Chapter 4 Static Models (OLS & Fixed-Effects) Results
4.5 Reliability of Static Models (OLS &FE) and Possible Solutions
The diagnostic tests (heteroscedasticity and autocorrelation) reject the null hypotheses that there is heteroscedasticity as well as autocorrelation in the data. The question is how reliable are the estimates from OLS and FE? What are the possible solutions to these problems? As argued by Baltagi (2008), the estimations of OLS are consistent but inefficient in the presence of heteroscedasticity and autocorrelation as the standard errors are downward biased and CLRM assumes that the disturbance terms are constant and independent across cross-sections and time. Similarly, FE estimation also assumes that the disturbance term
v
it is identically distributed and independent ofv
it for alli
andt
.Since our estimates (OLS & FE) are inefficient, we now try to find solutions to these problems. One prominent solution to heteroscedasticity suggested by Gujarati (2012) is to assign weights to each observation in the data. He argues that observations from a population with less variability should carry more weight and those coming from a population with greater variability should have less weight in the regression. In other words, the weights should be inversely related to the standard deviation of the observations. Simple OLS and FE cannot incorporate these weight phenomena but this problem, however, can be overcome by running Generalized Least Squares (GLS), which assigns weights to each observation and solves the problem of heteroscedasticity. Similarly, the problem of autocorrelation, according to Baltagi (2008) and Gujarati (2012), can be solved in few ways such as adding more independent variables, data transformation such as taking logarithms and using lags of dependent variable as regressors.
However, we suspect another missing link between IC and firm performance. This missing link is the potential existence of an endogeneity problem that is mainly because of simultaneity or reverse causality in the IC - firm performance relationship. In the literature, the focus has been on a one way relationship, i.e., how does IC efficiency affect the financial performance of the firm? But there is a possibility that IC efficiency is also being affected by past firm performance, which is the case with simultaneity. If simultaneity exists (a cause of endogeneity), then the usual static models such as OLS and FE (this issue is further discussed in chapter 5) do not generate BLUE estimations (Wintoki et al., 2012) rather, the Dynamic Panel Data (DPD) estimator should be used (Gujarati, 2012). However, no existing study in the literature has explored the dynamic nature of the relationship between IC and firm performance.
In the next chapter, we test whether the relationship between IC and firm performance is dynamic and how this relationship should exactly be estimated.
4.6
Chapter Summary
This chapter reports the results of descriptive statistics, static OLS and Fixed-Effects (FE) estimation models. Mean IC efficiency scores “measured in terms of VAIC” vary from 5.08 to 9.28 with an overall mean of 7.90 for all five developed markets in this current study. Among the five developed markets, Australia scores highest and Austria lowest, which implies that Australian firms use IC more efficiently than the other four developed markets. The mean IC efficiency scores are consistent with those reported by Joshi et al. (2013) for the Australian financial sector (8.82), however, the scores are higher than those reported by Chen et al. (2005) for Taiwan (5.49). In terms of human capital efficiency in developed markets, the mean scores vary from 4.13 to 8.06 with an overall mean of 6.66. Australia once again tops the list with Austria at the bottom, which means firms in Australia use human capital more efficiently than the other four developed markets. The SCE scores vary from 0.46 to 0.58 with an overall mean of 0.51 among the five developed markets. The CEE scores among the five developed markets vary from 0.39 to 0.98 with an overall mean of 0.62. The mean CEE scores for Singapore (0.39) are lowest, which implies that physical capital is no longer considered a major contributor towards firm value in Singapore.
Among emerging markets, the mean IC efficiency scores vary from 5.10 to 9.18 with an overall mean of 7.10. The mean scores are consistent with those reported by Pek (2005) for Malaysia (7.11) but higher than those reported by Pal and Soriya (2012) for India. The mean HCE scores in emerging markets vary from 4.52 to 8.19 with an overall mean of 6.20. The HCE for South Africa is lowest (4.52), which is similar to Firer and Williams (2003)’s argument that South African firms still rely on physical capital for value creation. Among the emerging markets, the structural capital scores in Table 4.2 vary from 0.59 to 0.86 with an overall mean of 0.69.
The IC efficiency scores for frontier markets vary from 4.21 to 11.26 with an overall mean of 7.26. The mean VAIC scores are slightly skewed towards the higher side because Saudi Arabia exhibited exceptionally high scores (11.26) compared with the other four frontier markets. In terms of human capital, Nigeria scored the lowest (1.49) and Saudi Arabia the highest (10.35). These high HCE scores contradict Kaplan and Norton (2004)’s argument that countries such as Saudi Arabia and Venezuela are rich in natural resources but make poor investment in their human capital. Our results provide evidence that in the 21st century firms rich in natural resources are significantly investing in their
human resources. In terms of the developed, emerging and frontier markets, the IC efficiency scores are highest in developed markets, which means that developed countries are most efficient in using IC for value creation.
This current study applies both Fisher-Type and Modified Fisher-Type tests to check for stationarity on the unbalanced panel data. From p-values in Table 4.4 the null hypothesis can be rejected at all
conventional significance levels in all the countries for all four dependent variables (ROA, ROE, ATO and P/B), which means that there is no unit root in our data. Pearson pairwise correlation results show that correlations among the regressors do not exceed 0.80, which implies that there are no issues of multicollinearity. The OLS results show that IC efficiency in terms of VAIC is positive and significant at 1% with ROA in all 15 markets in this study. This shows that IC resources contribute significantly towards value creation of firms, which endorses the RB theory. Individual components of the VAIC model show that only SCE and CEE are significant (at 10% or less) with ROA in most markets whereas HCE is either negative or insignificant in nine markets in the study. Our robustness checks indicate that IC is significant only with ROE but insignificant with ATO and the P/B ratio.
Fixed-effects analysis of the relationship between IC and firm performance shows that VAIC is positive and significant (at 5% or less) in all developed, emerging and frontier markets. Individual component analysis produces similar results to OLS. HCE is once again negative or insignificant with ROA whereas SCE and CEE are significant (at 5% or less) in almost all markets. Advanced diagnostic tests, such as the Bruesch-Pagan test for heteroscedasticity and Wooldridge test for autocorrelation, reject the null hypotheses which means that there is heteroscedasticity and autocorrelation in the data.
As argued by Baltagi (2008), the estimations of OLS are consistent but inefficient in the presence of heteroscedasticity and autocorrelation as the standard errors are downward biased and CLRM assumes that the disturbance terms are constant and independent across cross-sections and time. Similarly, FE estimator also assumes that the disturbance term
v
it is identically distributed andindependent of
v
it for alli
andt
. These problems can be solved in many ways such as through theapplication of GLS, taking first difference or data transformation. However, we suspect another econometric problem, i.e., the presence of endogeneity. The literature has so far considered the IC and firm performance relationship as one way but we look at it from another angle, i.e., firm performance might also affect IC and this is simultaneity. As argued by Gujarati (2012), the application of OLS or FE produces biased, inconsistent results in the presence of endogeneity (mainly because of simultaneity). In the next chapter, we test whether the relationship between IC and firm performance is dynamic and how this relationship should exactly be measured.