The competition level of commercial banking industry in USA : a theoretical and empirical study

University of Amsterdam

Economics MSc

Track Industrial Organization, Regulation and Competition Policy

Master Thesis

The Competition Level of Commercial Banking Industry in USA

A Theoretical and Empirical Study

Name: Kong Xiangyu

Student number: 10664718

Student email: kongxy@outlook.com

Supervisor: Zheng Jindi

Abstract: In this study, I improved the regression model of Boone Indicator (BI), which is a new indicator of competition level. After that, I provided BIs of 48 states in USA for each year from 1984 to 1993 and the long-term BIs of those states between 1984 and 1993. I analyzed the BIs across years and states. The result showed that BIs had an increasing trend during that decade. The BIs in some agricultural states tended to be steady and the BIs in some industrial states were sharply waving. That matched the reality. At the same time, I compared BIs with concentration level and Panzar-Rosse (1982) approach, both of which are important measurements of competition level. I found the concentration level failed to describe the competition level, and Panzar-Rosse approach was not suitable because the banking market during that decade deviated from the equilibrium.

Table of contents

1 Introduction 4

2 Literature review 6

3 The model 8

3.1 Regression model of Boone Indicator 8

3.2 Index of efficiency 9

4 Sample selection 12

5 Result and discussion 13

5.1 Short-term result 13

5.2 Long-term result 23

6 Limitation 25

6.1 Limitation of Boone approach 25

6.2 Limitation of this research 26

7 Conclusion 26

Reference 27

Appendix 29

1

Results of Boone Indicator every year every state 29

2

Result of long-term BI for every state 47

3

Result of PR approach 58

1. Introduction

Competition level is an important aspect to a market. It can influence the efficiency of

the industry and affect price, quantity and quality of products or services, which in

turn influence the social welfare. In additional to these, the degree of competition in

banking industry matters for some extra reasons. Banking industry has a close

relationship with other industries, so the behavior of banks would lead to deeper

changes than many industries. The competition may influence interest rate of loan

(Hoff, 1995; Repullo et al., 2008), degree of moral hazard, stability of industry (Vives,

2010), born rate of local firms (Patti et al., 2000; Cetorelli et al., 2003) economic

growth (Jayaratne and Strahan, 1996; Levine, Loayza and Beck, 2000; Collender and

Shaffer, 2003) and so on, although not all relationships are clear enough.

An index to measure the level of competition is necessary. Scholars prefer using

Herfindahl–Hirschman Index (HHI) to describe the competition. It is defined as the

sum of the squares of the market shares of all firms (or the 50 largest firms) within the

industry. A high HHI indicates a high level of concentration and, traditionally believed,

a low level of competition caused by a strong monopoly power. It is widely used in

competition and antitrust law and also technology management.

However, concentration is uncorrelated with level of competition sometimes. That

makes HHI fail to measure competition level. Suppose there are two cities. City A

contains only a few firms, while City B has many firms. Both cities have 5 banks with

the same market share (20%). Hence, HHI is equal to 0.2 (or 2000 if we times it by

10000) in either city. If we assume that the two cities are two independent markets

and every firm of the two cities has the same demand of financial service (ignoring

the financial service to individuals or households here), we can easily find the

competition level of banking industry in City A is higher than it in City B. however

the level of concentration (HHI) is the same. In another case, if an efficient bank takes

over an inefficient bank, the competition level tends to increase in the market.

However, this merger raises the concentration level at the same time, which means a

higher HHI. These problems threat the validity of using HHI as a measurement of

competition.

Using empirical research, Claessens and Laeven (2003) found there was no strong

evidence that competitiveness negatively relates to banking market concentration.

They applied Panzar-Rosse (1982) approach to describe the competition level of

banking sector. The PR approach is another important way to describe competition

level. This approach is based on a notion that, in a competitive market, the cost and

revenue would increase by a same amount when the input price increases. While the

sum of the factor price elasticity of a monopolist's reduced form revenue equation

must be non-positive (Panzar and Rosse, 1987). Thus, they can research on the

relationship between input price and revenue and obtain an “H-statistic” to measure

competition level.

The PR approach is greatly helpful when measuring the level of competition in

banking sector, but it still has some serious problems. First, the PR model is valid

only if the market is in long run equilibrium. Short run equilibrium is also acceptable

but it needs some extra data and very complex statistic procedures (Claessens and

Laeven, 2003). Second, the demand elasticity should neither be too low nor be

negative (Bikker, Shaffer and Spierdijk, 2009). Third, the regression model should

have some bank-specification control variables. However, which variables should or

should not be put into the model is still in fierce debate.

The PR approach provided a valuable idea that we can measure the level of

competition with bank-level (firm-level) data. Boone (2008) constructed his way to

measure the differences in performance in terms of profits of banks, which reflect the

level of competition. According to Boone, a higher competition level of a market

means that a bank with higher efficiency tends to have higher profit. The relationship

Boone Indicator (BI).

In this paper, I first discussed the literature about BI and showed the theoretical model.

Based on former researches, I improved the regression model to obtain BI, using

return on assets (ROA) as an independent variable. I discussed the advantages and

disadvantages of different approaches and chose the most suitable one. After running

the regressions, I obtained BIs of 48 states in USA for each year from 1984 to 1993

and the long-term BIs of those states between 1984 and 1993. By analyzing the result,

I found the concentration level failed to describe the competition level and PR

approach was not suitable because the banking market during that decade deviated

from the long-run equilibrium. The competition level described by BIs met parts of

the reality.

2. Literature review

Boone Indicator is based on a notion that an efficient firm should have a better

performance (compared to other firms) when it is in a competitive market.

Specifically, the difference between an efficient firm and an inefficient firm in a

competitive market tends to be larger than the difference between an efficient firm

and an inefficient firm in a low-competitive market.

Following Boone et al. (2005) and van Leuvensteijn et al. (2007), the demand

function for bank i can be written in this way.

i i j j i

p

a bq

d

q

≠

= −

−

∑

(1)

An assumption of this function is that the products of different firms (banks) are

slightly different. That is the reason why the quantity of other firms’ products has less

influence on price (d is smaller than b). To simplify the model, we assume the

influences of other banks’ q on the price are the same (d remains the same for every j

≠i).

A bank will set a quantity to maximize its profit

(

)

i

p

i

c q

i i

p =

−

(2)

The model supposes the products have a constant marginal cost, c

i

.

If there are N banks in this market, the equilibrium of output should be:

(2 /

1)

(2 /

1)

[2

(

1)](2 /

1)

i j j i

b d

a

b d

N

c

c

q

b d N

b d

−

−

+ −

+

=

+

−

−

∑

(3)

Combine equations (1) (2) and (3):

(2 / 1) (2 / 1) (2 / 1) (2 / 1) ( ) [2 ( 1)](2 / 1) [2 ( 1)](2 / 1) i j i j j j i j i j i b d a b d N c c b d a b d N c c a b d q c b d N b d b d N b d p ≠ − − + − + − − + − + = − − − + − − + − −

∑

∑

∑

(4)

We can find there is a relationship between profit and marginal cost. The profit here is

the variable profit excluding entry cost. That means a firm (bank) will enter the

market only when

π is larger than or equal to the entry cost ε. Based on Boone’s

theory, competition increases for two reasons. First, competition increases when close

substitutes exist for bank products and when banks interact more aggressively (the

difference between b and d becomes smaller). Second, competition increases if entry

costs go down.

Leuvensteijn et al. (2007) indicated that there is also a relationship between cost and

market share. We can show it by transforming equation (3).

(2 /

1)

(2 /

1)

N(2 /

1)

(2 /

1)

i j j i i j j j j

b d

a

b d

N

c

c

q

s

q

b d

a

b d

c

−

−

+ −

+

=

=

−

−

−

∑

∑

∑

(5)

it gives no significant improvement to Boone’s original approach and, the most

important, it has a serious problem (I will discuss it later.). Hence, I still treat equation

(4) as the theoretical model and focus on the relationship between cost and profit of

banks.

The theoretical model mentioned above can also be used to explain why

widely-applied measures of the Herfindahl–Hirschman Index may fail as reliable

competition indicators. The theoretical assumption of HHI is based on a Cournot

model with symmetric (same) firms. The decrease of entry barriers tends to reduce

HHI. However, an increase in competition through a rise in d reallocates output to

more efficient banks that already had higher output levels, because banks have

different efficiency levels. Thus, the rise in competition raises the HHI. The effect of

competition on Lerner Index may also be perverse in this industry. Generally, heavier

competition reduces the price-cost margins of all banks. However, banks with higher

efficiency tend to have higher price-cost margins. That makes the increase of their

market share increase the Lerner Index of the market, which is contrary to common

expectations.

