Institutions Rule in Export Diversity

Abstract

We used the data of 133 countries to explore the determinants of export diversity among candidate variables that are clearly exogenous or endogenous but with a well-established instrument. These candidates include geography, resource abundance, and institutional quality. Our results suggest that institutional quality is important in determining export diversity, and a country with better institutions has a more diversified export structure. Once institutional quality is controlled for and instrumented with European settler mortality, variables such as geography and resource abundance, which significantly determine export diversity in the baseline regression, are no longer significant. The results are still robust even using alternative measures of institutional quality. They also indicate that the mechanism whereby institutions largely determine export diversity has been presented under the guise of natural resource abundance and geography. Therefore, to some extent, the “natural resource curse” is not a curse, but simply two sides of the same coin.

1. Introduction

An economy with greater export diversification is more resistant to exogenous international shocks and has better economic growth performance. Understanding what determines the export diversification of a country is vital for long-term economic development. There are many studies that explore the relationship between export diversity and per capita income. After Imbs and Wacziarg [1] discovered the U-shaped pattern between sectoral concentration and per capita income, Klinger and Lederman [2] found a similar U-shaped pattern between export concentration and per capita income, and Cadot et al. [3] also found a hump shape of export diversification along the diversification path. However, income is an endogenous variable. The question, then, is what fundamentally shapes export structure? Resource abundance automatically qualifies as a candidate because of its importance for some countries’ exports (for example, oil-rich countries), and major determinants of income, such as institutional quality and geography, should also be considered, as they can shift economic structure and change the way an economy is involved in international trade. In this paper, we intend to examine the respective contributions of variables such as resource abundance, institutional quality, and geographic characteristics in determining export diversity. Following Rodrik et al. [4], we only target variables that are clearly exogenous, such as geography and resource abundance, or endogenous but with a well-established instrument, such as institutional quality with European settler mortality as an instrument [5]. Such a design helps us address endogeneity problems.

According to institutional quality, Acemoglu et al. [5] (hereafter AJR) proposed a theory of institutional differences among countries colonized by Europeans, where different types of colonization policies determine the different sets of institutions in these colonized countries. They argued that the potential settler mortality influenced the colonization strategy, and the early institutions brought by settlers persist to the present in the ex-colonies even after their independence, which accounts for a large part of variations in income per capita. Colonies with higher potential settler mortality were associated with the formation of a more extractive state and worse institutional quality, lacking protection for private property or checks and balances against government expropriation. Based on the theory, we propose that the export diversity of a country was influenced by its early institutions shaping its early patterns of economic development and trade. Since the main purpose of the extractive state was transferring as much of the recourses from the colony to the colonizer, a country with worse early institutions faced more difficulty in developing competitive industries and was more likely to form a trade pattern of exporting its natural resources regardless of its resource endowment and thus less diversified exports. Present institutional quality also has a positive effect on export diversity as a result of institution persistence.

To test the proposed relationship between institutions and export diversity, and to determine the contributions of the aforementioned variables, we used the data of 133 countries. Specifically, export diversity is measured by the 2018 export diversification index of each country, while institutions and other key variables are averaged over the sample period from 2002 to 2017 to address the endogeneity problem and instrument institutional quality with European settler mortality. When we use European settler mortality as the instrument of institutional quality, the sample is restricted to the same “base sample” as in AJR’s work, consisting of 64 countries with European settler mortality data. In the benchmark results, we found that the coefficients of institutional quality, resource abundance, and geography (the latitude of a country’s capital) are significant. Specifically, the improvement in institutional quality has a negative effect on the export diversification index, for which a lower value corresponds to more diversified exports. This confirms that a country with better institutions has a more diversified export structure. Resource abundance has a positive effect on the export diversification index, indicating that the exports of resource-abundant countries are less diversified, while geography has an opposite effect.

To test the robustness of our baseline results, we used different measures for institutional quality and export diversification, and the results are quite similar to our benchmark results. Finally, in consideration of the endogeneity of institutional quality, we instrumented it with European settler mortality and found that the coefficients of institutional quality are robustly significant, but geography and natural resource abundance are no longer significant in determining export diversity. Our results show that institutional quality “trumps” all other variables in determining export diversity, just as it “rules” in determining income in Rodrik et al. [4] (2004). To enhance export diversity, it is crucial for a country to improve its institutions for more secure property rights and less distortionary policies.

