Tobin’s Q as dependent variable

Table 5 shows the results of the linear regressions with Tobin’s Q as the dependent variable and book leverage as independent variable. The model represents the results of eight different regressions and shows that in all the models leverage has a significant negative impact on Tobin’s Q at the 0.01 significance level. Model 6, with all control variables included, reports as significant negative impact of leverage on Tobin’s Q at the 0.01 significance level. Table 5 shows that a one standard deviation increase in book leverage is associated with a -0.190 point decrease in Tobin’s Q. Model 7 shows that the exclusion of current ratio bearly influences the regression results and therefore, the correlation between current ratio and book leverage has no impact on the regression results. Table 6 shows the regression results of Tobin’s Q with market leverge as dependent variable. Here also, all regressions show a significant negative impact of leverage on Tobin’s Q. The results show that a one standard deviation increase in market leverage is associated with a -0.290 point decrease in Tobin’s Q (model 6). This negative impact is in line with the research of Bae, Kim, and Oh (2017) for their sample of 1481 firms traded on the US stock market for the period 1970-2011. King and Santor (2008) also found a negative impact of leverage on Tobin’s Q for Canadain firms. Therefore, when a firm increases their leverage, so adding more debt to their capital structure, this lowers their financial performance in terms of Tobin’s Q.

Concerning the control variables, size has no significant impact on Tobin’s Q in model 2 in table 5, but has a significant positive impact at the 0.01 level in model 6. This is the result of; ‘omitted variable bias’. The omitted variable bias can occur when in a statistical model one or more relevant variables are left out. This variable is known as the confounding variable and forces the model to attribute the effect of the omitted variable to the other variables in the model, which biases the coefficient estimates2_{. This bias results in the fact that the model} attributes the effect of the missing variable to the estimated effects of the variables that are included. It can have the following effects:

37 - Overestimate the strength of an effect.

- Underestimate the strength of an effect. - Change the sign of an effect.

- Mask an effect that actually exists

There are two conditions that must hold for omitted variable bias to exist3_: - The omitted variable must be correlated with the dependent variable

- The omitted variable must be correlated with one or more other explanatory/ independent variables

To explain the phenomeon see figure 3, where Y is the dependent variable and A and B are two independent variables. Area 1 is the impact of variable A on Y, area 3 is the impact of variable B on Y. When you include variable A and omitted variable B in the regression, the impact of variable A is explained by areas 1 and 2, not alone by area 1. When you only add variable A in the regression (figure 4) you explain the impact of variable A on Y by areas 1 and 2 while area 2 actually belongs to both variables A and B. This means the impact is biased because area two actually belongs to both variables A and B and not only to variable A.

Y Y

1 3 1

2 2

A B A

Figure 3: Omitted variable bias Figure 4: Omitted variable bias

The insignificance of size in model 2 and its significant impact in model 6 is a result of omitted variable bias. There is a confounding variable in model 6 that affects both Tobin’s Q and size. This basically occurs because of correlation, when you only include one of two strongly correlated variables, this one variable will show ‘absorb’ the effect of the variable that is left out. It will display its own effect and that of the other variable. These two effects might cancel each other out, which results in a seemingly insignificant prediction. Therefore, both correlated variables should be included together and both variables will display their own effect. Because size has the highest correlation (-0.313**) with business risk among the variables these variables are included together in model 8. The results of this regression show that when these variables are included together both variables are statistically significant at

38 the 0.01 level and display their real effect. When using market leverage as independent variable size is positive significant on Tobin’s Q in model 2 and 6. Tangibility has a significant negative impact on Tobin’s Q in model 3 and model 6 at the 0.01 level. In table 6 this omitted variable bias problem also occurs, where tangibility is significant in the full model but not in model 3. Therefore, in model 8 tangibility and current ratio are added together because they have the highest collinearity. When adding these variables together, they both show their real effect and tangibility and current ratio are both significant at the 0.01. The negative impact of tangibility on Tobin’s Q suggests that the more tangible assets a firm has, the lower it’s Tobin’s Q is. The other two control variables, business risk and current ratio both have a significant positive impact on Tobin’s Q in all the models at the 0.01 significance level. The positive impact of current ratio, in table 5 and 6, on Tobin’s Q suggests that firms that have a higher ratio of current assets to current liabilities, and therefore are better able to meet their short term debt obligations have better financial performance in terms of Tobin’s Q. The positive impact of business risk on Tobin’s Q suggests that firms that are more risky have a higher Tobin’s Q, this results are significant at the 0.01 level when using book leverage as well as market leverage.