3. The model

3.1.Regression model of Boone Indicator

According to Boone, competitive level of a market can be estimated in this way.

Suppose there are three firms, with efficient levels and profit n

1

,

π

1

；n

2

,

π

2

and n

3

,

π

3

,

respectively. n

1

< n

2

< n

3

. The comparative level is:

3 1 2 1

BI

p p

p

p

−

=

−

₍₆₎

In a more comparative market, this figure would be larger. The efficiency level of a

firm can be described by the average variable cost. Based on this notion, a regression

model can be set.

ln( )

p

_i

= +

α β

ln(n )

_i

+

ξ

_i

π

i

refers to the profits of bank i. n

i

refers to the efficiency of bank i. The β here is

the Boone Indicator we need. It shows that how many percentages of the profit tend to

increase if 1% of the efficiency rises for banks in this market. A large Boone Indicator

indicates the profit of banks in this market is very sensitive to the efficiency, which

means a high degree of competition here.

Leuvensteijn et al. (2007) applied the relationship between market share and

efficiency. They built their regression model in this way:

ln(ms )

_i

= +

α β

ln(n )

_i

+

ξ

_i

(8)

ms

i

refers to the market share of bank i.

The problem of this model is that it ignores the size of market. When the aggregate

demand drops (assume the decrease is homogenous and the profit of every bank is

still positive after the decrease), market share and efficiency of every bank will not

change. The beta here remains the same but obviously the competition becomes

fiercer. Thus, this approach has a similar shortage as HHI. Because of this problem, I

follow the original idea of Boone and use equation (6) as my empirical model instead

of using Leuvensteijn’s approach.

3.2.Index of efficiency

Now, I should choose an index to describe the efficiency of a bank. In Boone’s work,

he indicated “cost per product” can stand for the efficiency. Hence, marginal cost may

be a good index. Another common used index for efficiency is return on assets. I will

discuss both of them and check which one is a better choice in this research.

A higher marginal cost indicates a lower efficiency. The regression model should be:

ln( )

p

_i

= −

α β

ln(mc )

_i

+

ξ

_i

same input price. The difference of marginal costs is only caused by different

efficiency.

Marginal cost is not available directly from data. We can only estimate it.

Transcendental logarithmic cost function can help me to do that. The reason why I

choose TLC function is that, according to Christensen etc. (1973):“It provides a local

second order approximation to any cost structure and allows a great variety of

substitution patterns than any other.”

I build the model based on the work by Guevara et al (2005). There are three cost

factors: labor (w1), funding (w2) and fixed asset (w3). Here Guevara set total cost (C)

as the proxy of production and include time trend to absorb the effect of the

technology change. Thus, the model goes as follow:

ln𝐶𝐶

𝑖𝑖

= 𝑎𝑎

0

+ 𝑎𝑎

𝑞𝑞

𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙 + 𝑎𝑎

𝑘𝑘

(𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙)

2

+ ∑

3𝑗𝑗=1

𝑎𝑎

𝑗𝑗

𝑙𝑙𝑙𝑙𝑤𝑤

𝑖𝑖𝑗𝑗

+

1₂

∑

3𝑗𝑗=1

∑

3𝑘𝑘=1

𝑏𝑏

𝑗𝑗𝑘𝑘

𝑙𝑙𝑙𝑙𝑤𝑤

𝑖𝑖𝑘𝑘

𝑙𝑙𝑙𝑙𝑤𝑤

𝑖𝑖𝑗𝑗

+

∑

3𝑗𝑗=1

𝑟𝑟

𝑗𝑗

𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑤𝑤

𝑖𝑖𝑗𝑗

+ 𝑢𝑢

1

𝑡𝑡𝑟𝑟𝑡𝑡𝑙𝑙𝑡𝑡 + 𝑢𝑢

2

𝑡𝑡𝑟𝑟𝑡𝑡𝑙𝑙𝑡𝑡

2

+ 𝑢𝑢

3

𝑡𝑡𝑟𝑟𝑡𝑡𝑙𝑙𝑡𝑡𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙 +

∑

3𝑗𝑗=1

𝛿𝛿

𝑗𝑗

𝑡𝑡𝑟𝑟𝑡𝑡𝑙𝑙𝑡𝑡𝑙𝑙𝑙𝑙𝑤𝑤

𝑖𝑖𝑗𝑗

+ 𝑙𝑙𝑙𝑙𝑢𝑢

𝑖𝑖

(10)

Restricted with linear homogeneity conditions:

∑

3𝑗𝑗=1

𝑎𝑎

𝑗𝑗

= 1; ∑

3𝑗𝑗=1

𝑟𝑟

𝑗𝑗

= 0; ∑

𝑗𝑗=13

𝛿𝛿

𝑗𝑗

= 0;

1₂

∑

3𝑗𝑗=1

∑

3𝑘𝑘=1

𝑏𝑏

𝑗𝑗𝑘𝑘

= 0

We can derive new equation after applying the linear homogeneity by substitution:

X

𝑖𝑖

= 𝑎𝑎

0

+ 𝑎𝑎

𝑞𝑞

𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙

𝐼𝐼

+ 𝑎𝑎

𝑘𝑘

(𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙)

2

+ ∑

3𝑗𝑗=2

𝑎𝑎

𝑗𝑗

𝑉𝑉

𝑖𝑖𝑗𝑗

+ ∑

𝑗𝑗=12

∑

𝑘𝑘=23

𝑏𝑏

𝑗𝑗𝑘𝑘

(

𝑙𝑙𝑙𝑙𝑤𝑤

𝑖𝑖𝑘𝑘

𝑙𝑙𝑙𝑙𝑤𝑤

𝑖𝑖𝑗𝑗

−

𝑙𝑙𝑙𝑙𝑤𝑤

𝑖𝑖12

) +

1₂

𝑏𝑏

22

�𝑙𝑙𝑙𝑙𝑤𝑤

𝑖𝑖22

− 𝑙𝑙𝑙𝑙𝑤𝑤

𝑖𝑖12

� +

1₂

𝑏𝑏

33

�𝑙𝑙𝑙𝑙𝑤𝑤

𝑖𝑖32

− 𝑙𝑙𝑙𝑙𝑤𝑤

𝑖𝑖12

� +

∑

3𝑗𝑗=2

𝑟𝑟

𝑗𝑗

𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑣𝑣

𝑖𝑖𝑗𝑗

+ 𝑢𝑢

1

𝑡𝑡𝑟𝑟𝑡𝑡𝑙𝑙𝑡𝑡 + 𝑢𝑢

2

𝑡𝑡𝑟𝑟𝑡𝑡𝑙𝑙𝑡𝑡

2

+ 𝑢𝑢

3

𝑡𝑡𝑟𝑟𝑡𝑡𝑙𝑙𝑡𝑡𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙 +

Where,

𝑋𝑋

_𝐼𝐼

= 𝑙𝑙𝑙𝑙𝐶𝐶

_𝑖𝑖

− 𝑙𝑙𝑙𝑙𝑤𝑤

_𝑖𝑖1

V

_{𝑖𝑖𝑗𝑗}

=

𝑙𝑙𝑙𝑙𝑤𝑤

_{𝑖𝑖𝑗𝑗}

-

𝑙𝑙𝑙𝑙𝑤𝑤

_𝑖𝑖1

And marginal cost should be:

Mc

i

=

_{𝜕𝜕𝜕𝜕𝜕𝜕𝜕𝜕𝜕𝜕i}𝜕𝜕𝜕𝜕 _{𝜕𝜕𝜕𝜕i}𝜕𝜕i

=

_{𝜕𝜕𝜕𝜕i}𝜕𝜕i

(𝑎𝑎

_𝑞𝑞

+ 𝑎𝑎

_𝑘𝑘

𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙 + ∑

3_𝑗𝑗=1

𝑟𝑟

_𝑗𝑗

𝑙𝑙𝑙𝑙𝑤𝑤

_{𝑖𝑖𝑗𝑗}

+ 𝑢𝑢

₃

𝑡𝑡𝑟𝑟𝑡𝑡𝑙𝑙𝑡𝑡) (12)

This approach of obtaining marginal cost has three problems. First, it assumes that the

flow of banking goods and services produced by a bank is proportional to its total

assets. That is against the theoretical model of this research that the quantity is

decided by the competition level of market and efficiency of banks. Second, this

approach assumes

𝜕𝜕 ln(𝐶𝐶) /𝜕𝜕ln (𝑙𝑙𝑙𝑙) remains the same for all the banks within this

market. This assumption is too strong and it can never meet the realities. Third, this

approach asks the input price for every factor. In all the researches I can found, they

use the ratio of personnel expense on total assets to stands for the price of labor.

