### 1. Introduction

*de facto*control rights over corporate management can make arbitrary decision-makings to maximize their utility (Almeida et al., 2011). Accordingly, many previous studies on emerging economies focus on the agency problem based on the ownership of controlling shareholders (Shleifer and Vishny, 1997; La Porta et al., 1999).

### 2. Background and Hypothesis Development

### 2.1 Managerial Ownership and Investment Inefficiency

#### 2.1.1 Overinvestment

#### 2.1.2 Underinvestment

### 2.2 Product Market Competition and Managerial Incentives

#### 2.2.1 Private Benefits

#### 2.2.2 Monetary Compensation

### 3. Data and Variables

### 3.1 Data

### 3.2 Variables

#### 3.2.1 Investment Inefficiency

*Investment*as the sum of capital expenditure and R&D expense divided by total assets. We estimate

*V/P*, a proxy for growth opportunity, firm value (

*V*) divided by the market value of equity (

*P*).4) Here,

*V=(1-αr)BV+α(1+r)X-αrd, α=(ω/(1+r-ω)), r=0.12*, and

*ω=0.62*.

*ω*is based on Ohlson’s (1995) abnormal earnings persistence parameter.

*BV*is the book value of common shares,

*d*is dividends, and

*X*is operating income after depreciation.

*Leverage*is total debt divided by total assets and

*Cash*is cash and cash equivalents to total assets.

*Age*is the natural log of the age of the firm,

*Size*is the natural log of total assets, and

*Stock return*is the stock returns for one year. We define the residual of equation (1) as investment inefficiency (

*Residual*).5)

##### (1)

*Residual Over*). As a measure of underinvestment, we set a variable that takes the absolute value if the residual is less than zero, and a value of zero if the residual is greater than zero (

*Residual Under*).

*Excess investment*). We calculate this variable as the firm’s investment expenditure minus the average industry investment expenditure, classified based on the 3-digit KSIC.

#### 3.2.2 Ownership6)

*Controlling ownership*) in the empirical analysis.

*Disparity*).

#### 3.2.3 Product Market Competition

*HHI*). HHI has a value between 0 and 1. A higher HHI means a concentrated market structure and low competition, while a low HHI implies that there are many firms with small market shares competing in an industry, indicating high competition.

*CR4*). If there is a large deviation in market share among firms in the industry, then competition among firms with a high market share will be intense, while firms with a low market share are more likely not to face competition. We use CR4 to reflect this property (Hou and Robinson, 2006; Giroud and Mueller, 2010). A higher CR4 indicates lower competition.

*Entry cost*). The weighted average is based on market share. A high level of this indicator means a relatively high entry barrier so that the magnitude of the potential competitive threat is small. The price-cost margin reflects product substitutability based on the industry’s profitability (Bena and Xu, 2017). It is the sum of sales divided by the sum of the cost of sales and selling and administrative expenses (SG&A) of firms in an industry (

*Price-cost margin*) (Karuna, 2007). The higher the price-cost margin is, the higher the monopoly profits compared to the input resources are, meaning that it is a relatively non-competitive market.

#### 3.2.4 Other Variables

*Size*). A high leverage ratio can cause underinvestment problems by increasing the bankruptcy risks (Myers, 1977). In terms of agency theory, a high debt ratio has the effect of preventing the overinvestment problem, as it reduces the free cash flow that managers can use (Jensen, 1986). Taking these effects into account, we include the debt-to-equity ratio in the model (

*Leverage*).

*ROA*). The firm chooses the optimal level of investment expenditure to realize growth opportunities and to ensure future sustainability. Even if managers have excessive investment spending from the agency theory perspective, it would be difficult to regard it as overinvestment if the goal is to realize growth opportunities. Thus, to control for growth opportunity, we include the market to book ratio, which we calculate as the market value of equity (the number of common shares*the stock price at the end of the year) divided by its book value (

*Market-to-book*) (Biddle and Hilary, 2006).

*Free cash flow*). To obtain a clearer result for the relationship between the ownership structure of controlling shareholders and their investment decisions, we control the controlling shareholders’ incentive to pursue private benefits unrelated to the ownership structure. Almeida et al. (2011) find that controlling shareholders may adjust discretionary accruals to avoid losses due to declining operating performance, which we can perceive as a typical agency problem. We calculate the discretionary accruals variable as the absolute value of the difference between net income divided by total assets and the operating cash flows divided by total assets (

*Accruals*).

