EARNINGS MANAGEMENT BY MERGER TARGETS: DISCRETION …

The Pennsylvania State University

The Graduate School

The Mary Jean and Frank P. Smeal College of Business Administration

EARNINGS MANAGEMENT BY MERGER TARGETS: DISCRETION

OVER THE LOAN LOSS PROVISION IN COMMERCIAL BANKS

A Thesis in

Business Administration

by

Bruce Bettinghaus

© 2000 Bruce Bettinghaus

Submitted in Partial Fulfillment

of the Requirements

for the Degree of

Doctor of Philosophy

December 2000

We approve the thesis of Bruce Bettinghaus

Date of Signature

____________________________________ ____________________ Anne L. Beatty Associate Professor of Accounting The PriceWaterhouseCoopers Faculty Fellow Thesis Advisor Chair of Committee ____________________________________ ____________________ Mark W. Dirsmith Deloitte & Touche Professor of Accounting ____________________________________ ____________________ James C. McKeown The Mary Jean and Frank P. Smeal Professor of Accounting Interim Chairman of the Accounting Department ____________________________________ ____________________ Mark J. Roberts Professor of Economics

iii

Abstract

This paper investigates the relationship between commercial banks’ accrual choices

and the likelihood of their takeover. I study a sample of 2,414 commercial bank holding

companies (banks) over the period of 1987 through 1998. The sample consists of 14,574

bank-years. I perform two complimentary tests to determine if the targets in bank mergers

are managing their loan loss provision downward in the period prior to the merger. In the

first test, I select a sample of 641 bank-years for banks in the year before they are the

targets in a merger. I test for the mean difference in the loan loss provision between these

target-banks and the rest of the sample after controlling for the economic determinants of

the loan loss provision. I find that both public and private intrastate targets as well as

private interstate targets all demonstrate a negative unexpected loan loss provision during

a time-period where there is a high probability of takeover. I do not find an earnings

management difference between public and private targets. In the second test, I select a

sample of 116 target banks that continue to be reported as a subsidiary to their new

parent. I test for a negative (positive) unexpected loan loss provision prior to (after) the

merger. There is some evidence that these targets do have a negative provision prior to

the merger, but they do not exhibit the expected reversing behavior.

iv

Table of Contents

List of tables .....................................................................................................................v

Acknowledgements .............................................................................................................vi

1. Introduction............................................................................................................. 1

2. Merger activity in the Banking Industry................................................................. 7

3. Hypotheses.............................................................................................................. 9

4. Empirical Methods ................................................................................................ 18

a. Modeling loan loss provisions .............................................................. 18

b. Test Variables ....................................................................................... 20

c. Firm-Specific Controls.......................................................................... 22

d. Macro-Economic Controls .................................................................... 25

e. Test for reversing behavior. .................................................................. 26

5. Sample Selection and Descriptive Statistics ......................................................... 28

6. Results .................................................................................................................. 32

a. The merger- file sample ......................................................................... 32

b. The status-change sample ..................................................................... 37

7. Sensitivity Analysis .............................................................................................. 39

a. Capital and Earnings ............................................................................. 39

b. Sample choices...................................................................................... 41

8. Contributions and Extensions ............................................................................... 43

a. Contributions ......................................................................................... 43

b. Extensions ............................................................................................. 44

References .................................................................................................................. 46

Appendix: Tables .............................................................................................................. 51

v

List of tables

Table 1 Probability of Being a Target........................................................................ 51

Table 2 Sample Selection Procedures ........................................................................ 52

Table 3 Distribution of all sample observations ........................................................ 53

Table 4 Distribution of target observations ............................................................... 54

Table 5 Distribution of bidder observations .............................................................. 55

Table 6 Descriptive statistics for all observations ..................................................... 56

Table 7 Descriptive statistics for target observations ................................................ 57

Table 8 Descriptive statistics for bidder observations ............................................... 58

Table 9 Regression results for merger- file sample .................................................... 59

Table 10 Total effect of unexpected provision ............................................................. 61

Table 11 Regression results for bidders and targets by year ........................................ 62

Table 12 Correlation between yearly coefficients and takeover probability ................ 63

Table 13 Distribution of the status-change targets and bidders .................................... 64

Table 14 Descriptive statistics for the status-change targets ........................................ 65

Table 15 Descriptive statistics for the status-change bidders ....................................... 66

Table 16 Regression results for the status-change sample ............................................ 67

vi

Acknowledgements

The road to the completion of a doctoral thesis can be treacherous and filled with unexpected surprises. I have had the benefit of working with great colleagues that have helped me traverse the rough spots, and warned me of the dangerous curves ahead. Two of my fellow doctoral students have been particularly helpful. Joe Weber and Ram Venkatamaran are like the Yin and Yang of my research conscience. Each is willing to spend the time to evaluate my work; Joe always finds the fatal flaw, while Ram attempts to convince me of my brilliance.

