4.2. Econometric results: Baseline estimation
I started the baseline estimation with a most parsimonious to a least parsimonious regression. The first estimation consisted of a simple bivariate regression of labour productivity growth rate on TCI. Then, I progressively include a set of confounding factors to test how stable the estimate of interest is with or without the control variables. In each estimation, I control for country fixed effects in all the models, and country-specific year fixed effects in column (1), time trend in column (2), time trend squared in column (3), region-specific time trend in column (4), and region-specific time trend squared in column (5).
Table 1 shows the regression results of the baseline estimation. The coefficient of TCI is negative and significant at 1% level, even when controlling for different sets of fixed effects. The sign and the direction of the coefficient are in line with literature and my expectation. The findings implies that a country defying its comparative advantage when designing an industrial policy induces a negative effect on aggregate labour productivity growth. In another term, a 10% increase in TCI decreases labour productivity growth by around 0.12 percentage point.
I then estimate the relation between the control variables and labour productivity growth rates. The estimation results of the quadratic relationship between labour productivity growth rates and economic development 1990-2017 are shown in Table A 1 (Appendix). Since GDP per capita and human capital index are highly correlated, they are used separately as regressors. A higher institutional quality and trade openness have no direct impact on labour productivity growth rate (column 1 to 10). As for human capital, it has an adverse impact on labour productivity, with an impact that vanishes from column 2 to 5. An increase in FDI net inflows contribute significantly to labour productivity growth rate. The domestic credit to private sector contributes negatively to labour productivity growth. The effect holds even when I control for GDP per capita. Lastly the quadratic relationship between labour productivity growth and economic development is supported by the estimation in Table A 1 as the coefficients of GDP per capita and squared GDP per capita are significant.
In order to test the linear and quadratic impact of CAF/ CAD development strategy on labour productivity growth, TCI and squared TCI are added to the baseline model. The results are presented in Table 2. This implies that the effect of TCI is not linear, since weakly defying the comparative advantage contributes to productivity growth while strongly defying it begins to have an adverse effect.
Table 1
Baseline estimation of TCI effect on labour productivity growth 1990-2017
VARIABLES
|
Dependent variable: labour productivity growth rate (%)
|
(1)
|
(2)
|
(3)
|
(4)
|
(5)
|
TCI
|
-1.182***
|
-0.828**
|
-0.828**
|
-0.865***
|
-0.870***
|
|
(0.326)
|
(0.327)
|
(0.327)
|
(0.325)
|
(0.325)
|
Constant
|
3.324***
|
2.924***
|
2.924***
|
2.965***
|
2.971***
|
|
(0.375)
|
(0.377)
|
(0.377)
|
(0.375)
|
(0.375)
|
Observations
|
2,656
|
2,656
|
2,656
|
2,656
|
2,656
|
R-squared
|
0.228
|
0.167
|
0.167
|
0.160
|
0.161
|
Country FE
|
Yes
|
Yes
|
Yes
|
Yes
|
Yes
|
Country Year FE
|
Yes
|
No
|
No
|
No
|
No
|
Time trend
|
No
|
Yes
|
No
|
No
|
No
|
Time2 trend
|
No
|
No
|
Yes
|
No
|
No
|
Region-specific time trend
|
No
|
No
|
No
|
Yes
|
No
|
Region-specific time2 trend
|
No
|
No
|
No
|
No
|
Yes
|
Note: The variable TCI is presented in log. Robust standard errors in parentheses. Significance level at *** p<0.01, ** p<0.05, * p<0.1. Source: Author’s calculation |
This implies that the effect of TCI is not linear, since weakly defying the comparative advantage contributes to productivity growth while strongly defying it begins to have an adverse effect. The effect holds when I control for GDP per capita which is also significant, indicating that it is not only driven by the economic development, but also the non-linear dynamics of TCI impact. However, the effects of both TCI and squared TCI vanish when I control for human capital index. Financial development, proxied by the domestic credit to private sector by banks, has a negative effect on labour productivity growth from column (1) to (10). This finding is in line with a study conducted by Ghani and Suri (1999) which, investigated the effect of capital accumulation and the Banking sector on productivity growth in Malaysia. In the case of Malaysia, rapid growth in bank lending was associated with falling total factor productivity. This finding suggests that financial development plays a crucial role in enhancing productivity growth only when capital is efficiently allocated to most productive activities. As expected, the estimated coefficient of FDI is positive and significant, suggesting that an increase in FDI net inflow in a country induces labour productivity growth.