*Age*). Dividend payments could be a signal to the stock market of stable profitability, so it may be associated with an increase in future investment expenditures. To control these effects, we add the ratio of cash dividends to equity (

*Dividend*). Corporate growth opportunities can be linked to both future growth potential and current growth levels. As we mentioned earlier, growth opportunities are the most important indicators for assessing the adequacy of corporate investment spending. Therefore, we use the average sales growth for the past three years as a proxy for the current growth level (

*Sales growth*). Lastly, liquid assets represent the size of the accumulated internal resources available to managers in the short term. Firms with larger liquid assets will have greater room to cover the investment costs. We control for this effect including liquid assets divided by total assets (

*Liquidity*).

*Stock return volatility*). Demsetz and Lehn (1985) argue that the ownership structure is endogenously determined. They suggest that if shareholders or outside investors are risk-averse, then the risk inherent in the firm would have a negative (-) relationship with the ownership structure. Kang et al. (2006) find such a relationship empirically in Korean firms. The controlling shareholders tend to avoid holding ownership in higher risky companies (higher stock return volatility). However, the stock return volatility is a result of the investment and cannot precede the investment. In particular, as stock return volatility is measured based on stock returns over the past five years, this precedent relationship is even more pronounced. Based on this logic, this study employs the volatility of stock returns as an instrumental variable. We discuss the adequacy of the instrument variable with the results of the empirical analysis.

*Dominant firms*).

### 4. Empirical Results

### 4.1 Descriptive statistics

*Controlling ownership*) and the

*HHI*. We set 10 intervals according to the size of each variable. The average controlling shareholders’ ownership is 25.03%, which is similar to that (21%) of Almeida et al. (2011). The largest number of firms is in the 0.0-0.1 of the ownership group. However, about 44.27% of firms are in the 0.2-0.5 section, and 11.43% of firms have more than 0.5 of the ownership, which implies that controlling shareholders have perfect control power. As controlling shareholders exercise their control rights not only through their direct ownership but also indirectly through affiliates, their actual influence on management activities will be greater. The results in <Table 1> clearly show the concentrated ownership structure of Korean firms, confirming the appropriateness of our assumption that the ownership structure of controlling shareholders has a significant influence on management decision-making, as they have

*de facto*control power.

##### <Table 1>

##### <Table 2>

*Residual*, a measure of investment inefficiency, is -0.0003.13) The average

*Residual Over*and

*Residual Under*are 0.0126 and 0.0129, respectively. The higher average of

*Residual Under*than that of

*Residual Over*reflects the passive investment expenditure of Korean firms, as Korea’s economic growth is low. However, the average

*Residual Over*of the group of overinvestment firms, 0.0354, is greater than the average

*Residual Under*of the group of underinvestment firms, 0.0201. This result suggests that the amount of overinvestment is larger than the amount of underinvestment.

*Controlling ownership*is 0.2503 and is smaller for overinvestment firms than for underinvestment firms. The difference is significant at the 5% level. This result is consistent with the theory. The average disparity between ownership and control (

*Disparity*) is 0.1534, implying that controlling shareholders can exercise 15.34% of control rights without direct ownership.

*HHI*, the measure of product market competition, is 0.0968. The average for the alternative measures of product market competition,

*CR4*,

*Entry cost*, and

*Price-cost margin*are 0.4281, 19.7540, and 1.0710, respectively. Similar to the

*HHI*, the values are higher for overinvestment firms. The average

*HHI 2digit*and

*HHI 4digit*are 0.0611 and 0.1327, respectively.

*Residual*and

*Controllership ownership*. However,

*Controlling ownership*is closely related to overinvestment (

*Residual Over*) and does not show a significant correlation with underinvestment (

*Residual Under*). This result supports Hypothesis 1-1. The correlation coefficient between

*Residual*and

*HHI*is not significant. The correlation coefficient between

*Stock return volatility*and

*Residual*(-0.0130) is insignificant, while that between

*Stock return volatility*and

*Controlling ownership*(-0.2090) is significant at the 1% level, suggesting that

*Stock return volatility*is economically and statistically appropriate as an instrument variable.

### 4.2 Main Results

#### 4.2.1 Regression results

*t*-1) before. Since our sample consists of panel data, errors due to the time-series correlation and heteroscedasticity may arise. Petersen (2009) suggests that using the firm-level clustered standard error can reduce such errors considerably. Therefore, we test our hypotheses using OLS regressions with a clustered standard error at the firm-level. To control for the year and industry effects to reduce the potential endogeneity, we include the year (Year effect,

*η*) and industry dummies (Industry effect,

*λ*) in the model. Equation (2) represents the main empirical model, and we provide the definitions of each variable in <Appendix>.

