The relationship between individual risk preferences and the probability to migrate

(1)

The Relationship between

Individual Risk Preferences

and the Probability to Migrate

Badzheva, Zlatina Veselinova

10398589

June 2015

Bachelor Thesis

BSc Economics and Business, Economics and Finance track

Thesis Supervisor: dr. Audrey Hu

(2)

1

Abstract

This paper investigates the relationship between the individual level of risk aversion and the propensity to migrate, using actual data. It extends the scope of existing empirical research from the intranational to international level. Information on participants’ migration history and preferences is collected through a survey, employing two separate risk measures – a hypothetical lottery experiment and a self-assessment question. The results support the hypothesis that risk attitudes are an explicit determinant of the migration decision and a higher level of risk tolerance is associated with a higher probability to change one’s country of residence. The estimates of the uncertainty measures are significant unconditional on the inclusion of other specifications in the model. From the personal characteristics used as controls, only age appears important for the migration decision. The findings have potential implications for authorities and policy-makers on various levels. Further empirical research on the topic of risk preferences and international migration is needed, with an extended number of determinants tested and a larger sample with more diverse demographic characteristics.

(3)

2

1. Introduction

The last few decades have been characterized by a steadily increasing international migration, a result from the globalization in general and the easier access to information and transportation in particular (Castles, 2002) 1. With migration being now far more structured and mass, it has the potential to affect established economic patterns and social, political and cultural structures and is therefore receiving increasingly more attention from regulatory bodies on country and global level. This creates a pressing need for a better understanding and quantifying of the migration decision and its determinants. Economics, psychology and sociology are amongst the front-runners exploring the phenomenon in depth, developing new theories, models and whole research areas such as the new economics of migration. They all accept migration to be a risky endeavor, characterized by uncertainty of foreign financial conditions, social networks, education, political systems and culture summed up as imperfect knowledge about the terms in the recipient destination (Jaeger et al, 2010) 2.

Similarly to migration, risk attitudes have the potential to cause significant changes on a local and global scale through their effect on personal economic behavior. They have repeatedly been proven to impact economic choices, such as propensity to save, portfolio selection, insurance, contracting and self-employment, and hence economic outcomes (Smith, 1979). Since the variability of risk-preferences might then lead to economic dissimilarities between regions and countries, it would be beneficial for policy setters to be aware of the average risk-aversion level of the population, especially when including a large share of immigrants.

Despite the great interest in both immigration and risk aversion separately, little academic focus has been put on the explicit interaction between the two. Already in 1982, Stark and Levhari urge for a better understanding of risk and migration’s relation, extending beyond the standard expected-income hypothesis (Stark and Levhari, 1982). The progress however has been slow and mostly unfruitful. Even though risk tolerance is theoretically hypothesized to have an important role as a determinant in migration models, there is relatively little direct evidence on its association with the migration decision (Jaeger et al., 2010). The largest body of research focused on internal movements in Germany and a number of LDCs but the amount of empirical studies concerned with quantifying the relationship is still quite limited. Nevertheless, it is widely accepted that the average “mover” should be more prone to risk as migration is usually associated with a higher uncertainty than non-migration (Jaeger et al., 2010).

Following the above notion, this paper aims to investigate the relationship between individual level of risk aversion and the probability for international migration, using actual data. It is therefore

1_{The terms “migration” and “immigration” are used interchangeably in the paper.}

2_{The terms “risk” and “uncertainty” are used interchangeably in the paper, despite Knight’s differentiation}

(5)

4 an extension of the existing research in two ways : first, it examines the role of uncertainty aversion as an explicit factor for the decision to move abroad, and second, it does so by collecting real data.

The remainder of the study is divided in 4 parts. Part 2 contains a brief overview of the connection between uncertainty and migration, as existent in the current academic literature. The experimental design used in the research is outlined in Part 3 and followed by a description of the data and analysis of the empirical results in Part 4. The final part concludes and includes implications and suggestions for further research. The survey used is presented in the Appendix.

2. Theory

2.1 Migration is Risky

Migration is characterized by risk and has the potential to generate risk in various dimensions for the migrants and non-migrants in both the recipient and originating country. These include, among others, “brain drains”, the creation of “outsider” minorities and human trafficking (Gibson, 2011). Focusing on the average voluntary migrant, migration is associated with less extreme uncertainties over outcomes, which nevertheless impact a wide number of aspects in the individual’s life.

