Competition and size: An empirical analysis of the Swedish banking sector

Competition and size

An empirical analysis of the Swedish banking sector

Cecilia Christensen

Master thesis I, 15 credits

Master’s Programme in Economics/ Master thesis in economics I, 15 credits

Abstract

This paper employs the Panzar Rosse model to examine the competitive state in the Swedish banking sector over the period 2009-2019. In order to account for the large asset differences, this study divides the banks into two subsamples: small banks and medium to large banks, which is done on the basis of the banks’ balance sheet totals. By data gathered from 32 banks with 438 observations the results show that large differences exist in the competitive level between the two subsamples: suggesting that the small banks are operating under perfect competition and the medium to large banks are working under monopolistic competition or a monopoly.

Keywords: Competition, Panzar Rosse, Sweden, banking sector, banking, Monopoly,

Monopolistic competition, Perfect competition, equilibrium test, tax-shift

Acknowledgement

First, I would like to extend my sincere gratitude for the guidance, helpful comments and

support from my supervisor David Granlund. I also want to thank all savings banks who

provided me with all the annual reports that were not accessible online. Finally, writing this

thesis would not have been possible without the support and encouragement from my family

and friends.

Table of contents

1. INTRODUCTION AND BACKGROUND ... 5

2. PREVIOUS STUDIES ... 7

2.1GENERAL FINDINGS WITH THE USE OF THE PANZAR ROSSE MODEL ... 7

2.2FINDINGS ON THE SWEDISH BANKING SECTOR ... 7

3. THE SWEDISH BANKING SECTOR ... 10

4. DATA AND VARIABLES ... 12

5. THE PANZAR ROSSE MODEL ... 14

6. EMPIRICAL RESULT ... 19

7. CONCLUSION ... 22

8. REFERENCES. ... 23

APPENDIX A - INCLUDED AND EXCLUDED BANKS ... 25

APPENDIX B - FORMAL DERIVATION OF THE H-STATISTIC ... 27

B1.PERFECT COMPETITION ... 27

B2.MONOPOLISTIC COMPETITION ... 29

B3.MONOPOLY ... 32

APPENDIX C – CORRELATION MATRIX ... 34

1. Introduction and Background

Competition in the banking sector is something that has been widely studied and discussed over the last years. The area is of growing interest for many reasons, mainly because the banks play a key role in our society. They supply credit, channel the monetary policy, enable savings, investments and payments. The banks can also have an impact on the financial stability, which is to be seen not in the least during recent global financial crisis in 2008. The collapse of the investment bank Lehman Brothers emphasized the problems with banks becoming ‘to large to fail’ and the long-lasting effects it can bring on the financial stability. However, even though there are many studies being performed in the area there are relatively few that analyze the competition among Swedish banks.

The studies that have been conducted also date back almost a decade, with Habte (2013) being the most recent. During this time the sector has transformed with large assets shifts among the banks. Also, the mortgage loans have increased at the same time as the sector have received a higher lending margin. All these mentioned events can shift and alter the competition, making it of interest to analyze the competitive level with new data.

Besides contributing to the literature by studying the competition among Swedish banks with more recent data this study also contributes by analyzing two different subsamples of the banking sector. The two subsamples are: small banks and medium to large banks, which are measured on the basis of the bank’s balance sheet total.

The only previous paper to my knowledge that have studied group-size differences in Sweden is Bikker and Haaf (2002). They performed a cross country study that included Sweden. Their paper did not discuss the results regarding the Swedish market separately, and only had a limited number of observations. For example, their group small banks only included 18 observations. This and the fact that the study was performed almost two decades ago makes it of interest to perform a new study. A new study analyzing the differences in competition within the Swedish banking sector between these two group sizes. By dividing the data into groups I am able to test the hypothesis that the small banks operate under more competitive forms while the medium to large banks operate

1 To my knowledge, Oxenstierna (2000) and Sjöberg (2004) and Habte (2013) are the only previous published studies whose focus is solely on the Swedish market. There has also been made cross-country studies including Sweden, whose focus is broader.

under near monopolistic conditions. Dividing the banks according to their size will also take better account for bank specific characteristics and the large differences in assets between the largest and smallest banks. Both attributes are believed to impact the analysis and the competitive degree in the sector.

The data included in this study has been collected by reviewing each of the 32 included banks’

annual reports between the period 2009-2019. Gathering the data from the annual reports has made it possible to choose the variables of most interest to apply the Panzar Rosse (P-R) model.

The model was developed by Rosse and Panzar (1977) and Panzar and Rosse (1982, 1987) and measures the sum of the elasticities of the reduced form revenue function with respect to its input prices. The level of competition is measured by what degree a change in the input prices affects the revenue. The model marks what level of competition the market is in and defines one of the three different levels: perfect competition, monopolistic competition or a monopoly.

This paper empirically analyzes the competition between the two groups small banks and medium to large banks over the period 2009-2019. The results clearly shows existing differences in the level of competition between the two groups. Further it suggests that the group small banks are working under perfect competition while the group medium to large banks are working under monopolistic competition or a monopoly. These findings are of interest and present also in the recent debate regarding the risk-taxes as proposed by the finance department on September 17

2020. The proposition concerns banks and other credit institutes from year 2022 (Finansdepartementet, 2020). In the present debate some mean that customers will bear all the cost of the new risk-tax put on the banks. This could be strengthened by the evidence found in this study for the group small banks but not for the group medium to large banks. Important to note is however that the group small banks only account for a total asset share of 6 % of the entire total assets from all banks included in this study.

The paper is dispositioned as followed: section 2 starts with a discussion on the previous studies

performed in the area, section 3 presents the banking sector in Sweden. Section 4 presents the

data and variables included in the model, section 5 explains the Panzar Rosse model. Followed

by a report of the results in section 6 and then ending with a conclusion in section 7.