2. Literature Review

Our paper is closely related to the literature on the determinants of export diversification. Recent studies have found that the degree of export diversification is mainly driven by the following factors. (i) Economic development and size of the economy [6,7,8,9,10,11,12,13,14]. Empirical findings have shown that the relationships between economic development and the level of export diversity have both monotonic and non-monotonic results. In addition to the effect of economic development on export diversification, Frenken et al. [15] and Saviotti and Frenken [16] also indicate that there is bi-directional causality between them. Similarly, there is no consensus on the effect of country size on export diversity. Larger economies demand more diversified products and are more likely to cause a greater diversification of production and exports. On the contrary, the scale effect would also imply more demand for conventional goods due to common cultural backgrounds and result in greater export concentration. (ii) Human capital [17,18,19,20,21]. Empirical studies have found a significant and negative relationship between human capital and the concentration of exports. (iii) Trade liberalization and trade openness [12,22,23,24,25]. Some studies have shown that higher trade openness results in more specialization in products with comparative advantages and less diversified exports, while other studies have shown that trade openness increases the number of trade patterns and trade opportunities, leading to more diversified exports. (iv) Geography [11,13,26,27]. Being far away from major markets has negative effects on the process of diversification. (v) Institutions [28,29]. Omgba [28] posited that the longer period between the beginning of oil production and political independence is, the higher its export diversification is, which suggests that better institutions, developed after independence, help diversification. Sheng and Yang [29] employed China’s data for the period of 1997–2007 and found that better institutions with greater ownership liberalization and judicial quality played an important role in improving China’s export diversity. (vi) Others. Some researchers [23,30,31,32] argue that the appreciation of the real exchange rate significantly constrains export diversification process, while real exchange depreciation promotes export diversity. According to natural resource abundance, many studies [33,34,35] assert that being rich in natural resources leads to greater concentration on primary commodities and lower export diversification. However, in this study, we observed a strikingly different result—when institutional quality is controlled and instrumented, natural resource abundance no longer makes a difference in export diversity. Our paper contributes to the literature by enriching the discussion of the role of institutional quality in determining export diversity and alleviating its endogeneity bias with European settler mortality as the instrument variable.

Research on the relationship between institutional quality and international trade performance is a rapidly growing area of study. Many scholars have found evidence that institutional quality is an important determinant of trade patterns [36,37,38,39,40]. Specifically, Francois and Manchin [38] declared that institutional quality not only makes a difference in exports, but also the propensity to export. However, recent literature has mostly focused on the influence of institutional quality on trade comparative advantages, trade patterns, and export product quality. There is little direct research on the relationship between institutional quality and export diversity. For example, some studies have proposed that institutional quality would affect the quality of contract implementation, and ultimately trade comparative advantages and trade patterns [41,42,43]. Similarly, Nunn [44] constructed a measure of contract intensity of industries and found that higher institutional quality with a better contracting environment is a source of comparative advantage and promotes more exports in industries with higher contract intensity. Another related research stream pays attention to the national level to explore the relationship between institutional quality and export product quality. Berkowitz et al. [45] stated that higher institutional quality enables a country to export more complex products and import simpler goods, while Ara [46] found a similar relationship between institutional quality and exports of institutional quality-intensive products. Lin et al. [47] used regional-level data from China and found that the improvement of institutional quality leads to higher quality of export products. Our paper contributes to the literature by directly probing the influence of institutional quality on export diversity, which has been left unexplored in these major studies.