However, the estimates from Table 2might be biased and doesn’t account for the lag value of labour productivity level in the equation 1. Therefore, the next analysis account for it and assess the robustness of the findings using System GMM.
Table 2
Estimation of the effects of TCI and TCI2 on labour productivity growth FE
VARIABLES
|
labour productivity growth rate (percent)
|
(1)
|
(2)
|
(3)
|
(4)
|
(5)
|
(6)
|
(7)
|
(8)
|
(9)
|
(10)
|
TCI
|
1.716
|
2.056*
|
2.056*
|
2.012*
|
2.011*
|
0.418
|
0.851
|
0.851
|
0.806
|
0.808
|
|
(1.146)
|
(1.192)
|
(1.192)
|
(1.186)
|
(1.186)
|
(1.064)
|
(1.107)
|
(1.107)
|
(1.101)
|
(1.101)
|
TCI2
|
-0.537*
|
-0.600**
|
-0.600**
|
-0.607**
|
-0.610**
|
-0.261
|
-0.431
|
-0.431
|
-0.427
|
-0.430
|
|
(0.285)
|
(0.296)
|
(0.296)
|
(0.295)
|
(0.295)
|
(0.266)
|
(0.275)
|
(0.275)
|
(0.274)
|
(0.274)
|
Arable land (% of land area)
|
-0.006
|
-0.008
|
-0.008
|
-0.004
|
-0.004
|
-0.011
|
-0.013
|
-0.013
|
-0.012
|
-0.012
|
|
(0.048)
|
(0.050)
|
(0.050)
|
(0.050)
|
(0.050)
|
(0.046)
|
(0.048)
|
(0.048)
|
(0.048)
|
(0.048)
|
Institution
|
0.003
|
0.010
|
0.010
|
0.002
|
0.001
|
0.005
|
0.030
|
0.030
|
0.020
|
0.019
|
|
(0.040)
|
(0.041)
|
(0.041)
|
(0.041)
|
(0.041)
|
(0.038)
|
(0.039)
|
(0.039)
|
(0.039)
|
(0.039)
|
Domestic credit to private sector by banks (% of GDP)
|
-0.026***
|
-0.031***
|
-0.031***
|
-0.031***
|
-0.031***
|
-0.020***
|
-0.020***
|
-0.020***
|
-0.020***
|
-0.020***
|
|
(0.006)
|
(0.006)
|
(0.006)
|
(0.006)
|
(0.006)
|
(0.005)
|
(0.005)
|
(0.005)
|
(0.005)
|
(0.005)
|
Population growth (annual %)
|
-0.079
|
-0.095
|
-0.095
|
-0.096
|
-0.096
|
-0.281***
|
-0.316***
|
-0.316***
|
-0.303***
|
-0.304***
|
|
(0.108)
|
(0.112)
|
(0.112)
|
(0.111)
|
(0.111)
|
(0.099)
|
(0.103)
|
(0.103)
|
(0.103)
|
(0.102)
|
Trade Openness
|
-0.705
|
-0.087
|
-0.087
|
-0.157
|
-0.149
|
-0.790
|
-0.114
|
-0.114
|
-0.212
|
-0.207
|
|
(0.646)
|
(0.661)
|
(0.661)
|
(0.658)
|
(0.658)
|
(0.597)
|
(0.609)
|
(0.609)
|
(0.606)
|
(0.606)
|
Foreign direct investment, net inflows (%GDP)
|
0.024***
|
0.020**
|
0.020**
|
0.021**
|
0.021**
|
0.015*
|
0.011
|
0.011
|
0.013
|
0.013
|
|
(0.009)
|
(0.009)
|
(0.009)
|
(0.009)
|
(0.009)
|
(0.008)
|
(0.008)
|
(0.008)
|
(0.008)
|
(0.008)
|
lngdpc
|
1.884***
|
0.839***
|
0.839***
|
0.795**
|
0.784**
|
|
|
|
|
|
|
(0.714)
|
(0.323)
|
(0.323)
|
(0.322)
|
(0.322)
|
|
|
|
|
|
Human Capital Index
|
|
|
|
|
|
-2.262**
|
-0.290
|
-0.290
|
-0.260
|
-0.275
|
|
|
|
|
|
|
(0.991)
|
(0.568)
|
(0.568)
|
(0.566)
|
(0.567)
|
Constant
|
-14.193**
|
-5.254*
|
-5.254*
|
-4.772
|
-4.676
|
9.567***
|
3.892**
|
3.892**
|
3.943**
|
3.991**
|
|
(6.611)
|
(3.138)
|
(3.138)
|
(3.126)
|
(3.129)
|
(2.774)
|
(1.909)
|
(1.909)
|
(1.900)
|
(1.902)
|
Observations
|
2,036
|
2,036
|
2,036
|
2,036
|
2,036
|
2,101
|
2,101
|
2,101
|
2,101
|
2,101
|
R-squared
|
0.260
|
0.202
|
0.202
|
0.192
|
0.192
|
0.233
|
0.175
|
0.175
|
0.165
|
0.165
|
Country FE
|
Yes
|
Yes
|
Yes
|
Yes
|
Yes
|
Yes
|
Yes
|
Yes
|
Yes
|
Yes
|
Country Year FE
|
Yes
|
No
|
No
|
No
|
No
|
Yes
|
No
|
No
|
No
|
No
|
Time trend
|
No
|
Yes
|
No
|
No
|
No
|
No
|
Yes
|
No
|
No
|
No
|
Time2 trend
|
No
|
No
|
Yes
|
No
|
No
|
No
|
No
|
Yes
|
No
|
No
|
Region-specific time trend
|
No
|
No
|
No
|
Yes
|
No
|
No
|
No
|
No
|
Yes
|
No
|
Region-specific time2 trend
|
No
|
No
|
No
|
No
|
Yes
|
No
|
No
|
No
|
No
|
Yes
|