1. Previous Studies

2.1 General findings with the use of the Panzar Rosse model

The first to apply the P-R model to the banking sector was Shaffer (1982). Using data on banks situated in New York he found evidence for monopolistic competition, which was later confirmed by Nathan and Neave (1989). Other well cited studies that followed are Molyneux et al. (1994), Vesela (1995) and Aktan and Masood (2010). Molyneux et al. found that banks in France, Spain, Germany and the United Kingdom are working under monopolistic competition, while the results suggested that the banking sector in Italy was working under a monopoly. Vesela established that the Finish banking industry over the period 1985-1992 was also composed by monopolistic competition. The study further concluded that with increased possibilities for the customers to compare loan offerings comes increased competition. Aktan and Masood instead performed a study on the Turkish banking sector over the period 1998- 2008 and like Vesela the study found that the Turkish Banking industry was working under monopolistic competition. Another important paper in the field is written by Claesson and Laeven (2004). Similar to Molyneux et al. they performed a cross-country study. Their results suggested a causal relationship between the existing regulations in a country and the competitive structure among the banks in that country. They also concluded that systems with higher levels of foreign bank entry and fewer barriers to enter were connected with higher level of competition.

2.2 Findings on the Swedish banking sector

The published studies that focus solely on the Swedish banking sector are Oxenstierna (2000),

Sjöberg (2004) and Habte (2013). All three papers conclude that the Swedish market is not

working under perfect competition. Sjöberg’s findings suggested that the Swedish sector is

composed by some intermediate level of competition i.e., a level between perfect competition

and cournot competition. Oxenstierna and Habte instead found that the sector was composed

by monopolistic competition. Unlike Sjöberg and Oxenstierna, Habte used the P-R model as is

commonly used to measure competition in the banking sector. She also divides the data

according to two bank types: savings banks and commercial banks. Dividing the data into these

groups has its disadvantages. The group savings banks only cover a small segment of the small

banks while the group commercial banks are all other banks. That include small, medium-sized

and large banks. This meaning that the data in the two groups should exhibit large differences in total assets that is not accounted for. Bearing this in mind, the results in her study suggest that the group commercial banks work under a higher level of competition than the group saving banks does. The study also concluded that the overall market is defined by monopolistic competition.

There are also a handful cross-country studies that include the Swedish market. Among these are Bikker and Haaf (2002) who performed a study that included 26 Swedish banks with 145 observations. Similar to this study they divided the data according to group-sizes, applying the groups: small banks, medium-sized banks and large banks.

The study did not discuss the results obtained from the Swedish sector, but the coefficients received in the result shows that medium-sized Swedish banks were working under a lower level of competition compared to the small- and the large Swedish banks. The results presented suggested that the hypothesis that medium-sized banks were working under perfect competition was rejected. The result further indicated that they were unable to reject the hypothesis that small- and large banks were working under perfect competition. Another cross-country study including Sweden was done by Bikker et al. (2012). They analyzed 63 countries over the period 1994-2004. Their study included data for 91 Swedish banks with 401 observations. The paper did not present any results specifically for the Swedish segment regarding the competitive state. They did however find evidence that the Swedish market is not in equilibrium. Another cross-country study that includes Sweden was that performed by Schaeck et al. (2009). Their aim was to assess if the effect of increased competition is related to decreased vulnerability in the financial system.

Their result showed that systematic crisis has an increased risk of happening when competition in the banking sector is scarce. By studying 45 countries over the period 1980-2005 they could get a broad picture on the overall banking sector. The coefficients received for the Swedish segment, indicated that the Swedish sector is composed by monopolistic competition. Carbo et al. (2009) findings are in line with Schaeck et al. (2009), pointing to that the competition in the Swedish banking sector is scarce. Their study finds large differences between Sweden compared to Portugal and Luxemburg which operate under a higher level of competition. They instead conclude that Sweden is closer to Greece with a low level of competition.

3 The sizes are in the study based the banks’ assets, where the group small banks have the lowest assets, the largest banks have the largest assets, and the medium-sized banks have assets in between the small banks and large banks.

Table 1 presents a summary of the earlier discussed papers who study the competition in the Swedish banking sector. The results regarding the existing studies are not entirely unanimous but lean towards that the sector is composed by monopolistic competition. However, most cross-country studies take a broader spectrum than those that solely analyze the Swedish segment. Also, the only study that have compared different bank-sizes is the one conducted by Bikker and Haaf (2002). Their result is however based on few observations, where their group small banks only had 18 observations.

Table 1: Summary of studies made on the Swedish banking segment.

Author

Time

period Method

Nr. of banks

Nr of obs.

Dependent

variable Findings

Oxenstierna (2000)

1989- 1997

Asymmetric game theoretic model.

Oxenstierna (1998) N/A

33 OC MC

Bikker and

Groeneveld (2000)

1989-

1996 P-R

27

140 log(II/TA) MC

Bikker and Haaf (2002)

1988-

1998 P-R 26

145 log(II/TA) PC/MC

Sjöberg (2007)

1996- 2002

Bresnahan and

Reiss entry model 90

631 ln(Q)

Intermediat e level of competitio n

Schaeck et al. (2009)

1980-

2005 P-R N/A N/A ln(II) MC

Carbo et al. (2009) 1995-

2001 P-R

49 N/A

log(TI) MC

Bikker et al. (2012)

1994-

2004 P-R

91

401 log(TR) N/A

Habte (2013)

2003- 2010

P-R and Boone

indicator model 59 436 ln(TR) MC

Note: II (interest income), TA (total assets), TI (total income), OC (operating cost), Q (output), MC (monopolistic competition), PC (perfect competition), P-R (Panzar Rosse model), N/A (not stated). All studies, except Oxentierna (2000) Sjöberg (2007) and Habte (2013), are cross-country studies. The data presented over the cross-country studies concerns only the Swedish segment.

2. The Swedish Banking Sector

According to the Swedish Banker Association (2020), the Swedish banking industry consisted of 125 registered PLC banks in 2019.

However, most of these are small and/or have little business in Sweden. To put this in context, the 32 largest banks total balance sheet accounts for 98 % of the balance sheet total for all listed banks and the three largest banks (SEB, Handelsbanken and Swedbank) account for 66 % of the balance sheet total among all listed banks (Swedish Banker Association, 2020).

