Does corruption cause encumbered business regulations? An IV approach

Abstract

In this research we attempt to understand the effects of corruption on business regulations measured by number of procedures, time and costs involved in satisfying governmental requisites. Using Two-Stage-Least Squares (2SLS) methods we lessen the problems associated with endogeneity and uncovered the existence of a robust, positive, statistically significant and quantitatively large direction channel from corruption to encumbered business regulations. These results suggest the need to deter corruption in order to improve business regulations. The quest to reduce corruption worldwide acquires new relevancy given the findings of Djankov et al. (2006) that countries with simple and transparent regulations grow faster.

Keywords:

• corruption
• business regulations
• instrumental variables

JEL Classification:

• D73
• K42
• O5

Notes

¹ The Doing Business database is available at www.doingbusiness.org.

² The procedure used was to subtract the sample minimum from each observation, divide this difference by the sample range and multiply the result by 10. All calculations are available upon request.

³ The TI corruption index used in this study is highly correlated with the World Bank's control of corruption index. For available years the correlation ranges from 0.95 to 0.98 (Treisman, 2007).

⁴ Other sources of endogeneity, not relevant for our specifications and samples, are auto-regression and unobserved heterogeneity.

⁵ A complete specification of equation in (1) is not required if the orthogonality assumption is satisfied. In this case any subsets of the determinants of the endogenous variables in the first stage are valid instruments for the consistent estimation of the population parameters in the structural equation (1).

⁶ We do not exploit the time dimension of the data applying panel data econometrics because the time series of corruption indices is relatively short. Additionally, TI alters the methods and sources over the years and therefore observed changes in the index may not correspond to changes of perceived corruption (Treisman, 2007).

⁷ The German Commercial Code was written after the unification of Germany by Bismarck in 1897.

⁸ Acemoglu and Johnson (2005) distinguish between ‘contracting institutions’ and ‘property rights institutions’. Contracting institutions regulate transactions between ordinary citizens such as contracts between creditors and debtors or protection of minority shareholders from expropriation by company insiders. Since contracting institutions deal primarily with relations between private individuals they could also be called horizontal institutions. Institutions constraining government and elite expropriation are called property rights institutions. Since they regulate the relation between governments (elites) and citizens, they are also called vertical institutions. Acemoglu and Johnson (2005) find that legal origin is a good instrument for contracting institutions but not for property rights institutions. On the contrary, settler mortality and population density in 1500 in the colonized regions are good instruments for property rights institutions but not for contracting institutions.

⁹ Hall and Jones (1999) also use ‘distance from the equator’ and ‘European language’ as instruments and uncover support for the institutions and policies theory of growth.

¹⁰ The only exceptions take place when corruption is instrumented with latitude, ethno-linguistic fractionalization and settler mortality rate; ethno-linguistic fractionalization, settler mortality rate and legal origin; and latitude, ethno-linguistic fractionalization, settler mortality rate and legal origin in which cases the F-statistics range between 8 and 9.86. Second-stage coefficient estimates identified using these instruments will be further tested in Section V given the potential presence of weak instruments according to Staiger and Stock (1997).

¹¹ Similar findings are reported in Djankov et al. (2002).

¹² In this regression when corruption is instrumented with latitude the coefficient estimate is not significant and when instrumented with ethno-linguistic fractionalization and with the linear combination of latitude and ethno-linguistic fractionalization the significance level is of 10% (p-values of 0.057 and 0.051, respectively).

¹³ This is the cost of avoiding the asymptotic bias of OLS.

¹⁴ Assuming that there is no heterogeneous response across countries, results are also quantitatively consequential. For instance, in the opening up a business regression when using as IVs for corruption ethno-linguistic fractionalization, the second-stage parameter estimate is 0.6144 which multiplied by the sample SD of corruption of 2.108 gives a change in the index of opening a business of 1.29. For concreteness, Russia's index of opening up a business is 1.3826 which corresponds to the place 62. Thus, if Russia could reduce its index of corruption one SD, its forecasted new level of 0.09 would represent an improvement of 59 places, similar to the level of regulatory transparency of Australia. Similarly, in the enforcing contracts regression the coefficient associated with the exogenous component of corruption is 0.4554 when it is instrumented with latitude and ethno-linguistic fractionalization. Diminishing corruption by one SD would reduce our index of enforcing contracts by 0.963. Specifically, if Panama were to reduce its corruption by one SD, would move up in the ranking of 50 places exhibiting a level of regulatory transparency in the area of enforcing contracts similar to Denmark. Overall, these findings indicate that with higher levels of corruption it takes more time, more procedures and more financial resources to open up a business and to enforce contracts.

¹⁵ The finite sample bias of 2SLS can be exacerbated as the number of instruments increases if they do not add much to the R-square of the semi-reduced form. What matters in this context is the marginal explanatory power of an extra IV after controlling for the other IVs and exogenous variables in the semi-reduced form. However, each valid additional over-identifying moment restriction can reduce the variance of the 2SLS estimates. This is the tradeoff faced when adding new IVs; see (Hahn and Hausman, 2005).

¹⁶ The critical values depend on the number of endogenous regressors and the number of instruments.

¹⁷ The null hypothesis of this test is that the bias of the 2SLS estimate is less than b% of the OLS bias. Failure to reject the null occurs if the F-statistic from the first-stage regression exceeds the critical values calculated by Stock and Yogo.

¹⁸ This argument is also valid for the Hansen (1982) OIR test.

¹⁹ Hansen's test generalizes Sargan's OIR test.