Note: The variable TCI is presented in log. Robust standard errors in parentheses. Significance level at *** p<0.01, ** p<0.05, * p<0.1. Source: Author’s calculation |
4.3. Econometric results: Dynamic estimation and extension analysis
Table 3 reports the estimates from the dynamic model illustrated by equation 2 for the period 1990 - 2017. Regression models (1) to (7) use the System GMM approach to obtain the estimates. The variables in model (1) include only the proxy for development strategy, TCI (log), and the initial labour productivity level (log), whereas model (2) account for the squared TCI (log) and then from model (3), other control variables that capture institutional quality, size of the population, foreign direct investment, trade openness and land size were progressively added. However, due the high correlation between financial development, proxied by the domestic credit to private sectors by banks (% of GDP), human capital index, GDP per capita (log) and the one-period lag of the initial level of labour productivity (in 2011 international PPP exchange rate), these variables were not included in the dynamic model.
Before interpreting the estimation results, it is worth mentioning that the diagnostic test of system GMM estimator was satisfactory. The autocorrelation test shows that the residuals are an AR(1) process which is what is expected. The test statistic for second order serial correlation is based on residuals from the first-difference equation and it rejects the null hypothesis of serial correlation of second order. The instrument set is valid as evidenced by the Hansen test of over-identified restrictions and the variables of interest have expected signs. We note across all these columns that the p values related to AR(1) test are 0. While the p-values of AR(2) test are higher than 10%. Moreover, the p-value of the Hansen test is always higher than 10% and the number of instruments used in the regressions is always lower than the number of countries. In the regression I used maximum of 2 lags of dependent variable as instruments and 2 lags of endogenous variables as instruments.
The results in column (1) of Table 3 indicate that the TCI has the expected negative effect and is highly significant. This finding supports the idea that the further a country pursued a CAD strategy, the worse the labour productivity growth during the period 1990 – 2017. From the estimates I can infer that a 10 percent increase in the TCI from the mean may result in approximately a 0.74 percentage point in the reduction in the country’s average annual labour productivity growth for the entire period 1990 – 2017. Unlike the results in Table 2, TCI2 is positive across model (1) to (7) but significant only in model 6 and 7 even when controlling for other covariates. This indicates a nonlinear relationship between the comparative advantage development strategy on labour productivity growth. As unexpected this implies that slightly defying the comparative advantage may have a diminishing contribution to labour productivity growth during the period 1990 – 2017, while beyond a certain point, strongly defying it yields positive contribution to growth. This is finding is puzzling for the advocates of the NSE who advise for very small to zero deviations from the comparative advantage, and also unexpected against Chang’s suggestion of rather an ‘inverted-U-shaped relationship’ between an economy’s deviation from comparative advantage and growth.
