Do banks treat balance sheet liabilities and off-balance sheet obligations equally?

(1)

Do banks treat balance sheet liabilities

and off-balance sheet obligations

equally?

Master Thesis

Student: Axel Jozef Brown Student Number: 10682716 Date: June 22th, 2015 Word count: 15,000

Education: MSc Accountancy and Control, variant Accountancy Amsterdam Business School

Institution: University of Amsterdam, Faculty of Economics and Business Supervisor: dr. A. (Alexandros) Sikalidis

(2)

Statement of Originality

This document is written by Student Axel Jozef Brown who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

Abstract

This research predicts that banks treat operational leases equally compared to balance sheet liabilities. This means that banks already adjust for operational leases as if they were capitalized. The purpose of this study is to provide a better understanding of the expected changes in the leases accounting rules. Within the new exposure draft all leases are recognized on the balance sheet and this will have a major impact on debt ratios. This research provides information on how banks account for operational leases at this moment.

This paper uses an empirical approach. By selection data, which include operational leases information and financial ratios that are used by credit rating companies in their credit assessments, the difference between the levels of operational leases are compared. Furthermore information on industries and credit ratings are included. With the use of industries and credit ratings, a comparison can be made between high and low level of operational leases for companies with similar circumstances.

Although some differences were founded, it was not sufficient enough to conclude that banks make a distinction between companies with a high and low level of operational leases. The comparison based on different industries and credit ratings did not found any differences. Especially since there is no difference for companies with similar circumstance, the conclusion can be drawn that banks do not treat operational leases differently compared to balance sheet liabilities.

The main limitation is that due to the unavailability of data that includes interest rates, a proxy for the interest rate has been used. Furthermore there were limitations due to the fact that credit assessment is very complex and that financial ratios are a small part of credit assessment.

Key words

(4)

1 Introduction:

For a long time there has been a lot of discussion going on between the ways in which a company should recognize lease obligations. A company can either use operational lease or financial lease. How a firm should account for their lease obligations is described in IAS 17. A short explanation between the differences of financial and operating leases depends on the risks and rewards. When all the risks and rewards are transferred with the lease obligation, it is classified as an financial lease. All other leases are classified as operating leases. Whether all the risks and rewards are transferred depends on;

- If ownership is transferred at the end of the lease term

- Option to purchase the asset at the end of the lease term for a price that is sufficiently lower than the fair value at that date.

- The lease term is for the major part of the economic life of the asset

- At inception the present value of the lease payments amounts to at least the fair value of the leased asset

- The lease assets are made so that only the lessee can use it.

So, whether a lease obligation is classified as an operating lease depends on the factors mentioned above. The difference between accounting for operating and financial leases as mentioned above is recognition on the balance sheet. This is the main reason for discussions. Especially operational lease where a company doesn’t recognize a lease obligation in the balance sheet causes a lot of discussion.

Wilkins and Zimmer (1983) investigated whether lenders understand the uses of alternative accounting methods. They find that alternative accounting methods such as disclosure in footnotes about operational lease obligations do not affect evaluations and decisions. The reason for this is that there are strict guidelines provided by banks in how to deal with credit risks caused by lease obligations.

(7)

Duke et al (2002) investigated the differences among firms in terms of their use of operating leases. By doing this they found a negative relation between effective tax rate and operating leases and a positive relation between owner concentration and restrictive financial contracting variables. However, they also found that different time periods could have a strong influence on operating leases. They found this because contrary to the predictions, the existence of management compensation schemes, based explicitly on return on capital, is not related to operational lease. Although they did find some evidence, it was not strong enough to state that there is a relation.

In a study to examine debt management and the use of lease in the UK Fawthrop and Terry (1975) found that lease financing would be a very important factor in debt management. However, this study is old and the circumstances have changed in today’s world because lease financing was rising at that moment. They already state that lease financing will play a big role in financial management. How, when, under what conditions and what kind of terms will become the main focus of debt management. Based on their findings they conclude that lease financing will always face psychological barriers.

It is clear that off-balance leasing leads to discussion and that a bank and other users of the financial statements account for this. However there has been no research to examine whether banks treat off-balance obligations and balance sheet liabilities equally. This research founds that different levels of operational leases do not cause a difference in the interest rate that a bank charges. Based on this the conclusion can be drawn that firms treat off-balance sheet obligations and balance sheet liabilities equally.

This study is structured as follows: After this first paragraph, the second paragraph will consist of a literature review regarding the focus points of this paper. The third paragraph discusses the chosen research design, the next paragraph provide the findings of this research. The last paragraph introduces the conclusion that can be drawn based on the findings and the limitations of this research.

1.2 Research question

My research question will be:

(8)

1.3 Motivation

On the 6th of May 2013 a new exposure draft for IAS 17 was published. This exposure draft contains a lot of changes that will have a great impact on the balance sheet. The exposure draft has been discussed a lot because there are many people against these big changes. This is also the main reason that the implementation of this exposure draft is deferred until probably 2017.

The main changes are made in recognizing a leased asset on the balance sheet. In the current situation there is a separation between operational leasing and financial leasing and only financial leasing is recognized on the balance sheet1 In the new exposure draft there is no longer a distinction in recognizing on the balance sheet or not. The new exposure draft forces a company to recognize all leases with a term of 12 months or longer on the balance sheet. This will have influence on both the preparers and the users of the financial statement. The aim of this exposure draft is to improve comparability and to show the lease liability on the balance sheet.

It is already investigated what kind of influences this may have on different areas like financial ratios, risk assessment and audits. Answering this research question will help to find out whether the new exposure draft have influence on banks etc. and whether it affects the financial statements. Therefore this research question will have a contribution to knowledge of the impact of the new exposure draft of IAS17. Moreover, when the new exposure draft will finally be implemented it could help users of the financial statements. The implementation of the exposure draft will cause a lot of changes in the balance sheet. Users of the financial statement may want to know how this will affect banks. The answer of this research question will provide them with this knowledge and is therefore very important. Also, this research question can be seen as a follow up to the paper of Altamuro et al. (2014). This research examined if banks incorporate operating leases in their credit assessments through the interest rate charged on loans.

1 http://www.ifrs.org/Current-Projects/IASB-Projects/Leases/Exposure-Draft-May-2013/Documents/ED-Leases-Standard-May-2013.pdf

(9)

2 Literature review

As mentioned before the aim of this paper is to describe whether banks treat off-balance leasing equally as off-balance sheet liabilities. By doing so there can be link made to the new exposure draft of the IAS 17 where all lease obligations are recognized on the balance sheet. But the most important aim is to describe off-balance leasing and the way they are treated.