As for the majority, the immigration interest arises from the desire to improve welfare and especially wealth, and is usually seen as an investment decision, depending on the difference between the income in the recipient country and the home country. Following standard economic theory, Bowles formulates the migration decision as weighing benefits versus costs, with the present value of the expected future income in the host country being the benefit of migration and the present value of expected income at home – the opportunity cost (Bowles, 1970). Assuming that the individual can better observe the income in his current location than the final destination, he cannot accurately assess the benefits of relocating and thus the net gain. Besides this pure financial risk, changing one’s country of residence accounts for uncertainty in other areas of life, including social networks, education, political systems, culture etc., which could be summed up as imperfect knowledge about conditions in the recipient destination. Even in the cases of already arranged occupation, accommodation and social circle in the recipient country, other uncertainties arise, such as problems with adaptation to the new environment (Haug, 2008). Another way to consider the risks is to consider an individual with utility depending on leisure and consumption. Taking after O’Connell’s views, that many characteristics of the end destination are only locally observable (the person learns about them only after the relocation), a reasonable assumption is that agents are better informed about factors in their own place of residence rather than the potential recipient country (O’Connell, 1997). This relative lack of knowledge over the elements of the utility function is another reason for deeming migration as fundamentally risky. On a more global perspective, uncertainty extends not only spatially but also in regard to time, being split between present and future. The future is uncertain per se but

(6)

5 migrants face additional challenges for expectation formation, as they also have to include the effect of the already unknown current factors (O’Connell, 1997).

These arguments lead to the expectation that the average “mover” should be more prone to risk as migration is usually associated with a higher uncertainty than non-migration. Therefore, the hypothesis of this research is that more risk – tolerant individuals will be more prone to migration.

2.2 Risk in Migration Models

The academic literature is abundant in attempts to explain and analyze migration through frameworks, models, notions and generalizations, but there is still very little progress on creating a coherent and robust theory (Bonin et al., 2012). Many of the contributions on the topic are stand-alone, instead of connected building blocks in the advancement of the migration research. A common feature for most is the solely implicit acknowledgement of risk as a factor playing a role, even though it is fundamental in the process of migration. Regardless of the perception of risk as important, there is very little definitive theory on its function and it is usually accounted for simply as a cost, rather than an explicit separate influence. This is the case with the neo-classical human capital theory, the oldest and best-known model of migration, associated with the works of Todaro and Sjaastad. It takes a disequilibrium theoretical perspective, assuming that spatial differentials of wages, earnings and income provide opportunities for utility gains and thus motivate relocation. On the micro level, it presents migration as an investment decision, dependent on a positive net return after a cost-benefit analysis (Sjaastad, 1962). It is the result of a rational agent’s desire to maximize utility by moving to a place where he will get a higher return for his labor, adjusted for various tangible and intangible costs, including uncertainty. As Borjas asserts, the immigration is lower the higher its costs, hence increasing risk would decrease the willingness to change country of residence. The measure of income, relevant to the individual, is the net present value of the expected cash flows over his lifetime, which he compares across destinations. Even though income maximization is not the sole factor accounting for utility maximization and hence for relocation, it is a necessary condition for reaching it and aids for an easier empirical testing of the theory (Borjas, 1989). Besides regarding it as a cost, uncertainty in the human capital theory is also implied in underlying concepts such as the “brain drain” (Bhagwati and Hamada, 1974). An issue with this basic migration explanation, as with many to follow, is its incompatibility with reality – if its predictions would hold, the number of migrants actually observed should be many times higher (Arango, 2000). In an attempt to reach a more realistic and practical representation, the new economics of migration shifts the focus from the individual to the household as a decision-making unit. It is one of the few areas with explicit theorizing of risk, making uncertainty minimization as important for the relocation decision as maximizing income. Central to the theory is the concept of diversifying income risk across markets, therefore influential is not only the returns’ geographical distribution, but also their correlation (Stark

(7)

6 and Levhari, 1982). The aim is spreading the sources of family income across uncorrelated markets. Unlike the human capital views, wage differentials adjusted for costs are not sufficient for the moving decision and it is not irrational for agents to choose destinations of lower pay and higher uncertainty, if this would lead to a decrease of overall household risk (Stark and Bloom, 1985). Besides whole theories, the literature includes several single hypothetical models aiming to explain migration in various settings and details, such as Beine et al’s presentation of a small open economy with two-overlapping generations. While the authors consider uncertainty potentially impactful, they assume it away by employing risk-neutral agents (Beine et al, 2001).