The professors that I have worked with at Penn State have all been encouraging and helpful in my progression towards the completion of this stage of my education. Mark Dirsmith has helped me to look at research from alternative perspectives. His thoughtful and provocative questions often lead to new ideas. I shall never look at a simple push-button ballpoint pen in the same light again. Mark Roberts has been a sure and steady guide through difficult territory. His ability to make the complex understandable has influenced both my research and my teaching.

Jim McKeown has had a big influence on the way I interpret the world that surrounds me. Jim’s critical analysis of all research spills over in all aspects of life. After being exposed to the level of scrutiny that he can bring to bear on empirical research, I find that I am wary of every fact and argument that I encounter. Everywhere I turn, I see that what most people might accept as a fact, I now interpret as an opinion.

Anne Beatty has treated me like a colleague from the moment I met her, I hope that someday I am worthy of that level of respect. She is the single most important influence on my development as a scholar. Her constant, arms-length supervision has allowed me to grow and learn as a researcher and a teacher.

My parents, Cay and Erv, are responsible for my academic achievements in more ways than I can count. Their willingness to help and encouragement of my return to college at the age of 31 gave me the ability and inspiration to succeed beyond my wildest dreams. In addition to this contribution late in my life, I am convinced that my early childhood played a role as well. From the early part of my childhood, my folks required me to keep a ledger of my finances that I had to present to my dad before I could receive my allowance. This was my first exposure to any type of financial record keeping. This was also my first experience with discretionary disclosure and earnings management. I do not think it is a mere coincidence that these are some of my research interests today.

My wife Trish deserves much of the credit for any success that I might achieve in my life. She has shown unwavering confidence in my academic abilities. I cannot count the times that I have complained to her that I was struggling to grasp some concept in my classes or studies, and that I would surely do horribly on the next exam, paper, or presentation. She would always calmly reassure me that I would pass, find clarity, or knock them dead. Luckily, she was right more often than not. Thank you Trish, your support has made my success possible.

1

Based on extant research, we came to the conclusion that

understanding management’s incentives is key to understanding the desire to engage in earnings management. In particular:

• Managers have strong incentives to “beat benchmarks,” implying that firms just beating benchmarks are potentially more likely to be engaging in earnings management.

• Managers of firms desiring to issue equity have strong incentives to boost stock price and hence engage in earnings management. (Dechow and Skinner, 2000, emphasis added)

1. Introduction

This thesis seeks to further our understanding of earnings management by

investigating firms at a time when they have strong incentives for management to

increase the value of the firm. Although many studies have investigated the relationship

between earnings management and the incentives of the capital markets, there are still

calls for additional research in this area [Healy and Wahlen (1999) and Dechow and

Skinner (2000)].

The main purpose of this study is to determine if commercial bank holding

companies (banks), that are targets in upcoming mergers, manage their earnings by

lowering their loan loss provision. I use two complementary tests to investigate this

question. In the first, I use a sample of 14,574 bank-years to estimate an expected value

for the loan loss provision. I then test for a difference between this expected value and the

value in 641 bank-year observations where the bank becomes the target of a merger in the

2

following year. This set of 641 target observations is for mergers where the target bank is

eventually absorbed into the new entity in the year of the merger. For the second test, I

also select a sample of 116 target observations where the acquirer continues to report the

target as a separate entity. With this sample, I again look for a difference between an

expected value and the target group. However, with this group I also track the differences

for a 5-year window surrounding the merger.

The targets in a merger should have the same incentive to increase their earnings as

those firms engaged in offering new equity to the market. This incentive to increase

earnings is essentially a method to increase the price of the shares sold, and there is

evidence that the benefits of this manipulation outweigh the costs [see for example

Rangan (1998), Teoh, Welch and Wong (1998a), Teoh et. al (1998b) and Teoh, Wong

and Rao (1998)]. This incentive to “fool” the new investors in a firm is the same in a

merger; the difference is that the new investor is a single firm, instead of a group of

diverse shareholders.

Managers face a variety of motivations when they make accounting choices.

Dechow and Skinner (2000) describes these choices as a continuum ranging from

conservative accounting to fraudulent accounting. Accounting for the provision for loan

losses could either be conservative or aggressive depending on management’s motivation

to increase or decrease earnings. Healy and Wahlen (1999) divide the motivations to

manage earnings into three main areas: capital market, contracting and regulatory. Under

each of these motivations the choices that management can take range from the neutral

3

application of accrual accounting to presenting an inaccurate picture of the firms

underlying economics. The incentives might be to mislead, conform, or simply meet

some pre-set goal or standard. The driving force behind any of the potential choices that

management might make will be a cost benefit tradeoff for management.