2.1 why firms use operating leasing

First thing that is important to know is why firms use operating leases. A company may choose leasing instead of alternative sources of finance for multiple reasons. It can for example be because lease finance can be obtained with greater ease and fewer restrictions, the lessor can borrow at a lower rate than the lessee, tax depreciation, a better leverage ratio or a manager self-interest (drury and Braund, 1990). However it is logical to expect that the most common reasons are taxation and the relative cost of leasing compared to other sources of finance. This is also in line with the evidence from Drury and Braund(1990).

2.1.1 Possibilities for operational leases

There are several reasons why a company may choose off-balance sheet financing instead of another ways of financing. Based on the research of Drury and Braund (1990) the three most important factors are implicit interest rate, corporation tax considerations and conservation of work capital. It is a common thought, that one of the mean reasons for companies to use off-balance leasing is that it is easier to arrange. Drury and Braund however find no evidence to prove this thought. It seems that the financial aspects are more important to not involve in off-balance sheet leasing. For example the most important reason for companies to use any other source of financing than leasing is that leasing is more expensive.

(10)

2.1.2 Industries

It is a well-known fact that different industries cause for different accounting techniques and different firm strategies. This is also the case with off-balance sheet liabilities. In some industries there is almost no use of operating leasing, while on the other hand in some industries almost every firm uses operating leasing. For example the data used by Altamuro et al (2014) shows that within the retail industry operating leases is used the most compared to other industries. This is confirmed and explained by Goodacre (2003). There is one important reason why operating leases are often used in the retail industry. This is because retail assets are relatively standard compared to the assets that are often used in other industries. Therefor it is easier to finance this by operating leases. So whether a company may choose to use operating leases also depends on the industry.

2.1.3 Taxation and costs

Another reason why a company may choose to lease an asset instead of buying can be explained by the costs and taxation. In some situations it can be better to lease instead of buying the asset. This depends on the cost of financing. When the finance rates for leasing and buying are the same buying an asset will provide tax benefits through high depreciation possibilities. When a taxpayer is not subject to tax, leasing the asset will be better. In this case the lessor will have the tax benefits and can pass this to the lessee with lower lease expenses. According to Ho and Khan (2010) tax consequences play a very big role in how an asset will be acquired.

2.2 What are the determinants for operating leases

Second Duke et al. (2002) also investigated firm’s determinants to choose off-balance leasing. In their research thee examined the relation between five determinants and the ratio of off-balance leasing. They find a positive relation between off-balance leasing and the existing of strict debt covenants, the extent to which a firm is closely held, effective tax rate and the debt/equity ratio. These findings are in line with the findings of Drury and Braund (1990). It is clear why firms may choose to use operating leases. And also that a lot off companies are using operational lease, for

(11)

example in 2006 in Germany 24,6 % of all investments were leasing (Fulbier et al, 2008). Fulbier et al. investigated the suggest changes of the Exposure draft and suggest that users of the financial statements treat on and off-balance sheet information differently. This is also a finding in the paper of Ge (2006). This research shows that investors appear to value operating lease activities as if they are positively associated with future operating performance. This shows that users of the financial statements do treat on and off-balance sheet information differently.

2.2.1 Debt Covenants

Operating leases maintain off-balance. This causes a lower debt ratio (total debt/ total assets) compared to finance leases or a loan to buy an asset. Banks often use debt covenants to protect themselves. A debt covenant is an agreement between a bank and a firm that they should operate with a few restrictions. A restriction can be that the debt ratio should stay under a certain percentage. Duke et al (2002) found a positive association between debt covenant and operating lease activities. In their research to the predictions of Walkman and Smith (1985) they find that a firm with strict debt covenants is more likely to have more operating lease activities. This is also in line with the research of El-Gazzar (1993) who also finds a relation between operating leases and debt covenants. Strict debt covenants can be a very important determinant for a firm to use operating leases.

2.2.2 Taxes

As mentioned before companies may be affected by taxes to use operational leasing. Duke et al (2002) confirmed this finding. In their research tax was concluded as one of the strongest determinants. They attempt different models to investigate a company’s determinants to use operational leases. In all but one model tax rates showed a significant result. Based on this it is save to state that taxes is very important determinant

(12)

2.3 credit raters and operating leases

One very important fact to keep in mind is the way credit raters account for off-balance sheet leasing. Credit raters are very important because they have a lot influence in how people react to a company. For example the research of Altamuro et al (2014) indicates that in their sample of 5,812 more than 50 % were deals where the information of a credit raters was used. This shows that banks often use information from credit raters in their risk assessments. Standard & Poor is one of the biggest credit raters. When they make a risk assessment they adjust all operational lease as if they were capitalized2. By doing so they want to improve clarity, consistency and comparability. Basically they are acting as if the exposure draft is already implemented.

2.3.1 Adjustments by credit raters’

Credit raters play an important role in this economy. Moody’s and Standard & Poors are probably the most famous credit raters. Both companies make adjustments for operational leases in their credit assessment. Fulbier et al. (2008) uses the Standard & Poors factor model, to adjust for their operational leases. Altamuro et al (2014) uses both of the companies to adjust for operational leases. This proves that the adjustments of credit raters are important. Especially when you keep in mind that a bank often uses credit ratings in their credit assessments.

2.4 Hypotheses 2.4.1 Hypothesis 1

Because banks also use credit ratings in their credit assessment one could state that indirectly banks treat off-balance sheet obligations and balance sheet liabilities equally. Hence credit raters treat off-balance sheet obligations as if they were balance sheet obligations. However, since it still is in a disclosure, and banks also have their

(13)

own way of handling credit risk, it is expected that they may treat this differently. A reason for this can be that banks may interpreter operational leases as a risk. This leads to the first hypothesis:

H1: Banks do not make a distinction between companies with a relatively high and low level of operational leases.

2.4.2 Hypothesis 2

There are however differences in industries and the usage of operational lease. Fawthrop and Terry (1975) already noticed differences in the UK. This is in line with the findings of Goodacre (2003) who focused his research on the lease finance in the UK retail sector, which is the industry that uses operational leases, the most. Altamuro et al (2014) also concluded this finding. In their sample it shows that operational leases compared to the total assets gives the highest percentage in retail industry. It is questionable whether banks make a difference between when operating leases are common in the industry and when it doesn’t occur that much. This leads to second hypothesis

H2: Banks do not make a difference between industries when they account for operational leases.