A more systematic investigation of the association of risk-aversion and immigration has started only recently, even though interest has not been previously lacking (Jaeger et al, 2011). Noteworthy is Hart’s effort in creating a series of economic models employing various theoretical assumptions within static and dynamic frameworks. He suggests the inclusion of risk-aversion measures in econometric models in place of the standard risk-neutral assumption but does not test them in practice since there is no suitable data which could allow for rejection or acceptance of the hypothesis (Hart, 1975). David takes the idea a step further, showing that it is not simply risk preferences, which are important for the migration decision, but their varying levels. Employing a partial equilibrium analysis, he disproves the usual risk-neutrality assumption in favor of risk-aversion (David, 1974). As with other models, it is problematic for empirical testing since its restrictive assumptions make it difficult to obtain a representative sample. Notable recent model of the migration decision is built by Heitmueller, who illustrates the moving or staying decision as a purchase of a lottery ticket with two possible outcomes – being employed or not. Incorporating varying degrees of risk tolerance and using actual data, he concludes that risk aversion is significant for the migration decision (Heitmueller, 2005).

2.3 Empirical Results in the Literature

Although the interest in migration, its determinants and implications is increasing, the theory in the field remains mostly untested. Regarding uncertainty, the number of empirical studies concerned with quantifying the relationship between risk aversion and relocating is even more limited, with the largest body of research being focused on internal migration. Urban-rural movements in LDCs are widely utilized to relate risk-aversion to the moving decision3. Stark and Levhari focus entirely on this connection, challenging Todaro’s theory on migration as motivated by the expected income differences in two regions. They call for the inclusion of risk as an explanatory variable of migration and prove that risk aversion is a strong force underlying the moving decision but although very challenging to model (Stark and Levhari, 1982). Smith also analyses migration with a two-region

(8)

7 model and finds significant differences in the level of risk-aversion across the two regions, hence a difference in risk tolerance of movers and stayers (Smith, 1979).

The most straightforward evidence and most extensive investigation of the topic is done with regard to the East-West German migration in the last two decades. Most studies are structured in an analogous manner, using data from the German Socio-Economic Panel, which provides direct measurements of risk and migration. The researches vary in the types of relationships between risk and migration that they test – from differences of risk attitudes between migrants and natives (Bonin et al, 2009); persistence of risk preferences in movers (Bonin et al., 2012); to the explicit association of risk tolerance and migration (Jaeger et al., 2011). A general conclusion they draw is that risk attitudes are significantly associated with the relocation decision.

2.4 Relation of Risk Preferences & Migration: Not so Straightforward

So far the discussion has been focused on return-maximizing individuals and more precisely on speculators, who are looking for improving their welfare rather than “hedging” their risks. As this is not always migration’s main purpose, it is useful to consider how relocation resulting from other stimuli might be connected to risk aversion.

A theory from the new economics of migration suggests that non- migration is also risky but constitutes different uncertainties. As Massey et al. explain, moving can be seen as a way to spread risk over uncorrelated markets (Massey et al., 1993). In these cases, when migration is assumed to be a way for diversification of family income, more risk-averse individuals would be expected to engage in relocating. As Smith suggests, the less risk-tolerant agents are more likely to change their country of residence, searching for locations with lower variance in income (Smith, 1979). However, foreign income volatility is usually not freely observable and thus constitutes part of the imperfect information which is the main factor making international migration risky in the first place.

The connection between risk aversion and immigration is unclear also because the direction of the effect is ambiguous. Cameron et al. provide experimental evidence that immigrants’ preferences are subject to convergence as a result of cultural integration. Using a sample of 300 students of Chinese origin currently residing in Australia, the authors examine, among other characteristics, the subjects‘ convergence of risk preferences due to exposure to Western education. The results are gender specific and are obtained under the implicit assumption that risk preferences in both the recipient and destination countries are homogeneous (Cameron et al, 2014). Most of the research on immigrants’ conversion however is developed with regard to social characteristics, such as altruism or individualism, and the findings are usually conditional on social status, social networks, community, cultural exposure, time spent and other, making it difficult to disentangle connections and build a comprehensive theory. Risk preferences, on the other hand, have been proven to be relatively stable over time (Jaeger et al., 2007). Bonin et al. research the risk attitudes of first and second

(9)

8 generation migrants in Germany and conclude that the first generation movers did not converge to the general risk-level of the natives, while the second –generation (German-born) have a risk level comparable to the local population (Bonin et al., 2010). This gives confidence that findings of a relation between risk and migration would be due to risk attitudes being a determinant of the migration decision rather than vice versa.

