Location intentions of immigrants at retirement: stay/return or go ‘back and forth’?

Abstract

This article analyses the intentions of immigrants regarding their residential choice after retirement. Three options are investigated: stay in the host country, go back to the origin country and adopt a mobile strategy, sharing time between the host and the origin country (defined as ‘back and forth’). For this purpose, we analyse a cross section of non-French nationals aged above 45 who are not yet retired. The impacts of several potential determinants of such intentions are investigated, with particular focus on the effect of the actual location of the respondents’ children. The estimation is conducted using a multinomial Logit model. Instrumental variable technique is used to address the issue of endogenous children's location with respect to parents’ location intention. Results suggest that the immigrants prefer to return rather than stay in the host country when their children live in the country of origin. However, children's location does not seem to affect the ‘back and forth’ strategy with respect to staying in the host country.

Acknowledgements

We are indebted to Graham Cookson, Christian Dustmann, Alan Manning, Arnaud Chevalier, Olivier Guillot and an anonymous referee for their very helpful comments and suggestions on previous drafts. We also thank participants at the Centre for Economic Performance annual meeting and the European Society for Population Economics (Bergen). Any remaining errors are ours.

Notes

¹ The use of panel data may lead to measurement errors, in particular owing to selective attrition. It is a priori much more difficult to be successful in tracking a migrant who decides to return to the origin country and then to come back in the host country.

² This is particularly important given that the health of older immigrants is significantly worse than comparable natives in European countries (Solé-Auró and Crimmins, 2008).

³ The empirical results reported in Dustmann (2003) are consistent with a model where parents caring for their children make different return plans according to whether they have sons or daughters. They return more often when they have daughters than sons. This is compatible with a story where children are causally related to intended locations of working aged parents.

⁴ A convenient way to include this behaviour is to simply assume that the number of contact enters positively into the utility function of parents.

⁵ It is interesting to note, however, that nonretired immigrants generate visits and travels from their family and relatives still living in the origin country (Prescott et al., 2005).

⁶ This is plausible as interaction with children is a service without any direct substitute (Ehrlich and Lui, 1991; Laferrère and Wolff, 2006).

⁷ This is the case of family altruism, which remains unobserved. Altruism is important as children will provide more upstream transfers and attention if they are more altruistic towards their parents, while altruistic parents will also experience more satisfaction from their children's contact.

⁸ Unobserved heterogeneity could be addressed by the use of panel data and fixed effect regressions. Also, measurement errors could be addressed by the use of a panel data set that tracks intention of parents and location of children over a longer period of time.

⁹ The GSOEP has been used for studying returns using the attrition property of the data set where a ‘moved abroad’ is indicated (Constant and Zimmermann, 2003; Dustmann, 2003). Burda et al. (1998) have also used the GSOEP to investigate moving intentions of East Germans to West Germany.

¹⁰ In France, the bulk of workers retire at the age of 60, and a significant proportion is retiring between 55 and 60 years. Self-employed people or shopkeepers tend to retire later.

¹¹ On this point, see Hausman and McFadden (1984). In our context, several Hausman tests of the IIA are possible. After having chosen a base category, two tests can be conducted by excluding each of the remaining categories to form the restricted model. Another test can be computed by changing the base category.

¹² The resulting costs (financial and health related) of a move for them seem too high at this later stage of life to be recovered by subsequent benefits.

¹³ We have also estimated the recursive model using a maximum likelihood method. For that purpose, we add to the MNL model an unobserved heterogeneity term (specific to each observation) and suppose that this perturbation and the residual of the Probit equation follow a bivariate normal distribution with unitary variances and a correlation coefficient taking a value in the range from −1 to 1. We estimate jointly the Probit and the MNL equations by maximizing the corresponding likelihood function. Residuals are integrated out numerically, since there is no closed form solution for the likelihood (Lillard and Panis, 2003).

¹⁴ To further investigate this result that better assimilated migrants are less likely to return, we have also estimated models with dummy variables for respondents holding the French citizenship and owning a home in the origin country. Although both covariates are most probably endogenous to the location decision, they are negatively and significantly associated with the return intention at retirement.

¹⁵ The corresponding statistic is equal to 61.8 with 13 degrees of freedom and thus highly significant.

¹⁶ It is not possible to control for unobserved heterogeneity by introducing fixed effects in the regression since the choice of a parent is identical across all children of the same family.

¹⁷ See Angrist and Pischke (2009) for further details on the exclusion restriction.

¹⁸ We find a value of 7.62 for the corresponding overidentification test, with a p-value of 0.02.