When it comes to setting an interest rate banks will also make adjustment for a firm’s financial condition at that moment. For example, a company with a going concern problem can expect to be charged a higher interest rate. As mentioned before, banks are using information from credit raters (Altamuro, 2014) in their credit assessment. A higher credit rating can be seen as equal to lower interest rate. This leads to last hypothesis

H3: Banks do not make a difference between companies with relatively high and small operational leases when they have a similar credit rating.

(14)

3 Data and method

3.1 Research methodology

In order to answer the research question there is a need of a comparison of financial statements with a lot of off-balance leasing and financial statements with a low number of balance leasing. For every firm the focus will be how much of off-balance leasing they have. Dividing the amount of off-off-balance leasing to the total asset can do this. Based on this percentage rate it will be possible to divide my sample into four groups from low to high. Off course there are always more factors that can influence credit ratings and loan interests and therefor influence the way to how a bank treats off-balance leasing. Factors like, terms of the loans, stock prices, amount of the loans, going concern, and the current situation of the financial statement of a firm can have influence. Investigating how credit raters select their credit ratings, and a distinction between the different ratings will provide an overview for healthy and unhealthy companies. This way a separation can be made between different kinds of groups and also find out how important off-balance leasing really is to credit raters and banks. Because the usage of groups gives the opportunity to find out whether there is big difference within groups and also between different groups. For example maybe banks may only pay extra attention to off-balance leasing when a company is currently unhealthy. The use of groups will help to find this out.

3.2 Sample data

Since this paper follows up onto the research of (Altamuro et. al, 2014) it is logical that the same database is used. Altamuro et al. (2014) based their research on a sample that focused on the period between the year 2000 and the year 2009. For this paper it is more sufficient to focus on a more recent period of time for example the years 2010 until 2013. This time period is after the financial crisis, a period where the banks were forced to maintain a very strict policy. It’s expected that this will cause for better result, since during this period lending money was very risky. And banks therefore increased interest rates. However, when the data in this period is not sufficient enough than I will use the same period as the paper of Altamuro et al (2014).

(15)

The primary source of data for this study is the DealScan Database. This database provides detailed information on commercial loans and is compiled by Loan Pricing Corporation. Due to subscriptions and accessibility the primary source of data for this study is Compustat. This database contains detailed information about a company’s quarterly and annual report. The data is mostly focused on Northern America and provides information until 1962. Although the information of this database is not as detailed about loans as the Dealscan Database, it contains a lot of information about annual reports, which can be used to transform it into the useful data. Due to the fact that the data needs to be calculated only data, which contains all the necessary information, is selected. Also this research focuses on the years after the financial crisis so only the years between 2009 and 2013 are selected. This resulted in 5995 observations. However this data had a lot of extreme observations. Therefore, the most extreme observations where deleted. The remaining 5883 observations were chosen.

The first hypotheses investigate whether the level of operational leases has influence on the interest percentage banks charge on loans. Therefore a comparison is made between the operational lease commitments and the total assets. The observations will be divided into four groups based on the quartiles.

Table 1: Quartiles based on OPTA

Quartile Opta percentages

1 2 3 4 0 – 2,753% 2,753% – 6,71% 6,71% – 17,94% 17,94% <

The second hypothesis is based on the industries. It investigates whether banks make a difference between different industries and the level of operational leases. Within in Compustat the sector code of S&P is used3. The observations are divided as followed.

3_{For specific information : See Appendix}

(16)

Table 2: Industries descriptive

Industry nr Number of Observations Mean

1 2 3 4 5 6 7 8 9 10 11 No industry code 197 86 325 495 854 332 628 517 91 1146 702 510 32,14 3,09 7,77 16,44 9,04 5,39 14,95 6,69 17,04 28,28 28,08 22,97

For this hypotheses there will be two different groups, where the will be a distinction between sectors with a High and low mean. Sectors with a high mean can be seen as specific industries where operational leases are very common. Sectors with a low mean can be seen as non-specific industries where operational leases are not usually. Table 3: Specific and non-specific industries

Group Industries

1- Specific industries 2- Non-specific industries

1, 4, 9, 10 and 11 2, 3, 5, 6, 7 and 8

The groups are divided as follows

Table 4:Specific and non-specific industries and high and low OPTA

Specific Industry Non-Specific industry High OPTA observations

Low OPTA observations

1642 989

1026 1716

The last hypotheses questions whether credit ratings have influence on the interest rate charges. So therefore a distinction is needed between high and low credit rating

(17)

and a high and low level of OPTA. Within the database of COMPUSTAT a list of credit ratings is included. This list consists of the following eight ratings.

Table 5: Credit ratings and number of observations Credit ratings Number of

observations A+ A A- B+ B B- C D Without rating 115 247 273 688 927 1267 1026 142 1198

To make a distinction between high and low ratings and high and low level of OPTA we will use the median of OPTA and of the Credit ratings. This will draw the line between High and Low and will divide the observations in the following four groups. Table 6:Groups based on credit rating and OPTA

Group Credit rating Level of OPTA

High rating, High OPTA High Rating, Low OPTA Low Rating, High OPTA Low Rating, Low OPTA

A+, A, A-, B+ & B A+, A, A-, B+ & B

B-, C & D B-, C & D 6,71 ≤ ≤ 6,71 6,71 ≤ ≤ 6,71

(18)

3.3 Research design

First an overview is given about which variables will be used during this research. Follows are the specific methodology that will be used for each hypothesis. By doing so it will become clear what the difference are between the hypotheses.

The dependent variable that will be used during this research is loanspread. This variable is the natural logarithm of the differences between the interest rate and LIBOR. The dependent variable is calculated in the following way. By dividing interest expenses for long-term debt by total long-term debt you can make a very useful proxy for the average interest rate. The usage of this dependent variable is consistent with prior research (Altamura, et al, 2014)

Since its very difficult to adjust for credit risk, and banks base their interest rate on credit risk. The independent variables will consist of six financial credit ratings that S&P uses to explain credit risk. These financial ratios are:

1. EBIT/Interest expenses EBIT_COV

2. EBITDA/ interest expenses EBITDA_COV 3. Total debt / total assets LEVERAGE

4. Total Debt/EBITDA DEBT_EBITDA

5. Operating cash flow- capital expenditure / total debt FREE_CASH 6. Funds from operations / Total Debt FFO

Further OPTA is very important. This is operating leases/ total assets. Also Industry and Credit ratings are included. Furthermore control variables such as revenue and total assets are included. These control variables are in line with prior research to control for firm size (Altamuro et al, 2014). The natural logarithm of each variable is used.