3. Experimental Design

As this study aims at providing empirical evidence on the theoretical relationship between risk and immigration, data collection is fundamental. For the purpose, a survey approach was employed which is a standard method for investigating risk preferences in the literature. The questions focus on demographic characteristics, risk attitudes and immigration.4

The demographic characteristics taken into account are age, gender, years of education (since beginning of elementary school), currently enrolled in full-time education and/ or being in full-time employment. These have been chosen as control variables as they have repeatedly been proven to influence migration. As Heitmueller sums up, the average migrant is a relatively well educated young male seeking better career and earning opportunities (Heitmueller, 2005).

With risk aversion being the main covariateof interest, measuring it has been a primary focus of the questionnaire. Several ways for determining individual risk preferences have been widely used with the real money at stake experiments seen as most reliable because they come closest to a real life decision situation. However, Dohmen et al. prove that the results from hypothetical payoff experiments come fairly close to the ones predicted by real payouts, suggesting they are a valid measure of behavior patterns (Dohmen et al, 2011). For the largest part, survey measures try to estimate a general risk-level based on hypothetical financial lotteries or health risks, ignoring context specific questions. In the present study two different risk-measures are employed to decrease the chance that the estimates obtained are solely results of the risk measure. The benefit of having both is that they are proven to be accurate in estimating general risk tolerance levels but use different approaches to determine them.

The first research method is based on Holt and Laury’s “Risk aversion and Incentive effects”, where the authors use a menu of 10 paired lottery choices for measuring risk aversion (Holt and Laury, 2002). In this approach the subjects do not receive any actual financial compensation. The individuals have the choice to participate in one of the two hypothetical lotteries (Option A or Option B) with potential payoffs as shown in Table 1. As it can be seen, the remuneration for Option A is much less variable compared to Option B, which is hence assumed to be the “risky” option. The last column (not provided to the participants) presents the expected payoff differences between the two

(10)

9 Options in each decision pair – looking at the first row, the expected benefit of choosing A over B is 1.17 $, but decreases with each subsequent decision. According to the rationality prediction, a person should switch from choosing Option A to choosing Option B once the probability of the high payoff increases enough. Therefore a risk-neutral individual would switch from A to B after the fourth decision and even the most risk-averse participant should pick Option B over A in row 10, since this provides a guaranteed payoff of 3.85 $ (Holt & Laury, 2002).

Table 1- The ten paired lottery-choice decisions (as presented by Holt and Laury)

Note: Adapted from “Risk Aversion and Incentive Effects” by C.A. Holt & S.K. Laury, 2002, American Economic Review, 92 (5), p.1651

Consistent with Holt and Laury, the number of not risky A choices will be used as representative for risk aversion – the more A choices the lower the risk tolerance, with four As followed by strictly B choices being the benchmark for risk-neutrality.

Even though Holt and Laury prove that scaling up the hypothetical payoffs by factors of 20, 50 or 90 does not alter behavior significantly, an equivalent increase in real payouts sharply decreases the willingness to take risks, which is a main limitation of survey approaches for measuring risk aversion (Holt & Laury, 2002). An experimental study offering real payouts would be an incentive compatible measure of risky behavior but due to the high costs and difficulties in implementation very few large scale studies of risk aversion exist.

As mentioned, to avoid outcomes stemming solely from the choice of a risk measure, a second procedure for determining risk attitudes was employed. It is a self-assessment question regarding the individual’s general willingness to take risks on a scale from 0 (not willing to take risks at all) to 10 (fully prepared to take risks)5. Dohmen et al. prove that it is a good predictor of actual individual risk tolerance level by comparing the responses from this survey item to results from real-steaks experiments.6 The authors also test the outcomes from the general question against results from highly contextualized questions regarding sports, career, leisure and health. They find that risk preferences

5_{The exact translation of the question from German is “ How do you see yourself? Are you generally a person}

who is fully prepared to take risks or do you try to avoid taking risks?”

