5. Empirical results

Panel regression analysis was used to investigate the tradeoff between WCC and the profitability of the 41 firms listed on BIST Industrial Index in Turkey. In the panel data analysis, variables include both time and cross section size. According to time and cross-section effects, it is determined that the model should be predicted to be one way or two ways. For this purpose, the LR test has performed with the maximum likelihood method, and the findings are given in Table 3. The calculated test statistics are interpreted according to the 1% significance level.

For the two-way effects test, the null hypothesis is formed no cross section and time effects in the model. Because the value of the test statistic for the two-way effect is 279.1188 at 1% significance level, the null hypothesis is rejected. This result shows that it is a two-way effect. Then, the presence of the cross section and time effects was tested separately with the movement from the findings that it was a two-way effect. The null hypothesis for cross section effect analysis is that the standard error of cross section is equal to zero. According to the analysis results, the null hypothesis is rejected at 1% significance level, since the value of the test statistic is 262.4951. In this case, there is a cross section effect in the panel data model. The existence of time effect was also examined, and the test statistic was calculated as 3.981432 at 5% significance level. According to this result, the null hypothesis cannot be rejected at the 1% significance level with no time effects.

Score test, Breusch-Pagan Lagrange Multiplier test, and Hausman tests were applied to identify the suitable model in the study. It was determined whether the analysis should be done


Table 3. Test results of cross section and time effects.

with the pooled model, the random effects model, or the fixed effects model. Both the score test and the Breusch-Pagan Lagrange Multiplier test analyze the pooled model against the random effects model. The null hypothesis suggests that the pooled model is appropriate, and that there is no random effect that reflects the existence of heterogeneity. Score test statistic was calculated as 8586.81 at 1% significance level, and the Breusch-Pagan Lagrange Multiplier test statistic was also estimated as 479.82 at 1% significance level. The null hypothesis is rejected relative to the 1% significance level. According to both tests, it is determined that the pooled model is not a suitable model. After it is defined that the pooled model is not suitable, it will be determined whether the model is a fixed effect model or a random effects model with the Hausman test. The test results are given in Table 4.

Because the Hausman test statistic was calculated as 25.46, the null hypothesis is rejected at 1% significance level. Hausman test shows that the model is a fixed effect model. The fixed effects panel data model results are given in Table 5.

The findings in Table 5 show that all predicted parameters and model are significant at 1% significance level. Modified Wald test was applied to examine heteroskedasticity in the model. The null hypothesis for the modified Wald test is constructed as:

$$H\_0 = \sigma\_i^2 = \sigma^2 \quad \text{for all } i \tag{11}$$

$$H\_1 = \sigma\_i^2 \neq \sigma^2 \tag{12}$$

The null hypothesis is rejected according to the test result at 1% significance level. There is a heteroskedasticity problem in the model. Autocorrelation was investigated with modified Bhargava et al., Durbin-Watson, and Baltagi-Wu LBI tests. The test result is assessed by


b = consistent under Ho and Ha, B = inconsistent under Ha, efficient under Ho, Ho: difference in coefficients not systematic. chi2(6) = (b-B)'[(V\_b-V\_B)^(�1)](b-B) = 25.46.

Prob > chi2 = 0.0003.

5. Empirical results

Table 2. Descriptions of the variables.

significance level with no time effects.

Table 3. Test results of cross section and time effects.

cance level.

Panel regression analysis was used to investigate the tradeoff between WCC and the profitability of the 41 firms listed on BIST Industrial Index in Turkey. In the panel data analysis, variables include both time and cross section size. According to time and cross-section effects, it is determined that the model should be predicted to be one way or two ways. For this purpose, the LR test has performed with the maximum likelihood method, and the findings are given in Table 3. The calculated test statistics are interpreted according to the 1% signifi-

The ratio of fixed assets to total assets (FATA) Fixed assets/total assets

Variables Formulas

Independent variables Cash conversion cycle (CCC) (receivables collection period + inventory

Inventory conversion period (ICP) (Inventories/cost of goods sold)365 Payables deferral period (PDP) Accounts payable/cost of goods sold)365

conversion period) payables deferral period

Short-term financial debts/short-term debts

Dependent variables Return on assets (ROA) Net profit/total asset

Control variables Sales growth (SG) Change in sales (%) The ratio of short-term financial debts to

short-term debts (FDSD)

212 Financial Management from an Emerging Market Perspective

For the two-way effects test, the null hypothesis is formed no cross section and time effects in the model. Because the value of the test statistic for the two-way effect is 279.1188 at 1% significance level, the null hypothesis is rejected. This result shows that it is a two-way effect. Then, the presence of the cross section and time effects was tested separately with the movement from the findings that it was a two-way effect. The null hypothesis for cross section effect analysis is that the standard error of cross section is equal to zero. According to the analysis results, the null hypothesis is rejected at 1% significance level, since the value of the test statistic is 262.4951. In this case, there is a cross section effect in the panel data model. The existence of time effect was also examined, and the test statistic was calculated as 3.981432 at 5% significance level. According to this result, the null hypothesis cannot be rejected at the 1%

Score test, Breusch-Pagan Lagrange Multiplier test, and Hausman tests were applied to identify the suitable model in the study. It was determined whether the analysis should be done

Tests Two-way effects Cross-section effects Time effects χ<sup>2</sup> testi 279.1188 262.4951 3.981432 Prob. 0.000 0.000 0.023

Table 4. Hausman test results.


F test stat. = 16.92 (prob. = 0.000).

Modified Wald test for groupwise heteroskedasticity: 918.72 (prob. = 0.000). Modified Bhargava et al. Durbin-Watson = 1.3899562. Baltagi-Wu LBI = 1.7238703. Pesaran test of cross sectional independence = 6.814 (prob. = 0.000). \*indicates significance at the level 1%.

Table 5. The fixed effects panel data model results.

comparing it with two values which indicate no autocorrelation. Since test statistics are smaller than 2, it can be said that it is autocorrelation. Pesaran test was performed to examine the crosssectional dependence in the model. The null hypothesis of no cross-sectional dependent is rejected at 1% significance level. For this reason, resistance fixed effect panel data model results were obtained by using in [37] estimator, which provided consistent estimates in the case of heteroskedasticity, autocorrelation, and cross-sectional dependent [35].

When the resistive fixed effects model presented in Table 6 is examined, it is seen that the coefficients do not change, but t statistics and confidence intervals calculated by using Driscoll and Kraay standard errors change. These estimates give consistent results in the case of heteroskedasticity, autocorrelation, and cross-sectional dependent.


F test stat. = 329.63 (prob. = 0.000). \*

Significance at the level 1%.

\*\*Significance at the level 5%.

Table 6. Resistance fixed effect panel data model.

According to the estimation results presented in Table 6, it was found that PDP, CCC, FDSD, and FATA have a negative effect on ROA. An increase of one-unit in PDP, CCC, FDSD, and FATA would induce a decrease of 0.0004057, 0.0004332, 0.0595096, and 0.2220324 on ROA, respectively. On the other hand, ICP and SG have a positive effect on the ROA. An increase of one-unit in ICP and SG would induce an increase of 0.000696 and 0.0847359 on ROA, respectively.

Although the studies in the literature are different in the way of both the country and the sector, similar results were obtained with other studies in the literature that a negative relationship exists between CCC which measures the efficiency of WCM, PDP, and ROA [6, 8, 22, 23, 29, 34]. Besides, the finding of this study is similar to in Ref. [27, 28] who report a positive relationship between ICP and ROA.