However, they do not have any clear relationship at all. That makes the approach

inaccurate.

Efficiency is the profit ability of a firm (bank) with limited resource. Thus, return on

assets (ROA) is also a good index for efficiency. ROA is calculated by dividing a

company's annual net income by its total assets. It indicates how profitable a company

is relative to its total assets. It can also describe how efficient a bank is (e.g. Spong et

al., 1995; Claessens and Laeven, 2003). An important shortage of ROA is that, it

cannot figure out the investment of fixed assets. For instance, if a bank purchases

many new computers, its ROA would looks very low that year (suppose the bank can

use these computers several years). That makes the efficiency of that bank seem low.

It would make a noise in the result of the regression. However, the dataset of this

research is large. That can help to cover the noise in the regression. Thus, I choose

ROA as the index of efficiency. The regression model is:

ln( )

p

_i

= +

α β

ln(ROA )

_i

+

ξ

_i

have the natural logarithm, so I have to amend the regression model in this way:

ROA

i i i

p

= +

α β

+

ξ

(14)

Here the beta is the Boone Indicator we need. I can run this regression for every state,

every year. However, in order to improve the degree of freedom, I prefer to use the

following regression model for every state.

𝜋𝜋

𝑖𝑖𝑖𝑖

= 𝛼𝛼 + 𝛽𝛽ROA

𝑖𝑖𝑖𝑖

+ ∑

𝑖𝑖=1,…,𝜕𝜕−1

𝛾𝛾

𝑖𝑖

𝑡𝑡

𝑖𝑖

+

∑

𝑖𝑖=1,…,𝜕𝜕−1

𝛿𝛿

𝑖𝑖

ROA

𝑖𝑖𝑖𝑖

𝑡𝑡

𝑖𝑖

+ 𝜉𝜉

𝑖𝑖𝑖𝑖

(15)

𝑡𝑡

𝑖𝑖

is time dummy variable.

𝛽𝛽 is the BI of year 0 and 𝛽𝛽 + 𝛿𝛿

𝑖𝑖

is the BI of year t.

I can also get the long term Boone Indicator to describe the competition level of that

decade for every state with this regression model:

𝜋𝜋

𝑖𝑖𝑖𝑖

= 𝛼𝛼 + 𝛽𝛽ROA

𝑖𝑖𝑖𝑖

+ ∑

𝑖𝑖=1,…,𝜕𝜕−1

𝛾𝛾

𝑖𝑖

𝑡𝑡

𝑖𝑖

+

+ 𝜉𝜉

𝑖𝑖𝑖𝑖

(16)

The beta here stands for the Boone Indicator of state i during the period.

4. Sample Selection

These data consist of financial data on individual banks from the Call reports released

by the Federal Reserve Bank of the U.S. for the period 1984-1993. Bank-level data

are involved in this research. The table below shows detailed information

Table 1

T

Year

S

State

II

Pre-tax Profit (000’s of dollars)

ROA

Return on Assets

w1

The ratio of interest expense to liabilities

w2

The ratio of personnel expense to total assets

w3

The ratio of fixed asset expenses over fixed assets (%)

y1

The ratio of equity over total assets (%)

y2

The ratio of loans to total assets

between 1984 and 1993. The reason why I choose data of the US to research this issue

is that the degree of competition in the US is more valuable in further research. In

some researches involved competition level, such as finding the relationship between

competition level and firm born rate, some data about institutions are necessary. But

institution data is difficult to measure or quantize, and it is always omitted. Different

states in America share similar institution. That can help researchers to deal with the

problem of omitted variables. The reason why I study on the data before 1994 is that,

banks in the US were forbidden to set a branch in other states before 1994. We can

treat every state as an independent market. I do not have the relevant branch-level

data after 1994, when bank-level data became meaningless for our research. Thus, I

have to just focus on the data before 1994.

This is some detailed information about the data:

Table 2

Variable

Mean

Max

Min

Standard

divination

II

1387.264

1214000

-1055767

17929.85

ROA

0.0056794

3.499669

-0.276897

0.0210596

w1

1.623674

2.448995

1.324207

0.2196333

w2

5.277747

7.342852

2.126793

1.082697

w3

44.22043

124.3419

25.17789

14.16273

y1

0.5362228

1.024696

0.0000659

0.1478484

y2

8.649051

98.54072

0.0068034

4.205253

5. Result and discussion

5.1.Short-term result

I will use equation (15) as the regression model of this research. Hausman test

indicates that fixed effect panel regression will be applied for all regressions. Detailed

Table 3

state 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1 20926 50371 30942 47188 40685 65916 45678 66358 82585 87498 2 923835 117338 123483 92820 413062 155461 174774 300269 27229 4 135337 153467 181782 135337 257099 247410 55316 105548 175856 147920 5 48432 192312 126249 77994 67998 73498 90172 59537 67531 157291 6 65327 86261 72790 259595 250616 212791 386500 -601090 8 27697 24835 28179 21875 31560 24901 39781 40685 48642 68001 9 117592 274729 460700 372409 1212038 1353143 444294 198736 498324 1130351 11 286166 438786 235121 441873 672140 1128815 1386108 1064127 520706 1260202 12 42668 71455 72906 61620 111267 137116 160419 151846 166241 264450 13 25937 25083 49429 29318 38715 42468 80111 102590 88417 63310 16 45874 -57061 -51838 39819 106047 198661 272559 382818 884003 572906 17 1017061 133342 114810 413052 52261 98765 132902 121467 123128 274421 18 41722 38704 40421 54936 41722 86331 165339 307917 160515 238452 19 44378 30439 51418 49263 52265 30327 62279 59159 20 24606 23535 17647 34375 25710 25933 51833 37810 83036 110168 21 22173 33688 41753 58211 57056 39586 82020 23914 103268 116755 22 51977 61257 56220 62943 79401 206545 105527 103914 71970 301234 23 16191 -59308 52425 770142 1291607 930856 801400 534573 578037 1253977 24 635352 218656 277292 426560 645524 362886 911530 1107405 307967 546172 25 -229088 139564 349767 197373 292997 1110175 318552 274423 -229088 -229088 26 73725 95706 118812 187594 114926 101755 115444 170214 162078 215959 27 49657 61922 49328 206500 187814 319571 146598 289001 655290 384650 28 49326 33122 36490 79235 70287 103709 93160 72656 60178 50808 29 10161 27944 19961 42884 57760 109043 87309 188253 79189 188741 30 36867 37394 35119 38790 73352 45208 39881 34930 17740 2863 31 23329 25077 21474 19186 34793 35380 35152 47136 119367 32 -105213 -10156 190140 6036 467947 516535 532849 761124 585621 621312

33 10931 79212 75852 187400 200369 128800 469738 133613 253222 34 248041 137567 268199 222857 308453 210574 910722 1993567 995017 1779157 35 52646 51573 87058 70408 93739 70309 107619 111337 155064 107316 36 1345424 472573 1510732 11323014 2015300 3807224 1800627 1273333 1582489 3952318 37 -13700 -21187 68065 54784 80593 91007 72786 63826 176182 428780 38 45513 32142 25579 35839 19440 17850 26064 29807 47816 39 9560 34841 60753 68445 135139 221719 149368 411666 562080 701793 40 29174 38681 33191 29881 30798 4166 129418 30163 46979 54658 41 18597 72585 18327 98359 133774 615061 208983 325346 1898502 730881 42 188093 516420 419983 821620 395056 656779 774203 45 99698 110809 137554 151665 61349 138147 164674 599056 112590 142272 46 48531 48171 46214 41010 54711 36107 33368 34706 81768 47 36521 60248 72734 52686 127082 76055 244259 115186 98687 151781 48 48739 52280 100769 106357 124762 230807 117377 178786 96018 209947 49 25444 40608 28163 41242 27060 38997 62022 108244 24725 -30144 50 36295 -71033 -70131 34151 -30737 71707 40252 263803 198636 307327 51 43001 53458 94269 80481 83269 68276 74407 130823 53 -30505 7325 17264 7403 -1574 31562 114412 26160 52368 264445 54 52530 40730 49488 40624 55202 41704 44574 57890 69501 127552 55 36802 43219 43047 51429 36680 34196 45095 63271 107941 141276 56 34740 53207 55775 29898 46100 83532 96549 95596 94351 54164

The empty blocks in the table are due to the lack of data. Three states have only a few

observations, which are not enough to run such a regression with so many variables

and interaction terms. We can find that the BI varied a lot not only across states but

also across years. The lowest BI (-601089.61) is due to State 6 (California), Year 1993.