(11)

10 are stable and highly (but not perfectly) correlated across contexts (Dohmen et al, 2011). This is important in the case of assessing migration probabilities, since relocating is characterized by various types of uncertainty. Sahm also provides empirical evidence for the existence of “constant relative risk aversion” over time, which is not significantly affected by changes of wealth, income or personal events and only moderately increases with age (Sahm, 2007). Jaeger et al. prove stability of uncertainty tolerance directly with regard to migration with ex ante and ex post regressions (Jaeger et al., 2010).

The survey was initially prepared and distributed online as it is easier for collecting responses from more geographic locations with a greater variety in individual characteristics hence reducing the selection bias. To diversify the sample further and not discriminate on the basis of technology access, printed copies were handed out. Selection bias is an important issue for this research as the existing time and budget constraints limit the possibilities for conducting a large scale experiment with an extensive number of participants.

4. Results and Discussion

4.1 Sample Statistics - Demographics

Out of 246 observations, only 204 responses were used for the analysis. The remaining 42 were discarded because the participants were outside of the age range (18-65), submitted incomplete surveys or were inconsistent in the Holt & Laury experiment (switching several times between Options A and B or choosing Option A in the last pair when B is the dominating Option).

Table 2 presents the share of respondents who have lived only in their own country of origin, have lived in two countries or have moved to three or more different countries. With 49% non-migrants and 51% migrants the sample is well balanced between movers and stayers.

Table 2 – Number of respondents according to number of migrations # Respondents Share

Non-migrants 100 49.02

Migrated once 64 31.37

Migrated more than once 40 19.61

Table 3 presents the share of female respondents, the ones currently in full-time employment and/ or in full-time education, per migration group. Out of all movers, exactly half are females and the percentage falls for non-migrants and the overall sample, 43% and 47% respectively. From the total number of participants approximately 48% are in full-time employment and 55% in full-time

(12)

11 0 5 10 15 20 25 18 20 22 24 26 28 30 32 34 36 38 40 42 44 46 48 50 52 54 56 58 60 62 N u m b e r o f re sp o n d e n ts Age in years Non-migrants Migrants

education, with one category not excluding the other (a number of participants reported being in both groups or currently in neither). Despite the fact that the sample is almost equally split between employed and unemployed, students and non-students, the division is rather skewed in the total migrants group with only 36% being full time workers and 65% in full time education. The non-migrants exhibit an opposite trend: 59% are full-time employed and 42% in full-time education. Table 3 – Shares of respondents according to gender, currently in full time employment and/or full time education Gender (share females) Full-time employment Full-time education Non-migrants 0.430 0.590 0.420 Migrated once 0.469 0.328 0.641

Migrated more than once 0.550 0.400 0.675

Total migrants 0.500 0.356 0.654

Total sample 0.471 0.475 0.544

Figure 1 graphs the age distribution of the respondents. The distribution of the overall sample is clearly skewed to the left, with the biggest share of participants being in their twenties. The variability in age is much higher for migrants, ranging from 19 to 62 years (19 to 56 for non-migrants) and is relatively more evenly distributed, with most participants being in their early twenties or late forties. The modal value of both groups is 21 years; however, the average mover is 28 years old: 7 years younger than the average non-mover.

(13)

12 0 5 10 15 20 25 30 35 40 45 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 N u m b e r o f re sp o n d e n ts Years of education 0 5 10 15 20 25 30 1 2 3 4 5 6 7 8 9 10 N u m b e r o f re sp o n d e n ts

Number of risky choices

Non-migrants Migrants 0 5 10 15 20 25 30 1 2 3 4 5 6 7 8 9 10 N u m b e r o f re sp o n d e n ts Risk index Non-migrants Migrants

Figure 2 illustrates the total sample’s years of education, which are approximately normally distributed with a mean of 16.7 years.

Fig. 2 – Years of education distribution of all participants

4.2 Sample Statistics – Risk Measures

The following two figures show the distribution of the responses to the two risk measures in the survey. Figure 4 presents the distribution of answers to the self-assessment risk question for non-migrants and non-migrants. For both movers and stayers the modal value is 5 on the scale, which represents risk neutrality. However for migrants the distribution is more skewed towards the right-tail, indicating that they are more prone to taking risks.