##### (2)

*High*is a dummy variable that takes the value of one when the competition measure is in the highest tertile (non-competitive product markets),

*Median*is a dummy variable that takes the value of one when the competition measure is in the median tertile, and

*Low*is a dummy variable that takes the value of one when the competition measure is in the lowest tertile (competitive product markets). Using the interaction variables between these dummy variables and controlling shareholder’s ownership, we analyze how the effect of controlling shareholder’s ownership changes depending on the level of competition, following Giroud and Mueller (2011). As <Table 1> shows, the division based on the tertile of competition is generally similar to the classification of the monopoly regulation,14) so we expect that the analysis will reflect the reality.

*Controlling ownership*is significantly negative (-) at the 5% level. This result implies that, as discussed theoretically, controlling shareholders with low ownership tend to overinvest to maximize their private benefits (Hypothesis 1-1). On the other hand, controlling shareholders with large ownership tend to underinvest as they are more risk-averse (Hypothesis 1-2).

##### <Table 3>

*Controlling ownership*and

*High*has a statistically negative (-) value, suggesting that controlling shareholders change investment expenditures to pursue private benefits only in non-competitive product markets. Notably, the magnitude and statistical significance of the coefficient of the interaction variable is larger and more significant than is the coefficient of

*Controlling ownership*in model (1). Thus, our results for the overall sample are mainly driven by the behavior in non-competitive product markets. The interaction variables between

*Controlling ownership*, and

*Median*and

*Low*are not statistically significant, implying that controlling shareholders do not change investment expenditures according to their ownership due to the disciplinary effect of competitive threats. The coefficient of

*HHI*is insignificant, meaning that we find a moderating effect of the controlling shareholders on management decisions, rather than a direct effect of competition. The above results support Hypothesis 2-1.

*Leverage*) has a significant negative (-) effect, which we can interpret as an increase in interest costs and a relatively high risk of bankruptcy, making aggressive investment difficult. The market-to-book ratio (

*Market-to-book*), which is a substitute for growth opportunities, has a positive (+) coefficient, as firms with relatively high growth opportunities actively expand investment spending.

*Free cash flow*also has a positive (+) impact because firms with high internal funding capacity can cover the cost of investment. The age of the firm (

*Age*) has a positive (+) coefficient, as older firms have low information asymmetry and can secure external financing.

*Liquidity*has a positive (+) effect because firms with a large number of assets to use in the short term can actively invest.

*High*), and model (4) shows the result of the sample of firms in competitive product markets (HHI lower 33%,

*Low*). The coefficient of

*Controlling ownership*in model (3) has a significantly negative (-) value at the 5% level, while the significance in model (4) disappears.

*Excess investment*) as the dependent variable. The coefficient of the controlling shareholders’ ownership in model (5) is negative (-), but not significant. The interaction variables between

*Controlling ownership*, and

*Median*and

*Low*are not significant. The results have the same interpretation as in the results of model (2).

#### 4.2.2 Overinvestment Versus Underinvestment

*Residual Over*. In model (1), the coefficient of

*Controlling ownership*is significantly negative (-) at the 1% level. In model (2), the interaction variable between

*Controlling ownership*and

*High*is significant and negative (-), but that between

*Controlling ownership*and

*Median*and

*Low*is insignificant. Models (3) and (4) focus on the underinvestment problem, and the dependent variable is

*Residual Under*. The coefficient of

*Controlling ownership*in model (3), as well as the coefficients of the interaction variables in model (4), are insignificant. These results imply that the results in <Table 3> are mainly due to the overinvestment problem. That is, the main path by which controlling shareholders pursue private benefits is overinvestment based on low ownership. The disciplinary effect of competitive threats affects this overinvestment behavior.

##### <Table 4>

*Residual Dummy*as a dependent variable to examine whether the incentive for overinvestment or underinvestment is more related to the controlling shareholders’ ownership and product market competition. The

*Residual Dummy*is a dummy variable that takes the value of one if a firm is an overinvestment firm. This approach focuses on the probability of over- or underinvestment rather than their scales. It compares the tendencies to overinvest or underinvest and alleviates the concern about overestimation due to the scale of over- or underinvestment. In model (5), the coefficient of

*Controlling ownership*is significant and negative (-), which suggests that controlling shareholders with low ownership tend to overinvest. In model (6), the coefficient of the interaction variable between

*Controlling ownership*and

*High*is significantly negative (-), implying that we mainly observe the overinvestment tendency found in model (5) in non-competitive industries.

