Page 94 - Azerbaijan State University of Economics

THE JOURNAL OF ECONOMIC SCIENCES: THEORY AND PRACTICE, V.83, # 1, 2026, pp. 82-106

market wage is‚ however‚ not self-obvious. Hence‚ we also use a regression model
controlling for country-level unobservable effects to study these relations.

6. REGRESSION ANALYSIS
We used linear regressions to model the relationship between unemployment and
remittances‚ separately by country‚ and pooled with country fixed effects. The regression
framework allows us to control for the initial level of country unemployment and
eliminate the possibility that our results are merely due to differences in unemployment
rates between countries. We first show the result of running the regressions separately for
each country‚ and then the pooled regressions with fixed effects.

Table 3 shows the output of four separate simple linear regressions (one regression
for each country) that predict the unemployment rate on the independent variable
remittances (% of GDP). The constant (intercept) and slope can be read off of the
output. The estimated constant (intercept) of the regression model should be
interpreted as the predicted value of unemployment when remittances take a value of
0 (that is‚ the case when there are no remittances). We interpret these slopes as the
change in the unemployment rate in percentage points following a one percentage
point change in the remittances-to-GDP ratio. In terms of the proportion of the
variations in unemployment rates explained by remittances‚ we report the R-squared
statistic for each specification.

Table 3: Country-specific regression of unemployment on remittances
Country Intercept (α) Slope β (Remittances effect) R² Sig. of β
Kyrgyzstan 8.723 (***) -0.1371 (***) per 1% remittances 0.750 p < 0.001
Moldova 1.753 (***) +0.1548 (***) per 1% remittances 0.569 p < 0.001
Nepal 10.488 (***) +0.0069 (n.s.) per 1% remittances 0.005 p = 0.79
Tajikistan 10.321 (***) +0.0430 (**) per 1% remittances 0.409 p = 0.003
p < 0.001, p < 0.01, n.s. = not significant. Coefficients are unstandardized. Each row
is an independent regression for that country. R² is the coefficient of determination.*

The unemployment-remittances relationship varies greatly for each country‚ as shown
in Table 3. Kyrgyzstan's average unemployment rate is negatively related to
remittances‚ and is statistically meaningful with a β of -0.1371. Importantly‚ our
results imply a 0.137 percentage points decline in the unemployment rate for every 1
percentage points increase in remittances as a share of GDP. Therefore‚ using our
estimated effect to extrapolate on a larger remittances growth (for example‚ from 10%
of GDP to a 30% of GDP)‚ we may estimate a decrease of 2.74 percentage points in
unemployment rate. An intercept coefficient of 8.723 implies that the unemployment
rate would be around 8.7% if remittances were zero. The model was then run‚ and it
found that when remittance levels are 32% of GDP‚ it would predict an unemployment

94