The regression results also show that the initial level of labour productivity have the expected sign and significant effect, institutional quality combines with openness to trade strongly contribute to labour productivity growth during the period of 1990-2017. The results are in line with previous literatures on institutional quality (Acemoglu et al., 2002, 2003), trade openness (Alcala and Ciccone, 2004; De Loecker, 2013; Frankel and Romer, 1999; Hall and Jones, 1999) and labour productivity growth. However, the coefficients of FDI are positive but insignificant, implying that FDI did not play any major role in contributing on average on productivity growth during the period 1990 – 2017.
Table 3
GMM-system estimation of TCI effect on labour productivity growth 1990-2017
|
Dependent variable: labour productivity growth rate (%)
|
VARIABLES
|
(1)
|
(2)
|
(3)
|
(4)
|
(5)
|
(6)
|
(7)
|
Lpxr (lag)
|
-4.987***
|
-4.606***
|
-5.647***
|
-5.325***
|
-5.679***
|
-6.102***
|
-7.109***
|
|
(1.378)
|
(1.280)
|
(1.952)
|
(1.538)
|
(1.566)
|
(1.661)
|
(2.091)
|
TCI
|
-7.453***
|
-11.272**
|
-11.814**
|
-9.356**
|
-9.769**
|
-10.882**
|
-13.567**
|
|
(2.273)
|
(4.394)
|
(5.302)
|
(4.123)
|
(4.121)
|
(4.301)
|
(5.501)
|
TCI2
|
|
1.263
|
1.388
|
1.020
|
1.050
|
1.235*
|
1.518*
|
|
|
(0.799)
|
(0.854)
|
(0.697)
|
(0.704)
|
(0.700)
|
(0.814)
|
Institution
|
|
|
0.429**
|
0.315**
|
0.321**
|
0.341**
|
0.433***
|
|
|
|
(0.195)
|
(0.134)
|
(0.134)
|
(0.130)
|
(0.160)
|
Population growth (annual %)
|
|
|
|
-1.240*
|
-1.273*
|
-0.900*
|
-0.994*
|
|
|
|
|
(0.743)
|
(0.714)
|
(0.529)
|
(0.544)
|
FDI
|
|
|
|
|
0.047
|
0.014
|
0.009
|
|
|
|
|
|
(0.043)
|
(0.023)
|
(0.023)
|
Trade Openness
|
|
|
|
|
|
3.403***
|
3.616***
|
|
|
|
|
|
|
(1.242)
|
(1.362)
|
Arable land (% of land area)
|
|
|
|
|
|
|
-0.121*
|
|
|
|
|
|
|
|
(0.066)
|
Constant
|
61.583***
|
60.038***
|
68.744***
|
65.760***
|
69.591***
|
71.840***
|
86.387***
|
|
(16.607)
|
(16.518)
|
(23.540)
|
(18.672)
|
(18.895)
|
(19.283)
|
(25.667)
|
Observations
|
2,554
|
2,554
|
2,405
|
2,403
|
2,385
|
2,341
|
2,235
|
Number of countries
|
102
|
102
|
97
|
97
|
97
|
97
|
97
|
Year FE
|
Yes
|
Yes
|
Yes
|
Yes
|
Yes
|
Yes
|
Yes
|
AR(1)
|
0.000
|
0.000
|
0.000
|
0.000
|
0.000
|
0.000
|
0.000
|
AR(2)
|
0.193
|
0.206
|
0.326
|
0.427
|
0.436
|
0.470
|
0.460
|
Hansen
|
0.425
|
0.344
|
0.445
|
0.452
|
0.467
|
0.532
|
0.496
|
Number of Instruments
|
33.000
|
34.000
|
35.000
|
36.000
|
37.000
|
38.000
|
39.000
|
Note: The variables lprx (lag), TCI are presented in log. Robust standard errors in parentheses. Significance level at *** p<0.01, ** p<0.05, * p<0.1. Source: Author’s calculation |
In order to enhance our understanding of this puzzling finding of TCI effect on growth, I conducted specific sub-periods analysis as suggested by Bruno et al. (2015), and also to test the extent to which TCI is a more stable determinant in explaining productivity growth unlike other conventional determinants which have proven to vary in significance and importance during the last decades (Dieppe, 2021). Therefore, the period of analysis is divided into three periods, the pre global financial crisis (1990 – 2007), during the financial crisis and the period of recovery (2007 – 2012) and the post global financial crisis (2012 – 2017). |