Instead looking at the banks’ assets. From the data gathered in this study it can be seen that the medium sized to large banks

have reduced their total asset share of all banks from 98 % in 2009 to 94 % in 2019. The group small banks

has during the same period increased their market share of total assets from 3 % in 2009 to 6 % in year 2019. That is, there has been a shift in the asset concentration among the banks, where the larger banks have over the mentioned period been losing asset shares to the smaller banks.

Figure 1. Data source: Swedish Banker Association (2020).

Another transformation for the banking sector is the increase in loans over this period. Looking at how the mortgage market has transformed during this decade it can be seen that both loans

4 PLC, public limited company.

5 The group medium sized banks are in this study banks with a balance sheets total between SEK 50 to 1,000 billion and the group large banks include banks with total balance sheet above SEK 1,000 billion.

6 Small banks are banks with a balance sheet total between SEK 1,000 million and under SEK 50 billion. More information regarding which banks are included and their size can be found in Appendix A.

0 500 1 000 1 500 2 000 2 500 3 000 3 500 4 000

2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019

Loans measured in Billion SEK

House Apartment Total

for houses and apartments from the public have increased rapidly. Figure 1 displays the change in the total loans from SEK 1,903 billion in year 2009 to SEK 3,555 billion in year 2019, which is an increase by almost 87 %. (Swedish Banker Association, 2020).

Figure 2. Data source: Finansinspektionen (2020).

During the same period Finansinspektionen (2020) reports that the banks margins on loans has also increased. Figure 2 shows a rising trend, starting from 0.59 % in December 2009 to 1.37

% in December 2019. That, which is an increase in 0.78 percent units, shows the rapid rise in the profitability for the banks generated by housing loans. An increased total lending for the banks at the same time as a higher lending margin could be incentives for new firms wanting to enter the market and therefore also shifting the competition.

Also, the Swedish finance department proposed on the September 17

2020 to implement a new risk-tax put on banks in year 2022. The objective is to strengthen the societies abilities to face crisis in the financial system (Finansdepartementet, 2020). In the recent debate around this area many asks if there will be a tax-shift. In other words, if the firms will put this increase in cost (that comes from the taxation) on the customers. The results found in this study should be able to give more direction in this present debate.

0,00 0,20 0,40 0,60 0,80 1,00 1,20 1,40 1,60 1,80 2,00

jan.-09 juli-09 jan.-10 juli-10 jan.-11 juli-11 jan.-12 juli-12 jan.-13 juli-13 jan.-14 juli-14 jan.-15 juli-15 jan.-16 juli-16 jan.-17 juli-17 jan.-18 juli-18 jan.-19 juli-19

Banks mark-up in procent

3. Data and Variables

This study uses unbalanced bank level panel data. The data include 32 banks whose aggregated balance sheet total account for 87 % of all registered Swedish banks’ balance sheets totals. The data is gathered from each banks annual report and contains 438 observations over the period 2009-2019. The choice of year is mainly due to that not all banks had released their annual report for year 2020. All numbers from the banks’ annual reports could be found in Swedish Krona except from IKANO Bank who reported their figures in Euro. The numbers there have been converted to Krona by applying the end year EUR/SEK exchange rate.

All Sweden’s listed PLC banks are found in the technical report handed out by the Swedish Bank Association (2019). The banks are in this study included from the report based on two criteria. The first criterion for inclusion is that the bank has a balance sheet total larger than SEK 1,000 million. The second criterion refers to that the bank has its main business in Sweden.

The reason for excluding banks with a balance sheet total less than SEK 1,000 is due to that these banks did not present the figures needed for this study, this together with the fact that their market share in Sweden is almost insignificant led me to exclude them. The reason for excluding banks based on the second criteria is done in order to avoid bias and to be able to give as accurate analysis of the competition in the Swedish sector as possible. I have therefore excluded Nordea, Santander and Danske bank etc.

For example, from Nordea’s annual report from year 2019 it can be seen that Nordea’s household lending on the Swedish market only covers 32 % of their total lending to households on all segments. Similar argument could be made for Swedbank and SEB who have a large business in the Baltic countries. However, their net revenue from business made on the Swedish market is at least 70 % of their respective total net revenue.

Table 3 below presents the descriptive statistics of the variables included in the analysis that measure competition according to the empirical model that will be presented in the next chapter. The definition of each variable is found in Table 2 below.

7 List of all included and excluded banks are found in Appendix A.

Table 2: Definition of variables Variables Explanation

TR Total Revenue. Used as dependent variable for the H-statistic test.

ROA Return on assets. Defined as net profit of each bank. Used as dependent variable in equilibrium test.

W

Personnel cost, calculated as ratio of personnel expenses to total assets. Used as explanatory variable.

W

Interest cost, calculated as interest expenses to total assets. Used as explanatory variable.

W

Other costs calculated as other operating expenses to total assets. Used as explanatory variable.

Q

Risk, calculated as ratio of equity to total assets. Used as control variable.

Q

Net loans, calculated as deposits and borrowing from the public to total assets. Used as control variable.

Q

Scale, calculated as the banks’ total assets. Used as control variable.

Table 3: Summary statistics

Variable Obs Mean Std. Dev. Min Max

TR 345 471.00 116.00 .441 50,100.00

ROA 348 165.00 4,540.00 -18.50 231.00

W

326 .011 .013 .000 .105

W

326 .011 .010 .000 .066

W

344 .008 .020 .000 .121

Q

345 .113 .099 .012 1.001

Q

345 .634 .423 .000 6.715

Q

345 267.00 708.00 1,230.00 3,070,000.00

Notes: All data is collected from the annual reports of each bank for the period 2009-2019. The

table displays nr of observations, mean, standard deviations, minimum and maximum for the

variables used in this study. The first two variables and the last variable in the table (TR, ROA

and TA) are expressed in million SEK. The definition of each variable is found in Table 2.

4. The Panzar Rosse model

There are many different models to analyze competition in the banking sector, the models are generally divided into structural approaches and non-structural approaches (Bikker, 2004).