Many studies have looked at discretion over the loan loss provision in a variety of

settings. Wahlen (1994) finds that a measure of unexpected loan loss provision is

positively related with future changes in cash flows and contemporaneous stock returns.

This is consistent with the market interpreting the increase in loan loss provision as a

signal of strength. This signaling story is contradictory to the idea of using the discretion

over the loan loss provision to increase earnings. Tests of the signaling story have

resulted in some mixed findings. Beaver and Engel (1996) find support for the idea that

the discretionary portion of the loan loss provision is positively related to market value.

However, Ahmed, Takeda and Thomas (1999) find that loan loss provisions are

negatively related to future earnings changes and contemporaneous stock returns. Their

claim is that the previous studies had not completely controlled for the economic

determinants of loan loss provisions.

Prior research that looked specifically at the earnings management of targets in

non-bank firms has not been able to consistently document that targets increase earnings

prior to the merger [e.g. Easterwood (1997) and Erickson and Wang (1999)]. In the

banking industry, Chamberlain (1991) looks at a sample of 192 bank mergers during the

period 1981-1987. She investigates earnings management using the provision, loan

4

charge-offs and securities gains. She finds “no anticipatory earnings management prior to

the merger”(p. 75). In this study, I focus on a single accrual in a single industry, instead

of relying on a measure of unexpected total accruals. Using this method I do find

evidence that some of the targets of mergers have unexpected income increasing accruals

in the period prior to their merger.

In addition to the results concerning the targets, this setting allows me to reexamine

the results relating to bidders as demonstrated by Erickson and Wang (1999). The

argument for bidders to manage earnings upward prior to the merger is analogous to that

for the targets. If a bidder can increase their share price prior to the transaction, and they

use those shares as a medium of exchange, then the earnings management can affect a

lower price for the acquisition. In the sample I examine there are 346 bank-years where

the bank is a bidder in at least one merger in the following year. Of these observations

74% of the banks explicitly report that they issued equity with respect to a business

combination, and another 18% issue other equity in the year of the merger. I find that

these bidder observations exhibit a lower loan loss provision during the period before the

year of the merger, but it is at marginally insignificant levels.

The data for my sample of bank holding companies come from the public

regulatory reports that all banks must file with the Federal Reserve. Because these data

are available for both public and private banks, I can also test for differences in earnings

management across these types of firms. There has been recent evidence that earnings

management incentives and outcomes differ across these two groups [Beatty and Harris

5

(1999) and Betty, Ke and Petroni (1999)]. In this setting, I do not find any significant

differences in unexpected loan loss provision between the public and private banks.

Another difference that this sample allows a test for is differences in type of

merger. During the sample period barriers to both intrastate and interstate mergers fell.

These two types of mergers might be seen as accomplishing different goals for the

participants. An interstate merger is a vehicle for expansion for an aggressive firm.

Alternatively, the trade press often describes an intrastate merger as a defensive move to

protect against such an aggressor. I test for differences across these two types of mergers

and find that the targets in the intrastate mergers exhibit earnings management behavior

consistent with increasing earnings, while those that are the targets of out-of-state bidders

do not.

This research contributes to the literature on earnings management in a number of

ways. First, I have a powerful setting to test for merger targets managing earnings prior to

the merger. Prior literature has predicted, but not found, this result [Easterwood (1997)

and Erickson and Wang (1999)]. Second, it adds to the literature concerning earnings

management surrounding an issue of equity [e.g. Rangan (1998), Teoh, Welch and Wong

(1998a), Teoh et. al (1998b), Teoh, Wong and Rao (1998) and Erickson and Wang

(1999)]. These studies have relied on an unexpected total accrual model to measure levels

of earnings management, while my study focuses on a single account. Because I use a

different method to test for earnings management and find consistent results, this lends

credibility to both types of findings. Finally, there are recent studies that find that public

6

banks manage earnings on an ongoing basis to both smooth earnings and avoid losses and

declines [Beatty and Harris (1999) and Betty, Ke and Petroni (1999)]. This study

examines the behavior of these two groups during an event, as opposed to ongoing

behavior. I do not find any significant differences between these two groups prior to a

merger.

The arrangement of the remainder of this thesis is as follows: Chapter 2 discusses

the background of the banking industry and why the sample time period is a fertile time

for mergers. Chapter 3 presents the hypotheses to be tested. Chapter 4 outlines the

empirical methods. Chapter 5 provides the sample selection and a description of the data.