This paper also contributes to the discussion of export diversification’s impacts on growth, which relates to the “natural resource curse”. There is extensive literature on the natural resource curse that we cannot completely cover here (see, for example, Sachs and Warner [48,49], Auty [50,51], Gylfason [52], Brunnschweiler and Bulte [53], Alexeev and Conrad [54], Frankel [55], Badeed et al. [56], etc.). Following Sachs and Warner [48], the natural resource curse is the negative correlation between natural resource wealth and growth rates for economies. Usually, resource wealth is described by direct measures of natural resources, such as the export of primary products (as a share of GDP) or rents from natural resources over GDP, but export diversity would also qualify [57]. Most studies suggest that export diversification is positive for growth [58,59,60,61,62,63,64], while other studies suggest a negative correlation [65,66,67,68]. Aditya and Acharyya [69] argue that there exists a critical level of export concentration, where concentration below this level hurts growth and concentration above it helps growth. According to our results and combined with AJR’s results, insufficient institutions lead to both export concentration and low income. Therefore, consistent with results of Bonaglia and Fikasaku [57], to some extent, the natural resource curse is not a curse, but simply two sides of the same coin. More specifically, institutions largely determine export diversity, but this mechanism has been presented under the guise of natural resource abundance.

The rest of the paper is organized as follows. Section 2 describes the research design and data. Section 3 presents and discusses the baseline results. Section 4 provides the robustness check of our baseline results, using alternative measures for our key variables and the instrumental variable estimation, and Section 5 concludes.

3. Materials and Methods

3.1. OLS Estimation

Figure 1 presents the clear negative correlation between the export diversification index in 2018 (lower value corresponding to more diversified exports) and our institutional quality indicator, the mean IQ variable (higher value corresponding to better institutional quality) for a sample of 133 countries from 2002 to 2017. It indicates that better institutional quality is associated with more diversified exports.

In our benchmark regressions, we employed the ordinary least squares (OLS) method to identify the impact of institutional quality on export diversity, using the following regression equation:

$$Y_i={\alpha }_0+\beta \times R_i+Z_i^{\prime }\times \gamma +{ϵ}_i$$

(1)

The dependent variable is the export diversification index in 2018, and the key independent variable ($R_i$) is institutional quality (IQ) averaged over the sample period from 2002 to 2017. $Z_i$ is a set of control variables that include geographic characteristics, resource abundance, and other variables, which we specify later, and ${ϵ}_i$ is the error term.

3.2. Variables and Data Sources

3.2.1. Export Diversity

We measure export diversity using the export diversification index (ediv) from the UNCTAD STAT database, which measures the absolute deviation of a country’s export structure from world structure. It takes values between 0 and 1, and a larger value indicates a country’s greater divergence from the world export pattern and lower export diversification. We also use two alternative sets of measures from the UNCTAD STAT database for export diversification. The first one is the export concentration index (ecie), which is a Herfindahl–Hirschman index normalized to be between 0 and 1 following SITC (Rev. 3). The larger the index is, the more concentrated and less diversified a country’s exports are. The second measure is the number of products exported (enpe) at the three SITC (Rev. 3).

3.2.2. Institutional Quality

Regarding data on institutional quality, we used the IQ variable (iq) following Frankel et al. (2014), which is computed by taking the average of four normalized variables (investment profile, corruption, law and order, and bureaucratic quality) from the International Country Risk Guide (ICRG) data. It reports a value for each country from 1 to 7, with 7 corresponding to the highest institutional quality. We also used two other sets of indicators for institutional quality, with a larger value corresponding to higher institutional quality. The first is to directly use the Corruption Index (corruption) from ICRG, as in Alesina et al. (2008). The second data source is the Worldwide Governance Indicators (WGI) from Kaufmann et al. (2009). The WGI come with 6 different indices: Voice and Accountability, Control of Corruption, Government Effectiveness, Political Stability and Absence of Violence/Terrorism, Rule of Law, and Regulatory Quality. In this paper, we mainly use the Rule of Law variable (g_rl).

3.2.3. Others

Referring to the literature, we controlled several variables of the countries in the regression, including population density (population_density), trade as a share of GDP (trade), and real GDP per capita (lgdp_pc) based on purchasing power parity (PPP), which is measured in constant USD. All these variables are from World Development Indicators. Another control variable is the natural resource endowment dummy variable (resource from Auty (2001), with 1 corresponding to resource-abundant countries. The alternative control variables are geographic variables, including latitude of capital scaled between 0 and 1 (lat_abst) and a set of continental dummy variables (Africa, Asia and other), with America as the omitted group.

We performed unit root tests for the time-varying variables, and the results suggest that they are all stationary. To address the endogeneity problem and instrument institutional quality with European settler mortality, we averaged institutional quality and other control variables over the period from 2002 to 2017. The specific definitions and descriptive statistics for the above variables are presented in

Table 1. Descriptive statistics.