The loan-spread regression is as follows:

Log (LOANSPREAD) = α0 + α1 EBIT_COV + α2 EBITDA_COV + α3 LEVERAGE + α4 DEBT_EBITDA + α5 FREE_CASH + α6 FFO + α7 INDUSTRY + α8 CREDITRATING + α9 Log_rev + α10 Log_ta

(19)

The first hypothesis questions whether banks treat off-balance lease differently compared to on balance sheet liabilities. By using the ratio of OPTA There will be 4 groups made based on the quartiles. By making a distinction between the groups and dividing them into four groups. It will be possible to see which groups explain the loanspread the best. In order to do these dummy variables will be used. Since the test is about the differences between companies with a high percentage of OPTA and companies with a lower percentage of OPTA we will use the last quartile as the reference group. It is expected that companies with a high level of OPTA explain the variance of the LOANSPREAD better so therefor it is the chosen reference group. The following dummies will be created.

Table 7: Dummies for hypotheses 1

DUMMY 1 DUMMY 2 DUMMY 3

Quartile 1 Quartile 2 Quartile 3 Quartile 4 1 0 0 0 0 1 0 0 0 0 1 0 Dummy 1: Refers to difference between quartile 1 and quartile 4

Dummy 2: Refers to difference between quartile 2 and quartile 4 Dummy 3: Refers to difference between quartile 3 and quartile 4

The second hypothesis questions whether banks also make differences between industries. In order to do this there will be usage of a distinction between specific industries and non-specific industries. Industries where the average OPTA is higher than the mean of 6,71 % are specific industries and where it is lower these are the non-specific industries. In order to control whether banks make a difference between industries two comparisons need to be made. The first one will compare if a company’s business is in a specific industry banks make a difference between high and low level of OPTA.

(20)

Therefor the following dummy will be made. Table 8: Dummy for specific industries

Specific industry DUMMY

High OPTA (>6,71) Low OPTA (<6,71)

0 1

High OPTA companies will be the reference group within in this hypothesis. Since this is the specific industry group it is expected that banks will not charge a higher interest rate when a company has a high OPTA because high OPTA is a common certainty in these industries. This dummy refers to difference between low levels of OPTA compared to high level of OPTA.

The other comparison that is necessary is the comparison within in the non-specific industries. In these industries high level of OPTA is not a common certainty. With the use of a dummy variable an comparison is made between companies with a high level of OPTA and a low level of OPTA.

Table 9: Dummy for non-specific industries

Non-specific industry Dummy

1 0

Within this comparison the reference group is low level of OPTA. A low level of OPTA is normal in these industries. Therefor this is the reference group. This dummy refers to difference between companies with a high level of OPTA compared to companies with a low level of OPTA. Another possibility as an extra comparison is to make a comparison between the companies with a high level of OPTA and

The third hypothesis question whether the banks will treat off-balance leasing equally to balance sheet liabilities when the firm has a high credit rating. In order to check this there need to be a definition when a credit rating is high and when it is low. This done by using the mean of the distribution of the credit ratings and the level of OPTA Again there will be 4 groups made.

(21)

Table 10:Groups based on credit rating and OPTA Group Content Group 1 Group 2 Group 3 Group 4

High OPTA, High credit rating Low OPTA, High credit rating High OPTA, Low credit rating Low OPTA, Low credit rating

Within this hypothesis the main goal is to control for credit ratings. When a company has a high credit rating, it means that they are more credible and that lending them money will become less risky for a bank. When a bank has less risk they will charge a lower interest rate. Therefor it is expected that a bank will not make a difference between levels of OPTA when a company has a high credit rating. Therefor the following dummy will control whether there is a difference between companies with a high credit rating and a different level of OPTA. Low OPTA will be the reference group.

Table 11: High credit rating dummy

High credit rating Dummy

1 0

When a company has a low credit rating this means a bank is exposed to a higher risk and will charge a higher interest rate. It is expected that when a company has a high level of OPTA and a low credit rating it may lead to a higher interest rate than when a company has a low level of OPTA. In this dummy low level of OPTA is the reference group.

Table 12: Low credit rating dummy

Low credit rating DUMMY 1

High OPTA (>6,71) LOW OPTA (<6,71)

0 1

(22)

3.4 Rationale

This research design is chosen for multiple reasons. Since this research follows up to the paper of Altamuro et al. (2014) it is desirable to follow their research design when it is possible. As mentioned in their paper credit assessments are very complex. In their paper they capitalize operational leases and question whether banks treat operational lease as being capitalized. Comparing financial ratios to the loanspread, and check if there is difference when operational leases are capitalized and when they remain off balance does this. This research questions whether banks treat operational leases and normal debt equally and therefore it is desirable to use almost the same variables hence the financial ratios. Furthermore the second and third hypotheses are also relevant to the Altamuro et al (2014) paper. The decision to maintain these hypotheses is very logical. It does make sense for banks to treat operational leases different when their field of operations is one where operational leases are very common. Also a company’s credit rating may affect the interest a bank charges. When a company has a high credit rating a bank is exposed to a lower risk and will charge a lower interest rate. Since credit raters such as moody’s and S&P capitalize all operational leases it is interesting to see whether this still may affect banks.

(23)

4 Results

4.1 Model testing and variable testing 4.1.1 Distribution of the dependent variable

Before any regression it is important to test the regression model by checking whether the dependent and independent variables are justified. The most important variable in this research is the dependent variable LOANSPREAD. This variable is the difference between the interest rates and LIBOR4. An important matter in a regression model is a normally distributed dependent variable. To test whether a population is normally distributed one can use the z score of the skewness. When a population is not normally distributed it may influence the outcome of several test, which may lead to a wrong interpretation of the results. When a population is not normally distributed or to get a better distributed population the natural logarithm of that population is often used. The dependent variable in this research is distributed in the follow way, also the natural logarithm of LOANSPREAD is included to check which one is more suitable. Table 13: Skewness of LOANSPREAD

LOANSPREAD LOG_LOANSPREAD Observations Mean Skewness Standard error of skewness 5995 2,66 3,485 0,032 5995 0,49 -0,918 0,032

To test for skewness the z score of skewness is used. Z-score of Skewness = _{!"#$%#&% !""#" !" !"#$%#!!}!"#$%#&&!!