Table 4 reports results of the estimation of the dynamic model for the period between 1990 – 2007, representing the pre-GFC. The results are similar to those of Table 3 with evidence of the negative effects of TCI on labour productivity growth and the non-linearity effect as shown by the significant level of TCI2. Both institutional quality and openness to trade significantly contribute to labour productivity growth, while there is no evidence of the effect of FDI during this period of 1990-2007. However, the coefficients of all significant variables are much larger than those of Table 3, implying the major role of these factor in the period of 1990-2007 compared to 1990-2017. The possible explanation of the positive and significant coefficient of institutional quality and trade during this period, is that in the 1990s and early 2000s, many countries embarked on some institutional reforms and trade liberalization advocated by the Bretton Woods institutions followed by the debt crisis. Institutional prerequisites such as protection of property rights, rule of law, efficient bureaucracy have been identified as factors limiting the influence of foreign aid, foreign investment, and education (Easterly, 2001). This explanation is supported by previous studies suggesting that lack of property rights protection hinders investment in both physical and human capital which are proximate determinants of economic growth (North and Thomas, 1973; Jones, 1981; North, 1981). As for trade openness, the literature theoretically and empirically provides evidence of its impact on growth. One way is, it can influence growth directly through absolute and/ or comparative advantage, and it can also increase efficiency indirectly through technology transfer, economies of scales, and competition with firms in domestic and international markets (Bloch and Tang, 2004). A possible reason for a non-significant effect of FDI during this period may be due to the environment. FDI inflows is conditioned by business-friendly environment. During this period, institutional quality may have not been high enough to spur the effect of growth, as suggested by Li and Tanna (2019) that countries that fall below a minimum level of institutional quality may have either a negative or statistically insignificant impact. The other way is this could be due to the lack of enough absorptive capacities (human capital) in developing countries which is one mechanism through which the gains of technology and the productivity Spillover effect associated with such investment are maximized.
Table 4
GMM-system estimation of TCI effect on labour productivity growth 1990-2007
VARIABLES
|
Dependent variable: labour productivity growth rate (%)
|
(1)
|
(2)
|
(3)
|
(4)
|
(5)
|
(6)
|
(7)
|
Lpxr (lag)
|
-7.049***
|
-6.099***
|
-8.684**
|
-8.859***
|
-11.217***
|
-12.377***
|
-12.094***
|
|
(2.613)
|
(2.166)
|
(3.779)
|
(2.898)
|
(3.595)
|
(3.923)
|
(4.553)
|
TCI
|
-11.169**
|
-20.527**
|
-22.822**
|
-19.912**
|
-25.182**
|
-27.739**
|
-30.450**
|
|
(4.410)
|
(8.338)
|
(11.142)
|
(8.499)
|
(10.304)
|
(11.465)
|
(14.255)
|
TCI2
|
|
2.986**
|
3.197*
|
2.714*
|
3.417**
|
3.660**
|
4.072*
|
|
|
(1.431)
|
(1.732)
|
(1.385)
|
(1.654)
|
(1.798)
|
(2.194)
|
Institution
|
|
|
0.706**
|
0.597**
|
0.705**
|
0.739**
|
0.787**
|
|
|
|
(0.331)
|
(0.242)
|
(0.294)
|
(0.310)
|
(0.349)
|
Population growth (annual %)
|
|
|
|
-2.128
|
-2.386*
|
-1.632
|
-1.708
|
|
|
|
|
(1.369)
|
(1.431)
|
(1.083)
|
(1.204)