Structural approach which comes from the structure-conduct-performance (SCP) paradigm by Mason (1939) and Bain (1956) measure competition from the structure of the market. The non- structural approach, from the New Empirical Industrial Organization (NEIO), uses bank level data to directly assess the degree of competition on the market (Leon, 2015).

This study applies the non-structural P-R model, which is one of the more commonly used measures of competition in the banking sector.

The model which was developed by Rosse and Panzar (1977) and Panzar and Rosse (1982, 1987) measures competition among banks by something they name the H-stastic. The H-statistic measures the sum of the elasticity of the bank’s reduced form revenue function to their input prices. Panzar and Rosse (1987) prove that it is from this measurement possible to analyze the competitive structure in the banking sector.

A value of H equal to 1 (H=1) indicates perfect competition, while a value below or equal to 0 (H ≤ 0) indicates monopoly and a value between 0 and 1 (0 < H < 1) shows monopolistic competition. Table 4 summarizes the interpretation of the H-statistic.

Table 4: Interpretation of Panzar Rosse H-statistic Values of H Level of competition

H = 1 Perfect competition. No barriers to entry or exit. All firms operate where prices are equal to average cost at its minimum.

0 < H < 1 Monopolistic competition. Many firms with low barriers to entry and exit. Characterized by firms offering products that are similar but not perfect substitutes.

H ≤ 0 Monopoly or perfect cartel. Few firms that have significant market power. Existing entry barriers and higher price level than under perfect competition.

8 See for example, Shepherd (1992, 1996), Molyneux et al. (1994), Bikker and Haaf (2002) and Claesson and Laeven (2004).

To give an intuition for how the test works it is important to first define the reduced form revenue function and the cost function.

R*

(y

,z

,n) = C*

(y

,w

,t

) , where Py = R*

(y

,z

,n) (1)

Where, R

, denotes the revenue of bank i, which is a function of the output for that bank (y

), the exogenous variables for the bank (z

) and the number of firms (n) operating in the market.

The aserix (*) marks equilibrium values. Revenue is in equilibrium (under perfect competition or monopolistic competition) equal to the cost (C

) firm i faces which is a function of that firm’s input prices (w

), and the firm’s vector of exogenous variables (t

) and output level (y

). The H- statistic test is based on economical findings that a firm with monopoly power reacts differently to an increase in input prices and will put a lesser part of that increase on the customers than a perfectly competitive do (Schaeck et al., 2009).

Figure 3 – Perfect competition

Panzar and Rosse (1987) and Vesela (1995) shows that H-statistic is equal to 1 (H=1) under perfect competition which means that a 1 % increase in input prices will increase revenue by 1

%. The result, which is quite intuitive, is explained by the cost function being homogenous of

degree one in input prices (w

) and that long-run competitive firms operate where prices (P) is

equal to average cost (AC) at its minimum. Figure 3 illustrates how an increase in input prices

will shift the AC curve up keeping output unchanged whilst increasing P equally. Therefore,

because the only variable changing in the reduced form revenue function are the prices (P)

(remembering here that R=Py), it will result in revenue increasing by the same percentage as the percentage increase in input prices. This is indicated by the H-statistic being equal to 1.

Figure 4 – Monopolistic competition and monopoly

In the case of monopolistic competition and monopoly, where the firm has market power, the firm operates where marginal revenue (MR) is equal to marginal cost (MC). The demand from that point defines the price level. If the demand is linear this mean that the increase in costs will lead to prices increasing by ½ of the increase in cost (dp/dMC=1/2). For example, figure (4) shows how an increase in MC from 2 units to 4 units (an increase by 2 units) will result in prices increasing by 1 unit. Therefore, (by remembering that Py=R) the percentual change in the revenue function will be increasing less than the percentual increase in the input prices. To put in other words, the sum of the elasticities from the reduced form revenue function is less than unity. For formal derivations, see Appendix B.

Panzar Rosse H-statisic is defined as the below elasticity measurement.

" = $ %&'

%(

(

&'

$

#%&

= $ %)*&'

%)*(

$

#%&

(2)

The derivative /TR//(

shows the change in the total revenue of bank i as the input prices (W

) is increased by one unit. K, marks the various input prices and the asterix (*) marks that

9 See Appendix B, Panzar and Rosse (1987) or Vesela (1995) for formal detailed derivations.

it is the long-run equilibrium level. Market power is defined by the percentage change in total revenue which follows a percentage increase in the input prices.

The econometric models used to predict the H-statistic value are many. Shaffer (1982) use total annual interest revenue as dependent variable, as do Bikker and Haaf (2002). Vesela (1995), Aktan and Masood (2010) and Bikker et al. (2012) instead applies total revenue. The latter seems to be the more frequently used dependent variable among the published articles in the field. Also, most of the papers using the P-R model proxies the input prices as expenses that can be found in the banks’ annual reports. Frequently used variables as proxies for input prices are personnel expenses, interest expenses and other expenses. Because these variables are the largest costs visible in the annual reports for the banks, they should be good proxies used for explaining the changes in input prices. The variables chosen in this study fall much in line with the study of Aktan and Masood (2010) that analyze the competition in the Turkish banking sector.

The econometric analysis to estimate the H-statistic in this study builds on the following log- log reduced form revenue function:

ln($%)_!" =∝_!" + +_#,-._#!"+ +_$,-._$!"+ +_%,-._%!"+ /_#,-0_#!"+ /_$,-0_$!"+ /_%,-0_%!"+ 1_!" (3),

where H= 0

+ 0

+ 0

defines the level of competition.

Where TR defines total revenue for bank i in period t. The W-variables define the bank specific

proxied input prices in period t: W

shows personnel related costs, W

the interest cost of the

bank and W

are the other non-personnel nor interest related costs of the bank. The control

variables are marked as Q. Q

is measured as equity to total assets and captures the risk behavior

of the bank. A higher equity ratio indicates, ceteris paribus, that the bank has less risky assets

in the form of debt and loans. Q

, reflects net loans measured as lending towards the public to

total assets. Because the banks take out a margin on their issued loans the larger Q

is the more

revenue the banks should receive all else being equal. The third and last control variable, total

assets (Q

), is included as a regressor to control for scale. Controlling for scale has a clear

importance since larger firms has a larger revenue, ceteris paribus. Dropping this variable is

therefore assumed to give poor result.

