used proxy for the quality of contract enforcement and the protection of property rights (see, e.g., a review by Nunn and Trefler, 2014). For the net FDI positions, negative reciprocity is more important than risk-taking, and its inclusion probability is also slightly higher than that of the rule of law. 4.2. Determinants of total net foreign asset positions Results of cross-sectional regressions for total net foreign asset position are presented in Table 4. The larger models are selected by the BMA method (specifications (1), (4), and (6)) and the smaller ones by the Schwarz information criterion (specifications (3), (5), and (7)). 15 There is a strong positive relationship between total net foreign asset position and patience (specifications (1)-(3)). 16 In other words, countries inhabited by patient individuals tend to have a positive net foreign asset position, whereas impatient populations tend to have a negative NFA position. This result is not driven by an outlier (see Figure A6 in the Appendix). 17 The Schwarz information criterion includes patience in the first best specification (column (3)). Patience is positively correlated with GDP per capita and with measures of institutional quality (see Table A3 in the Appendix), but 14 There are three cases in which excluded control variable is statistically significant at the 10% level: Private credit in specifications (1)-(2) in Table 4 and Child dependency ratio in specification (3) in Table 6 (see Table A12 in the Appendix). However, none of these undermine our main results. 15 This also holds true for Tables 5-6. 16 The assumption of linearity holds true for all other dependent variables and the three preferences measures from the GPS (i.e., the quadratic terms of preference measures are not statistically significant in Tables 5-6). 17 Results for the economic preferences excluding the United Arab Emirates are presented in Table A9 in the Appendix. The total net foreign asset position of the United Arab Emirates in 2019 is a potential outlier (see Table A8 in the Appendix). If the United Arab Emirates is excluded, the quadratic term of Patience is not statistically significant at the 5% level (specification (2) in Table A9 in the Appendix) in which case the relationship between the total NFA position and patience is not only positive but also linear. M. Nieminen and K. Kuziemska-Pawlak Journal of International Money and Finance 146 (2024) 103130 7 the variance inflation factors (VIFs) of our preference measures are below 3 in Tables 4-6. This means that the results on preference measures do not suffer from multicollinearity. For the most part this holds true also for the control variables as the VIFs are below 10, which is a commonly used threshold value for severe multicollinearity. 18 In Table 4 the only exception is Child dependency ratio in specifications (1)-(2), (4) and (6). Based on specification (1), a one standard deviation increase in patience is associated with a 0.223 larger net foreign asset position (ratio to GDP). The coefficients of patience in columns (1) and (3) in Table 4 are well in line with the coefficient average of patience over all models, which is 0.553 in Figure A3 in the Appendix. Consequently, we find strong support that Hypothesis 1 – “There is a positive relationship between time preference and net foreign Table 2 Posterior inclusion probabilities for preference measures. Dependent variable: Total net foreign asset position in 2019 Measure of preferences: Patience 0.86 Risk-taking 0.45 Neg. reciprocity 0.31 Pos. reciprocity 0.17 Altruism 0.09 Trust 0.12 Control variables: Yes Yes Yes Yes Yes Yes Dependent variable: Net risky position in 2019 Measure of preferences: Patience 0.97 Risk-taking 0.60 Neg. reciprocity 0.57 Pos. reciprocity 0.20 Altruism 0.08 Trust 0.08 Control variables: Yes Yes Yes Yes Yes Yes Dependent variable: Net FDI position in 2019 Measure of preferences: Patience 1.00 Risk-taking 0.41 Neg. reciprocity 0.73 Pos. reciprocity 0.29 Altruism 0.08 Trust 0.09 Control variables: Yes Yes Yes Yes Yes Yes See Table 1 for the set of control variables. All countries for which we have data are included (70 countries). Table 3 Posterior inclusion probabilities for the explanatory variables of net positions in 2019. Total net foreign asset position Net risky position Net FDI position (1) (2) (3) (4) (5) (6) (7) (8) (9) Measure of preferences: Patience 0.86 0.97 1.00 Risk-taking 0.45 0.60 0.41 Neg. reciprocity 0.31 0.57 0.73 Control variables: Log GDP per capita 0.64 0.71 0.85 0.17 0.46 0.38 0.09 0.19 0.17 GDP per capita growth 0.12 0.10 0.11 0.50 0.17 0.20 0.56 0.23 0.26 Fiscal balance 0.90 0.95 0.96 0.13 0.12 0.12 0.08 0.08 0.08 Fuel exports 0.30 0.15 0.14 0.31 0.31 0.34 0.12 0.14 0.12 Financial openness 0.21 0.32 0.41 0.08 0.09 0.10 0.24 0.33 0.31 Old dependency ratio 0.34 0.18 0.21 0.10 0.10 0.15 0.09 0.19 0.13 Child dependency ratio 0.34 0.39 0.25 0.10 0.14 0.17 0.10 0.15 0.21 Population growth 0.62 0.86 0.88 0.12 0.18 0.30 0.09 0.12 0.16 Voice and accountability 0.60 0.17 0.15 0.12 0.12 0.12 0.08 0.17 0.25 Rule of law 0.12 0.14 0.13 0.14 0.64 0.71 0.09 0.62 0.65 Private credit ratio 0.20 0.09 0.09 0.10 0.09 0.09 0.08 0.27 0.19 Control variables are country-level averages during the period 2010–2019. All countries for which we have data are included (70 countries). 