Variable Name	Definition	Data Source	Obs	Mean	SD	Min	Max
ediv	Export diversification index	the UNCTAD STAT database	133	0.656	0.172	0.238	0.927
ecie	Export concentration index	the UNCTAD STAT database	133	0.316	0.214	0.053	0.981
enpe	Number of products exported	the UNCTAD STAT database	133	196.489	59.753	10	260
iq	Average of four normalized variables (investment profile, corruption, law and order, and bureaucratic quality)	ICRG database	133	4.241	1.241	1.027	6.809
corruption	Corruption Index	ICRG database	133	2.585	1.122	0.419	5.823
g_rl	Rule of Law Index	WGI database	133	−0.009	1.027	−2.323	1.988
lat_abst	Latitude of capital scaled between 0 and 1	Acemoglu et al. (2001) [5]	130	0.300	0.197	0.000	0.722
population_density	People per sq. km of land area	WDI database	133	230.833	849.652	1.757	7031.250
trade	Trade as a share of GDP	WDI database	132	67.733	42.801	20.179	343.512
lgdp_pc	log PPP GDP per capita	WDI database	130	9.383	1.186	6.784	11.443
resource	Natural resource endowments dummy	Auty (2001) [50]	71	0.775	0.421	0	1
africa	Africa dummy	Acemoglu et al. (2001) [5]	130	0.285	0.453	0	1
asia	Asia dummy	Acemoglu et al. (2001) [5]	130	0.246	0.432	0	1
other	Other continental dummy (excluding America)	Acemoglu et al. (2001) [5]	130	0.023	0.151	0	1
logem4	log European settler mortality	Acemoglu et al. (2001) [5]	74	4.567	1.342	0.936	7.986

Note: Definitions and sources for these variables are described in Section 2.

.

We present the correlation coefficients of these variables in

Table 2. Correlation matrix.

Variables	ediv	ecie	enpe	iq	Corruption	g_rl	lat_abst	Population_Density
ediv	1.000
ecie	0.733 *	1.000
enpe	−0.697 *	−0.642 *	1.000
iq	−0.610 *	−0.429 *	0.570 *	1.000
corruption	−0.511 *	−0.369 *	0.444 *	0.901 *	1.000
g_rl	−0.618 *	−0.453 *	0.560 *	0.962 *	0.907 *	1.000
lat_abst	−0.556 *	−0.396 *	0.471 *	0.612 *	0.537 *	0.610 *	1.000
population_density	−0.128	−0.063	0.136	0.222 *	0.201 *	0.227 *	−0.095	1.000
trade	−0.252 *	−0.104	0.225 *	0.282 *	0.188 *	0.292 *	0.105	0.700 *
lgdp_pc	−0.589 *	−0.319 *	0.628 *	0.803 *	0.650 *	0.778 *	0.624 *	0.201 *
resource	0.381 *	0.272 *	−0.252 *	−0.105	−0.189	−0.135	−0.028	−0.372 *
africa	0.470 *	0.388 *	−0.527 *	−0.444 *	−0.360 *	−0.438 *	−0.468 *	−0.128
asia	−0.018	0.046	0.151 *	−0.002	−0.118	−0.043	−0.070	0.278 *
other	0.018	−0.042	0.064	0.248 *	0.264 *	0.253 *	0.065	0.038
logem4	0.543 *	0.426 *	−0.469 *	−0.633 *	−0.582 *	−0.632 *	−0.535 *	−0.244 *
Variables	trade	lgdp_pc	resource_abundence	africa	asia	other	logem4
ediv
ecie
enpe
iq
corruption
g_rl
lat_abst
population_density
trade	1.000
lgdp_pc	0.344 *	1.000
resource	−0.211 *	−0.111	1.000
africa	−0.230 *	−0.688 *	0.218 *	1.000
asia	0.206 *	0.129	−0.408 *	−0.360 *	1.000
other	−0.041	0.155 *	- 1	−0.097	−0.088	1.000
logem4	−0.217 *	−0.696 *	0.197	0.587 *	−0.162	−0.340 *	1.000

Note: * Significance at p < 0.1. ¹ There is no correlation between resource and other dummy variable because all countries that have access to the data of the resource variable come from either Africa, Asia, or America.