The econometric analysis based on equation (3) is estimated separately for the two subsamples small banks and medium to large banks. By doing so I can account for that the coefficients differ largely between the two groups. The model would be wrongly specified if it were to be estimated for all banks in the sector as if the coefficients are the same for all banks, when they are in fact not.

The benefit with using the P-R model is mainly its easy applicability. The model requires only few variables for few banks. Schaeck et al. (2009) also point out that the model is analytically strong when derived from the banks’ revenue function. The model does however have downsides. Leon (2015) raises the problem of the interpretation of the H-statistic when a H- value is less than zero. Shaffer (1983) and Bikker et al. (2012) also finds that a negative H- statistic can be the reason for a high level of competition in the short run or when the average cost is constant. From the result presented in table 5 it can be seen that none of the received H- statistics are negative, therefore this should not cause a problem.

Also, for the P-R model to be applicable the data gathered from the banks must be observations from a long-run equilibrium. This is tested with the help of ‘E-statistic’, which is commonly used in other similar studies when testing for equilibrium. The test was first used by Shaffer (1982). Other studies that followed are Molyneux et al. (1996), Claesson and Laeven (2004), Aktan and Masood (2010) and Shaffer and Spierdijk (2015) etc.

The test for long-run equilibrium is estimated with the below econometric model:

ln(ROA)it = ∝_!" + 4_#,-._#!"+ 4_$,-._$!"+ 4_%,-._%!"+ 5_#,-0_#!"+ 5_$,-0_$!"+ 5_%,-0_%!"+ 1_!" (4),

where E = 2

+ 2

+ 2

= 0 signals that the market is in equilibrium.

The test follows quite same framework as the H-statistic measurement. However, instead of ln(TR) as dependent variable this model use the natural logarithm of Return on Asset (ROA)

as dependent variable. (ROA)

denotes the net return on assets for firm i in period t and is

calculated as the banks net profit. The test confirms that the zero hypothesis is true, that the

market is in equilibrium when E = 0. Alternatively, when E ≠ 0 the market is not in equilibrium

which is an indicator that the H-statistic performed to measure the competition could be biased.

5. Empirical Result

This section presents the empirical findings that build on the estimations of equation (3) and (4) presented in the previous section. Separate results are estimated for the two subsamples small banks and medium to large banks. The reason for not estimating the large banks as one category and the medium-sized banks as one is because the large banks only had around 30 observations. By adding the two groups together the number of observations instead increased to around 88 and therefore provided more validity to the results. Equation (3) is estimated with ordinary least square (OLS) using the fixed effects estimator. The regression is also clustered on the banks sizes and performed with robust standard errors. Equation (4) is conducted with the same estimation but without robust standard errors. The presence of heteroskedasticity is tested by Breusch-Pagan /White’s test. The test only found presence of heteroskedasticity when running equation (3), which is the reason for only regressing this equation with robust standard errors. Also, the Wooldridge test for serial correlation is conducted and found evidence in favor of, which is accounted for by clustering the data on the various bank-sizes. Further on, appendix C provides a correlation table proving that multicollinearity was not a problem.

As discussed in the earlier section, the H-statistic test defining the level of competition is performed under the assumption that the sample is in long-run equilibrium. Table 5 presents the results on the estimation of equation (4), the equilibrium test. The observed E-statistic for the group small banks is 0.335 and for the group medium to large banks the observed value is 0.661. The hypothesis that the market is in equilibrium (E=0) could therefore not be rejected even on a 10 % significance level for either group. The banking sector in Sweden can thus be confirmed to be in equilibrium.

The competitive conditions are estimated with equation (3), where the natural logarithm of

total revenue is used as dependent variable. The results presented in table 5 shows that

personnel cost (W

), interest cost (W

) and other cost (W

) are all positive for the group small

banks. The coefficients for estimating the input prices (W) for the group medium to large banks

are also positive, all except for the interest cost (W

) which has a value of minus 0.02. The

control variable scale (Q

) is as expected positive for both groups but higher for the group small

banks indicating that scaling up the size has a higher effect on the revenue for the group small

banks than for the group medium to large banks. Noticeable from the result is also that the

received significance level is higher for the group small banks. This could perhaps be explained

by that group medium to large banks exhibits larger differences in assets and perhaps other aspects than the group small banks do.

Turning to the results regarding the H-statistic. From table 5 it can be observed that there are significant differences in the H-statistic between the two groups. The group small banks receive a H-statistic value of 1.435, while the group medium to large banks receive a H-statistic value of 0.169. From the Lincom test conducted, the hypothesis that the market is defined by perfect competition (H=1) cannot be rejected even on a 10 % significance level for the group small banks. The results for the medium to large banks does however point in the opposite direction.

For that group the hypothesis that the market is working under perfect competition (H=1) is rejected on a 1 % significance level. Instead, it is for this group not even on a 10 % significance level possible to reject that the market is working under a monopoly (H=0). The 5 % confidence interval for the medium to large banks states that the true H-statistic value lies between minus 0.212 to plus 0.551. Indicating that the true value could both be in the range for monopoly and monopolistic competition.

The results found in this study are similar to Bikker and Haaf (2002), who also found evidence for a higher level of competition for the small banks and lower level of competition for the medium-sized banks. In difference to their result this paper can provide evidence suggesting that both medium-sized to large bank work under lower level of competition. Bikker and Haaf (2002) could in difference not reject that the large banks operate under a perfect competition.

According to Vesela (1995) larger possibilities for the customers to compare the banks services is connected to a higher level of competition. Sweden is technologically advanced, which can perhaps explain the results indicating perfect competition for the small banks. The market for the larger banks has also transformed, where the medium to large banks have been losing asset shares to the small banks. This could indicate that the medium to large banks have been facing increasing competition during the period 2009-2019. A further comparison with other studies that have not divided their data on group-sizes is difficult to do. The reason is that those estimations are for the entire banking sector in Sweden and not divided on group-sizes.