Fig.4 - Migrants and non-migrants general Fig.5 - Migrants and non-migrants risk attitudes question (self-reported) risk attitude from lottery

(14)

Figure 5 illustrates the distribution of risky choices made by the participants. It shows a similar trend to the one from the risk index: the responses of the stayers are more skewed towards the left tail, indicating lower willingness to take risks. The mode is 6 risky choices for migrants, standing for risk neutrality. For non-migrants an equal number of individuals chose the risky option 5 and 6 times hence most participants exhibit risk neutrality or low level of risk aversion. There are several features worth noticing – first, more stayers than movers chose Option B 10 times and showed the highest level of risk tolerance; second, no migrants made only one risky choice; and third – no one from the sample made 9 risky choices.

4.3 Regression Analysis

Table 4A presents marginal effects from estimating regression models with number of countries the individual lived in as the dependent variable. Columns (1) and (2) present marginal effects of ordered probit model with the self-reported level of risk preferences (Risk index) as the risk measure, while in columns (3) and (4) the risk measure included is the number of risky choices from the Holt and Laury lottery. The last two columns present estimates from OLS regression, used as a control.

Table 4A- Migration probabilities regressions estimates

Covariates (1) (2) (3) (4) (5) (6) Risk index .0306 * (.0162) .0291 * (.0170) .0438 (.0272) # Risky choices .0174 (.0161) .0293 * (.0165) .0417 (.0256) Age -.0079 (.0051) -.0092 * (.0051) -.0104 (.0075) -.0121 (.0073) Female .0833 (.0680) .0748 (.0650) .1344 (.1095) .1246 (.1061) Years of education .0016 (.0156) .0030 (.0154) -.0036 (.0241) -.0062 (.0237) Full-time employed .0631 (.1246) .0927 (.1256) .1068 (.2103) .1442 (.2109) Full-time education .0941 (.1227) .1124 (.1228) .1907 (.2148) .2166 (.2142) R2 0.0087 0.0417 0.0026 0.0412 0.0774 0.0757

Note: The entries in columns (1) through (4) are marginal effects from ordered probit regression, at sample means. Dependent variable in all regressions is the number of countries the individual has lived in. Standard errors are presented in parenthesis. For columns (1) through (4) R2 _{stands for Pseudo- R}2_.

Coefficient’s significance: *p < .1; ** p < .05; *** p < .01

Columns (1) and (3) include only the marginal effects of an ordered probit regression with the number of countries the individual lived in as the dependent variable and a risk measure as a covariate. The regressions differ in the magnitude of their effect and only the risk index (self-reported

(15)

14 risk level) appears significant at the 10% level and indicates that an increase in the general risk tolerance of a person would lead to a 3, 06% increase in the probability of migration. Regressions (2) and (4) include the additional control variables age, gender, full time employment, full time education and number of years of education. For the model using the risk index, the risk measure is the only variable significantly impacting migration, with almost the same increase in probability as the specification with no additional variables: 2, 91%. The result is similar for column (4). The number of risky choices becomes significant for the migration probability and its marginal effect is very close to the one of the risk index: 2, 93% increase in the probability of moving for 1 more risky choice. Age is also significant at the 10% level and one year increase decreases the probability of changing country by 0, 92%. For both specifications no other variables play a role in explaining the migration decision. The linear specifications in the last two columns explain almost 8% of the variation in the sample but produce no statistically significant estimates for any of the variables. The two risk measures predict more than 4% increase in migration probabilities for one unit increase in risk tolerance with p-values of 10,9% for the Risk index and 10,4% for the number of risky choices, which are not significant at the 10% level, even though fairly close.7

The reason for the similarities in the results of regressions (2) and (4) is the relatively high correlation in the two risk measures (ρ = 0, 44). This is well above the correlations Dohmen et al. and Ding et al. found between their hypothetical investment decisions and general risk attitudes (ρ = 0.15 and ρ =0.26 respectively) (Ding et al., 2010). This could be a result of the participants being more aware of their own risk-attitudes and reporting them truthfully or due to selection bias in the sample.

Overall, correlations between all variables are relatively low 8. Following standard theory, age, risk and gender should be strongly associated and thus the coefficient on the respective risk measure in the regression should decline, while in this sample it increases from 1, 74% to 2, 93%. The only high correlations are between age and full time education (ρ = 0, 82) and age and full time employment (ρ = 0, 79). To test whether significance is impacted by these correlations, the regressions are repeated by omitting various specifications. Estimates are reported in Table 4B.