, showing that the different measures of export diversity are highly correlated. Similarly, we see that there are highly positive correlations among different measures of institutional quality. Examining the correlations between institutional quality and export diversity, we find that even measured in different variables, they are all highly correlated. The results are consistent with the implication of Figure 1, that better institutional quality is correlated with more diversified exports. There are highly negative correlations between the European settler mortality and different measures of institutional quality, suggesting that higher European settler mortality is associated with worse institutional quality.

4. Results

Table 3. Determinants of export diversification index.

	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)
	Full Sample	Base Sample	Full Sample	Base Sample	Full Sample	Base Sample	Full Sample	Base Sample	Full Sample	Base Sample
iq	−0.0845 ***	−0.0732 ***	−0.0614 ***	−0.0693 ***	−0.0514 ***	−0.0722 ***	−0.0619 ***	−0.0712 ***	−0.0468 ***	−0.0420 *
	(−8.5717)	(−4.2632)	(−5.1648)	(−3.9296)	(−3.7511)	(−4.0207)	(−5.2454)	(−3.4265)	(−2.8446)	(−1.6862)
lat_abst			−0.2520 ***	−0.0615	−0.1205	−0.0283	−0.1628 **	−0.1249	−0.2183 **	−0.0568
			(−3.3830)	(−0.4814)	(−0.9988)	(−0.1988)	(−2.0443)	(−1.0052)	(−2.4827)	(−0.4367)
resource					0.1093 ***	0.0658 *
					(3.5979)	(1.9417)
africa							0.0868 ***	0.0351
							(3.0978)	(1.2221)
asia							0.0259	−0.0618
							(0.8286)	(−1.1106)
other							0.1937 ***	0.1690 **
							(3.7583)	(2.1441)
population _density									3.97 × 10−6	5.87 × 10−7
									(0.2190)	(0.0221)
trade									−0.0004	−0.0001
									(−0.8931)	(−0.1657)
lgdp_pc									−0.0226	−0.0367 **
									(−1.4148)	(−2.0690)
_cons	1.0142 ***	0.9982 ***	0.9911 ***	0.9940 ***	0.8487 ***	0.9438 ***	0.9306 ***	0.9990 ***	1.1563 ***	1.2213 ***
	(26.2851)	(16.2974)	(25.7031)	(16.4790)	(15.0963)	(12.1648)	(20.5357)	(14.6228)	(10.9721)	(9.1830)
N	133	64	130	64	71	54	130	64	126	63
r2	0.373	0.329	0.430	0.331	0.299	0.339	0.484	0.440	0.443	0.372
Mean VIF	1.00	1.00	1.55	1.36	1.06	1.06	1.52	1.43	2.32	2.59

Note: The dependent variable is export diversification index (higher value corresponding to less diversified exports). In regressions with continent dummies, America is the omitted group. Columns (1) and (2) have no control variable, while Columns (3)–(10) control for different sets of variables. Columns “Full Sample” and “Base Sample” show the results of regressions in the full sample and base sample, respectively. Robust t statistics in parentheses, *** p < 0.01, ** p < 0.05, * p < 0.1.

reports ordinary least squares estimation results of export diversity and institutional quality (Equation (1)). Column (1) suggests that the correlation between export diversification index and institutional quality is strong and statistically significant in the whole sample, and better institutional quality is correlated with more diversified exports. Column (2) confirms that this is also the case for the same correlation in the base sample composed of 64 countries, which is the same as the “base sample” in AJR’s work. These countries were ex-colonies for which we have settler mortality data. Following AJR, we also added normalized latitude as a control in Column (3) and dummies for Africa, Asia, and other continents in Column (7), with America being the omitted group. We can see that both latitude and being in Africa or other continents play a role in determining export diversification, where latitude significantly improves the export diversity of a country, but the latter makes a country’s exports less diversified. We repeated the same regression but in the base sample and report the results in Columns (4) and (8). The results suggest that the other continents dummy still stands out as being less diversified in exports; however, the Africa dummy and latitude variable are no longer significant. We also include resource abundance because it may directly change export structure, and the results for the whole sample and the base sample are listed in Columns (5) and (6). These variables constitute the “horse race” between potential determinants of export diversity using OLS, where IQ and the newly added resource abundance are statistically significant. Resource abundance is positively correlated with the export diversification index, while IQ and normalized latitude are negatively correlated. Lastly, we added population density, trade openness, and log PPP GDP per capita for the whole sample in Column (9) and for the base sample in Column (10). In the whole sample, all the newly added variables are insignificant, while in the base sample, the coefficient of log GDP per capita is significantly negative. Normalized latitude has a significantly negative correlation with export diversification index in the whole sample, which disappears in the base sample. The results in Table 2 show that regardless of the control variables, the coefficient of our target variable IQ is always negative and statistically significant, indicating that better institutional quality makes the exports of a country more diversified.

