Discussion Paper

No. 2016-23 | June 08, 2016
A Comparative Analysis of Forced Migration: Cold War Versus Post-Cold War Eras

Abstract

This paper conducts a comparative analysis of forced international migration between two historical periods: 1969-1990 representing the Cold War era, and 1991-2012, the post-Cold War era. The determinants of refugee migration over the two periods are assessed and compared using a panel data analysis for a sample of 125 countries. In order to control for unobserved country-specific effects and the joint endogeneity of the explanatory variables, the Arellano-Bond Dynamic Panel GMM (Generalized Method of Moments) estimator is used. Overall, the results suggest that significant changes have taken place with both the set of explanatory variables and their individual impact on refugee migration over time. Also noteworthy is the significance of the two flight facilitators – the Internet and the telecommunication devices – included in the model to explain the refugee migration dynamics during the post-Cold War period.

JEL Classification:

F22, D74, L86, L96, C23

Assessment

  • Downloads: 347

Links

Cite As

Bilol Buzurukov and Byeong Wan Lee (2016). A Comparative Analysis of Forced Migration: Cold War Versus Post-Cold War Eras. Economics Discussion Papers, No 2016-23, Kiel Institute for the World Economy. http://www.economics-ejournal.org/economics/discussionpapers/2016-23


Comments and Questions


Anonymous - Referee Report 1
June 20, 2016 - 09:55

The manuscript empirically investigates the determinants of international forced migration for two separate periods – 1969-1990 and 1991-2012 – using a panel of 125 countries. Results suggest that some determinants and their magnitudes differ across two periods. The manuscript needs substantial revisions. Below I provide my concerns.

Introduction
1. ...[more]

... Why sudden drop in refugee stock in early 1990s? What do negative refugee flow and refugee stock numbers suggest? Does your sample include the former Soviet Union and former communist countries? From the list of sample countries in the appendix table it does not appear so. The former communist countries are important for explaining the story in this manuscript, including the graph; ignoring these countries may lead to a selection bias that is similar to the one author(s) of this manuscript guarded against on p. 4, while criticizing the literature.

2. Motivation on why the two periods should be examined separately should be spelled out.

Methodology
3. The missing values are filled through interpolation. Footnote 1, p.5, suggests that the results without interpolation do not differ. I like to see those results reported in the appendix.

4. What is the proportion of zeros in the dependent variable? The dependent variable is a count – the number of refugees. Therefore, count regression models such as Poisson or Negative Binomial estimators (depending on your data) are appropriate. Why a count regression was not applied? I am interested in seeing the count regression results.

5. Further, the Dynamic Panel GMM estimator is not reliable with well-known issues. First, be careful with overstating that the endogeneity of variables are addressed; the assumptions the dynamic GMM estimator makes regarding the excludability restriction of the internal instruments should be discussed and justified. Second, I did not find any information and justification on exogenous and endogenous variables and the lag structure in the estimated models. Third, are the results based on a two-step efficient GMM estimator or one-step GMM estimator? Have you used Windmeijer’s robust standard errors in the second step? Fourth, the Dynamic Panel GMM estimator suffers from instrument-proliferation problem as the number of time-periods increases, which is the case in this paper. How is the problem of too many instruments addressed? More important, how many instruments does each regression model have? The validity of results depends on this.

6. Given the issues surrounding dynamic panel GMM estimator, the fixed-effects regression with lagged dependent variable should be estimated and the results should be reported for comparison.

Results
7. First, the comments in methodology part should be addressed. Then, the findings should be explained in more details. For example, why income is significant for the later sample period, while is not significant for the earlier sample period? Other results should also be explained. Further, when comparing the estimated coefficients across periods, a test should be conducted to see whether the differences (e.g., civil war) are statistically significant.


Anonymous - Referee Report 2
July 04, 2016 - 14:38

See attached file


Bilol Buzurukov and Byeong Wan Lee - Reply to Referee Report 1
August 23, 2016 - 12:21

See attached file


Bilol Buzurukov and Byeong Wan Lee - Reply to Referee Report 2
August 23, 2016 - 12:22

See attached file