|
FDI
|
|
|
|
|
0.121
|
0.025
|
0.006
|
|
|
|
|
|
(0.132)
|
(0.054)
|
(0.064)
|
Trade Openness
|
|
|
|
|
|
6.994**
|
5.536*
|
|
|
|
|
|
|
(2.908)
|
(2.852)
|
Arable land (% of land area)
|
|
|
|
|
|
|
-0.225
|
|
|
|
|
|
|
|
(0.146)
|
Constant
|
88.648***
|
84.916***
|
108.780**
|
111.059***
|
139.380***
|
148.205***
|
153.050***
|
|
(31.676)
|
(29.344)
|
(47.100)
|
(36.117)
|
(44.326)
|
(46.821)
|
(57.941)
|
Observations
|
1,534
|
1,534
|
1,435
|
1,433
|
1,415
|
1,376
|
1,366
|
Number of countries
|
102
|
102
|
97
|
97
|
95
|
93
|
93
|
Year FE
|
Yes
|
Yes
|
Yes
|
Yes
|
Yes
|
Yes
|
Yes
|
AR(1)
|
0.000
|
0.000
|
0.000
|
0.000
|
0.000
|
0.000
|
0.000
|
AR(2)
|
0.190
|
0.249
|
0.555
|
0.792
|
0.947
|
0.941
|
0.856
|
Hansen
|
0.040
|
0.015
|
0.054
|
0.087
|
0.171
|
0.273
|
0.079
|
Number of Instruments
|
33.000
|
34.000
|
35.000
|
36.000
|
37.000
|
38.000
|
39.000
|
Note: The variables lprx (lag), TCI are presented in log. Robust standard errors in parentheses. Significance level at *** p<0.01, ** p<0.05, * p<0.1. Source: Author’s calculation |
Table 5
GMM-system estimation of TCI effect on labour productivity growth 2012-2017
VARIABLES
|
Dependent variable: labour productivity growth rate (%)
|
(1)
|
(2)
|
(3)
|
(4)
|
(5)
|
(6)
|
(7)
|
Lpxr (lag)
|
-2.559*
|
-2.287*
|
-2.173
|
-1.999**
|
-1.769*
|
-1.858**
|
-2.000**
|
|
(1.304)
|
(1.162)
|
(1.327)
|
(0.992)
|
(0.906)
|
(0.931)
|
(0.973)
|
TCI
|
-2.378
|
8.417**
|
13.434***
|
13.114***
|
14.249***
|
14.353***
|
16.027***
|
|
(2.052)
|
(4.064)
|
(5.070)
|
(4.390)
|
(4.841)
|
(4.945)
|
(5.058)
|
TCI2
|
|
-4.140***
|
-5.663***
|
-5.158***
|
-5.463***
|
-5.558***
|
-6.188***
|
|
|
(1.428)
|
(1.909)
|
(1.488)
|
(1.629)
|
(1.673)
|
(1.815)
|
Institution
|
|
|
0.180
|
0.070
|
0.043
|
0.052
|
0.074
|
|
|
|
(0.130)
|
(0.077)
|
(0.069)
|
(0.070)
|
(0.074)
|
Population growth (annual %)
|
|
|
|
-1.071***
|
-1.038***
|
-0.994***
|
-1.033***
|
|
|
|
|
(0.387)
|
(0.373)
|
(0.337)
|
(0.333)
|
FDI
|
|
|
|
|
0.042*
|
0.034*
|
0.033**
|
|
|
|
|
|
(0.025)
|
(0.017)
|
(0.015)
|
Trade Openness
|
|
|
|
|
|
0.836
|
1.019
|
|
|
|
|
|
|
(0.698)
|
(0.734)
|
Arable land (% of land area)
|
|
|
|
|
|
|
-0.027
|
|
|
|
|
|
|
|
(0.035)
|
Constant
|
30.628*
|
21.636
|
15.920
|
15.456
|
12.132
|
12.329
|
12.703
|
|
(15.510)
|
(13.712)
|
(14.048)
|
(11.309)
|
(10.587)
|
(10.366)
|
(11.410)
|
Observations
|
612
|
612
|
582
|
582
|
582
|
581
|
485
|
Number of countries
|
102
|
102
|
97
|
97
|
97
|
97
|
97
|
Year FE
|
Yes
|
Yes
|
Yes
|
Yes
|
Yes
|
Yes
|
Yes
|
AR(1)
|
0.000
|
0.000
|
0.000
|
0.000
|
0.000
|
0.000
|
0.000
|
AR(2)
|
0.189
|
0.185
|
0.227
|
0.323
|
0.327
|
0.390
|
0.440
|
Hansen
|
0.495
|
0.499
|
0.480
|
0.488
|
0.315
|
0.310
|
0.413
|
Number of Instruments
|
33.000
|
34.000
|
35.000
|
36.000
|
37.000
|
38.000
|
39.000
|
Note: The variables lprx (lag), TCI are presented in log. Robust standard errors in parentheses. Significance level at *** p<0.01, ** p<0.05, * p<0.1. Source: Author’s calculation
The same regressions are run to test the effect of TCI during the transition period 2007 – 2012. The estimation results of the dynamic model for the period 2007 – 2012 are reported in appendix (Table A 2). There is no significance evidence of the impact of TCI on labour productivity growth. During this period, labour productivity growth is only affected by the initial value of labour productivity level (lag).