That is the only negative BI of California. The highest BI (11323014) is due to State

36 (New York), Year 1987. That is the only BI higher than 5000000 in the table.

Most indicators are positive. Only 17 indicators are negative, which theoretically

means that the more efficient banks obtained less profit, in those 17 markets. This

situation is possible. If all firms in a market have decreasing marginal profits

according to scale (because of higher management cost of more complex firm

structure), BI would become negative. Additionally, the BI can be less than zero when

the scales of efficient firms are limited (because of the legislation, the policy or the

unobtainability of some essential factors). However, an important shake in the

accounting data of some huge banks in a market can also make the coefficient

deviating from the competition level, such as merger, failure or large investment in

fixed capital.

The graphs below can help us roughly compare BI across states and years.

0 50 00 00 0 10 00 00 00 B o on e I n di c at or 1984 1986 1988 1990 1992 1994 Year

Graph 1

Graph 2-10

Graph 20-28

Graph 38-43

We can intuitively observe a slightly positive relationship between BI and time in the

first graph. In the line charts we can find that the BIs varied a lot in some states, and

remained quite stable in others. BI of most states waved upward. Most states with

steady BIs are agricultural states such as South Dakota (state code 46), Mississippi

(state code 28) and Montana (state code 30). Most states with sharply waving BIs are

industrial states, such as Massachusetts (state code 25) and New York (state code 36).

In order to test whether the positive relationship between BI and Year is statistically

meaningful, I would like to apply some simple regressions.

st st

BI

= +

α β

Year

+

e

₍₁₇₎

1993 1985 st t t st t

BI

α

β

YearDummy

e

=

= +

∑

+

(18)

In this regression, all available BIs of every state, every year are involved. The sample

is equal to the group. Thus, Hausman test is unnecessary here. I will still use fixed

error panel regression. The result is below:

Table 4

(16) (17) VARIABLES BI BI d85 -26,583 (38,850) d86 14,586 (37,662) d87 255,090 (216,973) d88 114,418** (56,033)

d89 198,427** (79,037) d90 152,704*** (51,904) d91 172,095*** (58,512) d92 166,537*** (61,375) d93 267,128*** (86,560) Year 25,820*** (6,411) Constant -5.110e+07*** 106,874* (1.275e+07) (53,236) Observations 466 466 R-squared 0.023 0.040 Number of state 48 48

Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

The result of model (17) indicates that BI significantly increases with Year. BI tends

to go up by 25820 every year. The result of model (18) shows a clear procedure that

BI increased step by step every year. In 1985, 1986 and 1987, BI was not significantly

different from what it was in 1984. In 1988 and 1989, the differences became

significant and after that, the differences were dramatically significant.

In order to robust the analysis, I run a Wilcoxon rank-sum (Mann-Whitney) test. The

result of this test shows the BI in 1984-1988 is significantly lower than in 1989-1993

(z=-7.034).

This meets the reality. During 1984-1993, the commercial banking sector in the U.S.

deteriorated dramatically. The competition was fierce and a large number of banks

break down. In the year of 1990, there were 536 bank failures. The number of

commercial banks decreased by almost one third, from 8157 (Year 1984) to5457

(Year 1993).

Graph 50 (Jones and Critchfield, 2005)

At the beginning of that decade, the price of oil increased sharply. That increased the

costs of many industries in America. In the middle of 1980s, dollar-yen exchange rate

reached 1:250. The appreciation of dollar decreased the international demand of

American products (Berger etc. 1995). As a result, many firms in America were under

production and the demand of loans decreased. Additionally, the interest rate

increased a lot. Because of Regulation Q, banks cannot provide interest to current

deposit and the interest rate for time deposits or

savings deposits was limited. The

return rate of bank was much lower than other financial institutions. Thus, banking

industry was not attractive for capitals and banks met strong competition from other

financial institutions (Demsetz etc. 1997). Lower demand for loans and stronger

competition from other financial institutions made a higher degree of competition in

the banking sector in USA.

industries, so the states with large amount of agriculture or mining industry were

affected less by the increasing price than the states with high level of industrialization.

That is a reason why some states had stable BIs but others’ BIs varied a lot.

At the same time, the concentration level did not decrease when the competition rose.

On the contrary, the banking industry began to be concentrated among several largest

financial institutions of the nation at the end of this period (Jones and Critchfield,

2005). The degree of concentration fails to describe the competition level here.

5.2.Long-term result

The data about the long-term competition level are in the appendix. The table below

shows the summary information of long-term competition level across states.

Table 5

state

BI-mean

BI-var

Long-term BI N

H

Mean 239774.3125 2.31232E+11

197760.0833

156.7145833 0.772458333

Max

2908303

8.94893E+12

2016105

962.4

2.466

Min

31117

97747958

6947

4.9

-0.971

In this table, BI-mean refers to the average of short-term BI for every state. BI-var

refers to the variation of short-term BI for every state. Long-term BI is obtained from

regression model (16). N refers to the number of observations per year. H is the

H-statistic of PR approach. I would like to compare BI with PR approach here. To get

H-statistic, I will apply the regression model below:

1 2 3 1 2

ln(

R

_it

/

TA

_it

)

= +

α β

ln( 1 )

w

_it

+

β

ln( 2 )

w

_it

+

β

ln( 3 )

w

_it

+

γ

ln( 1 )

y

_it

+

γ

ln( 2 )

y

_it

+

ξ

_it

(19)

Equation (19) is the regression model of Bikker, Jacob, Shaffer and Spierdijk’s work

(2012). In the model, R is the revenue of bank i, year t, TA is the total assets, w1 is the

ratio of interest expense to liabilities, w2 is the ratio of personnel expense to total

assets, w3 is fixed asset expenses over fixed assets (%), y1 is equity over total assets

(%), and y2 is the ratio of loans to total assets. I run the regression with fixed effect.

1 2 3

H

−

statistics

=

β β

+

+

β

(20)

The larger the H-statistics is, the higher is the competition level. Theoretically

speaking, the maximum of H is 1, which means perfect competition.

When I arranged the data, I found the states with more observations seem to have

more stable BIs. In order to proof that, I run a regression with this model:

ln(BIvar/ BImean)

_i

= +

α β

ln(

nb

)

_i

+

ξ

_i

(21)

BIvar is the variation of BI of State i among the 10 years. BImean is the mean of BI of

State i among the 10 years. nb is the number of banks per year of State i. I use natural

logarithm to make the distribution of residues normal. This is the result of the

regression.

Table 6

(21) VARIABLES ln(nb) -0.536*** (0.156) Constant 13.30*** (0.701) Observations 48 R-squared 0.128

Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

The result indicates that the number of banks has a significant effect on the stability of

competition level. A state with more banks changed its BI less. This is easy to

understand. In a market with only a few banks, the behavior of every bank may

influence the competition level a lot. But in a huge market, the influence can be

diluted.

The correlation between BI and H-statistics of PR approach is quite low, only -0.1118.

The p-value of correlation is 0.4493, which indicates BI and H-statistics have

insignificant correlation. An important reason is that PR approach is not suitable here.

PR approach asks for long term equilibrium. However, during that decade, “the

structure of the U.S. banking industry indeed underwent an almost unprecedented

transformation—one marked by a substantial decline in the number of commercial

banks and savings institutions and by a growing concentration of industry assets in a

few dozen extremely large financial institutions.” (Jones and Critchfield, 2005)

Because the situation is against the assumption of PR approach, PR approach fails to

describe the competition level here. That is an important reason why some H-statistics

in the table are larger than 1.

6. Limitation

6.1.Limitation of Boone approach

During this study, I find some limitations of Boone approach.

A shortage of Boone approach is that, it is difficult to find an index to describe the

efficiency of a firm. The marginal cost is difficult to obtain and ROA can be affected

by fixed costs.

Additionally, we can only get Boone Indicator in a market with enough observations

for regression. However, we always care the market with only a few competitors,

especially during policy making. Markets with fewer competitors tend to contain

more monopoly power and markets with enough observation are often competitive

markets. That limits the practicability of Boone Indicator.

6.2.Limitation of this research

This study can only show that

t

he competition level described by BIs meets some

parts of the situation but cannot provide any proof of the validity of the competition

degree described by BIs.