The results found in this study are interesting in the present debate regarding the risk-tax

proposed by the Swedish finance department. The debate refers to who will bear the cost of

this proposed taxation on the firms. Some mean that there will be a tax shift, resulting in

customers bearing all cost of the proposed tax. The result from this study suggests that this

could be the case if the risk-tax is put on the group small banks but could not be confirmed for

the group medium to large banks. Important to note is however that the small banks in this study only account for a total asset share of 6 % of the total assets of all banks and the group medium to large banks accounts for all other asset shares. Therefore, even though the observations in the group small banks are many in the number their market share of total assets is small.

Table 5: Equilibrium and H-statistic results

Small Banks Medium to Large Banks

Variable ln(TR) ln(ROA) ln(TR) ln(ROA)

Personell Cost (W

) 1.332** 0.389* 0.104 0.500

(.534) (.231) (.146) (.281)

Interest Cost (W

) 0.024 -0.107* -0.021 0.037

(.035) (.063) (.070) (.205)

Other Cost (W

) 0.079** 0.053 0.086 0.124

(.038) (.083) (.067) (.077)

Risk (Q

) 0.099 0.576 0.158 0.341

(.157) (.454) (.161) (.366)

Net Loans (Q

) -0.030 0.023 0.167 0.172

(.037) (.048) (.130) (.283)

Scale (Q

) 1.386*** 1.099*** 0.854** 0.465*

(.306) (.202) (.302) (.810)

H-statistic 1.435 0.169

(.555) (.161)

E-statistic 0.336 0.6614

(.248) (.432)

P value H=0 0.017 0.330

P value H=1 0.444 0.001

P value E=0 0.190 0.170

Observations 237 230 88 86

Note: * ten percent significance level, **five percent significance level, ***one percent significance level. The table presents the coefficients with standard errors in parenthesis for the estimation of equation (1) and (2) between the group small banks and the group medium to large banks. All variables are defined in table 2. Additional control not displayed in the table is: a dummy variable capturing the size of the banks (small or medium to large). The dependent variables are: the natural logarithm of total revenue, ln(TR), and the natural logarithm of return on assets, ln(ROA).

6. Conclusion

Many studies have attempted to determine the competitive state among banks, however only a few have been performed in the Swedish market. Of them only one study were to my knowledge done by analyzing the Swedish banks based on their size. This paper aims to contribute to the literature by separately for the Swedish market analyze the competitive state between two subsamples: small banks and medium to large banks, over the period 2009-2019.

The findings concludes that the Swedish market is in a long-run equilibrium. The results also clearly point towards the existence of differences in the competitive level between the two groups. From the empirical analysis it was not possible to reject that the group small banks are operating under perfect competition. For the group medium to large banks, it was instead not possible to reject the hypothesis that they are operating under monopolistic competition or a monopoly. The results found in this study are partly in line with Bikker and Haaf (2002), who also divided the Swedish banks based on their size and found a higher level of competition among the small banks. In contrast to that study these results present evidence that both medium-sized and large banks are working under a lower level of competition, while Bikker and Haaf could not reject that the large banks are operating under perfect competition.

As is usually the case, this area could benefit from further studies. An interesting approach

would be to apply other methods than the P-R method and compare the different result to

estimate the competitive level. Another aspect that would be of interest is to analyze what effect

the technological advancements have had on the competitive state in Sweden. This under

assumption that with larger technological advancements comes increased possibilities for the

customers to compare the banks and with that also a higher level of competition.

7. References.

Aktan, B. and Masood, O. (2010). The state of competition of the Turkish banking industry:

An application of the Panzar‐Rosse model. Journal of Business Economics and Management, 11(1), pp.131-145.

Bain J. (1956). Barriers to new competition, Cambridge. Mass. Harvard University press.

Bikker, J. A. and J. M. Groeneveld. (2000). Competition and concentration in the EU banking industry. Kredit und Kapital 33(1), 62.

Bikker, J. A. and K. Haaf. (2002). Competition, concentration and their relationship: An empirical analysis of the banking industry. Journal of Banking & Finance 26(11),2191–2214.

Bikker, J. A. (2004). Competition and Efficiency in a Unified European Banking Market (Cheltenham: Elgar Publishing. (2004).

Bikker, J.A., Shaffer, S. and Spierdijk, L. (2012). Assessing competition with the Panzar-Rosse model: The role of scale, costs, and equilibrium. Review of Economics and Statistics, 94(4), pp.1025-1044.

Carbo, S., D. Humphrey, J. Maudos, and P. Molyneux. (2009). Cross-country comparisons of competition and pricing power in European banking. Journal of International Money and Finance 28(1), 115–134.

Claessens, S. and L. Laeven. (2004). What drives bank competition? some international evidence. Journal of Money, Credit and Banking 36(3), 563–583.

Coccorese,P. (1998). Assessing the Competitive Conditions in the Italian Banking System:

Some Empirical Evidence,BNL Edizioni BNL Edizioni. HighBeam Research.

Finansinspektionen. (2021). Bankernas marginal på bolån. Statistical report. Finansinspek- tionen.

Finansdepartementet. (2020). Departementsserien och promemorior, rättsliga dokument.

Finansdepartementet.

Habte, O. (2013). Competitive conditions in the Swedish banking system. Konkurrensverket research, global assets.

Leon, F. (2015). Measuring competition in banking: A critical review of methods.

Mason, E.S. (1939). Price and production policies of large-scale enterprise. The American economic review, 29(1), pp.61-74.

Molyneux, P., Lloyd-Williams, D.M. and Thornton, J. (1994). Competitive conditions in

European banking. Journal of banking & finance, 18(3), pp.445-459.

Molyneux, P., Thornton, J. and Llyod-Williams, D.M. (1996). Competition and market contestability in Japanese commercial banking. Journal of Economics and Business, 48(1), pp.33-45.

Nathan, A. and Neave, E.H. (1989). Competition and contestability in Canada's financial system: Empirical results. Canadian Journal of economics, pp.576-594.

