Results

OLS and robust regression results are given in Table 3 and Fixed and Random effects models are given in Table 4. In model 1, every variable is significant under 1% level therefore, the overall significance of the model (F-stat) implies a good fit. Diversification (-0.032) and R&D expenditure (-0.270) have negative signs. Therefore, as all other factors are kept constant while increasing the sales diversification ratio or increasing the allotted resources for the Research and Development department, the return on assets is expected to decrease. On the other hand, higher capital expenditure is related with higher return on assets. The R-squared is 5% which is a relatively low number. The R-squared around 20% would be satisfactory in a typical social science study, thus our models have a limitation in explaining the overall variance of the dependent variable.

In models (2) and (2a), the outliers are left out not resulting in a substantial change in the generated outputs. Sales diversification is negative and significant, R&D expenditure is again negative but not significant and capital expenditure is at similar levels. Asset diversification variable (2a) is not significant but it is a negative and very small number.

As the interaction terms are included, the overall explanation of variance of dependent variables does not virtually increase. On the contrary, the significance of variables disappear and the only remaining significant explanatory factor remains is capital expenditure which is significant in all regressions. Capital expenditure ranges from 0.29 to 0.37 across six models.

Robust regression generates a significant negative diversification estimate (-0.021) and significant capital expenditure (0.29). The last regression in Table 3 includes the industry dummy variables that substantially increase the R-squared to 0.30, having an F-stat of 14.14. Regression generates a negative Research and Development expenditure estimate and positive Capital expenditure estimates both of which are at similar levels as compared with the previous estimates. Therefore, taking into the industry characteristics account, the dummy variables strengthen the overall explanation of the variation in returns across firms. Different industry characteristics covary with return levels, as well as different firm strategies do in terms of resource allocation.

Fixed and random effects are utilized in addition to several OLS and robust regression analysis. In the fixed effects model, diversification is estimated to be negative. However, it is statistically insignificant. Research and development expenditure significantly undermines returns. Capital allocation variables are significant under 1% level. Overall model's significance (F-stat: 33.61) indicates a good fit of the set of the variables. However, the R-squared is as low as 6%.

In the random effects model, diversification is not statistically significant whereas R&D (-0.2366) and capital expenditure (0.3675) are significant under 1% level. The R-squared within is just below 6% and the R-squared between and overall is 4%.

Fixed effects and random effects are used in our panel data regression analyses, and the Hausman test (p > 0.1373) revealed that using random effects allows the analysis to fit the data better. Nevertheless, the outcomes of both models are largely similar with the exception of lower coefficient of R&D expenditure in the random effects model. Both random and fixed effects models results are provided.