18 High variance inflation factors are reported below the particular tables. M. Nieminen and K. Kuziemska-Pawlak Journal of International Money and Finance 146 (2024) 103130 8 asset position” – holds true for the total net foreign asset positions in our sample. This is theoretically plausible because time preference, which we measure by patience, is a decisive factor in saving and investment decisions. Further robustness checks are discussed in Section 4.5. Based on specifications (4)-(7), there is a statistically significant positive relationship between risk-taking and total net foreign asset position, but not between negative reciprocity and total net foreign asset position. However, risk-taking is not selected by the Schwarz information criterion in the first best specification. 4.3. Determinants of net risky positions The results on the net risky position are presented in Table 5. For patience and risk-taking the results are very similar to those in Table 4: there is a strong positive linear relationship between the dependent variable and these two preference measures (specifications (1)-(4) in Table 5). However, unlike in Table 4, the Schwarz information criterion includes risk-taking in the first best specification of net risky positions (column (4) in Table 5). These results are not driven by potential outliers (see Figures A9-A10 in the Appendix). 19 There is no indication of severe multicollinearity in Table 5. Based on specification (1), a one standard deviation increase in patience is associated with a 0.186 larger net risky position (ratio to GDP). Based on specification (3), a one standard deviation increase in risk-taking is associated with a 0.090 larger net risky position (ratio to GDP). The coefficient average of risk-taking over all models (0.219 in Figure A4 in the Appendix) is smaller than the coefficients of risk-taking in Table 5. This is because the posterior inclusion probability of risk-taking is below one (0.60 in Table 3). Consequently, we find strong support that Hypothesis 1 – “There is a positive relationship between time preference and net foreign asset position” – holds true for the net risky position in our sample. Similarly, we find strong support that Hypothesis 2 – “There is a positive relationship between risk preference and net risky position” – holds true in our sample. This is theoretically plausible because risk preference, which we measure by risk-taking, is a fundamental determinant of portfolio allocation. Again, further robustness Table 4 Determinants of total net foreign asset positions in 2019. Dependent variable: Total net foreign asset position (1) (2) (3) (4) (5) (6) (7) Measure of preferences: Patience 0.604*** (0.223) 0.797*** (0.207) 0.610*** (0.219) Patience^2 −0.738** (0.336) Risk-taking 0.350** (0.173) 0.551*** (0.188) Neg. reciprocity 0.334 (0.226) 0.330 (0.224) Control variables: Log GDP per capita 0.227 (0.145) 0.229 (0.140) 0.304** (0.127) 0.260* (0.153) 0.317** (0.141) 0.402*** (0.099) Fiscal balance 7.618** (3.000) 8.607*** (2.876) 7.614** (3.049) 9.422*** (3.314) 8.639*** (2.976) 9.967*** (3.222) 10.223*** (3.331) Financial openness −0.068 (0.061) −0.086 (0.059) −0.096* (0.054) Child dependency ratio −0.626 (1.001) −0.639 (0.938) −1.178 (1.048) −2.465*** (0.685) −0.706 (0.983) Population growth 0.244 (0.161) 0.271* (0.150) 0.180** (0.081) 0.340** (0.154) 0.395*** (0.146) 0.357** (0.153) 0.292*** (0.088) Voice and accountability −0.213** (0.084) −0.181** (0.077) −0.238*** (0.080) Constant −2.172 (1.587) −2.094 (1.531) −3.093** (1.235) −2.292 (1.689) 0.590*** (0.194) −3.018* (1.537) −4.030*** (0.935) R 2 0.501 0.541 0.497 0.477 0.449 0.478 0.472 SIC rank 1 2 2 # Obs. 70 70 70 70 70 70 70 Control variables are country-level averages during the period 2010–2019. Heteroscedasticity-robust standard errors are in parentheses. *, ** and *** denote statistical significance at the 10 %, 5 % and 1 % levels, respectively. Multicollinearity was tested using the variance inflation factor (VIF). In specifications (1)-(2), (4) and (6), the VIFs of Child dependency ratio were high (between 8.83 and 10.15). In other specifications and for other variables, VIFs were below 7. In specifications (1), (4), and (6), the five control variables with the highest posterior inclusion probability were included (see Table 3). Specifications (3), (5), and (7) were selected by the Schwarz information criterion (SIC). 19 Results for the economic preferences excluding the United Arab Emirates and the Netherlands are presented in Table A10 in the Appendix. Net risky positions of the United Arab Emirates and the Netherlands both in 2019 are potential outliers (see Table A8 in the Appendix). The results remain. M. Nieminen and K. Kuziemska-Pawlak