7. Conclusion

This study provides both short-term and long-term Boone Indicators of 48 states of

the USA, during 1984 and 1993. The result shows that the competition level tends to

increase during that decades. BIs in industrial states varied within a wider range than

BIs in agricultural states did. the concentration level fails to describe the competition

level and PR approach is not suitable because the banking market during that decade

deviated from the long-run equilibrium. The competition level described by BIs was

in line with some parts of the reality.

Based on my result, many empirical researches about degree of competition in

banking sector can be retested. Some relative studies are: the relationship between

competition level of banking industry and interest rate of loan (Hoff, 1995; Repullo et

al., 2008), the relationship between competition level of banking industry and stability

of it (Vives, 2010), the relationship between competition level and born rate of local

firms (Patti et al., 2000; Cetorelli et al., 2003) and the relationship between

competition level and economic growth (Jayaratne and Strahan, 1996; Levine, Loayza

and Beck, 2000; Collender and Shaffer, 2003).

Reference

Boone, J. 2008. A new way to measure competition, Economic Journal, 118, 1245–61.

Claessens and Laeven. 2003. What Drives Bank Competition? Some International Evidence, Journal of Money, Credit and Banking, Vol. 36, No. 3, Part 2.

Gual, J. 1999. Deregulation, integration and market structure in European banking, Journal of the Japanese and International Economies, 13, 372–96.

Johnson and Rice. 2007. Assessing a Decade of Interstate Bank Branching, Federal Reserve Bank of Chicago, WP 2007-3.

Leuvensteijn, Sørensen, Bikker & Rixtel. 2013. Impact of bank competition on the interest rate pass-through in the euro area, Applied Economics, 45:11, 1359-1380

Vives, X. 2010. Competition and stability in banking, CEPR Policy Insight No. 50.

Ray, Subhash C. 1982. A translog cost function analysis of US agriculture, 1939–77. American Journal of Agricultural Economics 64.3: 490-498.

Van Leuvensteijn, Michiel. 2008. The Boone-indicator: Identifying different regimes of competition for the American Sugar Refining Company, 1890-1914. No. 08-37.

Bonaccorsi di Patti, Emilia, and Giovanni Dell'Ariccia. 2004. Bank competition and firm creation. Journal of Money, Credit, and Banking 36.2: 225-251.

Cetorelli, Nicola, and Philip E. Strahan. 2006. Finance as a barrier to entry: Bank competition and industry structure in local US markets. The Journal of Finance 61.1: 437-461.

Martinez-Miera, David, and Rafael Repullo. 2010. Does competition reduce the risk of bank failure?, Review of Financial Studies 23.10: 3638-3664.

Hoff, Karla, and Joseph E. Stiglitz. 1998. Moneylenders and bankers: price-increasing subsidies in a monopolistically competitive market. Journal of Development Economics 55.2: 485-518.

Spong, Kenneth, Richard J. Sullivan, and Robert DeYoung. 1995. What makes a bank efficient? A look at financial characteristics and bank management and ownership structure. Federal Reserve Bank of Kansas City, Financial Industry Perspectives, Decembe: 1-19.

Panzar-Rosse model: The role of scale, costs, and equilibrium. Review of Economics and Statistics 94.4: 1025-1044.

Jones K D, Critchfield T. 2005. Consolidation in the US banking industry: Is the ‘long strange trip’about to end[J]. FDIC Banking Review, 2005, 17(4): 31-61.

Johnson, C.& Rice, T. 2007. Assessing a Decade of Interstate Bank Branching, Federal Reserve Bank of Chicago working paper

Shaffer, S. 1994. Bank Competition in Concentrated Market, Business Review, March/April Christensen, L.R., Jorgenson, D. &Lau, L.J. 1973. Transcendental Logarithmic Production

Frontiers, Review of Economics and Statistics 55, 28-45.

Guevara et al. 2005. Market Power in European Banking Sectors, Journal of Financial Services Research 27:2 109–137

Casu, B. &Gerardone, C. 2006. Bank Competition,Concentration and Efficiency in the Single European Market, The Manchester School Vol 74 No. 4 Special Issue

Angelini, P. &Cetorelli, N., The Effects of Regulatory Reform on Competition in the Banking Industry, Journal of Money, Credit, and Banking, Vol. 35, No. 5

Cetorelli, N. &Stranhan, P.E., Bank Competition and Industry Structure in Local U.S. Markets, The Journal of Finance, Vol. LXI, No. 1

J.A. Bikker and K. Haaf, Measures of competition and concentration in the banking industry: a review of the literature, Research Series Supervision no. 27, De Dederlandsche Bank Valverde et al. 2003. Deregulation, Bank Competition and Regional Growth, Regional Studies,

Vol. 37.3, pp. 227–237.

Berger, Allen N., Anil K. Kashyap, and Joseph M. Scalise. 1995 The Transformation of the U.S. Banking Industry: What a Long, Strange Trip It’s Been, Brookings Papers on Economic Activity 2.

Demsetz, Rebecca S., Mark R. Saidenberg, and Philip E. Strahan. 1997. Agency Problems and Risk Taking at Banks. Federal Reserve Bank of New York Staff Report 29.

Appendix

1. Results of Boone Indicator every year every state

t85 refers to the time dummy of year 1985 and r85 refers to the interaction term between roa and t85.

(1) (2) (3) (4) (5)

State 1 State 2 State 4 State 5 State 6

VARIABLES II II II II II roa 20,926*** 923,835*** 135,337* 48,432*** 65,327** (5,226) (72,065) (74,352) (15,363) (29,928) t85 -223.8 281.1 -1,312 (147.0) (993.3) (1,165) t86 102.7 796.6 966.7 -673.3 504.5** (106.7) (1,032) (1,452) (415.0) (217.0) t87 -21.89 528.3 549.5 -58.64 1,777 (151.3) (339.2) (2,543) (365.9) (1,435) t88 80.16 481.4 -61.38 -34.59 4,235 (89.34) (1,013) (1,971) (186.9) (3,926) t89 -132.6 -2,129* -13,081 -77.65 2,943* (195.1) (1,035) (9,209) (129.0) (1,668) t90 111.2 861.7 -8,298* -99.19 3,005* (160.4) (943.9) (4,852) (164.8) (1,627) t91 -52.33 819.2 -3,683 400.0** -3,111 (312.6) (1,114) (2,519) (176.6) (4,020) t92 -131.6 -1,173 -2,572 403.6 (208.5) (876.8) (2,098) (261.4)

t93 -58.95 2,902 698.1 -392.5 6,411 (217.3) (2,621) (5,242) (322.9) (4,904) r85 29,445** 18,130 143,880 (11,850) (51,345) (121,748) r86 10,016 -806,497*** 46,445 77,817* 20,934 (8,499) (76,547) (68,078) (42,513) (19,734) r87 26,262** -800,352*** 29,562 7,464 (13,208) (83,866) (37,750) (22,592) r88 19,759** -831,015*** 121,762 19,566 194,268* (8,542) (95,714) (110,959) (17,735) (102,066) r89 44,990** -510,772*** 112,073 25,066** 185,290* (17,966) (130,378) (183,501) (11,284) (112,288) r90 24,752* -768,374*** -80,021 41,740*** 147,465* (14,449) (133,747) (85,444) (15,738) (82,113) r91 45,432* -749,061*** -29,789 11,105 321,174 (27,172) (109,885) (66,830) (15,694) (221,927) r92 61,659*** -623,566*** 40,519 19,099 (16,213) (99,734) (156,943) (22,246) r93 66,572*** -896,606*** 12,583 108,859*** -666,416 (17,934) (205,913) (284,064) (30,207) (623,911) Constant 354.0*** -485.4** 3,945** 191.7 974.6 (63.03) (153.0) (1,760) (119.3) (1,480) Observations 1,239 44 275 1,364 1,878 R-squared 0.402 0.873 0.149 0.277 0.012 Number of Id 194 10 52 150 370

Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

(1) (2) (3) (4) (5)