Oxenstierna, G.C. (2000). Testing for market power in the Swedish banking oligopoly.

Södertörns högskola.

Panzar, J.C and Rosse, J.N. (1982). Structure, conduct and comparative statistics, Bell Laboratories Eco-nomic Discussion Paper No. 248

Panzar, J.C. and Rosse, J.N. (1987). Testing for" monopoly" equilibrium. The journal of industrial economics, pp.443-456.

Rosse, J.N. and Panzar, J.C. (1977). Chamberlin vs. Robinson: an empirical test for monopoly rents. Bell Laboratories.

Schaeck, K., M. Cihak, and S. Wolfe. (2009). Are competitive banking systems more stable?

Journal of Money, Credit and Banking 41(4), 711–734.

Shaffer, S. (1982). A non structural test for competition in financial markets. In Bank Structure and Competition, Conference Proceedings, Federal Reserve Bank of Chicago, 1982 (pp. 225- 243).

Shaffer, S. (1983). Non-structural measures of competition: Toward a synthesis of alternatives. Economics Letters, 12(3-4), 349-353.

Shaffer, S., & Spierdijk, L. (2015). The Panzar–Rosse revenue test and market power in banking. Journal of Banking & Finance, 61, 340-347.

Shepherd, W. (1982) “Economies of scale and monopoly profits”, in Industrial Organization, Antitrust, and Public Policy, J.V. Craven (ed.), Boston, Kulwer Nihoff.

Shepherd, W. (1986) “Tobin ́s q and the structure-performance relationship: reply”, American Economic Review 76, 1205-10.

Sjöberg, P. (2004). Market power and performance in Swedish banking. Department of Economics, Goteborg University.

Swedish Banker Association. (2020). Bank- and finance statistics 2020. Technical report,

Swedish Bankers Association.

Appendix A – Included and excluded banks

Table A1: Banks included in the study

1.SEB

17.Bluestep Bank

2.Svenska Handelsbanken

18.Forex Bank

3.Swedbank

19.Sparbanken Skåne

4.Länsförsäkringar Bank

20.Sparbanken Sjuhärad

4.SBAB Bank

21.Varbergs Sparbank

6.Landshypotek Bank

22.Sparbanken Rekarne

7.Skandiabanken

23.Sparbanken Alingsås

8.Nordnet Bank

24.Sparbanken Skaraborg

9.IKANO Bank

25.Sparbanken Lidköping

10.Volvofinans Bank

26.Bergslagens Sparbank

11.Resurs Bank

27.Sparbanken Eken

12.Avanza Bank

28.Sparbanken Göinge

13.Collector Bank

29.Södra Dalarnas Sparbank

14.Nordax Bank

30.Vimmerby Sparbank

15.Marginalen Bank

31.Ölands Bank

16.ICA Banken

32.OK-Q8 Bank

Note: The banks are divided into three different sizes, large, medium-sized and small banks. L denotes that the bank is a large bank. M denotes that the bank is a medium-sized bank and S denotes that the bank is a small bank. The grouping is based on the banks’ balance sheet total.

Medium sized banks are in this study banks with balance sheet totals between SEK 50 to 1,000 billion. Large banks have a total balance sheet over SEK 1000 billion to 2400 billion, and small banks have a balance sheet total above SEK 1,000 million and under SEK 50 billion.

Table A2: Banks excluded in the study with total balance sheet over SEK 1,000 million Nordea Bank

Santander Consumer Bank Danske Bank, filial

SEB Kort Bank (could not find any specific numbers/ annual report related to this unit) BNP Paribas

Ålandsbanken DNB Bank

Notes: Banks with a total balance sheet above 10.000 million Swedish Kr. Excluded due to their business in Sweden relatively small and not possible to distinguish between numbers from business on the Swedish market. For example, from Nordea’s annual report from year 2019 it can be seen that Nordea’s household lending on the Swedish market only covers 32 % of their total lending to households on all segments.

Table A3: Banks excluded in the study with total balance sheet under SEK 1,000 million or had missing balance sheet totals.

Crédit Agricole CIB Standard Chartered Bank AG Goldman Sachs Bank Europe

Bank of China UBS Europe HSBC France Bank

Toyota Kreditbank Aareal Bank HSBC Private Bank

Citibank Europe Landesbank Hessen-Thüringen J.P. Morgan AG

TF Bank NatWest Markets N.V Bank J.P. Morgan Bank Luxemburg

Erik Penser Bank Adyen Nordic Bank J.P. Morgan Europe (UK)

Tjustbygdens Sparbank Bank of America Merrill Lynch J.P. Morgan Securities plc

MedMera Bank Barclays Bank Ireland Klarna Bank

Lån & Spar Sverige Bigbank Northern Trust Global Services

PBB Deutsche Pfandbriefbank BRAbank Sverige Northmill Bank

Svea Bank Carnegie Investment Bank Renault Finance Nordic bank

Societe Generale Bank Credit Suisse Standard Chartered Bank (UK)

Deutsche Bank Express Bank (SevenDayBank)

Notes: Banks with a balance sheet total less than SEK 1,000 million. Not included in the

study due to that they did not present the figures needed over the Swedish market.

Appendix B – Formal derivation of the H-statistic

The formal derivations seen below are done by following Panzar and Rosse (1987) and Varian (1995).

B1. Perfect competition

In equilibrium the revenue minus the cost should be equal to zero, or put differently revenue is in equilibrium equal to the cost (C). R, denotes the revenue of a bank as a function of its input prices (w) and a vector of exogenous variables (z). The revenue can be written as the price (P) times output (y).

56(7, 9) − ;(6

, 7, 9) = 0 → 56(7, 9) = ;(6

, 7, =), (5)

7ℎ@A@ 56(7, 9) ≡ '(7, 9) (6)

Differentiating equation (5) w.r.t y gives the below expression, where P is the marginal revenue and C

is the marginal cost.

5(6

, *, 9) − ;

(6

, 7, 9) = 0 → 5 = ;

(7)

Totally differentiating equation (5) and (6) w.r.t w, y, p and solving with Cramer’s Rule gives equation (8), where EF

marks the optimal quantity of production factor i.

