Out of the shadows or into the dark? Economic openness, IMF programs, and the growth of shadow economies

By way of definition, the shadow economy comprises those activities that go unreported, unregulated, and untaxed (Feige 1989, 1990; Schneider and Williams 2013). The most common shadow sector activities include otherwise legal but unrecorded and untaxed commercial transactions, unreported labor, and smuggling and illicit commerce. The terms of shadow economy and informal economy are used interchangeably throughout the manuscript.

Despite the dynamics supported by the above arguments, it should be noted that the impact of globalization (in general) and foreign capital (in particular) on labor rights remains contentious (i.e., Blanton and Blanton 2016; Mosley 2010).

For comprehensive, critical overviews of the different methodologies, see Orsi et al. (2012) and Elgin and Oztunali (2012).

To check the robustness of our findings, we report additional ordinary least square models using the Schneider et al. data for the 1999–2007 period in the online appendix. We find mostly similar results in these additional models, but there are some differences that are likely related to the far shorter period of analysis.

We employ these two trade and investment measures instead of the KOF Actual Flows index variable (Dreher 2006), as KOF Actual Flows includes two other measures (portfolio investment and income from foreign nationals) that are not germane to our proposed theory.

When we replace trade with two variables that capture imports and exports separately, we find similar results. That is, both imports and exports of goods and products significant lower the extent of shadow economic activity. We also tested the possible effect of FDI outflows and found no statistically significant association between FDI outflows and the growth of informal economies.

In the online appendix, we also report models with two additional proxies for political stability and quality of governance, Control of Corruption and Bureaucratic Quality. More official corruption and less efficient bureaucracies might create more opportunities for state officials and private economic agents to engage in informal economic activity. Both variables are from the International Country Risk Group (ICRG) dataset (Knack and Keefer 1998). The corruption variable varies from 0 to 6 with higher scores denoting less corruption in the government. The bureaucratic quality variable ranges from 0 to 4 with higher scores indicating more efficient bureaucratic establishment. Our main findings remain largely unaltered in the models estimated with these additional control variables. We do not control for the corruption and bureaucratic quality variables in the main analysis, as both variables are available for a shorter time period (post-1984) and fewer countries in the ICRG dataset, which reduces the sample size by roughly a third.

On a more micro level, Tajima (2014) shows how transitions into democracy can result in increased criminality as “the coercive grip of the state loosens” (p. 4). Indeed, Vreeland (2008) argues that the democracy measures themselves are defined in reference to the presence of violence, with the result being that more violence is apparent in the middle range of the democracy measures.

Stata’s xtabond2 command is used to estimate the GMM model. In terms of model specifics, we derive the instruments using principal components analysis option with a one-year lag (e.g., Bai and Ng 2010). We also use the orthogonal transformation to the generated instruments, as it has been found to provide less biased and more stable instruments (Hayakawa 2009). Models also use robust standard errors with the two-step estimation option, as it provides increased efficiency (Roodman 2009). For the purposes of the models, we set each of our key independent variables, as well as our measures of economic growth and income, as endogenous. As lagged values tend to be weak instruments, GMM generates a large number of instruments, which raises two potential limitations: first the model requires a large sample size, and ultimately the total size of the instruments should be less than the number of groups included in a given model (i.e., Asiedu and Lien 2011; Baum et al. 2003). The former is not a problem in our analysis, and we limit the number of lagged values used in our instruments to keep the total number of instruments below the number of groups analyzed in our models. Results of the GMM models, included in Appendix Tables 3 and 4, largely support the findings of the main analyses. Though not reported in the main analysis to save space, as a further test of robustness, we also run models using two-stage least squares (2SLS) regression and continue to find significant support for our hypotheses in those models. More detailed explanations of the 2SLS models, as well as the results, appear in the online appendix.