• No results found

II. METHODOLOGY OF OIL PRICE AND FINANCIAL VARAIBLES IMPACT ON

2.4. Research paper’s methods and its characteristics

2.4.2. Panel data characteristics

Panel data is data which is multi-dimensional, including measurement over time (Verbeek, 2012). In our case, each object has their own specific characteristics, but all of them are from the Norwegian oil industry. Based on the literature analyzed, most analyses take place over 20 years, but we focus on a 10 year period because many of our listed companies only began issuing stocks 10 years ago, hence our focus on the period between 2007 and 2016 and the 21 companies chosen (from Table 1). We do not include Aker Solutions in this model due to lack of financial variable data.

Moreover, some financial variables (To be discussed) are transformed by using natural logarithms and calculating their growth (Formula 2). Overall, the number of observations is 178 and based on Park (2011), our observation number is suitable for significant panel data analysis. Therefore, the panel data should fully represent our selected objective.

The assumptions for regression analysis of panel data are almost the same assumptions as those previously mentioned: normality, linear in parameters, no perfect collinearity, zero condition mean, no serial correlation. The violation of those assumption would disrupt efficiency of our model and our selected variables would be biased. However, there is one assumption not mentioned in the previous section – homoscedasticity. This assumption means that random distribution in the relationship between the independent factors and the dependent factors is the same among all values of the independent factors (Wooldridge, 2014). If there is failure of the homoscedasticity assumption, then we say that the errors are heteroskedastic (Wooldridge, 2014). To avoid this kind of problem, we should use generalized method of moments (Forward – GMM) (Newey, 2007). The violation of that assumption could mislead the result we would get from regression analysis of our panel data. So, we will use the Breusch-Pagan test for checking if the error term is homoscedastic or heteroscedastic (Schmidheiny, 2016). If the value p is higher than 5 percent, then the error term is homoscedastic, and we do not need any adjustments to fix the problem. Otherwise, we need to use robust standard errors for fixing heteroscedasticity problem in our model (Williams, 2015). The robust standard errors would decrease the model’s efficiency, but it would remain unbiased. Moreover, the panel face problems with zero conditional mean violation. Therefore we will use generalized method of moments (Forward – GMM) to fix this issue (Newey, 2007).

𝑅eturns = β

0

+ β

1

Cash +

β2P

B

+

β3P

E

+ +β

4

ROE + β

5

ROA + +β

6

FAT +

β

7

CAPEX + β

8

Dividends + EG + U (3)

• Returns – Companies’ average yearly returns

• Cash – Cash ratio

• P/B – Price to book value

• P/E – Price to earnings

• ROE – Return of equity

• ROA – Return of assets

• FAT – Fixed assets turnover

• CAPEX – Capital expenditure to sales

• Dividends – Dummy variable of dividends

• EG – Enterprise value growth

The model includes 8 independent variables and one dependent variable (oil companies average yearly returns). The price to book value and price to earnings presents companies’ business value. Additionally, the ratio helps to reject or to accept the 8 hypotheses which relate to stock price and book value or price to earnings ratio. Price to book ratio is transformed by using natural logarithms and calculating percentage change during the analyzed period due to acceptance of previously mentioned assumptions. The price to earnings is transformed into a dummy variable because many companies had negative profits, therefore the ratios are a dummy variable. The zero values have data on which companies in that period do not have any information about their ratio value and one value show scenarios when the ratio is positive. Positive interaction between price to book or price to earnings and stock price would help to accept eight hypotheses.

Return on equity and assets (ROE and ROA) represents companies’ profitability. The ROE indicates how a profitable company is relative to that company’s total assets. In other words, it shows the company’s management efficiency in using its assets to get earnings. ROE ratio shows companies’ investment efficiency in using its capital to get profits. Both of those variables are transformed using natural logarithms and calculating percentage change. Therefore, a strong statistical and positive relationship between ROE or ROA and stock price would help to approve or deny the sixth hypothesis.

Cash ratio represents companies’ liquidity. The ratios are calculated by dividing cash or near to cash times (such as a short-term government bonds and other short-term financial assets) and the company’s current liabilities. Fixed asset turnover represents companies’ efficiency, and the ratio of net sales to net fixed assets. In other words, this ratio shows how companies are able to generate net sales from their net fixed assets. Both of those ratios are transformed by using natural logarithm and calculating percentage change. Therefore, the relationship between cash ratio and stock price should show if the fourth and seventh hypotheses are accepted or rejected in our case.

Capital expenditure to sales (CAPEX) is calculated by dividing capital expenditure by sales.

This measurement should also be transformed by using natural logarithm and calculation of percentage change over the analyzed period. Hence, a positive and statistical relationship between stock price and capital expenditure to sales with stock price growth will help to accept or deny the ninth hypothesis.

Enterprise values were found in the Bloomberg database and they measure specific companies’ theoretical takeover price. Enterprise value has to be transformed to enterprise value growth and which is accomplished by using natural logarithms. The positive statistical value of enterprise growth and stock price should help to accept the tenth hypothesis.

Finally, the dummy variable distinguishes between companies paying and not paying dividends and as such it has only two values – 0 and 1. The value of 1 unit means that a company in a given year paid dividends and 0 means that company did not pay any dividends. In other words, the dummy variable would deny or approve the hypothesis that companies which pay dividends increase their stock value. Again, a positive and statistically significant interaction of stock price and dividend payment shows that we can accept the fifth hypothesis.

Figure 9. Scheme of empirical analysis goals

Note: complied by the author.

Figure 9 presents conclusions of all methodological parts and presents the scheme of empirical analysis. In this work, we will use two regression models. Panel data regression results could be misleading due to heteroskedasticity. Hence, we will use the Breusch-Pagan test, which shows if our analyzed data is heteroskedastic or homoscedastic, and GMM method for fixing zero conditional mean voilation. If the data is not homoscedastic, then we use robust standard errors to fix data. The same situation could be present in our time series data regression, except that it would be impacted by autocorrelation. Therefore, we will use the Durbin-Watson test for checking if selected variables are impacted by autocorrelation or not. If we find out that some regressions violate the no serial correlation assumption, then we would use Prais-Winten estimators, which fix regression’s standard errors. Moreover, we use ARCH and GARCH for increasing less consequences from autocorrelation.

After using tests for better quality of our selected regressions, we can accept or reject our selected hypothesis. The time series 1st hypothesis would be accepted if a majority of company’s regressions have a positive and statistically significant relationship between oil price and stock price and the 2nd and 3rd hypotheses would be accepted if the relationship between stock price and exchange rate or OPEC decisions would be negative and statistically significant. Panel data regression will show if we accept or reject the last seven hypotheses. We would accept the hypotheses if the regression results would show positive and statistically significant relationships between stock price and selected financial variables. Therefore, the scheme fully presents the research paper’s empirical analysis goals.

However, our research paper has some limitation as all other research papers. Hence, we will present several limitations. First limitation could be that OPEC increase their oil supply amount only 18 times in overall all data. Moreover, the effect on oil price could be before or after announcement of oil production increment. The second limitation could be that oil price has lagging or leading impact on oil price. Therefore, the regressions results could be misleading. The third limitation could be that yearly financial variables has impact on stock price only on that day, which they announcement of those financial variables, and not on yearly stock price. the final limitation would be failure of one the assumption of regression analysis. Overall, those limitation could disrupt regression results and therefore interpretation of those results would be misleading.

III. EMPIRICAL RESULTS OF OIL PRICE AND FINANCIAL