### 4.3 Robustness check

#### 4.3.1. Alternative Model Specifications

*Residual*as a dependent variable. Similar to the main result, the coefficient of

*Controlling ownership*is significantly negative (-) in model (1), and only the coefficient of the interaction variable between

*Controlling ownership*and

*High*is significant and negative (-) among the interaction variables in model (2). This result is the same as that in <Table 3>. Models (3) and (4) use the

*Residual Over*and models (5) and (6) use the

*Residual Under*as the dependent variable, respectively. The results of models (1) and (2) are consistent with those in models (3) and (4), reaffirming that controlling shareholders’ pursuit of private benefits based on their ownership is mainly related to overinvestment.

##### <Table 5>

*Residual*as a dependent variable, the coefficient of

*Controlling ownership*is negative (-), but insignificant. However, in model (2), the coefficient of the interaction variable between

*Controlling ownership*and

*High*is significantly negative (-). Models (3) and (4) are the results using

*Residual over*as a dependent variable. The results are similar to those in <Table 4>.

*Residual*) and

*Controlling ownership*, we conduct the analysis using a one-period time-lag between the two variables. Nevertheless, given the claims in prior studies that the ownership of controlling shareholders could be determined endogenously, it cannot be completely free from the endogeneity problem (Demsetz and Lehn, 1985). Model (1) presents the results of the first stage model. The coefficient of

*Stock return volatility*is significant and negative (-). In general, the instrument variable is adequate by the rule of thumb if its t-value is 3.3 or higher.

*Stock return volatility*fulfills the criteria, thus confirming its adequacy. Models (2) and (3) are the results using

*Residual*as a dependent variable. The coefficient of

*Controlling ownership*in Model (2) is not significant, but the coefficient of the interaction variable between

*Controlling ownership*and

*High*in Model (3) is significant. This result supports our main finding. Models (4) and (5) use

*Residual Over*and models (6) and (7) use

*Residual Under*as the dependent variable, respectively. Only the coefficient of the interaction variable between

*Controlling ownership*and

*High*in model (5) is negative (-), confirming that controlling shareholders’ pursuit of private benefits is mainly related to overinvestment.

#### 4.3.2 Alternative Measures of Product Market Competition

*CR4*, models (4) - (6) use

*Entry cost*, and models (7) - (9) use

*Price-cost margin*to measure the level of competition. In most models, we find a negative (-) influence of

*Controlling ownership*on investment inefficiency mainly in non-competitive markets (

*High*), which disappears in competitive markets (

*Low*). In addition, investment inefficiency is mainly related to overinvestment, but not to underinvestment. These results confirm that our main results do not change with the selection of the measure of competition and are robust.

##### <Table 6>

*HHI*calculated by applying another industry classification standard. Models (1) - (3) use the 2-digit KSIC (

*HHI 2digit*) and models (4) - (6) use the 4-digit KSIC (

*HHI 4digit*) to estimate

*HHI*. The results are consistent with those in <Table 3> and <Table 4> and confirm the robustness of the main results.

### 4.4 Controlling Shareholders’ Incentive for Pursuing the Private Benefits

*Residual*as a dependent variable. In model (2), the coefficient of the interaction variable between the

*Disparity*and

*High*has a significant and positive (+) value, while that of the other interaction variables is not significant. This result suggests that the overinvestment behavior occurs in non-competitive markets, and disappears in competitive markets. This result supports our main findings. Models (3) and (4) are the results using the separated sample approach. Similar to model (2), the results also show that controlling shareholders overinvest in non-competitive markets based on their ownership-control disparity.

##### <Table 7>

### 4.5 Alternative Explanations: Predation Risk Versus the Agency Problem

*Dominant firms*).

*Controlling ownership*and

*High*have significantly negative (-) values, while those of other interaction variables are not significant. This result indicates no difference from our main results after dividing the sample based on market power. Rather, the result clarifies that there is no room for interpretation based on predatory risk, as the magnitude and statistical significance of the coefficient of the interaction variable between

*Controlling ownership*and

*High*are greater in the sample of dominant firms.

##### <Table 8>

*Accruals*) as a proxy for the agency problem. Panel B presents the results of the main analysis in two separate samples. Models (1) and (2) present the results using firms with larger discretionary accruals; that is, with a high probability of agency problems, and models (3) and (4) present the results using firms with smaller discretionary accruals; that is, with a low probability of agency problems. Our main results are observed only in the sample with a high probability of agency problems.