The results for the period 2012 – 2017, representing the post-crisis period or post-GFC, are reported in Table 5. TCI shows the expected sign, but it is not significant. The squared TCI is negative and significant from model (2) to (7). Unlike the results of TCI2 in Table 3, the nonlinear relationship during this period implies that defying comparative advantage positively contribute to labour productivity growth during this period 2012 – 2017, and beyond a certain point the gain decreases and becomes negative. This finding is in line with Chang’s argument of the ‘inverted-U-shaped’ relationship between deviating from comparative advantage and growth rate (Lin and Chang, 2009).
The regression results also show that the initial level of labour productivity have the expected sign and significant effect, FDI plays a positive and significant role in increasing labour productivity growth during this period. Unlike during the pre-GFC, the positive and significance effect of FDI during the post-GFC could be explained by developing countries having reached an acceptable level of absorptive capacities, or other characteristics such openness to trade as suggested by Balasubramanyam et al. (1996) and institutional quality having set up a low level of corruption and property rights. However, institutional quality and trade openness have no significance effects on labour productivity growth during the period of 2012 -2017. As mentioned, and evidenced earlier, almost 2 decades may have been a long and good period to have a certain level beyond which the marginal effect of these two variables may be small or insignificant.
To go further, I augment the dynamic model with an interaction term of TCI (log) and the regions dummy, and the results are shown in appendix (Table A 3). The relationship between TCI (log) and labour productivity growth in seven (7) different regions was tested, through the use of separate sets of dummy and interactions term for East Asia and Pacific (EAP), Europe and Central Asia (ECA), Latin America and Caribbean (LAC), Middle East and North Africa (MNA), North America (NAM), South Asia (SAR), and Sub-Saharan Africa (SSA).
The results in column 1 show that the interaction term between TCI (log) and EAP, LAC, SAR, SSA is negative and significant whereas it is negative but not significant in MNA. However, the interaction term between TCI (log) and ECA, NAM is positive and significant. In further attempt to explore the validity of the non-linear effect of TCI, I added the squared value of TCI in model 2. The results in column 2 show positive and negative significant coefficients of TCI and the squared TCI, for EAP, ECA, and NAM, indicating in fact that the effect is not linear for countries in those regions. To show this, I graph the predictive marginal effects of the changes in TCI for the results in column 2, in Figure 2. It confirms the linear effects in SSA, SAR, but a nonlinear and non-significant effect in MNA and LAC. However, it shows that slightly challenging the factor endowment increases labour productivity growth in EAP, ECA and NAM. Nevertheless, above a certain threshold of TCI (log), productivity growth tends to decrease. The turning point beyond which it has a negative impact on productivity growth in those regions are .73, .98 and .78 respectively in EAP, EAC and NAM. A better understand of the results above would require an income level analysis, as regions are heterogenous with countries of different level of development.
The relationship between the TCI and labour productivity may be heterogeneous with respect to the country’s economic development level. Therefore, I rerun the regression on four income groups by adding an interaction term of TCI and income dummy. The results of the relationship between TCI (log) and labour productivity growth in four (4) different income groups, through the use of separate sets of dummy and interactions term for High income, Upper middle income, Lower middle income, and low-income countries are shown in appendix (Table A 4).
The results in column 1 show that the interaction term between TCI (log) and high-income countries is positive and significant similar to the findings of Bruno et al. (2015), while it is not significant in upper middle income. The interaction term between TCI (log) and lower middle- and low-income countries is negative and significant. To test whether the non-linearity of effect holds or differs from income group, I include an interaction term between the squared value of TCI and income dummies in model 2.
The results in column 2 show positive and negative significant coefficients of the interaction term between TCI and the squared TCI, for high income and upper middle-income group, indicating that challenging the comparative advantage in high and upper middle-income groups increases the average level of labour productivity. Nevertheless, above TCI (log) value of 1.02, labour productivity growth during the period 1990 – 2017 tends to decrease in high income countries, while above a TCI (log) value of 1.03 labour productivity tends to decrease in upper middle-income countries.
As for lower middle- and low-income countries, the results reported in Table A 4 show that the coefficients are both negative and significant for TCI. There is no evidence of a nonlinear relationship between TCI and labour productivity growth in both lower middle- and low-income countries. This finding implies that defying comparative advantage might not be sustainable for countries in these income groups, as illustrated in Figure 3.