The fact that most coefficients in the regressions are not statistically different from 0 is unexpected because the variables have been repeatedly proven to have an effect on migration (Bowles, 1970). Especially interesting are the age and gender specifications – while the age has the usual negative relation with migration probability (even though not significant), being a female is estimated to increase the probability of migration by approximately 8%, which contradicts the general view that males are more prone to relocating. This could once again be a result of a biased sample due to its small size and low variability of respondents’ characteristics. As seen in the descriptive statistics, more than 55% of the individuals are in the 19-25 age group and 63% of them have already

7_{P-values not reported in the paper.} 8_{Correlations not reported in the paper.}

(16)

15 migrated at least once. Another reason could be omitted variable bias. Potentially impactful additional specifications would be marital status, income and nationality, as used by Jaeger et al. (Jaeger et al, 2010).

The reason for the number of years of education not appearing an important determinant could lie in the fact that the respondents come from a variety of educational systems. While for some 15 years of schooling is equivalent to high school education, for others this is the time taken for obtaining a Bachelor’s degree.

Table 4 B- Migration probabilities regressions estimates (continuation table 4A)

Covariates (7) (8) Risk index .0296 * (.0170) # Risky choices .0282 * (.0165) Age -.0090 ** (.0029) -.0101 *** (.0028) Female .0834 (.0681) .0746 (.0657) Years of education Full-time employed Full-time education R2 0.0406 0.0394

Note: The entries in columns (7) and (8) are marginal effects from ordered probit regression, at sample means. Dependent variable in all regressions is the number of countries the individual has lived in. Standard errors are presented in parenthesis. For columns (7) and (8) R2 stands for Pseudo- R2_.

Coefficient’s significance: *p < .1; ** p < .05; *** p < .01

Column (7) presents the marginal effects of an ordered probit with independent variables age, risk index and gender. Omitting years of education, full-time employment and full-time education leads to age now being significant at the 5% level and accounting for 0, 9% decrease of the probability to migrate for each extra year. The risk index is still having an effect at the 10% level and again accounting for almost 3% increase in the probability to move. Including the omitted variables one at a time does not lead to any statistically significant changes in the coefficients, with the exception of the regression with full-time education, in which age becomes insignificant for the moving decision. 9

Column (8) repeats the procedure this time using the number of risky choices as a risk measure. The results are similar – the coefficient of risk aversion is statistically different from zero and the increase in the risky choices by one leads to almost 3% increase of probability to change

9

(17)

16 country of residence. One year increase in age still accounts for approximately 1% decrease in the same probability, but this time its significance is at the 1% level.

5. Conclusion and Implications

This paper investigates the empirical relationship between the individual level of risk aversion and the probability for international migration. It contributes to the limited amount of experimental research focused on risk as an explicit determinant of the moving decision by broadening the scale from the intranational to an international level. It does so through the use of actual data collected with a survey.

In order to estimate the relationship between risk preferences and migration, participant’s level of uncertainty tolerance is determined using two validated risk measures. The first is a hypothetical payoff financial lottery, developed by Holt and Laury; the second is a self-reporting question on the individual’s general level of risk preferences. The outcomes from testing the results from both methods against the number of countries each person has lived in, provide evidence that uncertainty is an explicit determinant of migration. It supports the hypothesis that less risk averse individuals would be more prone to changing their country of residence.

Contrary to existing research, most of the other variables in the model have no significant role in explaining the moving decision. The importance of the age of the individual varies conditional on the inclusion of years of education and the binary full time employment and full time education factors, with which it is highly correlated. These results can potentially be attributed to the limited size of the sample and possible selection bias. For further research it is advisable to use a more diverse sample in terms of personal characteristics and include additional explanatory variables.

The findings of this study have implications for countries’ economic policy setting. As the level of risk tolerance impacts individual’s economic choices, it can lead to structural changes within countries, especially in the cases of a large immigrant population. Lower risk aversion can lead to an increase of entrepreneurship activity and investment levels, a shift of consumer choices, consumption-savings preferences and changes in the propensities to take insurance. Therefore being informed about the prevalent risk attitudes in the nation can help governments exploit more effectively the possibilities to further economic development.