%6

%7

= [EF

− 6 %E F

%6 ] 6;

(8)

By applying that revenue is equal to prices times output, equation (6), to equation (5) and then differentiating the revenue function (R) w.r.t 7

we get equation (9). Here the expression EF

seen in equation (9) comes from Shephard’s Lemma, EF

=

!

.

%'

%7

= ;

%6

%7

+ %;

%7

= ;

%6

%7

+ EF

(9)

Multiplying all factors in eq (9) with 7

and summing gives equation (10)

$ 7

%'

%7

= ;

$ 7

%6

%7

+ $

-

"%&

-

"%&

-

"%&

EF

7

(10)

Substituting the expression from equation (8) into equation (10) and dividing by R yields equation (11), then rewriting equation (11) gives equation (12).

" = $

-

"%&

7

' %'

%7

(11)

= ;

' $[

-

"%&

7

EF

− 6 %E %6

6;

] + $[

-

"%&

EF

7

' ] (12)

Rewriting equation (12) and simplifying with the usage of equations (5)-(7) finally yields equation (16):

= ;

'6;

[7

EF

− 6;6] + ;

' (13)

= 5

'6;

[' − 65] + ;

; (14)

= 5

'6;

[' − '] + 1 (15)

" = 1 (16)

B2. Monopolistic competition

Starting with the optimality condition here as well, that firms operate where its marginal revenue is equal to its marginal cost (equation 17) and earn in long-run equilibrium zero profit (equation 18).

'

(6F, *, 9) = 5(6F, *, 9) = ;

(6F, 7, 9) (17)

'(7, 9) = ;(6F, 7, 9), 7ℎ@A@ 56 = ' (18)

Differentiating equation (17) w.r.t 7

and solving using Shepherds Lemma and the chain rule gives:

%'N

%7

= ;

%6F

%7

+ %;

%7

= ;

%6F

%7

+ EF

(19)

The term, EF

, defines the optimal quantity of production factor i. Multiplying equation (19) with

and summing over all factors yields equation (20), rewriting this expression gives equation (21):

" = $

-

"%&

7

'N %'

%7

(20)

= ; 'N + ;

'N $ 7

%6F

%7

-

"%&

(21)

The next step is to retrieve an expression for

!

. This is done by total differentiating equations (17) and (18) w.r.t 6F, n and 7

and then solve for

!

using Cramer’s rule. Doing so

finally gives equation (27).

('

− ;

)/6F + '

/* − ;

/7

= 0 (22) ('

− ;

)/6F + '

/* − ;

/7

= 0 (23)

Writing equations (22) and (23) in matrix form yields:

P ('

− ;

) '

0 '

Q

⎣ ⎢

⎢ ⎢

⎡ /6F /7

/*

/7

⎦ ⎥ ⎥ ⎥ ⎤

= P ;

;

Q (24)

Finally solving the expression for

!

by applying Cramer’s rule gives equation 27:

0)1

0,_!

=

X ;

'

;

'

X, 7ℎ@A@ |ℎ| = X ('

− ;

) '

0 '

X = '

['

− ;

] > 0 (25)

→ %6F

%7

= ;

'

− '

;

|ℎ| (26)

→ %6F

%7

= '

[%E F

%6 − 5EF

]

|ℎ| (27)

Applying above expression for

!

from equation (27) to equation (21) then gives equation (28), which can be rewritten to equation (29):

" = ; 'N + ;

'N $ 7

%6F

%7

-

"%&

(28)

= 1 +

;

['

∑ 7

(%E F

%6 ) − '

∑ 7

EF

]

'N|ℎ| (29)

Simplifying equation (29) and using that the derivative of EF

(the cost minimizing input

factor) w.r.t y (the output level) multiplied with 7

(the input prices) equals the marginal cost

(∑ 7

\

] = ;

) and that sum of the input prices times the optimal output of production factor i equal the total cost (∑ 7

EF

= ;) gives equation (30):

= 1 + ;

['

;

− '

;]

'N|ℎ| (30)

Simplifying equation (30) with the use of equation (17) and (18) gives equation 31:

= 1 + '

['

'

− '

']

'N|ℎ| (31)

Then applying the inverse demand function and that 65 = ', the term in the brackets ['

'

− '

'] can be written as follows as follows:

[56

+ 65

]^5 + 65

_ − 56(6

5

+ 5

+ 65

) (32)

= 5

6

+ 6

5

5

− 55

6

(33)

Here the term 5

6

becomes zero. The intuition behind this result is that market output in a monopolistic equilibrium will be fixed. This due to that there exist and operate only a fixed number of firms in the long-run monopolistic equilibrium, an entry of a new firm can

therefore only happen with the exit of another firm. Therefore, in the long-run equilibrium the term 6

is equal to zero (6

= 0). The equation then becomes:

6

5

5

− 55

6

(34)

Substituting the above expression into the brackets in equation (31) yields equation (35):

" = 1 + '

[6

5

5

− 55

6

]

'N|ℎ| (35)

Here Panzar and Rosse (1987) makes the assumption that the elasticity of the assumed demand facing the firm w.r.t number of new entering firms (

= [6

5

5

− 55

6

]) is negative, hence the H-statistic for monopolistic competition results in the below:

" < 1 (36)

B3. Monopoly

First, the firm’s profit is written as a function of output (y), a vector of exogenous variables that shift the firm’s revenue function (z), a vector of m input prices (w) and a vector of exogenous variables that shift the firms cost function (t):

a (6, 9, 7, =) = ' − ;

The cost function and revenue function are specified as followed:

' = '(6, 9)

; = ;(6, 7, =)

The next step is to define the following expressions:

6

= bAc. dbE

[a(y,z,w,t)] and

6

= bAc. dbE

[a (6, 9, (1 + ℎ)7, =)], with the scalar h ≥ 0

I also define:

'

= '(6

, 9) ≡ '

(9, 7, =), b*/ '

= '(6

, 9) ≡ '

(9, (1 + ℎ)7, =)

The following is then true: