Influence of credit crunch on performance of Norwegian companies, before and after financial crisis in 2007

Transcription

1 Influence of credit crunch on performance of Norwegian companies, before and after financial crisis in 2007 GRA 1900 Master Thesis BI Norwegian Business School Simon Kramarič Irena Kustec Programme: Master of Science in Business and Economics, Major in Finance Master of Science in Financial Economics Supervisor: Charlotte Østergaard Date of submission:

2 ABSTRACT This thesis investigates the change in performance of Norwegian companies as a consequence of banks reduced approved loans during the 2007 crisis. The first research question in the paper examines whether the companies performance, measured by level of investment and return on assets (ROA), declined due to the credit crunch. Our analysis provides significant results to confirm our prediction. In the second research question we focus on the different effect of the credit crunch on listed and non-listed companies. We find that even though both groups of the companies were affected by the credit crunch, the performance of listed companies suffered less. For our third research question we find no significant results that would allow us to conclude that the credit crunch affected high levered companies more than low levered companies. i

3 ACKNOWLEDGEMENTS We would like to thank our supervisor Charlotte Østergaard for her motivation, guidance and support through the entire process of writing the thesis. With her help our work was possible and much more enjoyable. We would also like to thank the Centre for Corporate Governance Research at BI for providing the data. Finally, we would like to dedicate this work to each other for being the best Team ever! ii

4 TABLE OF CONTENTS 1 INTRODUCTION THEORY EFFECTS OF FINANCIAL CRISIS ON REAL SECTOR IN NORWAY HOW WERE THE EFFECTS DIFFERENT FOR LISTED AND NON- LISTED COMPANIES HOW WERE THE EFFECTS DIFFERENT FOR COMPANIES WITH HIGH AND LOW LEVERAGE METHODOLOGY DATA VARIABLES Dependent variables Independent variables Dummy variable Control variables HYPOTHESIS AND MODELS FOR THE EMPIRICAL ANALYSIS Fixed effect model Hypothesis Hypothesis Hypothesis EMPIRICAL ANALYSIS DESCRIPTIVE STATISTICS RESULTS Hypothesis Hypothesis Hypothesis MODEL RESTIRCTIONS CONCLUSION REFERENCES APPENDICES Table of contents iii

5 TABLE OF TABLES Table 1: Summary statistics Table 2: The effect of bank debt leverage on company s performance with interaction between leverage and time period Table 3: The effect of listing status on company s performance with interaction between listing status and time period Table 4: The net effect of year and listed dummy interaction on investments and ROA Table 5: The effect of leverage characteristic on company s performance with interaction between time period and leverage characteristic Table 6: The net effect of year and leverage characteristic dummy interaction on investments and ROA iv

6 1 INTRODUCTION Many research studies have been done in the past, leading to the conclusion that financial crisis or banks in distress have a negative shocks on real sector. Ben Bernanke argues in his research that economic institutions have the ability to affect cost of transactions between lenders and borrowers and thus either make the process easier in normal times or in distress times make it harder. He also argues that financial intermediaries that may perform well in normal times can be counterproductive in times of distress, when shocks or policy mistakes drive the economy (Bernanke, 1983). But there are also other studies done, suggesting the correlation between distressed banks and bad consequences in real sector is not as straight forward as one might think. Banks are not so special in countries with developed capital markets, because when banks stop lending, capital markets prove to be a good substitute, with variety of non-bank funding sources to companies such as venture capital, corporate bonds and other types (Greenspan, 1999). Also sufficient competition from capital markets should prevent misallocation of funds from banks to unprofitable investments and reduce the financial crisis impact on real sector. Meaning that if banks are not the only provider of funds, in order to stay competitive they must choose to support good investment opportunities, which effectively deals with misallocation of bank funds (Rajan and Zingales, 1998)? Our contribution in this paper is to empirically test to what extent did the financial sector contributed to worsen economic situation in Norway based on data from before and after 2008 financial crisis. Only one study has been done since the crisis, by Hetland and Mjøs in Most of previous studies were done on data from prior financial crisis and many used the sample of listed companies, because they used a stock price as a key indicator. However there have been many studies done on 2008 crisis, but they were not done for Norway, because data used was from other countries. We on other hand are using a Norwegian data and a much bigger sample of companies for the empirical analysis, because we include non-listed companies as well. This 1

7 allows us to compare the effects of financial crisis on listed and non-listed companies. Another difference between our study and studies already made is that we check the credit crunch impact not only on investments, but also on return on assets as an indicator of companies performance and profitability. We intended to take into account the employment level as well, but due to noisy and incomplete data this was not possible. Our first research question is to empirically test if the financial crisis had a negative impact on the real sector of the economy. The second research question is to look into the difference of the impact on listed and non-listed companies and the third research question is to look into the difference of impact on high and low levered companies. We expect to find differences between companies, because transparency 1, which is different for listed and non-listed companies, is an important factor when raising capital from bank and non-bank sources. Also leverage is an important indicator of companies health and can be a big constrain to raise new needed capital. It is crucial to understand the differences and impacts that the crisis had on companies, so that future actions by the companies themselves and regulators might help to mitigate future possible shocks to the economy. In the theory chapter we discuss our research questions in more details and look into the theoretical part of the subject. Then in third chapter we write about the data and models that we use in the fourth chapter Empirical analysis. In the final chapter we provide concluding remarks based on what the empirical analysis shows and discuss any theoretical inconsistency with the analysis. 2 THEORY In this chapter we introduce our research questions and relevant theory in three separate sections. Based on the literature and different prior studies we write 1 Listed companies are generally more transparent compared to non-listed, because regulations are stricter for listed companies. This means that listed companies are forced to release more information to the public compared to non-listed companies and thus increasing its transparency from market perspective. 2

8 predicted conclusions for each research questions. Later on we challenge these predictions in the chapter empirical study. 2.1 EFFECTS OF FINANCIAL CRISIS ON REAL SECTOR IN NORWAY Our basic research problem is to empirically test the effects of the 2008 financial crisis on the real sector in Norway, meaning how the shocks in financial sector transformed to non-financial companies. We analytically test this on data for Norwegian companies and banks. There are many theories predicting that shocks to banks will be transferred to the real sector, making banks somewhat special. On one hand banks are not special in the sense they could be substituted with a range of other institutions that already provide the majority of services provided by banks. But on other hand banks are special because of the government support and extra regulations that reduces corporate governance control and increases government influence, meaning that the government controls the behaviour of banks through imposed regulations (England, 1991). Also Nunami (2011) argues that banks are special because they are the core of payment instruments that support all economic activities and are the credit intermediary in any economy, because although we have witness a rise of shadow banks that started to acquire market share of normal banks, they were first to fall in the financial crisis, because they were not supported by deposit insurance, which is one of the important components in so called safety net that signals and assures financial stability. Also they could not ask for help from lenders of last resort to help them with liquidity and credibility problems that all banks faced in the turmoil of the crisis. Nunami also strongly believes that key conclusions in an essay Are banks special? by E. Gerald Corrigan, that Mr. Corrigan wrote more than 30 years ago and concluded that banks are indeed special, still hold even with significant changes to financial environment since then. According to James (1987), banks are special, because they give out unique loans to companies that endure high information asymmetry, which other lenders do not. So this makes the banks specialist in this manner and as researchers found, preferred source of debt. In their analytical work they found a statistically significant positive relationship between announcement of bank loans and the 3

9 stock price reaction of the company and on other hand a negative relationship if new debt came from the source, for example, from insurance company. Besides these theoretical references on why banks are special, there are also some studies that suggest banks are special, for example a study of Continental Illinois Bank during its de facto failure and subsequent rescue by Federal Deposit Insurance Corporation (FDIC). Authors analyse share price effect of companies that had lending relationship with the bank. They find that banks impending insolvency had a negative effect on stock prices and a FDIC rescue a positive effect. Based on that, they conclude that borrowers incur significant costs due to reduction to bank durability as they are important banks stakeholders (Slovin, Sushka and Polonchek, 1993). Many studies done on the financial crisis in Japan in 1990s also support the idea that banks when in distress have a significant effect on the economy. Banking problems were identified as one of the key contributor to poor performance of Japanese economy at that time and also a serious drag that prevented faster recovery of bad economic situation (Hoski and Kashyap, 2000). Also it was argued that Japans central bank and financial keiretsu systems left corporate governance in the hand of creditors rather than shareholders. Thus governance practices did not assign effective control rights to residual claims, which lead to misallocation of capital that eventually lead to liquidity problems of Japan economy (Morck and Nakamura, 1999). At that time banks played a much more important financial intermediation role in Japan than in other countries like United States or United Kingdom and were the main provider of loans to Companies in Japan. And because banks had lent large amounts, securing the loans with real estates which prices slowly declined, they run into troubles which at least partly contributed to Japans economic disruption at that time (Bayoumi, 1999). Previous studies of financial effects of the recent crises support each other in claim that the shocks were transferred to real sector. Because of this it is not hard to assume that banks are indeed special, as they are the core of the economic transactions. By assuming this, we also assume that any shocks to financial sector can be seen in the real sector. A recent study done on European countries showed that in the 2008 financial crisis, banks credit supply was 4