State 8 State 9 State 11 State 12 State 13

VARIABLES II II II II II roa 27,697*** 117,592 286,166 42,668** 25,937** (4,980) (150,387) (171,758) (19,635) (12,166) t85 -10.15 -1,238 -53.91 228.5 280.0 (46.60) (1,634) (3,022) (159.9) (216.6) t86 -43.33 -2,664 3,197 653.8** 107.3 (52.05) (2,474) (1,912) (258.7) (345.3) t87 -88.39 -1,573 -491.4 959.8** 538.1* (65.49) (1,878) (2,803) (484.2) (324.1) t88 -71.66 -8,167* 2,331 1,228** 636.8* (79.23) (4,508) (1,805) (580.2) (356.3) t89 -50.47 -13,001** -1,419 271.3 697.7* (82.57) (6,489) (2,265) (1,001) (401.5) t90 -194.8* -12,467* -6,215 -578.8 170.9 (111.5) (6,881) (4,937) (1,211) (512.5) t91 -70.65 -11,684* -6,231 316.4 46.85 (93.80) (5,880) (10,558) (820.8) (566.1) t92 59.50 -6,416 -2,826 1,500 457.3 (115.9) (5,389) (3,002) (1,136) (529.9) t93 -40.79 -3,010 -11,032 2,400* 812.8* (149.6) (5,645) (7,614) (1,389) (472.3) r85 -2,862 157,137 152,620 28,787 -853.3 (4,278) (171,191) (480,682) (21,602) (11,557) r86 481.2 343,108 -51,046 30,238** 23,492 (3,623) (258,679) (194,379) (14,451) (24,003) r87 -5,822 254,817 155,707 18,952 3,381

(4,825) (199,322) (302,076) (17,282) (12,355) r88 3,863 1.094e+06** 385,974 68,599* 12,778 (4,936) (519,293) (278,333) (35,640) (13,447) r89 -2,796 1.236e+06 842,649* 94,448* 16,531 (5,754) (883,446) (490,064) (50,941) (13,654) r90 12,084 326,703 1.100e+06 117,751* 54,174* (8,499) (287,854) (989,115) (66,383) (29,021) r91 12,987 81,145 777,961** 109,178*** 76,653** (9,328) (179,643) (324,298) (35,880) (38,147) r92 20,944** 380,732 234,540 123,573*** 62,480** (8,626) (339,129) (247,560) (31,510) (25,909) r93 40,303*** 1.013e+06* 974,036 221,782** 37,373* (13,693) (598,420) (627,441) (90,758) (20,754) Constant 204.1*** 7,643** 5,913 1,370** 527.8 (53.53) (3,564) (3,540) (552.0) (331.3) Observations 2,732 464 186 2,811 3,778 R-squared 0.311 0.186 0.381 0.035 0.045 Number of Id 429 76 24 476 503

Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

(1) (2) (3) (4) (5)

State 16 State 17 State 18 State 19 State 20

VARIABLES II II II II II

roa 45,874 1.017e+06 41,722*** 44,378*** 24,606***

t85 -470.1 8,581 112.7 60.26 (942.0) (8,508) (82.69) (55.84) t86 -754.8 8,778 151.4 86.40 (1,191) (8,377) (92.48) (74.76) t87 553.7 5,519 162.6* 196.5*** -3.389 (1,111) (5,164) (91.69) (58.67) (89.19) t88 422.9 9,864 45.14 78.26 (1,513) (9,178) (108.5) (80.38) t89 2,002 9,396 140.0 119.7 108.1 (1,331) (9,029) (120.4) (82.36) (90.79) t90 1,486 8,947 -731.7 82.20 -129.1 (1,394) (8,885) (455.5) (122.4) (133.5) t91 1,123 9,134 -1,977* 320.4*** 83.92 (1,417) (8,718) (1,017) (73.54) (115.3) t92 -2,161 9,631 -436.1 92.84 -254.0 (3,432) (9,024) (454.3) (141.2) (242.2) t93 1,148 8,622 -430.5 162.0* -440.3 (4,195) (8,337) (1,043) (83.97) (310.9) r85 -102,935 -883,719 -3,017 -1,071 (120,008) (886,093) (7,586) (6,058) r86 -97,712 -902,252 -1,301 -6,960 (134,582) (878,519) (8,380) (7,952) r87 -6,054 -604,010 13,214 -13,939 9,769 (124,270) (588,602) (8,988) (8,464) (13,135) r88 60,173 -964,800 7,040 1,104 (141,406) (927,119) (13,210) (7,711) r89 152,787 -918,296 44,610*** 4,885 1,327 (155,683) (927,069) (14,040) (12,538) (8,880) r90 226,685 -884,159 123,617*** 7,887 27,226**

(172,508) (929,339) (43,965) (12,009) (13,018) r91 336,944* -895,594 266,195** -14,052* 13,204 (186,586) (911,994) (107,499) (7,707) (9,708) r92 838,129 -893,933 118,793*** 17,901 58,430** (532,902) (914,896) (39,162) (16,736) (22,604) r93 527,032 -742,641 196,730** 14,781 85,562** (504,933) (814,999) (93,697) (9,909) (38,494) Constant 2,486* -8,834 560.0*** 70.31* 134.2* (1,263) (8,884) (92.50) (41.86) (75.38) Observations 160 6,594 1,849 2,235 3,136 R-squared 0.348 0.033 0.170 0.434 0.096 Number of Id 23 816 280 342 402

Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

(1) (2) (3) (4) (5)

State 21 State 22 State 23 State 24 State 25

VARIABLES II II II II II roa 22,173*** 51,977*** 16,191 635,352 -229,088 (6,625) (9,886) (60,366) (454,795) (267,705) t85 -14.56 -86.02 1,735 4,724 -2,380 (74.77) (90.30) (2,960) (3,215) (2,697) t86 -23.45 -523.3*** 935.9 4,974 -3,299 (132.5) (163.8) (1,794) (3,686) (3,182) t87 -140.8 -645.9*** -6,516 3,734 -3,598 (132.9) (210.7) (4,747) (3,605) (3,853)

t88 -56.37 -766.5** -10,087 2,858 -1,210 (128.0) (314.8) (9,241) (2,872) (2,754) t89 213.6 -1,322* -3,938 5,110 -17,080 (159.7) (772.6) (3,486) (3,218) (11,650) t90 -433.9 -276.8 -1,655 -4,171 -7,244** (285.8) (167.8) (2,680) (5,713) (3,251) t91 201.7 -126.1 -3,756 1,395 -4,110 (266.3) (162.6) (3,965) (7,326) (2,758) t92 -349.4* 652.5** -3,306 4,762 (194.2) (326.3) (4,104) (4,572) t93 -389.0 -1,976 -6,748 2,486 48,858 (323.1) (2,277) (5,562) (3,336) (33,856) r85 11,514** 9,280 -75,499 -416,695 368,651 (5,769) (7,790) (239,851) (330,789) (324,922) r86 19,579* 4,242 36,234 -358,060 578,854 (10,327) (10,988) (132,153) (340,009) (433,657) r87 36,038*** 10,966 753,950 -208,792 426,461 (12,542) (15,479) (469,852) (300,903) (347,413) r88 34,883*** 27,423 1.275e+06 10,172 522,085 (9,444) (17,859) (866,380) (241,091) (395,407) r89 17,413 154,568 914,664** -272,466 1.339e+06 (13,953) (97,391) (426,737) (277,152) (1.028e+06) r90 59,847*** 53,550*** 785,208 276,178 547,640 (22,846) (15,726) (514,310) (520,860) (358,474) r91 1,741 51,936** 518,381 472,053 503,510 (16,935) (23,281) (349,811) (477,223) (335,618) r92 81,095*** 19,992 561,845 -327,385 (16,519) (34,558) (350,280) (389,582) r93 94,581*** 249,257 1.238e+06* -89,180

(31,798) (206,678) (595,020) (288,776) Constant 617.9*** 481.4*** 3,428* -2,902 4,965 (76.60) (117.8) (1,672) (3,967) (3,145) Observations 1,822 1,540 125 492 546 R-squared 0.123 0.122 0.421 0.108 0.143 Number of Id 207 220 21 64 107

Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

(1) (2) (3) (4) (5)

State 26 State 27 State 29 State 30 State 31

VARIABLES II II II II II roa 73,725*** 49,657** 10,161* 36,867*** 23,329*** (24,859) (21,047) (6,123) (6,562) (4,638) t85 -62.61 83.84 -8.447 -40.67 (176.0) (152.0) (91.52) (83.56) t86 -139.4 261.2 280.0** -38.39 58.98 (202.8) (161.3) (128.9) (75.84) (41.22) t87 -896.0 -1,246 102.2 10.60 122.3** (1,122) (1,221) (140.3) (85.47) (59.12) t88 389.1 -1,158 105.6 -314.9** 190.0*** (467.0) (1,183) (159.1) (133.9) (59.08) t89 903.3* -2,244 -280.4 -12.59 84.50 (471.3) (2,440) (512.8) (74.93) (96.16) t90 1,148** -135.6 -50.26 222.1 16.98 (577.5) (276.1) (285.2) (201.3) (208.3) t91 842.2 -773.9 -931.2 192.4** 163.9**