The z-score test is a two-sided test with α = 5%, so in order not to be significant skewed the z score needs to be <1,96 (Z=0,025)

Table 14: outcome of the Z-score skewness test

(24)

So, based on this test the dependent variable is LOG_ LOANSPREAD since both the variables are significantly skewed and this variable is less skewed. Mainly because there are almost 6,000 observations. The population of a large number of observations is almost always skewed based on the Z-score test. However since the population is very large normality can be assumed. Also the plot of the populations showed a bell curved posture. Therefor LOG_LOANSPREAD is chosen.

4.1.2 Distribution of dependent variables

The distribution of the dependent variable is also important to understand the regression model, and to know which test to use. This regression model consists over multiple dependent variables. A skewed variable may cause a biased regression model. Therefor it is important to know if the variables are skewed. Again this can be done by using the z score of skewness, by looking at the number of skewness and the plot of the variable. In this model all the models score negatively on the z-score test. This means that all variables are significantly skewed. The variables however show a normal bell curved shape. There are also variables for which the skewness is below 1, but are still significantly skewed. Since the number of observations is large, normal distribution may be assumed. However it is necessary to pay extra attention on the distribution of the regression to make sure that the regression model is not biased. Also performing some non-parametric is necessary.

4.1.3 Anova test

In this research the usage of OPTA is very important. Also for the second and third hypotheses credit ratings and a distinction between industries is important. Before the use of the regression analyses it is interesting to see whether these variables are related to each other. In other words are their differences between the groups of these variables. If there are differences between the groups, are they also significant? A good way to test this is with a One-Way Anova. This test controls for the differences

LOANSPREAD LOG_LOANSPREAD Z-score Significant 108,91 Yes -28.69 Yes

(25)

between the means between groups. The first outcome of the one-way anova gives an overview of the groups and their means.

Table 15: outcome descriptive Opta groups and loanspread

Group N Mean Std deviation Std Error

1 2 3 4 1471 1470 1472 1470 2,745 2,266 2,601 2,992 2,568 2,280 2,759 3,256 0,067 0,059 0,072 0,085

As shown in the table there is some difference between the means of the groups. However they appear to be at a minimal level. The table shows the differences between the means on loanspread. This table is included because it is easier to interpret than the natural logarithm of the loanspread. The outcome for the natural logarithm of loanspread is basically the same. Looking at the means it is remarkable that the first group has a higher mean than the second group. From the second group until the last group there is an increase in the mean.

A first impression on these numbers only may cause the following assumptions. When a companies has a very low level of operational leases, this may indicate that the company is not doing well and therefor other companies are not willing to start a operation lease contract with them. The fact that those companies are not performing well may be the reason for the high mean of the first group because a bank charges a higher interest rate. For the other groups the level of Opta, may indicate the increase in the mean. However these are only assumptions. Follows is the Test of Homogeneity of variances, this test tells you whether there is a equality in variances between the groups.

Table 16: Outcome levene test log_Loanspread

Levene Statistic Sig.

1,853 0,135

The null hypothesis in this test is that there is equality between groups. The null hypothesis is rejected when there is a 95% confidence interval. Since the result of this test is not sufficient enough to reject the null hypotheses, equal variance is assumed.

(26)

The information whether variance is equally is important for the interpretation and the selection of the test to follow. At third is the actual Anova Test. An Anova test gives a random idea about the difference between the groups. It consist of the following hypotheses

H0 = There is no difference between the OPTA groups

H1= There is at least one difference between the OPTA groups

Table17: Anova outcome opta groups and loanspread

Sum of Squares

DF Mean Square F Sig.

Regression Residual Total 49,748 6861,744 6911,491 3 5789 5882 16,583 1,167 14,028 ,000

Oneway ANOVA LOG_Loan by Optagroup

Again the null hypothesis is rejected with a 95% confidence interval. Since the outcome of this test has a p-value of 0,00 a confidence level of 95% is reached and the null hypotheses is rejected. So there is at least one difference between the groups. By dividing the sum of squares between groups through the total sum of squares one can calculate how much influence there is.

In this case:

!",!"#

!"##,!"# = 0,0072

Different groups explain a really small percentage. In the test of homogeneity the conclusion was drawn that the variance is equally. With a post hoc test the difference between the groups can be defined. One post hoc test that assumes that variances are equally is Bonferonni. The Bonferonni test gives the following output

(27)

Table 18: Bonferonni test on Optagroups and Log_Loan

Opta group Opta group Mean difference Sig.

1 2 3 4 2 3 4 1 3 4 1 2 4 1 2 3 -,212 ,112 -,014 -,212 -,100 -,227 -,112 ,100 -,127 0,014 ,227 ,127 ,000 ,030 ,1 ,000 ,071 ,000 ,030 ,071 ,009 1 ,000 ,009 The mean difference is significant at the 0.05 level

As mentioned above the Anova test showed at least one difference between the groups. As shown in the table above based on the test of Bonferonni the conclusion can be drawn that there is a significant difference between Group 1 and Group 2 and 3, also there is a significant difference between the second group and the fourth group.. So there seem to be a difference between the groups however whether this may affect the regression is the real question.

For the second and third hypotheses the differences between the groups of credit ratings and industries are also interesting to investigate. The groups with different credit ratings are divided as follows.

(28)

Table 19: descriptive of credit ratings and loanspread

Group N Mean Std deviation Std Error

No rating D C B- B B+ A- A A+ 1198 142 1026 1267 927 688 273 247 115 3,26 3,95 4,02 2,54 1,90 1,60 1,64 1,45 0,87 3,39 3,43 3,95 1,87 1,56 1,20 ,97 0,89 0,62 0,09 0,29 0,12 0,05 0,05 0,05 0,06 0,06 0,06

There seems to be something going on because of the descending of the mean when the credit rating goes up. Again the homogeneity test needs to conclude if the variance is equally or not.

Table 20: levene test for credit ratings and log_loanspread Levene statistic

113,106

Sig. 0,000

In this case the variance is not equally since there is a 95% confidence interval that the variance cannot be assumed to be equal. A very logical reason for this is the difference in the size of the groups. The Anova test gives the following output.

Table21: Anova outcome credit ratings and log_loanspread

Sum of Squares

(29)

As expected there is at least one group that is significantly different than the other groups. However since the sum of squares from between the groups is much higher compared to the opta groups, the influence of the different groups will be much bigger.