10 restricted because of banks high leverage that lead to real effects in the whole economy (Buca and Vermeulen, 2012). But the study of Norwegian banking crisis in the period of 1988 to 1991 done by Ongena, Smith and Michalsen has provided with arguments that at that time banks were not so special in sense that the clear collapse of banking system, when banks either defaulted or had to be bailed out by the government, did not transform to the real sector as one would expect. The real sector felt the shock, but because the crisis was local from global perspective, companies had the possibility to acquire the needed capital outside of Norway. So we believe the correlation of shocks between financial and real sector was stronger in the recent global financial crisis (Ongena, Smith and Michalsen, 2000). Study of the sensitivity of credit supply to bank financial conditions in 16 emerging European countries before and during the financial crisis in 2007 revealed strong evidence that companies access to credit was indeed affected by the distress in the financial system, meaning that companies were more credit constrained if the bank experience a decline in equity and Tier 1 capital requirements according to Basel rules (Popov and Udell, 2010). Another study by Chava and Purnanandam found that when bank experienced capital shocks, this transformed to the companies that rely primarily on banks for capital in a way that companies experienced decreased performance and a decline in their capital expenditure and profitability (Chava and Purnanandam, 2009). This means that previous studies are consistent in supporting a fact that shocks to financial sector, especially if shocks are global, transfer to real sector in a form of credit crunch. So by doing the empirical analysis we hope to get the same conclusion, that real sector in Norway was indeed affected by the distress in banks during the latest financial crisis. A good support to our theory is a previous study by Hetland and Mjøs (2012) on Norwegian real sector using the similar data that we use. But with a significant difference in the setup of the study, because their focus was on the difference of effects of credit availability on financially constrained and financially unconstrained companies in real sector. Constrained meaning that the company already has some debt and thus is constrained by the bank, to regularly make payments or acquire additional debt, resulting that the company 5

11 is more careful. Authors were not using only the financial reports, but also a survey done among the companies. They found exactly what we are predicting to find, but also that effects of credit crunch were stronger for unconstrained companies compared to constrained ones, that normally hedge themselves against future cash flows shortfalls. Although we are doing similar study, we are expecting a bit different results because of differently defined dependent variable. 2.2 HOW WERE THE EFFECTS DIFFERENT FOR LISTED AND NON-LISTED COMPANIES After we establish that the shocks in financial sector during the financial crisis had been transferred to the real sector of economy, we ask ourselves if the negative impact they had, were different for listed companies in comparison to non-listed companies. This provides a suggestion on which company s legal organizational form and therefore different regulation, meaning disclosure requirements that force listed companies to be more transparent than non-listed which may lower the problem of information asymmetry, had cooped better with the disturbance in credit supply 2. From perspective of lenders the major cause for the difference between companies is information asymmetry. Information and incentive problems in the capital markets affect investments of companies. For example on one had we can have a manager that has incentive to raise the value of the company and has a plan to undertake many investment projects that have a positive net present value. Although this should not be a problem with lender as they would benefit from higher value of the company and from making more profit from issuing more debt, there is an issue of additional risk involved with any additional investment that the company makes. On one hand there is no problem issuing new debt for new investment, because lender asses the riskiness of the company and thus adjust the required interest rate. But on other 2 Credit supply is important for company s long and short term business. For long term business, companies use credit to fund their investments and thus generate future growth and raise possibility of acquiring additional debt. But for short term perspective, companies use credit to fund their daily operations. So if the bank suddenly stops supplying the needed credit to the company or just raises the cost of borrowing, this can mean that the company will either be in trouble of not being able to do daily business or will lose the ability to grow. In many cases if credit supply is interrupted, bout scenarios happen and the company suffers. 6

12 hand all the previous loans are already negotiated and in place according to past riskiness of the company. So even if the lender would have benefit from issuing new debt, they would lose on value of debt already issued to the company. And as mentioned before information asymmetry has a significant role as it makes the situation between the lender and the borrower even harder. Nevertheless, companies that have a stronger connection with a bank or in other words the information asymmetry is lower compared to companies that have weaker connection with a bank are more likely to raise needed capital (Hoshi, Kashyap and Scharfstein, 1991). Companies, listed and non-listed, that suffer from information asymmetry tend to have decreased productivity and profitability due to competitive pressure, than companies that has less problems with information asymmetry (Schoubben and Van Hulle, n.d.). Based on previous studies we are expecting to find and conclude that listed companies, that have more strict regulation on what information they have to supply to the market, were in a better position in raising capital during the crisis, because there was less information asymmetry factor between them and banks. Although we are expecting a positive difference, we are a bit reserved in the speculation, because of Accounting and Bookkeeping Act in Norway for listed and non-listed companies that makes Norway unique in comparison to other countries and reduces information asymmetry. 2.3 HOW WERE THE EFFECTS DIFFERENT FOR COMPANIES WITH HIGH AND LOW LEVERAGE The final research question is to empirically test if the shocks in the financial sector were transferred differently to companies with different leverage level. To distinguish between high and low levered companies, we divide all companies in the sample into three equal groups based on leverage and then compare a group with highest and lowest leverage. Although there is no negative relationship between leverage and growth if the company has good investment opportunities, the relationship is significantly negative if growth opportunities are not recognized by the capital markets or are not valued sufficiently (Lang, Ofek and Stulz, 1994). If the company has 7

13 already substantial leverage, only really good investments will be supported by the capital markets with additional funds to the company. But the fact of information asymmetry, this process is not so straight forward and companies have a hard time convincing lenders that the investments are good. Simplified relationship argued by authors is, that the higher the leverage, the better must additional investment be, for the companies to get funds from lenders. A more recent study done on Canadian publicly traded companies finds similar results, that financial leverage has a negative effect on company s growth. And the effects are significantly stronger if the company has low growth opportunities (Aivazian, Ge and Qiu, 2003). In times of crisis we can expect even more information asymmetry between company s projects and lenders, meaning that growth opportunities are not valued correctly. Also in times of crisis when all the economy is in a downstate, we do not expect many good growth opportunities. So because one of ours dependent variables, when looking at effects on companies, is investment we are expecting to find that effects of financial crisis were more severe for high levered companies than for low levered companies. But one might argue that high levered companies were smarter so they had more good investment opportunities before crisis and thus have good opportunities even in the crisis. But this is just speculation and thus it is even more interesting to see what the data shows. Researchers agree that companies with lower level of operating leverage usually experience lesser variability in their ROA than companies with higher levels of operating leverage. Lang, Ofek and Stulz found that low levered companies have higher exposure to systematic risk and lower probability of distress than high levered ones, which implies a negative relationship between expected returns and level of leverage (Lang, Ofek and Stulz, 1994). Also a study of Austrian high levered companies in hotel industry showed that smaller reductions in leverage were followed with larger increases in ROA (Giroud, Mueller, Stomper and Westerkamp, 2011). Even though ROA eliminates the effect of leverage, when a company finances itself with debt, we still expect to find significant difference between effects of financial crisis on high levered companies compared to low levered companies. 8

14 3 METHODOLOGY This chapter explains the methodology used for the analysis. The first part describes data and how it is selected, following by definition of chosen variables used in the regression analysis. At the end of this chapter, statistical model is discussed in details. 3.1 DATA Like Hetland and Mjøs (2012) did in their study of financing during the crisis, we will also use detail microeconomic company-level data for Norwegian companies from year 2004 to 2010 which is made available by the Centre for Corporate Governance Research (CCGR) at the Department of Financial Economics at BI Norwegian Business School. All Norwegian companies, private and public, are required to file their annual financial accounts with the Register of Company Accounts. Used accounting database includes a lot of relevant information about companies, such as balance sheets, profit and loss accounts, selected items from the notes to the accounts and five-digit industry codes with legal forms. Therefore we believe accuracy of the data is good and that we have enough data to compare listed with non-listed companies and high leveraged companies with low leveraged. Based on our hypothesis we will set up criteria that will help us narrow the sample used in the regression models. First we will choose a time period of The pre-crisis variables will be set to the end of year Norwegian economy had returned to positive GDP growth before the end-ofyear 2009 which Hetland and Mjøs (2012) used as the end of the financial crisis. But since the idea of this paper is to investigate how the crisis influenced the companies activities during and after the crisis, we will set the post-crisis variables to the end of year We will then have 3 years of observations for pre-crisis period and 3 years for post-crisis. This timing is likely to give a good reflection of the occurred events since our data is of annual frequency. Furthermore, for examining how the credit crunch caused by banks affected real sector, we restrict our sample to only non-financial companies. We do that by excluding all companies in the financial services industry according to 9

15 NACE codes (see Appendix 1). To avoid potential misclassification errors, we exclude from data companies with no borrowing from banks (companies with zero debt). Our final sample contains 27,309 unique companies and 119,834 company years of data, of which we have 57,144 observations for the pre-crisis period of 2005 until 2007, and 62,690 observations for years from 2008 to In addition, we also focus solely on the 24 manufacturing industries defined with NACE codes (see Appendix 2 for details). In this final pool of manufacturing companies we are left with 20,308 observations of 4,106 unique companies. This data subset contains 10,151 observations for pre-crisis period and 10,157 observations for post-crisis period. For all statistical analysis we use Stata, the data analysis and statistical software. 3.2 VARIABLES This section describes variables in the statistical analysis and the proxies that will be used to obtain the results. Variables chosen and their mathematical expressions are based on earlier empirical studies. The specific reason for choosing them will be explained. We use company-specific variables (variables that relate to observations within a company) that are most commonly used in previous empirical testing. For more detail construction of all the variables see Appendix Dependent variables First we define our dependent variable. In different studies different proxies have been used as a measurement of the effect of financial crisis on company s performance. Studying how the distress in banks affected their borrowers in Japan during the 1990s, Akiyoshi and Kobayashi used the Total Factor Productivity level for each observed listed company (Akiyoshi and Kobayashi, 2010). Even though, when studying the performance of listed companies (Ongena, Smith and Michalsen (2000), Chava and Purnanandam (2009)) realized stock returns seems to be the most straight forward measurement. But since we are not just looking at the effect on the listed companies but also on non-listed, we will choose a different measure. Hoshi, Kashyap and Scharfstein (1991), Buca and Vermeulen (2012), Hetland and Mjøs (2012) all used function of investments as their dependent variable. 10

16 The first two sets of researchers used normalized investments, while Hetland and Mjøs (2012) used investments in PME (plant, machinery, equipment) and investments in buildings. We follow similar approach that we wanted to complement it with human resource variable. Companies react differently on credit crunch and we believe that annual investments and employment growth can explain how companies adapted. When company is limited in raising funds, it can no longer invest in new projects or support the existing ones, therefore investments decrease. This can also lead to lower number of employees. But since data for company s employment is very noisy, proper analysis cannot be conducted. Thus our first dependent variable is defined as investment-capital ratio:. As an indicator of company s profitability we define another dependent variable. Return on assets ROA tells us how the company uses its assets and determines whether the company is able to generate adequate return on these assets. ROA is computed as: Independent variables We test whether high levered companies have higher cost of capital and are therefore most affected by the credit crunch. In the first test we follow the model of Buca and Vermeulen (2012), arguing that bank debt always matter, and we define an independent variable as bank debt leverage, which captures the information about the company's level of risk. In order to avoid reverse causality, we lag our dependent variable for one year. Leverage at the end of time t-1 is thus defined as liabilities to financial institutions (FI Liab i,t-1 ) normalized by company s capital (K i,t-1 ):. Companies bank debt leverage is used as a proxy for asymmetric information between the borrower and the lender. To see whether there is a significant difference between listed and non-listed companies we use a different proxy, 11

17 dummy variable D i,t listed. This dummy variable equals to 1 if company i is listed on Oslo Børs or Olso Axcess, a Norwegian regulated and licensed market, in year t and 0 otherwise. It is more complicated to define the independent variable for our third research question of the role of credit crunch on real sector. We want to know whether companies that went into crisis with higher level of leverage were harder hit by the credit crunch. In order to answer this question we define leverage characteristic dummy variable. To do that, many researchers use percentile approach, where statistical distribution of companies leverage ratio is used. Minton and Wruck (2001) definition of low levered company was the bottom 20% of the distribution of all companies. We will follow similar approach. To define whether company is low or high levered we will first compute the industry adjusted leverage ratio in the year 2005 (see Appendix 4). Then we divide our distribution into three thirds, and omit the middle one. We define leverage characteristic dummy variable D i low as a dummy variable with value of 1 if company i is in the bottom third of the distribution. The reason for using industry adjusted leverage ratio is that we want to neutralize permanent differences in the amount of debt caused by industries specific characteristics. Industry adjustments are made based on annual means for all companies with the same two-digit NACE code Dummy variable Our questions refer to how credit crunch affected the real sector during the crisis. We compare company s performance before and after the crisis with introducing a dummy variable D i,t 2008 for the company i in year t, which equals to 1 if the observation happened before year 2008 (i.e. observations from year 2005 to 2007) and 0 otherwise Control variables We also have to define control variables, i.e. variables that can affect the relationship between the dependent and independent variables. If left out, the effect of the omitted variable increases the correlation between independent variable and the error term, which results in omitted variable bias. Our main concerns are company s size and cash flows. There are a lot of reasons to believe that small businesses do not get the credit they need during the crisis period. Prior study found that the impact of the credit crunch in the late 90s in 12

18 Japan had non-negligible effect on aggregate investment of smaller non-listed companies (Ogawa, 2003). Akiyoshi and Kobayashi s (2010) defined company s size as logarithm of total assets, Chava and Parnanandam s (2009) as logarithm of sales. Our proxy for company s size is defined as: where REV i,t stands for company s annual revenue. Fazzari, Hubbard and Petersen (1988) found that investments are sensitive to cash flows. Cash flows thus became customary in the financing constraint literature, so like Buca and Vermeulen (2012) we also use cash flow capital ratio (CF i,t / K i,t ) as our control variable. 3.3 HYPOTHESIS AND MODELS FOR THE EMPIRICAL ANALYSIS This section will state hypotheses tested in this paper and explain the statistical model applied to the panel dataset in order to conduct a simple multivariate OLS regression Fixed effect model The model and the following assumptions are mainly derived from the book of Wooldridge (2012). Sometimes when conducting an OLS regression on the panel dataset, the results might suffer from omitted variables, which results in bias results. Meaning that, a lot of explanatory variables, which are strongly exogenous, are left out in the regression analysis and thus they cannot be directly controlled for. We assume that these unobserved factors are mostly company-specific factors that do not vary much over time. To discard all variation between individual companies and use only variation over time within an individual company, we impose time independent effects for each company and use fixed effects model as our statistical model. A general panel data regression model with fixed effects is then written as The main insight for the fixed effect estimator is that if we difference the average of all company s observations, the,. cancels out: 13

19 . The fixed effect estimator is then obtained by an OLS regression of on. The choice of method to avoid biased results depends on general assumptions which we will have to test. Since even then we can experience some problems, we will try to complement the model to some extent Hypothesis 1 Credit crunch affected real sector s performance during the crisis. First hypothesis connects credit availability with real sector s performance during the crisis. As already mentioned, real sector s performance is measured as investment-capital ratio and as return on assets ROA. As independent variable we use bank debt leverage LEV i,t-1, which we combine with dummy variable D i,t Since the effect on the dependent variables of one predictor depends on the value of other, we introduce the interaction term, which is represented by their product that is by the variable created by multiplying them together. Z is vector of control variables. To compare the regression coefficients between two groups (between pre-crisis and post-crisis period) we use two-way interaction model: where α i is company i s unobserved fixed effect and 14 an error term for company i. The chosen model is strongly believed to capture the essence of the subject under study. The constant term in the model α represents the intercept of omitted group (value of dummy variable equals zero), in our case these are observations from post-crisis period. Coefficient β 1 corresponds to the difference of the intercepts from the separate group analysis. The information about slopes of regression lines is gathered in coefficients β 2 and β 3. β 2 is the slope of omitted group, while difference in means between levels of LEV i,t-1 and D i,t 2008 is quantified by the value of β 3. We expect the regression coefficient for pre-crisis and post-crisis period will differ (slopes of regression lines will differ), which will be proven by significance of coefficient β 3. Detailed algebraic model description is available in Appendix 5.,

20 It is already argued that global shocks to financial sector are transferred to real sector (Chava and Purnanandam (2009), Popov and Udell (2010)). And even though the similar study on Norwegian companies during the Norwegian banking crisis in 1990s was already done by Ongena, Smith and Michalsen (2000); we expect our results to differ from theirs since the current financial crisis had major impact on global economy Hypothesis 2 Non-listed companies were more affected by the credit crunch. In addition to test the difference in the effect of the credit crunch on listed and non-listed companies we use a listed dummy variable D i,t listed as an independent variable. Interaction between explanatory variables is represented by their product. Z is vector of control variables. Model that we will use is defined as: where is company i s unobserved fixed effect and an error term for company i. Similar to model for hypothesis 1, the constant term α represents the intercept of omitted group (value of both dummy variables equal to zero), in our case this are observations of non-listed companies from post-crisis period. Coefficient β 1 measures the net effect of crisis on non-listed companies. The same effect for listed companies is measured by the sum of coefficients β 1 and β 3. To compare the difference between these two effects we use coefficient β 3. We expect negative and significant coefficient β 3, which would confirm our hypothesis that non-listed companies were more affected by the credit crunch. Detailed algebraic model description is available in Appendix 6. Ongena, Smith and Michalsen (2000) already showed that lower information asymmetry brings to fewer problems in raising new capital. We believe that listed companies have more restricted regulations which results in lower information asymmetry. Schouben and Van Hulle s study suggests that listed companies performance either stagnates or even increases. Our hypothesis is also supported by Hetland and Mjøs paper where they came to the conclusion that effect of credit crunch was stronger for financially unconstrained companies and also for non-listed ones. 15

21 3.3.4 Hypothesis 3 High leveraged companies were more affected by the credit crunch. Model used to test our third hypothesis is identical to previous model. We only replace independent variable with D i low dummy variable that specifies whether company is low or high levered. Our model is presented by equation: The constant term α represents the intercept of omitted group (value of both dummy variables equal to zero), in our case this are observations of high levered companies from post-crisis period. The net difference in performance for high levered companies in the pre- and post-crisis period is measured by the coefficient β 1. The same effect for low levered companies is measured by the sum of coefficients β 1 and β 3. Like when testing hypothesis 2, we use coefficient β 3 to compare the difference of crisis effect between two defined groups of companies, high and low levered. In order to confirm our hypothesis that low levered companies were less affected by the credit crunch, we expect significantly negative coefficient β 3. Detailed algebraic model description is available in Appendix 6. As already discussed above, due to increased information asymmetry during the crisis years, we are positive to find negative correlation between financial leverage of the company and its investments and employment growth. Our assumptions are made in correlation with results obtained by Aiyazian, Ge and Qui (2003) and Sharpe (1994). 16

22 4 EMPIRICAL ANALYSIS This chapter focuses on descriptive statistics and the final regression outputs. First summary statistics of dependent and explanatory variables are discussed, following by analysis of regression results from fixed effect estimation in order to accept or reject our hypotheses. 4.1 DESCRIPTIVE STATISTICS In this section we present basic descriptive statistics on the total sample and on manufacture subsample. Table 1 provides summary statistics of dependent and independent variables we use in our analysis. Detailed summary statistics, separated for listed/non-listed companies and for high/low levered companies are available in the Appendix 7 and Appendix 8. In our full sample of companies that exclude companies operating in financial industry we observe 24,444 unique companies from year 2005 to year The average bank leverage (liabilities to financial institutions) in this period is percent with a large dispersion between minimum and maximum value. As a consequence of credit crunch mean values of leverage are lower in postcrisis period (3.646) than in pre-crisis (3.993). We also observe that investments and return on assets are dependent on the time period. Before the crisis occurred mean values were twice as high as they were after the crisis. Table 1: Summary statistics Summary statistics for selected financial characteristics we use in the regression analysis. Reported are average and median value with standard deviation, minimum and maximum value for the full time period and two sub periods, pre-crisis and post-crisis period. N indicates the number of observations, the number of unique companies is reported in parenthesis. The summary statistics are shown both for both samples; Panel A shows summary statistics of our full sample including only non-financial companies with non-zero bank debt and Panel B sample of companies in manufacturing industry. Investments are measured as investmentcapital ratio, ROA as earnings divided by total fixed assets, and leverage is represented by normalized total liabilities to financial institutions. For proxy for size Log (REVi,t) is used. All financial characteristics are obtained from CCGR database, measured in Norwegian kroner, and are measured at fiscal year-end. 17

23 Panel A: full sample N mean median st.dev. min max full period ( ) 119,834 (24,444) investments ,657 ROA ,771 2,113 leverage ,564 9,500 lagged leverage ,500 pre-crisis period ( ) 57,144 (22,250) investments ,657 ROA ,866 1,402 leverage ,500 lagged leverage ,863 post-crisis period ( ) 62,690 (23,596) investments ,449 ROA ,771 2,113 leverage ,564 8,833 lagged leverage ,500 Panel B: manufacturing industry N mean median st.dev. min max full period ( ) 20,308 (4,106) investments ROA leverage ,551 lagged leverage ,257 pre-crisis period ( ) 10,151 (3,715) investments ROA leverage ,551 lagged leverage ,257 post-crisis period ( ) 10,157 (3,639) investments ROA leverage lagged leverage The same happens in our manufacturing sample where we observe 4,106 unique companies, but here the changes are even larger. Investments drop from to 0.198, while ROA changes from to Here we also observe that both, minimum and maximum values are lower. Similar to the full sample, 18

24 also companies in manufacturing industry have in average larger leverage in the pre-crisis period (2.717) than in the post-crisis (2.293). Looking at summary statistics separated for listed and non-listed companies (Appendix 7) we notice that all selected accounting variables are lower for listed than non-listed companies. But we must bear in mind that such comparisons are very difficult due to small number of listed companies (29) in comparison with non-listed (27,289). An interesting thing we observe for listed companies in both of our samples is that average leverage increases in the postcrisis period; from to in the full sample, and from to in manufacturing sample. This increase is not significantly high, but leverage mean does not decrease. There are some interesting findings comparing summary statistics of low and high levered companies (look Appendix 8). According to our definition of low levered companies it is not surprising that average leverage of low levered companies is lower than of high levered ones. Low levered companies also have lower rates of investment. The difference is bigger in post-crisis period. In the pre-crisis period our full sample average investments of high levered companies is and for low levered In the post-crisis period average values are and Difference is even bigger for manufacturing sample. The reason why low levered companies tend to have lower investment rates may lay in the fact that these companies had few investment opportunities in the past. The picture is somewhat similar with return on assets. But here the ROA of low levered companies decreases after the crisis, and increases for highly levered ones. ROA of low levered companies in the post-crisis period is negative for both of our samples ( for the full sample and for manufacturing one). Another interesting finding is that the average and median leverage of low levered companies increases from pre- to post-crisis period while it decreases for high levered companies. This happens for both of our samples. In the full sample, leverage increases from to for low levered companies, and decreases from to for high levered. In the sample containing only companies operating in the manufacturing industry differences are even bigger; leverage increases from to for low levered companies, and decreases from to for high levered. 19

25 Therefore it seems that high levered companies were more affected by the credit crunch, following our third hypothesis we expect their performance suffered more than performance of low levered companies. 4.2 RESULTS In this section we present regression results on our sample and connect them with our research questions and hypotheses Hypothesis 1 The regression output using fixed effect estimation is presented in Appendix 9, summary is given in Table 2. In this regression independent variable is regressed on each of dependent variables (columns (1) and (3) show results using investment-capital ratio as dependent variable, while columns (2) and (4) use return on assets) for our two samples; first for the sample of non-financial companies with bank borrowings (columns (1) and (2)), and for companies in manufacture industry (columns (3) and (4)). In our regression standard errors are made robust to control for heteroskedasticity. The test statistics of all four estimations results suggest that there is no problem with specification of our model nor with used instruments, thus that all coefficients in the model are different than zero. Table 2: The effect of bank debt leverage on company s performance with interaction between leverage and time period Dependent variable is investment-capital ratio INV i,t / K i,t-1 (model (1) and (3)), and return on assets ROA (model (2) and (4)). Models (1) and (2) include non-financial companies with nonzero bank debt, while models (3) and (4) run regression only on companies in manufacturing industry. Dummy variable D i,t 2008 equals 1 for observations before the beginning of year Control variables used are log (REVi,t) as proxy for company s size, and current and lagged normalized cash flows. All explanatory variables are observed annually. Each regression includes constant term. In the table coefficients of fixed effect regression are reported, standard errors are in parentheses. Regression is absorbed for company s fixed effect and implemented with robust option, which enables standard errors to take into account possible issues concerning heterogeneity and lack of normality. Sample period is

26 variable coefficient (1) (2) (3) (4) 2008 D i,t β ** -0.83*** -0.77*** (0.10) (0.22) (0.18) (0.23) LEV i,t-1 β *** 0.18*** ** (0.01) (0.02) (0.07) (0.12) D 2008 i,t * LEV i,t-1 β *** *** 0.67*** (0.04) (0.09) (0.12) (0.16) variable coefficient (1) (2) (3) (4) log (REV i,t ) γ *** *** (0.04) (0.08) (0.06) (0.09) CF i,t / K i,t γ ** * (21.76) (136.89) (97.37) (271.65) CF i,t-1 / K i,t-1 γ * * (41.59) (232.76) observations 110, ,390 19,243 19,476 number of groups 25,765 25,982 4,036 4,058 Prob > F *, **, and ***, significant at the 10, 5, and 1 percent level, respectively. What we find is that the only time β 1, which captures the information about the difference in intercepts of pre-crisis and post-crisis period, is positive and significant when we regress lagged leverage on company s return on assets for our full sample (model (2)). This implies that before the crisis companies ROA was higher when debt was zero or close than in period after the crisis occurred. For only manufacturing companies (models (3) and (4)) holds that both, investments and ROA, are significantly negative at 1 percent level, which means that levels of both, investments and ROA, are lower after the crisis. Before year 2007 economic conditions were in favour of bank borrowing, thus companies gained from being levered which confirms also the market-timing theory. In order to answer our research question we focus more on analysing coefficients β 2 and β 3. For better understanding of regression results we also provide graphical presentation of regression lines. Regression lines are obtained by running the same regression as above but without including control variables. Results of these regressions barely differ and are summarized in Appendix 10 and graphs are available in Appendix 11. First we focus on regression on our full sample, so on results from models (1) and (2). In both models β 2 is significantly positive at 1 percent level, meaning 21

27 that lagged leverage has positive effect on level of investment and ROA in the period after the crisis. Coefficient β 3 is also positive, but significant only in model (1), where positive effect of lagged leverage on investments is even stronger in the years before the crisis. Reason why this occurs is that economic conditions before the end of the year 2007 were very good, which for companies meant that besides easier access to banks borrowings, they also had a lot of other different funding possibilities, including strong internal funding due to big profits of most companies in the period of economic boom. In the model (2), where we regress lagged leverage on ROA, positive effect of lagged leverage does not significantly change. The results are a bit different for the sample containing only companies in manufacturing industry (models (3) and (4)). For these companies the overall effect of lagged leverage is significantly positive in pre-crisis period; for both models sum of coefficients β 2 and β 3 is significantly different and higher than zero. This tells us that companies invested more when lagged leverage is higher, which is not surprising since leverage is mostly used for investments. In the post-crisis period the positive effect of lagged leverage is diminished. Coefficient β 2 is negative at 1 percent significance level in model (4), which implies that the lagged leverage has negative effect on ROA. This suggests that for companies in manufacturing industry higher lagged leverage also meant lower performance in the years after the crisis. In model (3) we find no evidence for significant effect of lagged leverage on investments in the post-crisis period. Our results confirm our first hypothesis that after the crisis hit, performance of the companies declined due to the bank credit crunch and also that companies in manufacturing industry, our subsample, suffered more Hypothesis 2 Before focusing on results of testing hypothesis 2 we ran an extra fixed effect regression to see whether non-listed companies in our samples have in general lower performance than listed ones. The regression is identical to one with which we tested hypothesis 1, time dummy variable is only substituted with listed dummy variable. Summary of regression output in available in Appendix 12. We focus purely on coefficient β 3. We notice that this estimator is significant only for model (1) with negative effect of lagged leverage on 22

28 investment-capital ratio on our whole sample. But the result of F-test exceeds the critical value, which is the indicator that statistical model does not fit well to the data set. For obtained results from other regressions we cannot conclude that listed companies have significantly different performance than non-listed in our observation period. In order to test the second hypothesis we use a two-way interaction in fixed effect regression. The regression output can be found in Appendix 13, summarized in Table 3. Like in regression for testing hypothesis 1, independent variable is regressed on each of dependent variables (columns (1) and (3) show results using investment-capital ratio as dependent variable, while columns (2) and (4) use return on assets) for our two samples; first for the sample of non-financial companies with bank borrowings (columns (1) and (2)), and for companies in manufacture industry (columns (3) and (4)). In order to control for heteroskedasticity we use robust standard errors in our regression analysis. The test statistics of estimations results of models (1), (2) and (4) suggest that there is no problem with specification of our model neither with used instruments, thus all coefficients in the model are different than zero. But the result of the F-test for model (3) exceeds the critical value of 0.05, which is the indicator that the model used is not suitable for our data, thus we cannot make any conclusions from that model. Our interest lies in finding if changes caused by the credit crunch in investment level and ROA are different between listed and non-listed companies. In order to easier interpret the results of our analysis, we combine the regression results with the substantive net effect of listing status on investments and ROA, which is presented in Table 4. Regression coefficient β 1 represents the net effect of crisis on investments and ROA of non-listed companies. For listed companies this effect is captured by the sum of coefficients β 1 and β 3. We also test with usual t-test whether the effect is statistically different from zero. As mentioned before, we are focusing only on models (1), (2) and (4) due to unsuitable coefficient results in model (3). First we focus on coefficient β 1 that tells us how big the change in level of investment and ROA in non-listed companies was in the period before and after the crisis. In all three analysed 23

29 models β 1 is positive and significant at 1 percent significance level. The positive net effect indicates that for non-listed companies performance significantly decreased due to the crisis. What we observe from Table 4 is that the net effect of the crisis of listed companies, captured by the sum of coefficients β 1 and β 3, is only significant in model (4). This significantly positive effect tells us that performance of listed companies in manufacturing industry before the crisis was higher than after the crisis. On the other hand, in our full sample of companies, excluding the ones in financial industry, (models (1) and (2)) level of investment and ROA did not significantly change. Table 3: The effect of listing status on company s performance with interaction between listing status and time period Dependent variable is investment-capital ratio INV i,t / K i,t-1 (model (1) and (3)), and return on assets ROA (model (2) and (4)). Models (1) and (2) include non-financial companies with nonzero bank debt, while models (3) and (4) run regression only on companies in manufacturing industry. Dummy variable D i,t 2008 equals 1 for observations before the beginning of year Dummy equals to 1 if company i is listed on Oslo Børs or Oslo Axcess, 0 otherwise. Control variables used are log (REVi,t) as proxy for company s size, and current and lagged normalized cash flows. All explanatory variables are observed annually. Each regression includes constant term. In the table coefficients of fixed effect regression are reported, standard errors are in parentheses. Regression is absorbed for company s fixed effect and implemented with robust option, which enables standard errors to take into account possible issues concerning heterogeneity and lack of normality. Sample period is variable coefficient (1) (2) (3) (4) 2008 D i,t β *** 0.42*** 0.20* 0.37*** (0.04) (0.07) (0.12) (0.11) listed D i,t β ** ** -0.36*** (0.09) (0.68) (0.04) (0.14) 2008 listed D i,t * D i,t β *** -0.78** -0.16* (0.06) (0.31) (0.09) (0.15) log (REV i,t ) γ *** *** (0.05) (0.21) (0.20) (0.14) CF i,t / K i,t γ (8.69) (696.86) (120.75) (190.82) CF i,t-1 / K i,t-1 γ (20.77) (159.82) observations 110, ,390 19,243 19,476 number of groups 25,765 25,982 4,036 4,058 Prob > F *, **, and ***, significant at the 10, 5, and 1 percent level, respectively. 24

30 The answer to the question about which companies, non-listed or listed, performance was hurt more in the crisis is given by the coefficient β 3. Comparing first the net effects in models (1) and (2) we observe negative and significant β 3 for both models. This means that in our full sample the change in performance, that was the consequence of credit crunch in the crisis, is significantly smaller for listed companies than for non-listed. Table 4: The net effect of year and listed dummy interaction on investments and ROA Net change reports how investments and ROA change due to the crisis for non-listed and listed companies. For first group of companies the information about the net change is already captured by regression coefficient β 1, the net change for listed companies is represented by the sum of coefficients β 1 and β 3. P-value is obtained after basic test of linear hypothesis that states that net change estimation equals zero. net change p-value model (1) non-listed companies (D_listed = 0) β 1 = listed companies (D_listed = 1) β 1 + β 3 = model (2) non-listed companies (D_listed = 0) β 1 = listed companies (D_listed = 1) β 1 + β 3 = model (3) non-listed companies (D_listed = 0) β 1 = listed companies (D_listed = 1) β 1 + β 3 = model (4) non-listed companies (D_listed = 0) β 1 = listed companies (D_listed = 1) β 1 + β 3 = The results differ in model (4). Coefficient β 3 is not significant, which indicates that there is no significant difference in performance s change for listed and non-listed companies in manufacturing industry. The performance of both groups of companies decreased at equivalent level in the years after the crisis peak. Thus we can conclude that for the full sample, excluding companies in financial industry, hypothesis 2 is confirmed. Even though in the subsample of companies in manufacturing industry, non-listed companies performance did not suffer more due to the credit crunch Hypothesis 3 Before testing our third hypothesis we run an extra fixed effect regression to see whether companies performance is different between companies we 25

31 identified as low and high levered. This regression is identical to one we used for testing hypothesis 1, time dummy variable is only substituted with low levered dummy variable. Summary of regression output in available in Appendix 14. First, we notice that coefficient corresponding with the leverage characteristic dummy variable is omitted from the regression. This is the consequence of strong relationship between the definition of low levered dummy variable and independent variable in our regression. But since we are interested in the difference in performance between low and high levered companies, we focus purely on coefficient β 3. We notice that this estimator is significant only for model (4). Negative sign of this coefficient tells us that ROA for low levered companies in manufacturing industry is significantly lower than for their high levered peers. For obtained results from other regressions we cannot conclude that low levered companies have significantly different performance results than high levered companies in our observation period. Similar as before we test our third hypothesis by using a two-way interaction between leverage characteristic and year dummy variable in fixed effect regression. Summary of the regression outputs (Appendix 15) is presented in Table 5. Like in regression for testing hypothesis 1 and 2, independent variable is regressed on each of dependent variables (columns (1) and (3) show results using investment-capital ratio as dependent variable, while columns (2) and (4) use return on assets) for our two samples; first for the sample of non-financial companies with bank borrowings (columns (1) and (2)), and for companies in manufacture industry (columns (3) and (4)). In order to control for heteroskedasticity we use robust standard errors in our regression analysis. The test statistics of all four estimations results suggest that there is no problem with specification of our model nor with used instruments, thus that all coefficients in the model are different than zero. Result of the F-test for model (3) suggests that model used is not appropriate for our data, following that results cannot be used in our analysis. The question we want to answer is whether the credit crunch changed the level of investment and ROA differently for companies we define as low and high levered. The substantive net effect of leverage characteristic on investments 26

32 and ROA is presented in Table 6. The net effect of high levered companies is captured in coefficient β 1 and the effect for low levered companies is captured by the sum of β 1 and β 3. In order to conduct our analysis we use t-test to test if coefficients statistically differ from zero. Again, we are looking only at models (1), (2) and (4). Coefficient β 2 is, because of the definition of leverage characteristic dummy variable, omitted from the regression. Table 5: The effect of leverage characteristic on company s performance with interaction between time period and leverage characteristic Dependent variable is investment-capital ratio INV i,t / K i,t-1 (model (1) and (3)), and return on assets ROA (model (2) and (4)). Models (1) and (2) include non-financial companies with nonzero bank debt, while models (3) and (4) run regression only on companies in manufacturing industry. Dummy variable D i,t 2008 equals 1 for observations before the beginning of year Dummy variable D i,t low equals 1 if company i s industry adjusted leverage ratio in the year 2005 is in the bottom third of industry adjusted leverage ratio distribution. Control variables used are log (REVi,t) as proxy for company s size, and current and lagged normalized cash flows. All explanatory variables are observed annually. Each regression includes constant term. In the table coefficients of fixed effect regression are reported, standard errors are in parentheses. Regression is absorbed for company s fixed effect and implemented with robust option, which enables standard errors to take into account possible issues concerning heterogeneity and lack of normality. Sample period is variable coefficient (1) (2) (3) (4) 2008 D i,t β ** 0.38* ** (0.09) (0.20) (0.15) (0.10) low D i,t β 2 (omitted) (omitted) (omitted) (omitted) 2008 low D i,t * D i,t β * 0.71 (0.11) (0.31) (0.16) (0.63) log (REV i,t ) γ *** *** (0.04) (0.29) (0.11) (0.13) CF i,t / K i,t γ (8.88) (719.04) (156.28) (209.34) CF i,t-1 / K i,t-1 γ (21.72) (196.87) observations 63,405 63,870 11,936 12,039 number of groups 12,998 13,015 2,335 2,339 Prob > F *, **, and ***, significant at the 10, 5, and 1 percent level, respectively. 27