5. Robustness Check

5.1. Alternative Measures

We used different indicators of institutional quality and alternative measures of export diversity for the robustness check. As previously mentioned, we used two different measures of institutional quality, Corruption and Rule of Law, and two different measures of export diversity, the export diversification index and the number of products exported. In

Table 4. Robustness check (alternative measures of both institutional quality and export diversity).

	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
	ediv				ecie		enpe
	Full Sample	Base Sample	Full Sample	Base Sample	Full Sample	Base Sample	Full Sample	Base Sample
iq					−0.0480 ***	−0.0584 ***	20.1571 ***	15.9053 *
					(−2.9527)	(−2.6944)	(4.7484)	(1.9976)
corruption	−0.0483 ***	−0.0524 **
	(−3.4152)	(−2.1128)
g_rl			−0.0767 ***	−0.0882 ***
			(−5.3491)	(−3.3998)
lat_abst	−0.2437 ***	−0.2183	−0.1626 **	−0.1302	−0.0965	−0.2137	26.6173	31.6559
	(−3.1692)	(−1.5290)	(−2.1074)	(−1.0853)	(−0.8715)	(−1.4110)	(0.8642)	(0.6845)
africa	0.0972 ***	0.0402	0.0827 ***	0.0337	0.1359 **	0.0556	−38.7019 ***	−24.0556
	(3.3176)	(1.3271)	(2.9243)	(1.1809)	(2.4359)	(1.1285)	(−2.7862)	(−1.5759)
asia	0.0120	−0.0818	0.0168	−0.0617	0.0744	−0.0636	6.5071	37.4244 ***
	(0.3705)	(−1.3085)	(0.5476)	(−1.0908)	(1.6507)	(−1.3518)	(0.6255)	(2.8809)
other	0.1682 ***	0.1294	0.1963 ***	0.1780 **	0.1045 ***	0.1292 **	−27.8062 *	−14.6158
	(2.8886)	(1.5214)	(4.4076)	(2.3455)	(2.7852)	(2.4964)	(−1.8395)	(−0.6198)
_cons	0.8188 ***	0.8664 ***	0.6712 ***	0.6965 ***	0.4900 ***	0.5754 ***	113.5455 ***	124.9430 ***
	(19.3665)	(15.9913)	(20.5949)	(18.9718)	(6.5070)	(6.6845)	(5.7757)	(4.2314)
N	130	64	130	64	130	64	130	64
r2	0.436	0.340	0.490	0.438	0.271	0.273	0.439	0.339
Mean VIF	1.43	1.38	1.53	1.45	1.52	1.43	1.52	1.43

Note: Dependent variables: Columns (1)–(4), export diversification index (higher value corresponding to less diversified exports); Columns (5) and (6), export concentration index (larger value indicating more concentrated exports); Columns (7) and (8), number of products exported. All these regressions control for the same set of variables: latitude of capital and continental dummy variables (for continental dummies, America is the omitted group), but with different measures of institutional quality. Columns (1) and (2), (3) and (4), and (5)–(8) measure institutional quality using Corruption, Rule of Law, and IQ, respectively. Columns “Full Sample” and “Base Sample” show the results of regressions in the full sample and base sample, respectively. Robust t statistics in parentheses, *** p < 0.01, ** p < 0.05, * p < 0.1.