(724.6) (1,274) (710.0) (87.66) (68.25) t92 735.9 -4,501 441.7 536.0*** 104.2 (540.5) (5,255) (415.8) (201.9) (78.09) t93 413.1 -1,857 -307.0 754.3*** -437.8 (521.9) (1,746) (767.7) (251.3) (392.4) r85 21,982 12,265 17,782** 526.9 (18,176) (14,977) (8,130) (6,917) r86 45,087** -328.3 9,800 -1,748 1,748 (20,288) (13,679) (6,013) (8,309) (6,371) r87 113,869 156,843 32,723*** 1,923 -1,854 (105,018) (146,401) (9,201) (7,187) (6,826) r88 41,201 138,157 47,599*** 36,485** -4,143 (42,022) (103,489) (14,615) (16,092) (5,821) r89 28,030 269,914 98,882** 8,341 11,464 (34,407) (238,984) (47,272) (7,218) (7,020) r90 41,720 96,941 77,148*** 3,014 12,052 (36,774) (60,065) (22,688) (23,844) (16,682) r91 96,489*** 239,344 178,092*** -1,937 11,823 (35,618) (215,112) (63,353) (7,487) (7,977) r92 88,353** 605,633 69,028*** -19,126 23,807** (40,908) (573,673) (23,562) (13,887) (10,255) r93 142,234*** 334,993 178,580*** -34,004** 96,038 (47,414) (266,621) (57,199) (14,602) (58,894) Constant 552.6* 124.5 471.4*** 35.06 72.70 (294.2) (312.8) (149.7) (68.73) (82.92) Observations 1,761 2,772 3,140 957 1,862 R-squared 0.099 0.036 0.115 0.332 0.085 Number of Id 275 399 494 117 276

*** p<0.01, ** p<0.05, * p<0.1

(1) (2) (3) (4) (5)

State 32 State 33 State 34 State 35 State 36

VARIABLES II II II II II roa -105,213 10,931* 248,041 52,646*** 1.345e+06 (71,477) (6,458) (162,441) (13,777) (860,704) t85 -1,380 -448.9 1,872* 15.87 11,789 (1,216) (307.7) (1,056) (173.5) (7,324) t86 -1,547 -175.1 1,806 -91.79 -277.9 (1,567) (254.5) (1,200) (191.9) (9,225) t87 -390.1 -1,129 1,419 22.17 -114,273** (1,018) (704.8) (1,586) (228.1) (53,708) t88 -2,819 -989.9** 2,406 -198.9 13,188 (2,103) (428.0) (1,582) (168.7) (9,712) t89 -1,823 -330.3 2,706* -31.88 -42,642 (2,574) (401.1) (1,554) (140.4) (30,083) t90 -5,550 195.6 -8,441*** -596.7 -10,551 (6,695) (991.0) (3,225) (699.3) (15,413) t91 -7,777 -8,183 -453.8 -1,636 (7,125) (6,268) (500.6) (10,205) t92 -5,576 93.60 -1,000 -796.1 17,480 (3,457) (783.5) (3,634) (1,024) (13,716) t93 -5,436 -314.9 803.8 256.5 13,725 (5,702) (1,302) (5,539) (383.8) (14,874) r85 95,057 68,281** -110,474 -1,073 -872,852 (89,748) (27,242) (108,806) (16,805) (715,409) r86 295,353 64,921*** 20,158 34,412* 165,307

(337,516) (19,936) (123,749) (18,245) (834,190) r87 111,249 176,469** -25,184 17,762 9.978e+06** (64,047) (70,562) (152,180) (18,686) (4.803e+06) r88 573,160** 189,438*** 60,412 41,094*** 669,876 (257,811) (50,771) (149,836) (13,470) (1.028e+06) r89 621,748 117,869*** -37,467 17,664 2.462e+06 (407,618) (34,353) (150,794) (13,513) (2.071e+06) r90 638,062 458,807** 662,681** 54,973 455,203 (465,567) (188,646) (332,292) (51,081) (834,830) r91 866,337 1.746e+06 58,691 -72,091 (593,626) (1.188e+06) (49,148) (842,536) r92 690,834 122,682 746,976** 102,419 237,065 (413,009) (92,270) (333,063) (73,553) (875,680) r93 726,525 242,291 1.531e+06* 54,671 2.607e+06* (556,177) (209,957) (814,628) (52,377) (1.488e+06) Constant 5,637*** 886.1*** 3,226** 417.4*** 6,472 (1,090) (202.5) (1,391) (105.5) (9,847) Observations 109 315 806 648 933 R-squared 0.244 0.401 0.237 0.243 0.173 Number of Id 16 57 128 76 134

Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

(1) (2) (3) (4) (5)

State 37 State 38 State 39 State 40 State 41

VARIABLES II II II II II

(134,397) (9,192) (30,103) (4,027) (68,066) t85 2,113 229.1 -295.6 143.5 (1,612) (224.4) (200.9) (262.3) t86 3,931 64.52 376.2* -316.5*** 363.5 (2,378) (72.76) (192.6) (102.4) (482.0) t87 3,797 182.1** 29.62 -267.6*** 142.5 (2,354) (80.77) (444.7) (97.14) (1,709) t88 5,614** 25.66 553.7 -191.4** 1,463 (2,725) (233.7) (846.7) (96.82) (978.6) t89 6,095** 360.9** -499.2 -17.21 -1,619 (2,825) (167.7) (2,068) (131.3) (3,522) t90 5,701** 467.4** -660.1 -612.2 263.9 (2,631) (197.1) (1,489) (384.3) (1,050) t91 4,592** 425.4** -2,138 -30.26 -1,616 (2,302) (205.1) (2,234) (172.4) (3,150) t92 8,328** 407.7 -2,348 -85.10 -14,592* (3,468) (356.1) (1,586) (212.9) (8,006) t93 9,049** 269.6 -2,839* -35.76 -2,299 (4,281) (390.6) (1,633) (254.1) (4,592) r85 -7,487 25,281 9,507 53,988 (95,582) (18,514) (14,110) (43,931) r86 81,765 -13,371 51,193*** 4,017 -269.5 (135,193) (8,073) (16,475) (4,936) (57,157) r87 68,484 -19,934** 58,885** 707.1 79,763 (127,782) (9,275) (26,677) (4,361) (98,565) r88 94,293 -9,674 125,579 1,624 115,178 (134,301) (18,427) (77,881) (4,280) (127,537) r89 104,707 -26,073*** 212,159 -25,008*** 596,464 (139,806) (8,570) (158,041) (4,083) (500,756)

r90 86,486 -27,664*** 139,809 100,244 190,387 (145,096) (9,435) (96,063) (72,627) (129,273) r91 77,526 -19,449** 402,107** 988.5 306,749 (140,867) (9,515) (173,203) (11,358) (253,545) r92 189,883 -15,706 552,520*** 17,805 1.880e+06* (174,724) (20,704) (161,939) (13,647) (972,136) r93 442,480 2,303 692,233*** 25,484 712,284 (361,951) (19,975) (213,163) (18,914) (625,393) Constant 4,233 -51.65 2,360*** 246.8** 2,582*** (2,566) (134.5) (359.9) (95.97) (759.1) Observations 679 679 1,488 2,522 308 R-squared 0.081 0.176 0.088 0.175 0.194 Number of Id 99 97 211 361 55

Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

(1) (2) (3) (4) (5)

State 42 State 45 State 46 State 47 State 48

VARIABLES II II II II II roa 188,093 99,698 48,531*** 36,521*** 48,739*** (158,827) (74,472) (13,256) (10,148) (7,538) t85 98.93 -150.0 -117.4 (675.8) (168.1) (78.94) t86 318.2 -139.9 -149.0 -866.6*** (1,126) (324.2) (261.4) (182.5) t87 -2,531 733.5 79.04 183.4 -1,375*** (3,235) (970.3) (205.0) (210.2) (393.3)

t88 2,028 228.4 -592.5 -1,183*** (1,225) (261.2) (565.9) (423.0) t89 -526.2 1,995 70.04 83.54 -1,159*** (3,360) (1,265) (381.7) (264.4) (395.2) t90 -5,880* 1,441 355.5 -1,972 -643.1** (3,252) (1,285) (334.2) (1,545) (312.9) t91 -8.952 -3,727 509.8 -170.4 -876.4 (3,078) (5,443) (439.6) (282.4) (569.9) t92 140.3 2,507* 577.7 373.2 176.3 (3,317) (1,437) (538.6) (383.8) (396.3) t93 -468.8 4,357 318.3 578.5 -344.7 (3,541) (2,842) (353.5) (445.5) (729.7) r85 11,111 23,727* 3,541 (53,504) (13,788) (5,983) r86 37,855 -360.3 36,213* 52,030** (80,627) (19,810) (21,274) (24,162) r87 328,326 51,967 -2,317 16,165 57,618*** (236,780) (73,802) (12,719) (16,319) (19,038) r88 -38,349 -7,521 90,561** 76,022 (56,569) (12,200) (44,964) (50,419) r89 231,890 38,449 6,179 39,534** 182,067* (271,129) (67,343) (25,244) (19,235) (103,242) r90 633,526** 64,976 -12,424 207,738 68,638** (319,693) (66,893) (13,680) (139,744) (26,843) r91 206,962 499,357 -15,163 78,665*** 130,047* (214,194) (624,936) (18,311) (24,392) (70,888) r92 468,686** 12,892 -13,825 62,166*** 47,279 (216,347) (52,854) (24,956) (21,167) (34,247) r93 586,110** 42,574 33,236 115,260 161,208

(247,901) (75,834) (38,310) (77,040) (106,052) Constant 2,326 801.9 38.96 749.4*** 937.4*** (2,288) (1,114) (284.1) (148.7) (138.4) Observations 1,067 428 562 1,554 9,624 R-squared 0.104 0.190 0.101 0.083 0.035 Number of Id 239 59 87 214 1,594

Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

(1) (2) (3) (4) (5)

State 49 State 50 State 51 State 53 State 54

VARIABLES II II II II II roa 25,444*** 36,295 43,001** -30,505 52,530*** (8,679) (34,986) (17,616) (20,861) (10,937) t85 -60.78 1,329* 338.0 146.5 (204.2) (730.4) (443.0) (135.5) t86 -172.7 1,431 267.8 181.9 (214.4) (1,233) (499.5) (159.6) t87 -738.3 377.3 76.97 617.2* 330.3** (485.3) (864.3) (282.0) (334.9) (152.7) t88 -57.58 1,584 361.2 1,357** 242.5** (305.9) (1,411) (243.3) (653.3) (122.3) t89 256.7 487.8 741.5* 2,133** 448.4*** (349.3) (516.4) (434.4) (1,001) (122.0) t90 524.1 -696.4 -2,034 2,707* 479.1*** (432.5) (435.2) (1,228) (1,378) (143.2) t91 -53.90 449.1 -629.5 2,274* 464.9***

(998.3) (694.9) (675.2) (1,186) (147.7) t92 1,528* -301.6 -1,009 2,845** 495.5*** (804.2) (452.7) (1,359) (1,198) (177.0) t93 2,804** -699.5* 197.2 3,207 46.26 (1,205) (382.9) (1,789) (1,984) (287.8) r85 15,164 -107,328 37,830 -11,800 (16,678) (70,978) (22,881) (11,571) r86 2,719 -106,426 47,770** -3,042 (10,326) (118,337) (20,445) (14,237) r87 15,797 -2,144 10,457 37,908* -11,906 (23,737) (86,640) (8,764) (21,488) (14,505) r88 1,615 -67,032 51,268** 28,931 2,672 (19,368) (103,817) (23,360) (19,129) (10,282) r89 13,553 35,412 37,480** 62,067* -10,826 (15,212) (45,575) (15,453) (32,992) (9,877) r90 36,578** 3,956 40,268 144,918 -7,956 (15,417) (35,905) (33,911) (108,146) (12,334) r91 82,800 227,508*** 25,274* 56,665** 5,360 (59,565) (56,059) (12,830) (25,048) (11,608) r92 -719.4 162,340*** 31,406 82,873** 16,970 (13,731) (46,008) (38,047) (32,837) (12,744) r93 -55,588* 271,031*** 87,821 294,951** 75,022*** (29,344) (76,795) (59,407) (145,554) (28,173) Constant 451.4* 792.4** 3,314*** 1,730** 94.00 (246.3) (288.3) (347.0) (688.8) (133.7) Observations 279 136 737 589 1,786 R-squared 0.251 0.850 0.040 0.083 0.259 Number of Id 54 18 114 91 233

*** p<0.01, ** p<0.05, * p<0.1 (1) (2) State 55 State 56 VARIABLES II II roa 36,802*** 34,740*** (9,258) (10,638) t85 22.82 -49.37 (41.47) (77.84) t86 95.20 -51.39 (92.21) (94.87) t87 43.08 -26.89 (89.63) (113.8) t88 245.4** 55.31 (118.4) (99.06) t89 262.6 -10.47 (220.4) (243.8) t90 293.1** -26.50 (136.7) (344.1) t91 111.2 367.1 (259.1) (380.0) t92 -147.0 80.73 (195.6) (399.7) t93 -380.4* 1,165 (212.1) (1,255) r85 6,418*** 18,467 (1,742) (15,125)

r86 6,245 21,035 (8,013) (23,959) r87 14,627 -4,842 (9,027) (14,608) r88 -121.4 11,360 (10,966) (9,997) r89 -2,605 48,792* (17,780) (29,137) r90 8,294 61,809* (11,821) (35,269) r91 26,469 60,856 (23,759) (47,973) r92 71,140*** 59,611** (18,076) (25,711) r93 104,475*** 19,425 (19,054) (77,072) Constant 152.0 3.661 (108.6) (111.9) Observations 2,950 491 R-squared 0.222 0.312 Number of Id 391 84

Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

2. Result of long-term BI for every state

(1) (2) (3) (4) (5)

State 1 State 2 State 3 State 4 State 5

VARIABLES II II II II II roa 48,830*** 151,736*** 169,703** 105,900** 169,022** (9,218) (20,589) (81,166) (48,827) (73,162) t85 29.36 312.3 -315.5 (46.81) (949.2) (343.0) t86 166.3*** -2,510 778.3 -143.0 390.6* (48.35) (1,597) (1,094) (140.6) (230.9) t87 201.8*** -1,688 1,558 140.9 1,592 (62.15) (1,732) (3,021) (131.9) (1,443) t88 251.1*** -958.6 -1,530 44.28 4,683 (52.43) (1,588) (1,405) (96.77) (4,244) t89 258.7*** -1,536 -15,018* 0.177 3,492* (70.92) (1,587) (8,660) (154.0) (1,796) t90 336.5*** -692.2 -6,693 152.5 3,233** (68.18) (1,460) (4,330) (117.6) (1,573) t91 325.0** -478.9 -3,249 365.9** -3,001 (129.8) (1,527) (2,607) (144.9) (3,932) t92 507.6*** -1,227 -2,851 331.4 (98.09) (1,491) (1,754) (253.1) t93 681.5*** -84.77 167.8 639.5*** 310.2 (99.02) (1,522) (3,431) (191.9) (3,056) Constant 106.6 1,640 4,139** -210.2 904.6 (89.17) (1,406) (1,782) (333.2) (1,497)

Observations 1,239 44 275 1,364 1,878

R-squared 0.363 0.795 0.142 0.223 0.010

Number of Id 194 10 52 150 370

Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

(1) (2) (3) (4) (5)

State 8 State 9 State 11 State 12 State 13

VARIABLES II II II II II roa 26,907*** 555,226** 848,801** 113,374*** 59,573*** (3,184) (239,167) (317,865) (26,733) (14,287) t85 -12.15 752.0 -452.0 411.7** 264.0** (38.58) (551.3) (1,401) (176.6) (122.2) t86 -47.14 326.6 3,780 870.4*** 376.6** (45.16) (843.0) (2,720) (271.0) (173.7) t87 -72.12 1,255 1,575 1,319*** 609.6*** (58.76) (1,130) (3,491) (502.2) (231.8) t88 -72.09 -1,236 5,069* 1,680*** 832.1*** (69.48) (1,637) (2,634) (597.8) (279.0) t89 -50.33 -13,174 546.6 810.1 947.5*** (69.55) (8,210) (2,103) (894.1) (339.0) t90 -150.3 -6,388 -11,535 -28.32 671.2* (107.0) (3,947) (9,096) (1,106) (356.8) t91 9.001 -4,054 -9,953 905.9 651.0 (55.79) (5,067) (6,206) (796.9) (395.3) t92 298.2*** -3,065 4,191 2,331* 1,112** (77.19) (4,631) (4,506) (1,219) (451.1)