33 First we focus on β 1 that tells us how investments and ROA changed due to the crisis for high levered companies. In all three models this coefficient is significantly positive, meaning that high levered companies performance measures were higher in the years before the crisis. The next question we want to know the answer to is how credit crunch affected investments and ROA for low levered companies. The sum of β 1 and β 3 is significantly positive at 1 percent level only for models (1) and (2), which indicates that for our full sample low levered companies performance decreased due to the crisis. For model (4) we do not have any evidence that performance of low levered companies in manufacturing industry changed significantly in the crisis period. Table 6: The net effect of year and leverage characteristic dummy interaction on investments and ROA Net change reports how investments and ROA change due to the crisis for high levered and low levered companies. For first group of companies the information about the net change is already captured by regression coefficient β 1, the net change for low levered companies is represented by the sum of coefficients β 1 and β 3. P-value is obtained after basic test of linear hypothesis that states that net change estimation equals zero. net change p-value model (1) high levered companies (D_low = 0) β 1 = low levered companies (D_low = 1) β 1 + β 3 = model (2) high levered companies (D_low = 0) β 1 = low levered companies (D_low = 1) β 1 + β 3 = model (3) high levered companies (D_low = 0) β 1 = low levered companies (D_low = 1) β 1 + β 3 = model (4) high levered companies (D_low = 0) β 1 = low levered companies (D_low = 1) β 1 + β 3 = At the end we need to compare the differences of crisis effect on low and high levered companies. This information is assembled in the coefficient β 3. β 3 is not significant in any of the models, meaning that because of the credit crunch during the crisis low levered companies performance did not suffer less than the performance of high levered companies. With such results we can reject our third hypothesis that high leveraged companies were more affected by the credit crunch. 28

34 4.3 MODEL RESTIRCTIONS In order to make proper conclusions we test our sample for heteroskedasticity, serial correlation and normality of residuals. Heteroskedasticity means that the time-varying error does not have constant variance. We must note that this does not result in biased parameter estimates, but it does result in biased standard errors which affects t-statistics and confidence intervals. Observing too high or too low significance can bring to wrong conclusions. One way to deal with heteroskedasticity is to use robust standard errors in our regression analysis. The robust option in Stata relaxes the assumption that the errors are identically distributed. This does not change coefficient estimates, but because the standard errors are changed the test statistics give us reasonably accurate t- and p- values. Despite using robust standard errors, we still test our sample for heteroskedasticity. Results of modified Wald test for groupwise heteroskedasticity in fixed effect regression model can be found in Appendix 16. For all hypotheses and models we can reject the null hypothesis of homoscedasticity, from which we can conclude that our estimates still suffer from heteroskedasticity. The reason for this to occur might be that some variables are omitted from the regression or that there is a measurement error within the variables. The presence of serial correlation (autocorrelation) is the cross-correlation of time-varying error with itself. Serial correlation causes the standard errors of the coefficients to be smaller than they actually are and higher R-squared. We use Wooldridge test for serial correlation in panel-data models, results are presented in Appendix 17. The autocorrelation is present only in models (2) and (4) of hypothesis 2. That kind of cyclical effect is quite common in time series data; reason for it can be the possibility of momentum factor on performance measures of listed and non-listed companies or uncorrected fitted model for testing the hypothesis. Tests for the rest of hypotheses and their models show not to be significant, hereby we fail to reject the no serial correlation hypothesis. We conclude that at this point we do not have enough evidence to suggest that first-order autocorrelation is present in our data. 29

35 At last, we test idiosyncratic errors for normality. For that we use skewnesskurtosis test. Both, skewness and kurtosis, show how the distribution of a variable deviates from the normal distribution. In Appendix 18 it is possible to see test results for all hypotheses and models. Since the p-values associated with combined chi-squared test statistic of skewness and kurtosis combined is very small, we reject the null hypothesis of normality. Therefore it cannot be said that the fixed effects estimator is normally distributed and neither that the t and F statistics have the exact t- and F distribution. Still we have to be careful when using skewness-kurtosis test. The test tends not to give good information how the data is violating normality and it tends to reject null hypothesis of normality very easily. Because of large number of data points available nonnormality has no real effect on the linear regression's tests, for which it is only possible to rely on asymptotic approximations. 5 CONCLUSION We start our research with three questions. First and main question, which was already researched by Hetland and Mjøs in 2012, is whether the recent financial crisis, consequently the credit crunch, had influenced the real sector of the economy in Norway. We find conclusive evidence that level of investment and ROA is significantly lower during and post crisis than prior the crisis. As this was expected because of the severity of the financial crisis, that had a global impact, we then turned our attention to the question, if the effects were different for listed and non-listed companies. Although we find that both groups of companies experienced the decrease in performance, we conclude based on the data, that listed companies suffered less compared to non-listed ones. This is because listed companies had other means at their disposal to acquire funding for potential investments and also banks favour lending to listed companies. Even though we are confident in our findings there are some limitations to our second research question. For the applicable model we find an evidence of autocorrelation and even though this does not affect the consistency of our results, we have to be careful not to make wrong inferences. The last and most challenging research questions, we wanted to answer is whether the effects of the financial crisis were different for high and low 30

36 levered companies. On one hand one would expect to find some evidence that favoured low levered companies, like Buca and Vermuelen did in But on the other hand there is also some evidence that favoured high levered companies. At the end when interpreting our results, we could not rule in favour of low or high levered companies, due to insignificant results. Although focusing our research only on Norwegian companies, we set to explore the financial crisis and credit crunch in a way that was not researched before. With our research we, one on hand give incentive to the companies to rethink their strategy and on other hand raise some important questions that need to be answered by future researches done either on Norwegian data or on data from other countries. Especially interesting would be to see a similar research done on other European and non-european countries that experienced much bigger shocks to the economy in the recent financial crisis and then compare the findings. 31

37 REFERENCES 1. Aivazian V. A., Ge Y., Qiu J., The impact of leverage on firm investment: Canadian evidence. Journal of Corporate Finance, Vol. 11, pp Akiyoshi f., Kobayashi K., Banking crisis and productivity of borrowing firms: Evidence from Japan. Japan and the World Economy, Vol. 22, pp Bayoumi T., The Morning After: Explaining the Slowdown in Japanese Growth in the 1990s. Working Paper. National Bureau of Economic Research. 4. Bernanke S. B., Nonmonetary Effects of the Financial Crisis in the Propagation of the Great Depression. The American Economic Review, Vol. 73, pp Buca A., Vermuelen P., Corporate investment and bankdependent borrowers during the recent financial crisis. European Central Bank. Found on 22 th November at: eu/events/pdf/conferences/rolecred/ Buca_Vermeulen.pdf?ae6b3dac 77cc26f74a144e ae4 6. Chava S., Purnanandam A., The effect of banking crisis on bankdependent borrowers. Journal of Financial Economics, Vol. 99, pp England C., Are banks special? Cato review of business & government. Found on 12 th December 2012 at: /pubs/regulation /regv14n2/v14n2-3.pdf 8. Fazzari S., Hubbard G. R., Petersen B. C., Financing constraints and corporate investment. National Bureau of Economic Research (NBER). Found on 21 th December 2012 at: sol3/papers.cfm? abstract_id= Giroud X., Mueller H.M., Stomper A., Westerkamp A., Snow and Leverage. The Review of Financial Studies, pp Greenspan A., Lessons from the Global Crisis. Remarks Made before the World Bank Group and the International Monetary Fund on September 27, Found on 11 th December 2012 at: m 11. Hetland O. R., Mjøs A., Financial Constraints in the Financial Crisis. 24th Australasian Finance and Banking Conference 2011 Paper. Found on 12 th December 2012 at: abstract_id= Hoski T., Kashyap A., Scharfstein D., Corporate Structure, Liquidity, and Investment: Evidence from Japanese Industrial Groups. The Quarterly Journal of Economics, Vol. 106, pp Hoshi T., Kashyap A., The Japanese Banking Crisis: Where Did It Come From and How Will It End? In: Bernanke, B. and Rogoff, K. (Ed.), NBER Macroeconomics Annual. National Bureau of Economic Research, Cambridge MA, Forthcoming. 32

