Do remittances affect output? Evidence from Lebanon

ABSTRACT

This paper examines the effects of remittances on output for Lebanon, a case with a unique setting: remittances to Lebanon are primarily from workers located in oil-producing Gulf States and thus influenced by oil prices. This setting enables us to estimate a structural VAR model using the price of oil as an instrument. We estimate the model with and without the instrument and compare the impulse response of remittances. Our key results are as follows: without the instrument, we find that a one-standard deviation shock in remittances increases industrial production by about 0.10% to 0.20%, a small and weakly estimated effect. In the instrumental variable setting, the magnitude increases by a factor of at least 5.8: the increase in output ranges from 0.72% to 1.11%. Our results imply that remittance effects on output computed using standard VARs are likely underestimated. Moreover, they suggest that the adoption of policies that facilitate remittance flows should be encouraged as they likely stimulate economic growth by more than previously detected.

KEYWORDS:

• Remittances
• structural vector autoregression
• impulse response function
• instrumental variable

JEL CODES:

• C32
• E32
• F16
• F22

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

¹ See, for instance, Chami et al. (2018), Clemens and McKenzie (2018), Acosta, Lartey, and Mandelman (2009), and Francois et al. (2022).

² Depken, Nikšić Radić, and Paleka (2021), Vargas-Silva (2008), and Akkoyunlu and Kholodilin (2008).

³ According to World Bank (2020), remittances have averaged about 20% of Lebanon’s GDP for the past decade.

⁴ The sample period ends in 2019 to avoid the inclusion of anomalous changes in industrial production and remittances following the COVID-19 pandemic.

⁵ To corroborate this observation, we obtained annual real GDP data for Lebanon from the St. Louis FRED dataset. The annualized version of the log of the IP variable and the log of real GDP display a correlation coefficient of 0.85.

⁶ See Appendix for sources.

⁷ Supplemental unit root tests, such as the Phillips-Perron test, had similar results.

⁸ Cheng, Han, and Inoue (2021) shows that inference in SVAR-IV models is still possible even if some of the variables are nonstationary.

⁹ The results are similar with additional lags.

¹⁰ Since we treat oil prices as exogenous to economic conditions in Lebanon, we assume that the IP’s structural shocks are not contemporaneously correlated with oil prices, a necessary assumption for identification.