Trade, growth and wage inequality in Bangladesh

Abstract

Using model selection techniques based on out-of-sample predictive ability criterion in a Vector Autoregression (VAR) framework, this paper empirically examines the causal relations among growth, trade, and wage inequality in Bangladesh between 1971 and 2000. There is some evidence of bi-directional causality between growth and inequality and between trade and growth. That growth causes trade and that trade causes inequality are robust results. Furthermore, evidence strongly suggests that investment is important for trade, and the terms of trade between agricultural products and manufacturing products is an important causal determinant of both growth and trade.

Key Words:

• Bangladesh
• trade
• trade openness
• wage inequality
• out-of-sample predictive ability
• mean squared forecast errors

Acknowledgements

The authors are grateful to Shahina Amin, Don Freeman, and two anonymous referees for their useful comments. Comments and questions from the members of audiences at the 31st Annual Meeting of the Missouri Valley Economic Association, the 75th Annual Meeting of the Southern Economic Association, and the 43rd Annual Conference of the Indian Econometric Society are greatly appreciated. Thanks are also due to Gabi Eissa for his research assistance. The usual disclaimer remains.

Notes

The New Industrial Policy (NIP) announced by the government in 1982 outlined reform measures that were aimed at promoting private sector-led industrialization. The Revised Industrial Policy (RIP) of 1986 re-emphasized the role of the private sector by further strengthening the incentives for private acquisition of public enterprises. Special incentives to attract foreign direct investment (FDI) and significant liberalization of import licensing were other measures that were intended to help the reform measures. The Industrial Policy of 1991 and trade policies of mid-1990s placed further emphasis on trade liberalization.

We use the term ‘causal relation’ in the sense of ‘Granger causality’ as defined in Granger (1969, 1980). By using the percentage share of exports and imports in GDP as the trade variable in our empirical analysis, we are examining the relations among trade openness, growth, and wage inequality. As we will discuss in the third section, this measure of trade openness captures the effect of trade liberalization policy. Although the fact that the average tariff rate in Bangladesh decreased from 100 per cent in early 1980s to about 26 per cent in the late 1990s (see Berg & Krueger, 2003) may indicate that trade liberalization has contributed to increased trade openness, a formal validation of such a link is outside the scope of this paper.

This method has recently been used by Krishna et al. (2003).

Other notable examples include Dollar and Kraay (2003, 2004), Edwards (1998), Frankel and Romer (1999), Harrison (1996) and Islam (1995).

Examples of recent works based on Kuznets' hypothesis include Chambers (2007), Lin et al. (2006), and Lopez (2006).

Some prominent empirical and theoretical studies include Aghion and Bolton (1997), Alesina and Rodrik (1994), Bandyopadhyay and Basu (2005), Banerjee and Duflo (2000), Banerjee and Newman (1993), Barro (2000), Bertola (1993), Castello and Domeneh (2002), Forbes (2000), Galor and Zeira (1993), Persson and Tabellini (1994).

Khan (1990) is an exception. He uses income elasticity of demand for food items to study income distribution.

Some studies also use log of per capita real GDP. Since our objective is to focus on the interaction among trade, growth, and inequality, we use growth.

The implicit assumption is that CPI for the working class reflects costs of living for workers engaged in manufacturing and construction, and CPI for rural families reflects the costs of living for workers engaged in agriculture and fishery, which are predominantly rural industries.

This is to say that, on average, during the sample period the real value of a US dollar is equivalent to the real value of 8 Bangladeshi taka: what a dollar can buy in the US is equivalent to what 8 taka can buy in Bangladesh. In other words, $1 can buy eight times more than what 1 taka can buy.

See Granger (1980), Ashley et al. (1980) for early advocates; and Chao et al. (2001) and the references therein for more recent advocates.

Our approach is very similar to Krishna et al. (2003)

For a discussion on the usefulness of ‘general-to-specific’ approach, see Hendry (1995).

There is no general rule as to how one chooses the maximum lag length to start with. Enders (2004: 192) suggests that one should ‘start with a relatively long lag length…’. Some researchers use the following rule of thumb: start with a maximum lag length equal to the cube root of the number of observation which is 3. ( ) in our case. We also use other information criteria, such as the Akaike Information Criterion (AIC) or Hannan – Quinn Criterion (HQC). Most times these criteria choose the same lag length. Even for cases with different lag lengths selected by different criteria the ADF test results are qualitatively similar.

In this form we are assuming that each element of y is an I(1) process and thus Δy is a vector of I(0) variables. In application, after determining the order of integration of each of growth, trade, inequality, investment, inflation, fiscal, terms of trade and real exchange rate, we will include the stationary forms of the respective variables in the vector Δy.

As we can see from the table, the MSFE for Model 23 is the same. However, this is because of rounding of the value to the two decimal places. At five decimal places, MSFE for Model 22 is 12.15921 and MSFE for Model 23 is 12.15928.

We also compare these best models with simple AR(1) models for growth, trade and inequality using a Diebold – Mariano (see Diebold & Mariano, 1995) type test. We use forecast errors from both models to construct the test statistics as follows (see Amato & Swanson, 2001; and MacCracken, 1999):

• Following a suggestion from Amato & Swanson (2001), we use unity as the 5 per cent critical value. We find that the best models outperform the simple AR model in all three cases.

For a study using disaggregate level data, see Salim (2003).

In fact, the trade ratio jumped from less than 20 per cent in 1991 to more than 30 per-cent in 1995.

We find that export is an I(1) process and import is an I(0) process.

For a discussion, see Enders (2004: 283 – 284).