Migrant Remittances and Physical Investment Purchases: Evidence from Kenyan Households

Abstract

This paper investigates the impact of remittances on household decisions to purchase physical investments in Kenya using household survey data. An instrumental variables approach is employed using rainfall variation and mobile network coverage as instruments to control for the endogeneity of remittances. The empirical evidence obtained is suggestive of remittances having a positive and significant effect on the decisions by households to purchase physical investments.

Acknowledgements

The author would like to thank Barry Reilly, and two anonymous referees of the journal for providing constructive comments on earlier drafts of this paper.

Disclosure statement

No potential conflict of interest was reported by the author.

Notes

1. We classify these commodities under ‘physical investments’ in the current paper.

2. These districts are Kiambu, Kisii, Machakos, Nakuru, Nairobi, Rachuonyo, Thika, Garissa, Lugari, Malindi, Embu, Siaya, Kilifi, Kakamega, Migori, Vihiga, and Mombasa.

3. The author would like to thank an anonymous referee of the journal for this suggestion.

4. The survey does not distinguish between the setting up of a business and the opening of a store. We classify these as physical investments as it seems reasonable that such activities will most likely involve the acquisition of some form of business inventory. Examples of common micro-business set-ups in Kenya include retailing, food and drink production and other artisanry (see, for example, Ronge, Ndirangu, & Nyangito, 2002). Moreover, studies in the literature typically classify business investments as physical investments (see, for example, Amuedo-Dorantes & Pozo, 2014).

5. After some deliberation housing purchase was excluded from the physical investments category. As highlighted by an anonymous referee of this journal, in aggregate data housing purchase is likely to follow a different stochastic process from that of physical investments because at the household level choices are very different from each other. Studies such as Osili (2004) study investment in housing separate from physical investment. In the current study, only a small sample of households purchased housing therefore we are unable to study this type of investment separately.

6. We use predicted expenditure values to account for the endogeneity of expenditure.

7. Unfortunately, we are unable to resolve this challenge as we cannot redress this feature of the dataset. However, we conduct some checks to ensure that the findings of our analysis are robust to the rescaling of remittances.

8. Only 71 households were eliminated due to missing observations. Households with missing observations appear to be random as there do not seem to be any systematic differences in characteristics between these and the rest of the households in the survey.

9. As highlighted by an anonymous referee of the journal, it is possible that a high variance of rainfall made some areas poor such that households may not afford to send migrants thus violating the monotonicity assumption. The variance of rainfall is plotted against the number of migrants by district to investigate this. Figure A1 in the Appendix reveals that there is no obvious correlation between migrant numbers and rainfall patterns except for two outliers representing Nairobi and Lugari districts which have a very large and small number of migrants, respectively. Nairobi district is an obvious outlier given its large population, strong economic performance, and high education levels relative to other districts. On the other hand, Lugari district is unique in that in 2009 it was a relatively new, largely rural district. It experienced large numbers of in-migration (about 40% of the population) due to its high agricultural potential (see Kenya 1999 Population and Housing Census: The Popular Report, Volume 10, 2002). This may explain the small number of households with migrants relative to other districts in the sample. Moreover, a statistically insignificant correlation coefficient (p-value = 0.3574) is obtained when the outliers are excluded.

10. Urban areas in Kenya have more money transfer agents such as Western Union and Money Gram and are also likely to have more financial institutions such as banks through which money can be transmitted easily. Jack and Suri (2011) show that M-PESA uptake was particularly large among Kenyans residing in rural areas.

11. It is acknowledged that this represents a modelling convenience as the aforementioned tests are only strictly appropriate for use when the first and second stage models have continuous dependent variables. However, we believe the approach provides some approximate insights regarding the validity of the instrument set used here and has been employed in the literature by studies such as Ambrosius and Cuecuecha (2016).

12. The F-test for instrument relevance falls below the rule-of-thumb of 10 suggested by Staiger and Stock (1994). Moreira (2003) proposes a Conditional Likelihood Ratio (CLR) test that is robust to the use of weak instruments within an IV framework. The key results obtained here do not alter when the CLR test is employed.

13. However, given the weak instruments, the power of the Wu-Hausman test is acknowledged to be poor and therefore we proceed to employ an IV strategy.

14. Conditional IV regression estimates that are robust to the use of weak instruments are comparable to the results obtained here.

15. We acknowledge that taking remittances to be equally distributed across the first and second six months is a strong assumption. However, the main results hold and may be less biased than prior to rescaling.