38 14. James C., Some Evidence on the Uniqueness of Bank Loans. Journal of Financial Economics, Vol. 19, pp Lang L., Ofek E., Stulz R. M., Leverage, investment and firm growth. Journal of Financial Economics, Vol. 40, pp Minton B. A., Wruck K. H., Financial Conservatism: Evidence on Capital Structure from Low Leverage Firms. The Ohio State University. Found on 21 th December 2012 at: /sol3/papers.cfm? abstract_id= Morck R., Nakamura M., Japanese Corporate Governance and Macroeconomic Problems. Harvard Institute of Economic Research. Discussion paper number Nunami T., Are Banks Special? : The Volcker Rule and Future of the US Banking Industry. National Graduate Institute for Policy Studies. Found on 12 th December 2012 at: au/pub/regulations/article%20by %20Mr.%20Nunami%20final.pdf 19. Ogawa K., Financial distress and corporate investment: The Japanese case in the 90s. The Institute of Social and Economic Research, Osaka University. Found on 21 th December 2012 at: papers.cfm?abstract_id= Ongena S., Smith D. C., Michalsen D., Firms and their Distressed Banks: Lessons from the Norwegian Banking Crisis ( ). Journal of Financial Economics, Vol. 67, pp Popov A., Udell G. F., Cross-border banking, credit access, and the financial crisis. Journal of International Economics, Vol. 87, pp Rajan R.G., Zingales L., Which Capitalism? Lessons from the East Asian Crisis. Journal of Applied Corporate Finance, Vol. 11, pp Schubben F., Van Hulle C., n.d.. The impact of asymmetric information on the relationship between competition and corporate performance. Found on 15 th December 2012 at: /4650/2/ pdf 24. Sharpe S. A., Financial Market Imperfections, Firm Leverage, and the Cyclicality of Employment. The American Economic Review, Vol. 84, pp Slovin M.B., Sushka M.E., Polonchek J.A., The Value of Bank Durability: Borrowers as Bank Stakeholders. Journal of Finance, Vol. 48, pp Wooldridge, J. (2012). Introductory econometrics: a modern approach, 4th edition. Canada: John Wiley and Sons 33

39 APPENDICES Table of contents Appendix 1: Exclusion of companies in financial service industry... 1 Appendix 2: List of industries used... 1 Appendix 3: Construction and definitions of the variables... 2 Dependent variables... 2 Independent variables... 2 Dummy variables... 3 Control variables... 3 Appendix 4: Industry adjusted leverage ratio... 3 Appendix 5: Algebraic explanation of two-way interaction model interaction between continuous and dummy variable... 4 Appendix 6: Algebraic explanation of two-way interaction model interaction between two dummy variables... 5 Appendix 7: Summary statistics listed / non-listed companies... 7 Appendix 8: Summary statistics low / high levered companies Appendix 9: Fixed effect estimation regression output hypothesis Appendix 10: The effect of bank debt leverage on company s performance without controlling for company s size and cash flows with interaction between leverage and time period summary of regression outputs Appendix 11: Graphical presentation of the two-way interaction Appendix 12: The effect of bank debt leverage on company s performance with interaction between leverage and listing status summary of regression outputs Appendix 13: Fixed effect estimation regression output hypothesis 2 (twoway interaction regression) Appendix 14: The effect of bank debt leverage on company s performance with interaction between leverage and low levered characteristic summary of regression outputs Appendix 15: Fixed effect estimation regression output hypothesis 3 (twoway interaction regression) Appendix 16: Modified Wald test for groupwise heteroskedasticity in fixed effect regression model Appendix 17: Wooldridge test for serial correlation in panel-data models.. 30 Appendix 18: Skewness and kurtosis test for normality Appendix 19: Preliminary Thesis

40 Appendix 1: Exclusion of companies in financial service industry For the purposes of our analysis we use NACE industrial classification codes from SSB Statistics Norway. Our database consists of 5- and 2- digit code. In order to exclude companies in financial service industry we identify NACE codes for these industries. Brief summary of industry NACE codes is captured in Table A1. Table Appendix1: Industries 2-digit NACE codes for companies we eliminate from our analysis Principal activity NACE code before 2009 Financial service activities, except insurance and pension funding Insurance, reinsurance and pension funding, except compulsory social security Activities auxiliary to financial services and insurance activities Real estate activities Source: Appendix 2: List of industries used NACE code after 2009 For the purposes of our analysis we use NACE industrial classification codes from SSB Statistics Norway. Our database consists of 5- and 2- digit code. In order to include only manufacturing companies we identify NACE codes for these industries. Brief summary of industry NACE codes is captured in Table A2. Table Appendix2: Industries 2-digit NACE codes for manufacturing companies used in analysis Principal activity NACE code NACE code before 2009 after 2009 Manufacture of food products Manufacture of beverages Manufacture of tobacco products Manufacture of textiles Manufacture of wearing apparel Manufacture of leather and related products 18, Manufacture and production of wood and cork etc Manufacture of paper and paper products Printing and reproduction of recorded media

41 Principal activity NACE code before 2009 Manufacture of coke and refined petroleum products Manufacture of chemicals and chemical products Manufacture of basic pharmaceutics products and pharmaceutics preparations Manufacture of rubber and plastic products Manufacture of other non-metallic mineral products Manufacture of basic metals Manufacture of fabrication metal products except machinery and equipment Manufacture of computer, electronic and optical products Manufacture of electrical equipment 31, 32, Manufacture of machinery and equipment N.E.C Manufacture of motor vehicles, trailers and semi-trailers Manufacture of other transport equipment Manufacture of furniture Other manufacturing Repair and installation of machinery and equipment Source: Appendix 3: Construction and definitions of the variables NACE code after 2009 Dependent variables INV i,t / K i,t-1 : Investment-capital ratio. Investment is measured as a change in the stock of tangible assets. Capital is measured as by total fixed assets. ROA i,t = NI i,t / K i,t-1 : Return on assets. Net income is referred to a net profit or loss for the year. Capital is measured as by total fixed assets. Independent variables FI Liab i,t : Sum of long-term (amounts becoming due and payable after more than one year) and short-term (amounts becoming due and payable within one year) liabilities to financial institutions. D i,t listed : Listed company dummy variable. Dummy equals to 1 if company i is listed on Oslo Børs or Olso Axcess, 0 otherwise. 33 2

42 { D low i,t : Company s leverage characteristic dummy variable. Company i is characterized as low levered if its industry adjusted leverage ratio in the year 2005 is in the bottom third of industry adjusted leverage ratio distribution. Dummy equals to 0 if industry adjusted leverage ratio lays in the top third of distribution (the middle third is omitted). Dummy variables D i,t 2008 : Pre-crisis dummy variable. { { Control variables LEV i,t-1 = FI Liab i,t / K i,t-1 : Liabilities to financial institutions divided by capital, which is measured as by total fixed assets. CF i,t / K i,t : Cash flow-capital ratio. Capital is measured as by total fixed assets. CF i,t-1 / K i,t-1 : Lagged cash flow-capital ratio. Capital is measured as by total fixed assets. Appendix 4: Industry adjusted leverage ratio Many research papers confirmed that industry factors affect company s financial structure. For that reason we use industry adjusted leverage ratio to determine whether company is defined to be low or high levered based on observations from Leverage ratio is defined as the long-term debt ratio: Industry adjusted leverage ratio is defined as leverage ratio for the given company minus the average leverage ratio value for all 2-digit NACE code 3

43 matched companies. In the case where company has listed more than one industry it operates in, we subtract their average leverage ratios. Appendix 5: Algebraic explanation of two-way interaction model interaction between continuous and dummy variable An interaction occurs when an independent variable has a different effect on the outcome depending on the values of another independent variable. In linear model presented bellow y is the dependent variable, while x is explanatory variable. W is a dummy variable that influence the regression of y on x. Omitted group in the model is the one for which dummy equals to 0. Interaction between x and w is represented by their product x*w. Algebraically such model is represented by the following equation: y = α + β 1 * w + β 2 * x + β 3 * x * w + ε. We can reorder the terms into two groups, terms that do not contain x and the terms that do. y = (α + β 1 * w) + (β 2 + β 3 * w) * x + ε The first grouping defines the intercept while the second grouping defines the slope of the simple regression lines. Equations for the separate group analysis: w==0: y = α + β 2 * x + ε w==1: y = (α + β 1 ) + (β 2 + β 3 ) * x + ε Meaning and interpretation of the estimates: α The intercept for the omitted group (w==0). In the separate group analysis this corresponds to the intercept of the omitted group (α = α(w==0)). 4

44 β 1 Difference in the intercepts between groups. This corresponds to difference of intercepts from the separate group analysis (β 1 = α(w==1) - α(w==0)). β 2 Slope for the omitted group (w==0). It equals to the coefficient of the omitted group from the separate group analysis (β 2 = β 2 (w==0)). β 3 Difference in slopes between groups. This equals to the difference of coefficients from the separate group analysis (β 3 = β 2 (w==1) β 2 (w==0)). If in the regression with two-way interaction coefficient β 3 is significant, this indicates that coefficients from the separate group analysis significantly differ, which causes regression lines to cross at some level. In the case where β 3 is not statistically significant, regression lines are parallel, meaning there is no interaction effect. Appendix 6: Algebraic explanation of two-way interaction model interaction between two dummy variables An interaction occurs when an independent variable has a different effect on the outcome depending on the values of another independent variable. In linear model presented bellow y is the dependent variable, while x is dummy independent variable. W is a dummy variable that influence the regression of y on x. Omitted group in the model is the one for which both dummies equal to 0. Interaction between x and w is represented by their product x*w. Algebraically such model is represented by the following equation: y = α + β 1 * w + β 2 * x + β 3 * x * w + ε. Equations for the separate group analysis: x==0, w==0: y = α + ε x==0, w==1: y = α + β 1 + ε x==1, w==0: y = α + β 2 + ε x==1, w==1: y = α + β 1 + β 2 + β 3 + ε 5

