Networks, human capital and export success: evidence from Bangladesh

Abstract

This research recognises firm heterogeneity theories and uses the extensive transaction-level data of customs to analyse the export success of Bangladesh at the level of annual activity of individual exporters during the period 2004–2011. It identifies the network of peer exporters as a significant factor that promotes export success in both direct and indirect channels. As well as the financial system reported in earlier studies, this study finds the formation of human capital as another influential candidate for an indirect channel. Our findings also suggest that at the exporter’s level, experience with destination and products, initial exports and the level of competency with products contribute significantly to the success of new exports. More importantly, for Bangladeshi exporters, the effect of destination experience is stronger than that of product experience.

Appendix

1.1 Part A: Probit model

To identify the determinants of export success, this research uses the following estimating equation:

$$E\left[ {y|x_{fpdt} ,z} \right] = P\left( {s_{fpdt} = 1} \right) = \varPhi \left( {\varvec{x}_{fpdt} \beta + \delta_{z} } \right)$$

(1)

Here, $\varPhi$ is a probit function. The dependent variable $s_{fpdt}$ is the first-year export survival of the new export (in firm–product–destination cells in a given year) that takes a value equal to ‘one’ if the export spell continues in the next year of entry, and ‘zero’ if it does otherwise. The vector of regressors $\varvec{x}_{fpdt}$ presents a set of explanatory variables and their interaction terms. Table 5 presents a complete list of these variables. HS-2 industry, destination and entry year fixed effects are denoted by the set of fixed effects $\delta_{z}$.

Since probit is a nonlinear model, we report the marginal effects of each variable of interest on $E\left( y \right)$ as $\frac{\partial E\left( y \right)}{{\partial x_{fpdt} }}$.

1.2 Part B: Calculating margins of interaction term

We have taken measures to correctly calculate the margins for the interaction term and its variables by considering the following nonlinear model which contains the variables $x_{1}$ and $x_{2}$ where $\varvec{x}_{\text{fpdt}} \epsilon \left( {\varvec{x}_{1} , \varvec{x}_{2} } \right)$ and their interaction term:

$$Ey|x_{1} , x_{2} ,z = \varPhi \left( {\beta_{1} x_{1} + \beta_{2} x_{2} + \beta_{12} x_{1} x_{2} + \delta z} \right) = \varPhi \left( A \right)$$

(2)

The corrected margins of the dependent variables $x_{1}$ and $x_{2}$ involved in the interaction term are:

$$\partial E\frac{{\left[ {y |x_{1} , x_{2} ,z} \right]}}{{\partial x_{1} }} = \varPhi^{\prime } \left( A \right) \times \frac{\partial A}{{\partial x_{1} }} = \phi \left( A \right) \times \left( {\beta_{1} + \beta_{12} x_{1} } \right)$$

(3)

and

$$\partial E\frac{{\left[ {y |x_{1} , x_{2} ,z} \right]}}{{\partial x_{2} }} = \varPhi^{{\prime }} \left( A \right) \times \frac{\partial A}{{\partial x_{2} }} = \phi \left( A \right) \times \left( {\beta_{1} + \beta_{12} x_{2} } \right)$$

(4)

In Eqs. (3) and (4), $\phi \left( A \right)$ is a standard normal probability density function.

Finally, the corrected margin for the interaction term is:

$$\partial E\frac{{\left[ {y |x_{1} , x_{2} ,z} \right]}}{{\partial x_{1} x_{2} }} = \beta_{12} \varPhi^{{\prime }} \left( A \right) + \left( {\beta_{1} + \beta_{12} x_{2} } \right)\left( {\beta_{2} + \beta_{12} x_{1} } \right)\varPhi^{{{\prime \prime }}} \left( A \right) = \beta_{12} \phi \left( A \right) + \left( {\beta_{1} + \beta_{12} x_{2} } \right)\left( {\beta_{2} + \beta_{12} x_{1} } \right)\left[ { - A\phi \left( A \right)} \right]$$

1.3 Part C: Selection effect

This research checks whether selection problem can arise from the demand shock of a product at a given destination which affects the baseline results. Due to this demand shock, more firms will start exporting the product to that particular destination and will survive if the shock persists. To address the effect of such a selection in the aforementioned results, this research uses the two-step Heckman selection model. The first step involves a probit model (Eq. 2) that accounts for the probability of export in a certain product–destination–year cell that has not served in the previous year.

$${ \Pr }\left( {(v_{fpdt} > 0 |v_{fpd,t - 1} = 0} \right) = \varPsi \left( {x_{fpdt} \beta + \gamma v_{pdt}^{\text{row}} + u_{fpdt} } \right)$$

(5)

Here, $v_{fpdt}$ is the volume of a product ‘p’ exported by the firm ‘f’ to the destination ‘d’ in the year ‘t’, ‘x’ is the vector for determinants of export used in this research, and $v_{pdt}^{\text{row}}$ is the volume of the total export by the rest of the world to the market ‘d’ in the year ‘t’.

The inclusion of the term $v_{pdt}^{\text{row}}$ in Eq. 5 indicates that if the demand shock is responsible for the new export and its subsequent survival, the demand shock can be captured by the export volume of that product from the rest of the world to that country because firms in other exporting countries might be more informed about the demand shock. In the second step, we use ‘the inverse mills ratio’ estimated from the probit model as an additional regressor as well as the other determinants for first-year export survival. Cadot et al. (2013) use a similar method to investigate the selection bias in the analysis of the export success in four African countries.