, we present the results for the multivariate regression, controlling normalized latitude and continental dummies. Columns (1)–(4) present the regression results for “Corruption” and “Rule of Law” in the full sample and the base sample, respectively. The patterns are similar: the coefficients for institutional quality are negative and statistically significant, consistent with our benchmark case. The coefficients for latitude are negative and statistically significant in the whole sample, but insignificant in the base sample. The performance of continental dummies is quite consistent with Columns (7) and (8) in Table 3, except the other continents dummy is insignificant in the base sample using Corruption as the measure of institutional quality. Then, we repeated the same procedures for “IQ” but used the export concentration index and the number of products exported to measure export diversity, and present the results in Columns (5)–(8). The coefficients for IQ are statistically significant and suggest that better institutional quality makes exports more diversified. The coefficients for latitude are no longer statistically significant in these regressions, while continental dummies still play a role. These results suggest that our baseline results are still robust even using alternative measures for our key variables.

5.2. Instrumental Variable Estimation

As discussed in Section 1, we intended to implement a “horse race” between potential determinants of export diversity. These candidate variables include geography, resource abundance, and institutional quality. Geography and resource abundance are naturally exogenous, but institutional quality is highly endogenous. AJR argue that European settler mortality is a suitable instrument for institutional quality. We followed their strategy and used the two-stage least squares (2SLS) method to examine institutional quality’s influence on export diversity.

In Figure 2, we present the reduced-form correlation between institutional quality and European settler mortality. It suggests that there is a clear negative correlation between European settler mortality and our measure of institutional quality, indicating that ex-colonies with higher European settler mortality have worse institutions today. We start by checking the IV results without control variables and present the results in Column (1) in

Table 5. IV regressions of export diversification index.

	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)
	IQ			Corruption			Rule of Law
Panel A: Two-Stage Least Squares
iq	−0.1216 ***	−0.1274 **	−0.1447 ***
	(−5.4752)	(−2.5388)	(−2.8729)
corruption				−0.1540 ***	−0.2301 *	−0.2138 **
				(−4.6956)	(−1.7882)	(−2.2098)
g_rl							−0.1477 ***	−0.1792 **	−0.1882 ***
							(−5.5523)	(−2.3928)	(−2.8412)
lat_abst		−0.0552	−0.0207		−0.0597	0.0680		−0.0196	0.0095
		(−0.3355)	(−0.1002)		(−0.2572)	(0.2239)		(−0.1068)	(0.0436)
resource		0.0657			0.0200			0.0478
		(1.5512)			(0.2657)			(0.9779)
africa			−0.0032			−0.0256			−0.0146
			(−0.0805)			(−0.4432)			(−0.3436)
asia			−0.0732 *			−0.1038 *			−0.0705
			(−1.7277)			(−1.9319)			(−1.6385)
other			0.3214 ***			0.4060 **			0.3427 ***
			(2.8452)			(2.3010)			(2.8621)
_cons	1.1687 ***	1.1397 ***	1.2649 ***	1.0636 ***	1.2115 ***	1.2071 ***	0.6543 ***	0.6025 ***	0.6466 ***
	(13.2797)	(6.0317)	(7.4825)	(13.1823)	(3.9386)	(6.2151)	(38.4224)	(11.6553)	(10.5899)
N	74	60	74	74	60	74	74	60	74
Panel B: First Stage for Institutional Quality
logem4	−0.5282 ***	−0.2861 ***	−0.3131 ***	−0.4171 ***	−0.1584 *	−0.2119 **	−0.4347 ***	−0.2034 **	−0.2407 ***
	(−6.6285)	(−2.9790)	(−3.0081)	(−5.2940)	(−1.9888)	(−2.1932)	(−6.2813)	(−2.5049)	(−2.7183)
lat_abst		1.2666	2.4443 ***		0.6816	2.0691 **		1.0991	2.0395 ***
		(1.2798)	(2.7393)		(0.7692)	(2.5058)		(1.3006)	(2.8327)
resource		−0.2486			−0.3359			−0.2766
		(−0.8220)			(−1.4055)			(−1.1804)
asia			0.1838			−0.0190			0.1553
			(0.5555)			(−0.0695)			(0.5584)
africa			−0.0080			−0.1100			−0.0668
			(−0.0331)			(−0.5041)			(−0.3568)
other			1.2612 ***			1.2492 **			1.0827 ***
			(3.8562)			(2.4882)			(5.4374)
_cons	6.3134 ***	5.0380 ***	4.7783 ***	4.3019 ***	3.1010 ***	2.9632 ***	1.7131 ***	0.5833	0.3877
	(15.5089)	(7.4862)	(8.5687)	(10.7795)	(5.3341)	(5.8153)	(5.0132)	(1.0480)	(0.8107)
r2	0.401	0.241	0.526	0.339	0.160	0.483	0.399	0.224	0.531