(18)

17

Appendix A

(19)

(20)

(21)

(22)

21

References

Arango, J. (2000). Explaining Migration: a Critical View. International Social Science Journal,

52(162), 283-296

Beine, M., Docquier, F., Rapoport, H. (2001). Brain Drain and Economic Growth: Theory and Evidence. Journal of Development Economics, 64, 275-289

Bhagwati, J.N. & Hamada K., (1974). The Brain Drain, International Integration of Markets for Professionals and Unemployment. Journal of Development Economics, 1, 19-42

Bonin, H., Constant, A., Tatsiramos, K., Zimmermann, K.F., (2009). Native-Migrant Differences in Risk Attitudes. Applied Economics Letters, 16(15), 1581-1586. Doi:10.1080/ 13504850701578926

Bonin, H., Constant, A., Tatsiramos, K., Zimmermann, K.F., (2012). Ethnic Persistence, Assimilation and Risk Proclivity. Journal of Migration, 1(5)

Borjas, G.J., (1989). Economic Theory and International Migration. International Migration Review,

23(3), 457-485

Bowles, S. (1970). Migration as Investment: Empirical Tests of the Human Investment Approach to Geographical Mobility. The Review of Economics and Statistics, 52(4), 356-362.

Cameron, L., Erkal, N., Gangadharan, L., Zhang, M. (2014). Cultural Integration: Experimental Evidence of Convergence in Immigrants’ Preferences. Journal of Economic Behavior &

Organization, 111, 38-58

Castles, S. (2002). Migration and Community Formation under Conditions of Globalization. International Migration Review, 36(4), 1143-1168. doi:10.1111/j.1747-7379.2002.tb00121.x David, P.A. (1974). Fortune, Risk and the Micro-Economics of Migration. In P.A. David and M.W.

Reder (Eds.), Nations and Households in Economic Growth. (21-73) New York: Academic Press

Ding, X., Hartog, J., Sun,Y. (2010). Can We Measure Individual Risk Attitudes in a Survey? Working paper

Dohmen, T., Huffman, D., Schupp, J., Falk, A., Sunde, U., Wagner, G.G. (2011). Individual Risk Attitudes: Measurement, Determinants, and Behavioral Consequences. Journalof the

European Economic Association, 9(3), 522-550.doi: 10.1111/j.1542-4774-2011.01015.x

Gibson, J., & Mckenzie, D. (2011). Eight Questions about Brain Drain. Journal of Economic

Perspectives, 25(3), 107-128. doi:10.1257/jep.25.3.107

Hart, R.A. (1975). Interregional Economic Migration: Some Theoretical Considerations (Part II).

Journal of Regional Science, 15, 289-305

Haug, S (2008) 'Migration Networks and Migration Decision-Making', Journal of Ethnic and

Migration Studies, 34:4, 585 — 605

Holt, C.A. & Laury, S.K. (2002). Risk Aversion and Incentive Effects. American Economic Review, 92 (5), 1644-1655. doi:10.1257/000282802762024700

Jaeger, D.A., Dohmen, T., Falk, A., Huffman, D., Sunde, U., Bonin, H. (2010). Direct Evidence on Risk and Migration. The Review of Economics and Statistics, 92(3), 684-689

Katz, E., & Stark, O. (1986). Labor Migration and Risk Aversion in Less Developed Countries.

Journal of Labor Economics, 4(1), 134-149

Knight, Frank H. (1921). Risk, Uncertainty, and Profit. Hart, Schaffner, and Marx Prize Essays, no. 31. Boston and New York: Houghton Mifflin.

Massey, D.S., Arango, J., Hugo, G., Kouaouci, A., Pellegrino, A., Taylor, J.E. (1993). Theories of International Migration: a Review and Appraisal. Population and Development Review, 19 (3), 431-466

(23)

22 O'Connell, P. (1997). Migration under Uncertainty: “Try your Luck” or “Wait and See”. Journal of

Regional Science, 37(2), 331-347

Sahm, C.R. (2007). How Much Does Risk Tolerance Change? (FEDS Working Papers 2007-66).

Retrieved from Federal Reserve website:

http://www.federalreserve.gov/PubS/feds/2007/200766/revision/200766pap.pdf

Sjaastad, L.A. (1962). The Costs and Returns of Human Migration. Journal of Political Economy, 70, 80-93

Smith, T. (1979). Migration, Risk Aversion, and Regional Differentiation. Journal of Regional

Science, 9(1)

Stark, O., & Bloom, D.E. (1985). The New Economics of Labor Migration. The American Economic

Review, 75(2), 173-178

Stark, O., Levhari, D., (1982). On Migration and Risk in LDCs. Economic Development and Cultural

Change, 31(1), 191-196

Todaro, M.P. (1969). Urban job Expansion, Induced Migration and Rising Undemployment in Less Developed Countries. American Economic Review, 59, 138-148

The relationship between individual risk preferences and the probability to migrate