Drivers and Dynamics of Internal and International Remittances

Abstract

This article analyses whether and how intra-household remittance volumes vary with the length of a migrant's absence, and whether the drivers and dynamics of remittance decay depend on the migrant's destination. We address these questions by using data from the 64th round of the Indian National Sample Survey, conducted between July 2007 and June 2008. We find that the average intra-household remittance function follows a curvilinear trajectory that is best approximated by a quartic (M-shaped) specification. The volatility of intra-household remittances is distinct across migration destinations, with international remittance flows being more volatile than internal intra-state or inter-state flows.

Notes

1. For instance, an international phone call between India and the United Arab Emirates can cost as much as 185 Rupees (approximately 3 US dollars) per minute (Vodafone India, 2012).

2. Given a household's utility function by with V^h , C^h , I^h denoting the household's utility, consumption, income, and being a fixed quantity of services or assets purchased by the migrant. The household would accept to provide these assets if its participation constraint holds, that is if , such that the minimal compensating transfer is given by showing the amount remitted is increasing with the recipients income (Rapoport & Docquier, 2005).

3. Due to the declining percentage of short-term (and high-remitting) migrants in cross-sectional survey data, average remittance volumes are initially higher and decrease afterwards. This might not be true for individual migrants who might remit more steadily over time.

4. A household is defined as a ‘group of persons who normally lived together and took food from a common kitchen’ (NSSO, 2010: 6)

5. Out migrants are defined as ‘former members of the household who had migrated out of the village or town in the past and were alive on the date of the survey’ (NSSO, 2010: 100).

6. Table A1 in the Online Appendix reports descriptive statistics on all variables of our dataset.

7. See discussion at the end of Section 3.

8. R_i and the lower tail of the distribution of u_i is truncated at zero. Hence the mean of . See Hoddinott (1992) for discussion.

9. The truncated regression is discussed in more detail in Greene (1990: 721-723). The truncation model estimated is truncated at R>0 using maximum likelihood estimation techniques.

10. IFMR's (2010) survey on remittances in India reports that 57 per cent of respondents most recently used an informal system (mostly hawala couriers) to remit money. The use of hawala couriers does not appear to change with the size of the remittance amount. Hawala is consistently described as more expensive than bank transfers: between 4 and 45 per cent, depending on remittance amount. The use of the courier system is frequently not the preferred transfer method: 49 per cent of respondents would actually prefer to use banks. Therefore, lack of opportunities to use more formal channels is related to remoteness and inconvenient access to the banking system.

11. Given income is increasing in age but is subject to diminishing marginal returns, we proxy this effect with a variable constructed from an interaction between age and the average education of the family and an interaction between age squared and the average education of the family.

12. We calculate the average level of education for each household.

13. Akee and Kapur (2012) test this hypothesis in India by cross-checking survey respondents' answers to remittance questions with data from the respondents' own bank account. They find large differences between self-reported frequency of remittances and actual deposits and between the average amounts reported and actually remitted: the modal value for self-reports is 12 deposits per year while the administrative data indicates only two to three deposits per year; the average self-reported amount remitted in a year is Rs. 345,938 compared to an actual average amount deposited per year of Rs. 677,269. Beyond this, Akee and Kapur also find evidence that misreporting is mean-reverting, such that high remittance earners under-report remittances, and low remittance earners over-report remittances, and also that more highly educated households tend to be more accurate at reporting remittances.

14. The probit and truncated regression outputs are not reported but are available on request from the authors.

15. The coefficient on ‘urban’ was -0.984 with a standard error of 0.0271, the coefficient on ‘being economically active’ was 2.310 with a standard error of 0.724. Full results are available on request from the authors.

16. Available from the authors on request.

17. The results of this logit regression are available from the authors on request.