Note: The dependent variable is export diversification index (higher value corresponding to less diversified exports). Columns (1)–(3), (4)–(6), and (7)–(9) measure institutional quality using IQ, Corruption, and Rule of Law, respectively. Columns (1), (4), and (7) have no control variable; Columns (2), (5), and (8) control for latitude of capital and resource abundance; Columns (3), (6), and (9) control for latitude of capital and continental dummy variables (for continental dummies, America is the omitted group). Panel A reports two-stage least-squares estimates, with institutional quality instrumented using log European settler mortality. Panel B reports the corresponding first-stage estimates. Robust t statistics in parentheses, *** p < 0.01, ** p < 0.05, * p < 0.1.

. The first stage result suggests that European settler mortality is highly correlated with institutional quality, and the second stage result confirms that institutional quality negatively changes the export diversification index. We further add normalized latitude and resource abundance as controls and implement our “horse race” in Column (2). The result is surprising in that the coefficients for latitude and resource abundance are completely insignificant. This forms noticeable comparisons from the OLS results using exactly the same variables (Columns (5) and (6) in Table 3). This suggests that institutional quality is much more important in determining export variety than resource abundance and geography. Furthermore, we add normalized latitude and continental dummies as controls in Column (3) and find the robust correlation between IQ and export diversification index. To check the robustness of these results, we used alternative institutional quality measures to repeat these empirical results. The results for “Corruption” are listed in Columns (4)–(6), and those for “Rule of Law” in Columns (7)–(9). The results are similar to the benchmark case and confirm our findings that institutional quality matters the most in determining export diversity among the pool of chosen candidates.

6. Conclusions

In this study, we investigated what determines the export diversity of a country. We restrained potential candidates to variables that are clearly exogenous, such as geographic characteristics and resource abundance, or endogenous but with well-established instruments (i.e., institutional quality). Our OLS results suggest that they are all significantly correlated with export diversity. However, once we properly instrumented institutional quality, the IV results show that institutions matter the most, while latitude and resource abundance do not affect export diversity at all. This surprising result suggests that institutions largely determine export diversity, but this mechanism has been presented under the guise of other variables in simple OLS. Our results enrich the discussion of the role of institutional quality in determining export diversity by alleviating the endogeneity bias, and supplement the discussion of the “natural resource curse”, indicating that to some extent, the natural resource curse is not a curse, but simply two sides of the same coin. To enhance export diversity, it is crucial for a country to improve its institutions for more secure property rights and less distortionary policies.

Export diversification is essential for economic development and productivity [62,63,70,71]. A more diversified export sector of a developing country will also spread macroeconomic risks and benefit its financial development, which in turn promotes the diversification and sustainable development of the economy [7,72]. Given the fact that China’s institutional quality is at a medium level based on the measures of institutional quality from the ICRG dataset in 2018, there is room for further improvement in its institutional quality. Our results suggest that to expand the types of products exported in China, which can ultimately promote its financial development and power its sustainable economic growth, it is key to improve institutional quality and provide a better contracting environment.

Although this work provides some empirical evidence that better institutional quality accounts for more diversified exports than natural resource abundance and geography, we ignore the dynamics of the relationship between institutional quality and export diversity. An interesting future extension is to consider the time-varying effects of institutional quality on export diversity and to explore how institutional differences influence export diversity. Another limitation of this study is that we emphasize the influence of institutional quality on export diversity measured by relevant indices, but pay little attention to its influence on the types of products exported. In the future, we can further explore how institutional quality affects the types of products exported at the national level, which may provide some policy implications for promoting relevant industries.