582,762

6911,491= 0,08

The different groups explain approximately 8% of the variance. Due to the fact that within this hypotheses the different credit ratings will be separated based on the median, it is not relevant to check which groups are significantly different. However since the levene’s test showed that equal variances are not assumed, a non-parametric test will be used to prove the difference between the groups.

The last hypotheses is the about the different industries. The groups of different industries are divided as follows.

Table 22: descriptive of industries and loanspread

Industry code N Mean Std deviation Std Error

No code 1 2 3 4 5 6 7 8 9 10 11 510 197 86 325 495 854 332 628 517 91 1146 702 3,88 2,15 3,14 2,29 3,15 1,90 2,70 2,71 2,44 2,83 2,70 2,53 4,06 1,82 1,68 1,89 3,76 2,01 1,95 3,35 2,16 2,13 2,43 2,39 0,18 0,13 0,18 0,10 0,17 0,07 0,11 0,13 0,10 0,22 0,07 ,009

Also between the different industries there seems to be a difference. The means of the different industries have a range between 1,90 and 3,88 so this will probably be significant. First comes the Homogeneity test.

(30)

Table 23: Levene’s test industries and log_loanspread Levene Statistic

17,093

Sig. 0,000

Also with the different industry codes the null hypotheses is rejected and equally variance is not assumed. Thus, also for this test a extra non-parametric test is needed. The Anova test gives the following output.

Table 24: Anova outcome industries and log_loanspread

Sum of Squares

Again the conclusion can be drawn that at least one of the different groups is different than the other groups. Due the fact that the p-value is below 0,05. Since the Sum of squares between groups is lower compared to the credit ratings it has less influence.

208,242

6911,491= 0,03

The different groups explain so only approximately 3 % of the variance

In order to make sure no wrong assumptions were made for every anova test described above also a non-parametric has been performed. The main reason for this is that within the second and third hypotheses equal variances were not assumed. The use of a non-parametric test helps to make sure there is a significant difference between the means when equal variance is not assumed. With the usage of the Kruskal-Wallis test the assumption of equal variances is tested. The non-parametric tests were used because there are different groups with non-equal group sizes. The outcome of the test was the same as with the tests above. So there is a significant difference between the means of the different credit ratings and the different industries.

So now it is clear that there is difference between the groups that will be investigated in the hypotheses. This is interesting for the outcome of the regression models. Now it is clear that there is a difference, the question that is still unanswered is whether the financial ratios of these groups also explain the loanspread differently.

(31)

4.1.4 Multi collinearity

A regression analyses defines the spread of a dependent variable based on the independent variables. It is a possibility that the independent variables are correlated, this means that when one independent variable changes not only the dependent variable changes but also another independent variable will change. This is called multi collinearity. Although this research really questions the predictive power of a whole model it is important to know how the independent variables react to each other. Since some financial ratios consist of the same values in their calculation it is expected that this will cause correlation. Detecting collinearity is done based on tolerance statistics and Variance Inflation Factor.

Tolerance = 1-R2 VIF= _!"#$%&'($!

In the model of this research the collinearity statistics are defined as follows.

Table 25: collinearity statistics

Variable Tolerance VIF

Log_revenue Log_total assets Credit rating Industry code EBIT_cov EBITDA_cov Leverage Debt_Ebitda Free_Cash FFO 0,106 0,116 0,747 0,867 0,045 0,042 0,953 0,999 0,150 0,133 9,452 8,625 1,339 1,153 22,141 23,557 1,049 1,001 6,669 7,519

(32)

As a rule of thumb the following numbers are used. Table 26: Rules collinearity

Serious collinearity Collinearity No collinearity Tolerance VIF 0 - 0,10 10 - < 0,10 – 0,20 5 – 10 0,20 < 0-5

As expected there are some variables with a high level of correlation. For example EBIT_COV and EBITDA_COV seem to have a very high correlation. This is very simple to explain since they are both divided through interest expenses and also because EBIT and EBITDA are definitely related.

4.1.5 Factor Analysis

As mentioned above this dataset consist of strongly related independent variables. Although it is about the whole model, it may be useful to provide structure. One way to provide structure is to analyze which effects there are in a model. In other words grouping variables together that belong together because they have almost the same outcome on the dependent variable. To find the components that belong together there are three important criterions:

1. A number of factors that explain more than 60% of the variance 2. Eigen values larger than 1

3. Elbow scree plot

However there are some initial checks that are important.

1. Sufficient sample size, the sample size needs to be 10 till 15 times larger than the variables

2. Items are correlated

3. A Kaiser-Meyer score larger than 0.5, This score indicates that there are strong shared correlations

4. Bartlett’s Test of Sphericity, the null hypotheses in this test is H0= variables are uncorrelated

(33)

Table 27: Outcome initial checks for factor analysis

Check Sort Outcome Results

1 2 3 4 Sample size Correlated items KMO test Bartletts test 5888 6 0,607 0,00 Sufficient Sufficient Above 0,5 so sufficient Reject H0 so sufficient

All initial checks are sufficient enough to continue the factor analysis. Doing the factor analysis gives the following output. Starting with the total variance explained and the Eigen values.

Table 28: Total variance explained and Eigen values

Components Eigen Value Variance explained Cumulative % 1 2 3 4 5 6 7 8 9 10 3,671 1,771 1,056 1,000 0,906 0,813 0,608 0,97 0,57 0,21 36,71 17,72 10,56 10,00 9,06 8,13 6,08 0,97 0,57 0,20 36,71 54,43 64,99 74,99 84,05 92,18 98,26 99,23 99,80 100,00

In the table above the total variance is given and also the Eigen values. This information needed for the first and second criteria. Based on this there are two different possibilities. The first criteria suggest that when cumulative at least 60% of the variance is explained, so this suggests that there are three components. The second criteria states that the number of components with Eigen values above 1 are the number of components in the model. So based on this criteria there are 4 components.

(34)

The last criterion is the elbow scree plot. The scree plot needs to look like an elbow and the component after the major drop is the number of components.

As shown in the graphic above the scree plot is shaped like an elbow. Also with these criteria it is hard to define how many components to select. This leaves some room for interpretation. Based on the criteria it is possible to either select three or four components, so it is necessary to look at the variables to decide how many groups there are and of which variables they consist.

By using rotation and factor analysis it can be decided how many components are logical based on the variables. After checking both models the conclusion was drawn that four components is the most suitable for these variables.

0 0.5 1 1.5 2 2.5 3 3.5 4 1 2 3 4 _{Component
number}5 6 7 8 9 10

Scree Plot

Eigen values

(35)

Table 29: Outcome factor rotation

Component 1 Component 2 Component 3 Component 4 EBITDA_COV EBIT_COV FFO FREE_CASH LOG_TA LOG_REV Credit rating Industrycode Leverage Debt_ebitda 0,893 0,886 0,874 0,811 0,926 0,920 0,589 0,766 0,536 0,998

Based on the outcome of the four components there can be four categories identified. The first component consists of four financial ratios that focus on the total debt, interest expenses and cash flows. So one may call this firm’s financial health, the second component consists of the natural logarithm of total assets and revenue and credit ratings. This component can be related to the firm size.

After comparing the credit ratings with the two relevant variables, the conclusion can be drawn that credit ratings have a significant influence on the size of the total assets and the revenue.

The third component consist of leverage and industry codes. Within this component there is also a significant influence. Industry code has a significant influence on the level of leverage. So this component can be called solvability.

The last component only consists of debt divided by the ebitda. With three components this variable was left behind because it is not related to the first three components. This is also very logical because this is the only variable that divides through earnings. It defines how much total debt there is compared to the earnings from one year. It can be seen as how many years until a company need to earn the total debt. So one may call this maturity.

(36)

Table 30: explanation components Component Sort 1 2 3 4

Financial well being Firm size

Solvability Debt maturity

4.2 Hypotheses 1

The first hypothesis is about the level of operational leases and the loanspread. To check whether there is a difference between the different groups of level of OPTA three dummies are created at first. The one-way anova suggested a significant influence between the second group and the fourth group. Since it seems logical that a company with a high level of OPTA is exposed to more risks a bank will charge them a higher interest rate. The fourth group is considered to be the reference group. At first the regression will be executed without the use of any dummies. This is done to see how strong the model is and whether it is significant. Running the regression without the dummies gives the following output.

Table 31: Regression outcome without dummies

Panel a: Model summary

R R Square Adjusted R

Square

Std. Error of the Estimate

1 ,555 ,308 ,307 ,902

Panel b: Anova outcome Sum of Squares

Regression Residual Total 2128,997 4782,494 6911,491 10 5872 5882 212,900 ,814 261,401 ,000

A. Dependent variable: LOG_loan

B. Predictors: free_cash, Debt_Ebitda, leverage, industriecode, log_ta, creditrating, EBITDA_CO, FFO, log_rev, EBIT_COV

(37)

Panel C: Coefficients

Unstandardized coefficients Standardized Coefficients Beta T Sig. B Std. Error Constant EBIT_COV EBITDA_COV Leverage Debt_Ebitda Free_cash FFO Log_tev Log_Ta Creditrating Industrie 1,326 ,100 -,107 ,011 ,000 -,905 1,164 -,117 ,039 -,074 ,031 ,043 ,003 ,003 ,005 ,000 ,072 ,077 ,014 ,013 ,007 ,004 1,826 -2,099 ,024 -,011 -,348 ,445 -,278 ,093 -,143 ,096 30,647 36,374 -40,154 2,183 -,987 -12,492 15,041 -8,357 2,935 -11,349 8,239 ,000 ,000 ,000 ,029 ,324 ,000 ,000 ,000 ,003 ,000 ,000 a. Dependent variable: LOG_Loanspread

The model explains approximately 30,7% of the variance and at least on independent variable is assumed to be not equal. So the main model is considered to be good. Since our primarily interest is whether there is a difference between the levels of OPTA, it is not interesting to see which variables are significant and which are not. Especially since it is preferable that this research maintains the formulas used in the Altamuro et al (2014) paper as much as possible. However still the model is tested with different variables and this one seems to be the strongest. For example the only variable that is not significant was excluded from the regression model, but this didn’t cause an improvement in the model so therefor it is not excluded.

As mentioned before the most logical way is to test whether the levels of opta will cause for difference in loanspread is by comparing the group with the highest level of OPTA to each group. With the use of dummies the regression model gives the following output.

Table 32 : Regression hypotheses 1 outcome with dummies

Square

(38)

B. Predictors: Dummy 3, free_cash, Debt_Ebitda, leverage, industriecode, log_ta, dummy 2, creditrating, EBITDA_COV, Dummy 1, FFO, log_rev, EBIT_COV

Unstandardized coefficients Standardized Coefficients Beta T Sig. B Std. Error Constant EBIT_COV EBITDA_COV Leverage Debt_Ebitda Free_cash FFO Log_tev Log_Ta Creditrating Industrie Dummy 1 Dummy 2 Dummy 3 1,374 ,100 -,107 ,011 ,000 -,908 1,172 -,127 ,051 -,074 ,031 -,090 -,066 -,086 ,043 ,003 ,003 ,005 ,000 ,072 ,077 ,014 ,013 ,007 ,004 ,037 ,035 ,034 1,826 -2,099 ,024 -,011 -,348 ,445 -,278 ,093 -,143 ,096 -,036 -,026 -,034 30,647 36,374 -40,154 2,183 -,987 -12,492 15,041 -8,357 2,935 -11,349 8,239 -2,417 -1,880 -2,552 ,000 ,000 ,000 ,029 ,324 ,000 ,000 ,000 ,003 ,000 ,000 ,016 ,060 ,011 a. Dependent variable: LOG_Loanspread

Since the only interest is the differences between the dummies only these coefficients are explained. As shown in the table dummy 1 shows a significant level of 0,016. This means that there is significant difference between the dummy 1 and the reference group. With dummy 1 being the first quartile of OPTA levels and the reference group being the last one this shows that there is a significant difference between the extreme values. There is another dummy that shows a significant difference, that is dummy 3. Dummy 3 refers to the differences between opta group 3 and 4. Therefor it is quite strange that this shows a significant difference. Also it is noticeable that the one way anova test showed a difference between the second group and the fourth group the regression shows no significant difference between these two groups. For a better

(39)

understanding it is necessary to test the regression with the first group as the reference group. The following dummies will be used.

Table 33: Dummies used for the second test in hypotheses 1

DUMMY 1 DUMMY 2 DUMMY 3

Quartile 1 Quartile 2 Quartile 3 Quartile 4 0 1 0 0 0 0 1 0 0 0 0 1

With the first quartile being the reference group, a conclusion can be drawn whether there is only a significant difference between extreme measures or that banks somehow seem to react to companies with a low level of OPTA. Doing the regression with the new dummies gives the following output.

