Working Paper Series: Special Edition of 2016 to 2018 Interns

IV.

Methodology

(i) Framework Based on the conceptual framework highlighted above, the dynamic panel specification is the most appropriate estimation technique applicable, since the data set consisted of 32 cross sectional data points with four 4-year averages non-overlapping periods. This approach accounts for any chronologic autocorrelation and reduce any potential spurious regression that can cause regression estimates to be inconsistent, paving the way for more accurate inferences. Baseline Model In light of the existing literature, the baseline specification model will take the form: _ = 0 + 1 _ , −1 + 2 + 3 X′ + + (3) To assess GDP growth, we rewrite equation (3) as: _ ℎ = _ − _ , −1 (5) 36 ∴ _ ℎ = 0 + ( 1 − 1) _ , −1 + 2 + 3 X + + + (6) Where ( 1 − 1) is the convergence coefficient, _ denotes the initial level of real per capita GDP, is equal to remittances in current levels and X is a vector of core regressors used to model per capita GDP excluding remittances as described in table 2. The variable is a set of time dummies incorporated to capture business cycle effects, is the unobserved country-specific effect, and is the error term. The subscript ‘i’ represents the different countries while ‘ t’ represents the 4-year average periods. Using a pooled OLS and a panel regression (fixed effects) in its core form is dubious because _ , −1 will correlate with . Therefore, the preferred estimation technique applicable here is the Generalized Method of Moments developed by Arellano and Bover (1995) and Blundell and (ii) = + (4)

36 David Romer in his book “Advanced Macroeconomics” reminded us that a key fact about growth rates is that the growth rate of a variable equals the rate of change of its natural log. The natural log was used in all models of this paper thus accounting for per capita GDP growth.

131