45 Meaning and interpretation of the estimates: α The intercept for the omitted group (x==0, w==0). In the separate group analysis this corresponds to the intercept of the omitted group (α = α(x==0, w==0)). β 1 Difference in the intercepts between omitted group and group, where x==0, w==1. This corresponds to difference of intercepts from the separate group analysis (β 1 = α(w==1) - α(w==0)). β 2 Difference in the intercepts between omitted group and group, where x==1, w==0. This corresponds to difference of intercepts from the separate group analysis (β 1 = α(x==1) - α(x==0)). β 3 Difference of the effect of independent variable w on dependent variable y between two groups, x==0 and x==1. In this regression where only intercept dummy variables are used we assume that slopes of regression lines are all the same. This means that dummy variables only shift the regression line intercept up and down. 6

46 Appendix 7: Summary statistics listed / non-listed companies Summary statistics for selected financial characteristics we use in the regression analysis. Based on whether company is listed or not, the table reports average and median value with standard deviation and minimum and maximum value for the full time period and two sub periods, precrisis and post-crisis period. N indicates the number of observations, that the number of unique companies is reported in parenthesis. The summary statistics are shown both for both samples; summary statistics of our full sample including non-financial companies with non-zero bank debt is presented in Panel A, and of the sample of companies in manufacturing industry in Panel B. Investments are measured as investmentcapital ratio, ROA as earnings divided by total fixed assets, and leverage is represented by normalized total liabilities to financial institutions. For proxy for size Log (REVi,t) is used. All financial characteristics are obtained from CCGR database, measured in Norwegian kroner, and are measured at fiscal year-end. Panel A: full sample listed companies non-listed companies N mean median st. dev. min max N mean median st. dev. min max full period ( ) 137 (29) 119,697 (27,289) investments ,656.5 ROA ,771 2,113.0 leverage ,564 9,500.0 lagged leverage ,

47 listed companies non-listed companies N mean median st. dev. min max N mean median st. dev. min max pre-crisis period ( ) 73 (28) 57,071 (22,227) investments ,656.5 ROA , leverage ,500.0 lagged leverage ,862.7 post-crisis period ( ) 64 (23) 62,626 (23,575) investments ,449.0 ROA ,771 2,113.0 leverage ,564 8,833.3 lagged leverage ,500.0 Panel B: manufacturing industry listed companies non-listed companies N mean median st. dev. min max N mean median st. dev min max full period ( ) 52 (10) 20,256 (4,098) investments ROA leverage ,551.4 lagged leverage ,

48 listed companies non-listed companies N mean median st. dev. min max N mean median st. dev min max pre-crisis period ( ) 28 (10) 10,123 (3,709) investments ROA leverage ,551.4 lagged leverage ,256.7 post-crisis period ( ) 24 (8) 10,133 (3,631) investments ROA leverage lagged leverage

49 Appendix 8: Summary statistics low / high levered companies Summary statistics for selected financial characteristics we use in the regression analysis. Based on whether company is defined as low / high levered, the table reports average and median value with standard deviation and minimum and maximum value for the full time period and two sub periods, pre-crisis and post-crisis period. Company is characterized as low levered if it s industry adjusted leverage ratio in the year 2005 is in the bottom third of industry adjusted leverage ratio distribution, and as high levered if it lays in the top third. N indicates the number of observations, that the number of unique companies is reported in parenthesis. The summary statistics are shown both for both samples; summary statistics of our full sample including non-financial companies with non-zero bank debt is presented in Panel A, and of the sample of companies in manufacturing industry in Panel B. Investments are measured as investment-capital ratio, ROA as earnings divided by total fixed assets and leverage is represented by normalized total liabilities to financial institutions. For proxy for size Log (REVi,t) is used. All financial characteristics are obtained from CCGR database, measured in Norwegian kroner, and are measured at fiscal year-end. Panel A: full sample low levered companies high levered companies N mean median st. dev. min max N mean median st. dev. min max full period ( ) 33,459 (6,673) 33,899 (6,685) investments ROA , ,113.0 leverage , ,500.0 lagged leverage , ,

50 low levered companies high levered companies N mean median st. dev. min max N mean median st. dev. min max pre-crisis period ( ) 18,060 (6,666) 18,105 (6,678) investments ROA , leverage , ,500.0 lagged leverage , ,862.7 post-crisis period ( ) 15,399 (5,402) 15,794 (5,539) investments ROA ,113.0 leverage , ,833.3 lagged leverage , ,500.0 Panel B: manufacturing industry low levered companies high levered companies N mean median st. dev. min max N mean median st. dev. min max full period ( ) 6,284 (1,185) 6,238 (1,185) investments ROA leverage lagged leverage

51 low levered companies high levered companies N mean median st. dev. min max N mean median st. dev. min max pre-crisis period ( ) 3,345 (1,184) 3,304 (1,184) investments ROA leverage ,551.4 lagged leverage ,256.7 post-crisis period ( ) 2,939 (1,024) 2,934 (1,023) investments ROA leverage lagged leverage

52 Appendix 9: Fixed effect estimation regression output hypothesis 1 Model 1: Dependent variable: investment-capital ratio Fixed-effects (within) regression Number of observations = 110,139 Group variable: company Number of groups = 25,765 R-sq: within = between = overall = F(6, 25,764) = 9.72 Prob > F = Obs per group: Min = 1 Avg = 4.3 Max = 6 corr(u_i, Xb) = (Std. Err. Adjusted for 25,765 clusters in company) Robust investments Coef. Std. Err. t P>t [95% Conf. Interval] 2008 D i,t LEV i,t D 2008 i,t * LEV i,t log (REV i,t ) CF i,t / K i,t CF i,t-1 / K i,t α i sigma_u = sigma_e = rho = (fraction of variance due to u_i) 13

53 Model 2 Dependent variable: return on assets Fixed-effects (within) regression Number of observations = 111,390 Group variable: company Number of groups = 25,982 R-sq: within = between = overall = F(5, 25,981) = Prob > F = Obs per group: Min = 1 Avg = 4.3 Max = 6 corr(u_i, Xb) = (Std. Err. Adjusted for 25,982 clusters in company) Robust ROA Coef. Std. Err. t P>t [95% Conf. Interval] 2008 D i,t LEV i,t D 2008 i,t * LEV i,t log (REV i,t ) CF i,t / K i,t α i sigma_u = sigma_e = rho = (fraction of variance due to u_i) 14

54 Model 3 Dependent variable: investment-capital ratio Data sample: manufacture industry Fixed-effects (within) regression Number of observations = 19,243 Group variable: company Number of groups = 4,036 R-sq: within = between = overall = F(6, 4,035) = Prob > F = Obs per group: Min = 1 Avg = 4.8 Max = 6 corr(u_i, Xb) = (Std. Err. Adjusted for 4,036 clusters in company) Robust investments Coef. Std. Err. t P>t [95% Conf. Interval] 2008 D i,t LEV i,t D 2008 i,t * LEV i,t log (REV i,t ) CF i,t / K i,t CF i,t-1 / K i,t α i sigma_u = sigma_e = rho = (fraction of variance due to u_i) 15

55 Model 4 Dependent variable: return on assets Data sample: manufacture industry Fixed-effects (within) regression Number of observations = 19,476 Group variable: company Number of groups = 4,058 R-sq: within = between = overall = F(5, 4,057) = Prob > F = Obs per group: Min = 1 Avg = 4.8 Max = 6 corr(u_i, Xb) = (Std. Err. Adjusted for 4,058 clusters in company) Robust ROA Coef. Std. Err. t P>t [95% Conf. Interval] 2008 D i,t LEV i,t D i,t * LEV i,t log (REV i,t ) CF i,t / K i,t α i sigma_u = sigma_e = rho = (fraction of variance due to u_i) 16

56 Appendix 10: The effect of bank debt leverage on company s performance without controlling for company s size and cash flows with interaction between leverage and time period summary of regression outputs Dependent variable is investment-capital ratio INV i,t / K i,t-1 (model (1) and (3)), and return on assets ROA (model (2) and (4)). Models (1) and (2) include nonfinancial companies with non-zero bank debt, while models (3) and (4) run regression only on companies in manufacturing industry. Dummy variable D i,t 2008 equals 1 for observations before the beginning of year All explanatory variables are observed annually. In the table coefficients of fixed effect regression are reported, standard errors are in parentheses. Regression is absorbed for company s fixed effect and implemented with robust option, which enables standard errors to take into account possible issues concerning heterogeneity and lack of normality. Sample period is variable coefficient (1) (2) (3) (4) intercept α * 0.24 (0.12) (0.06) (0.22) (0.28) 2008 D i,t β * (0.12) (0.22) (0.06) (0.28) LEV i,t-1 β *** -0.89*** -0.58** (0.24) (0.24) (0.16) (0.27) D 2008 i,t * LEV i,t-1 β ** 0.13** 0.06** -0.11** (0.01) (0.05) (0.02) (0.15) observations 119, ,834 20,426 20,426 number of groups 27,309 27,309 4,157 4,157 Prob > F *, **, and ***, significant at the 10, 5, and 1 percent level, respectively. 17

57 Appendix 11: Graphical presentation of the two-way interaction Graphs below show the regression lines from all four regressions without introducing control variables for two time periods, pre-crisis and post-crisis period. Predicted variable is called prediction. Figure 1a Dependent variable: investment-capital ratio Figure 1b Dependent variable: return on assets Figure 1c Dependent variable: investment-capital ratio Data sample: manufacture industry Figure 1d Dependent variable: return on assets Data sample: manufacture industry 18