Table 33: Second regression hypotheses 1 outcome with dummies

Square

1 ,556 ,309 ,307 ,902

B. Predictors: Dummy 3, free_cash, Debt_Ebitda, leverage, industriecode, log_ta, dummy 2, creditrating, EBITDA_COV, Dummy 1, FFO, log_rev, EBIT_COV

(40)

Unstandardized coefficients Standardized Coefficients Beta T Sig. B Std. Error Constant EBIT_COV EBITDA_COV Leverage Debt_Ebitda Free_cash FFO Log_tev Log_Ta Creditrating Industrie Dummy 1 Dummy 2 Dummy 3 1,284 ,100 -,107 ,011 ,000 -,908 1,172 -,127 ,051 -,074 ,030 ,025 ,004 ,090 ,051 ,003 ,003 ,005 ,000 ,073 ,078 ,015 ,015 ,007 ,004 ,035 ,036 ,037 1,830 -2,103 ,024 -,011 -,349 ,448 -,302 ,122 -,142 ,093 ,010 ,002 ,036 25,361 36,391 -40,516 2,165 -,984 -12,491 15,067 -8,441 3,477 -11,267 7,974 ,708 ,114 2,417 ,000 ,000 ,000 ,030 ,325 ,000 ,000 ,000 ,003 ,000 ,000 ,479 ,909 ,016 a. Dependent variable: LOG_Loanspread

So after doing the second regression it is clear that there is difference between the extreme measures. Extreme measures in the sentence that they are either in the first quartile or in the last quartile. There was also a difference between the third and the fourth group. A regression with the third hypotheses as reference group also showed this difference. Since this difference is remarkable and also not expected it will be explained with extra testing. This is done in paragraph 4.5.1

4.3 Hypothesis 2

It is well known fact that there are differences between industries. For example companies in the oil industry are very different compared to companies in the retail industry. These differences can consist of the level to what content a company uses operational leases. For this hypothesis the data is divided in to two datasets. The first one contains all companies that where qualified as being in a specific industry. The second one contains all companies that where qualified as being in a non-specific industry. An industry is called a specific industry when the mean of the level I of OPTA is larger than 20.

In order to answer this hypothesis in a good matter three sub hypotheses are created to help answer this question. At first the dataset with specific industries is

(41)

used. With the use of dummies it is possible to see whether banks focus on the level of OPTA. Also the comparison of this model can be made with the whole model by looking at the R2_{. Running the regression analysis with the dummy gives the} following output.

Table 34: Regression hypotheses 2 specific industries

Square

1 ,595 ,354 ,351 ,864

B. Predictors: free_cash, Debt_Ebitda, leverage, industriecode, log_ta, creditrating, EBITDA_COV, Dummy 1, FFO, log_rev, EBIT_COV

Unstandardized coefficients Standardized Coefficients Beta T Sig. B Std. Error Constant EBIT_COV EBITDA_COV Leverage Debt_Ebitda Free_cash FFO Log_tev Log_Ta Creditrating Industrie Dummy 1 1,151 ,087 -,101 ,112 ,000 -1,679 2,105 -,141 ,083 -,064 ,021 -,009 ,079 ,004 ,004 ,018 ,000 ,161 ,166 ,024 ,023 ,010 ,005 ,010 1,468 -1,907 ,103 -,020 -,496 ,661 -,323 ,186 -,118 ,065 -,004 14,603 21,867 -26,906 6,190 -1,254 -10,445 12,715 -5,971 3,557 -6,640 4,021 -,247 ,000 ,000 ,000 ,000 ,210 ,000 ,000 ,000 ,000 ,000 ,000 ,805 a. Dependent variable: LOG_Loanspread

(42)

Table 34 shows that the Adjusted R square is slightly bigger compared to the standard model. Also is shown that the dummy is not of significant influence on the reference group. This means that there is no difference between companies with a high level, and a low level of OPTA within companies that perform in a specific industry. The outcome of this is pretty understandable. Since almost all the companies seem to be having a high level of OPTA, it would be not logical if a bank charges a higher interest rate when a company has a high level of OPTA.

This may be different for the non-specific industries. Since high level of OPTA is not a common certainty in these industries, it may be possible that banks account for this by charging a higher interest rate. Therefore the second dataset is used. Running the regression with this dataset gives the following output.

Table 35: Regression hypotheses 2 non-specific industries

Square

1 ,603 ,363 ,361 ,854

B. Predictors:, Free_cash, Debt_Ebitda, leverage, industriecode, log_ta, creditrating, EBITDA_COV, Dummy 1, FFO, log_rev, EBIT_COV

(43)

Unstandardized coefficients Standardized Coefficients Beta T Sig. B Std. Error Constant EBIT_COV EBITDA_COV Leverage Debt_Ebitda Free_cash FFO Log_tev Log_Ta Creditrating Industrie Dummy 1 1,209 ,139 -,153 -,002 ,000 -1,737 2,426 -,085 ,003 -,060 ,060 -,024 ,089 ,005 ,005 ,005 ,000 ,144 ,150 ,022 ,021 ,009 ,011 ,037 2,095 -2,403 -,005 ,002 -,423 ,589 -,191 ,008 -,111 ,095 -,011 13,538 29,512 -33,322 -,344 ,148 -12,043 16,161 -3,864 -,159 -6,585 5,716 -,644 ,000 ,000 ,000 ,731 ,882 ,000 ,000 ,000 ,873 ,000 ,000 ,520 a. Dependent variable: LOG_Loanspread

As noticed this dummy is also not significant. So there is also no significant difference between companies in a non-specific industry. However it is remarkable that the adjusted R square increases compared to previous dataset and compared to the standard model. This may indicate several things, it is possible that banks adjust for more risky industries, or maybe there are some industries that are performing not so well. It is important to control why this change exists. Before this can be seen as a difference it is important to check whether this change is significant or just incidental.

To control this the standard dataset is used. By making two dummies, the comparison can be made for companies in a specific, non-specific industry and without an industry code. Before an analysis is made of the industries, this comparison is necessary. It may be possible that the increase of the adjusted R square is just incidental. Another possibility is that this increase is caused by the fact that the datasets are not even. The first dataset contains fewer observations than the last one so this may cause a change. As mentioned two dummies where created, the reference group is the specific industry. This group is chosen because within this industry it is expected that companies have a higher level of operational leases. Running this regression gives the following output.

Do banks treat balance sheet liabilities and off-balance sheet obligations equally?