Board political connections and financial fraud: The case of business groups in South Korea

Abstract

Are business groups more likely to commit financial fraud when they are politically connected? Many past studies and anecdotal evidence have pointed out that business groups may behave more opportunistically when they are politically connected. However, the nature of the business group-government relationship evolves amid governance reform in many countries, making it difficult for business groups to abuse their political connections. In this research, we examine the business group-government relationship through the lens of social exchange and find a deterring effect of political connections on a connected business group’s propensity to commit fraud. Our results indicate that business groups are less likely to commit financial fraud when the extent of their political connection is high. By doing so, politically connected firms can prove themselves as legitimate business partners for the government and can more effectively secure a position to leverage their political connections. We additionally find that such an exchange relationship is weakened under a business group-unfriendly government due to their hostile relationships.

Appendices

Appendix 1

Table 4

Table 4 Description of control variables for fraud detection and commission

Appendix 2

We denote ${Fraud}_{it}^{*}$ as firm $i$’s propensity to commit fraud at time $t$ and ${Detect}_{it}^{*}$ as firm $i$’s likelihood of being detected for fraud at time $t$.

$${Fraud}_{it}^{*}={X}_{F,it}{\beta }_{F}+{\mu }_{it}$$

(1)

$${Detect}_{it}^{*}={X}_{D,it}{\beta }_{D}+{v}_{it}$$

(2)

Each ${X}_{F,it}$ and ${X}_{D,it}$ represents a row vector with variables that explain firm $i$’s propensity to commit fraud and the potential for being detected, respectively. ${\mu }_{it}$ and ${v}_{it}$ are zero-mean disturbances with a bivariate normal distribution and variances, which are also normalized to unity because variances are not estimable (Khanna et al., 2015; Shi et al., 2017). The correlation between the two is $\uprho$. Following previous studies, we define fraud commission as ${Fraud}_{it}^{*}=1$ if ${Fraud}_{it}^{*}>0,$ and ${Fraud}_{it}^{*}=0$ otherwise. Similarly, ${Detect}_{it}^{*}=1$ when ${Detect}_{it}^{*}>0$, and ${Detect}_{it}^{*}=0$ otherwise (Khanna et al., 2015). Note that we only observe when fraud is both committed and detected, that is, ${Observe}_{it}=1$, only when ${Fraud}_{it}^{*}=1$ and ${Detect}_{it}^{*}=1$, but ${Observe}_{it}=0$, when either ${Fraud}_{it}^{*}$ or ${Detect}_{it}^{*}$ is zero.

$${Observe}_{{\varvec{i}}{\varvec{t}}}={Fraud}_{{\varvec{i}}{\varvec{t}}}{Detect}_{it}$$

(3)

φ is the function for the bivariate standard normal cumulative distribution, and the empirical model for ${Observe}_{{\varvec{i}}{\varvec{t}}}$ is

$$P=\left({Observe}_{it}=1\right)=P\left({Fraud}_{it}{Detect}_{it}=1\right)=\varphi \left({X}_{F,it}{\beta }_{F},{X}_{D,it}{\beta }_{D},\uprho \right)$$

(4)

$$P=\left({Observe}_{it}=0\right)=P\left({Fraud}_{it}{Detect}_{it}=0\right)=1-\varphi \left({X}_{F,it}{\beta }_{F},{X}_{D,it}{\beta }_{D},\uprho \right).$$

(5)

The log likelihood function for the model is as follows:

$$L\left({\beta }_{F},{\beta }_{D},\rho \right)=\sum \mathrm{log}\left(P\left({Observe}_{it}=1\right)\right)+\sum \mathrm{log}\left(P\left({Observe}_{it}=0\right)\right).$$

(6)