Chapter 6 discusses the results of the empirical tests. Chapter 7 provides some sensitivity

analysis. Chapter 8 concludes and describes future research extensions.

7

2. Merger activity in the Banking Industry

The merger activity in the banking industry has increased within the last 10 years.

This increase in merger activity is the result of changes in regulation allowing mergers. In

the 1970s, the industry consisted of many small banks operating independently. Most

states restricted the ability of these banks to branch or otherwise expand. Most states did

not allow out-of-state banking firms to own or operate banks within the state boundaries.

By the middle of the 1990s, the industry had shifted from this type of structure to one that

allowed relatively free geographic expansion both within and across state boundaries.

The culmination of this change was passage of the Riegle-Neal Interstate Banking and

Branching Efficiency Act of 1994 (IBBEA). This federal law permits bank holding

companies to expand on a national basis beginning in June 1997.

Prior to the passage of the IBBEA, the regulatory control of the banking industry

structure was at the state level. Each state went through a number of deregulatory steps to

loosen the restrictions on bank consolidation. These restrictions can be divided into either

intrastate or interstate types. A common first step was to allow bank holding companies

to own banks anywhere within state boundaries. Alternatively, a state could allow

intrastate branching; either of these two deregulations would allow a banking firm to

control bank locations across the state.

The next regulatory change was to allow out-of-state competitors to buy in-state

banks. This was achieved by allowing the operation of interstate bank holding companies

(BHCs). The states first allowed these interstate BHCs from a restricted set of

8

neighboring states. These were known as regional reciprocal agreements. Many of these

regional reciprocal agreements included national trigger provisions that would allow the

measure to include all states after a period of a year or two. In addition, some of these

trigger provisions were structured so that as soon as any other state allowed the home

state banks to buy its banks, then the home state would act in kind.

The result of these changes is that the banks’ expectations about new competitors

were increased. As the states allowed out-of-state holding companies to compete, those

in-state banks that felt the need to be bigger to protect themselves from the inevitable

newcomers looked around for partners. In this way, both the intrastate and the interstate

deregulations helped to drive the intrastate mergers.

9

3. Hypotheses

Healy and Wahlen (1999) define earnings management as using the discretion

over financial reports to purposely mislead some party about the underlying economic

performance of the firm. In the case of a merger target managing earnings upward to

increase its price, the parties being misled are the bidder firm’s management and

shareholders. The merger agreements of these banks include a discussion of the process

used to value the target bank. An external financial advisor performs these valuation

calculations. In general, the statements seem to indicate that the fairness of the price is

evaluated by the relative multiples on earnings, capital, deposits and total assets. In

addition, the public targets usually have included a comparison of their stock valuation.

The act of managing the loan loss provision down will have the effect of increasing

earnings, assets, capital and firm value.

Extant research is unclear about the effect on firm value due to discretionary

changes in earnings components. Wahlen (1994) and Beaver and Engel (1996) both find

that the market rewards lower earnings when the loan loss provision is the instrument of

discretion. Ahmed et al. (1999) calls those results into question by providing a different

measure of discretionary loan loss provision. Chamberlain (1991) suggests that bank

manager’s motivations could cause them to use their discretion in either direction prior to

the merger. She tests a two-sided hypothesis that there will be an unexpected loan loss

provision prior to the merger and that it will reverse after the merger. In her sample of

192 mergers, she finds no evidence of earnings management prior to the merger. Teoh et

10

al. (1998a) and Rangan (1998) both offer evidence that non-bank firms that issue new

equity are able to successfully manipulate firm value over the short run. But the

conflicting evidence from the banking literature makes it unclear whether the use of the

provision to increase earnings would result in an increase in firm value as has been

shown in non-bank firms. This thesis does not test this question; rather I seek to

determine if merger targets lower their provision prior to the merger, which would be

consistent with an attempt to increase earnings to influence firm value in the short run.

This leads me to my first hypothesis.

H1A: Bank holding companies that are targets in a merger in a calendar year will have a negative unexpected1 loan loss provision at the end of the previous year.

The next three hypotheses are all concerned with finding the differences across

different types of banks, mergers and periods. I divide the banks into public and private, I

split the mergers between intrastate and interstate, and I separate the periods into a high

probability and low probability of merger. Each of these divisions should affect either the

target’s incentives and/or ability to manage earnings.

Beatty and Harris (1999) find that public banks engage in more earnings

management than do private. Their finding is consistent with public firms having a

1 In this paper I use the term unexpected loan loss provision to represent a difference from the mean loan loss provision after controlling for the economic determinants of the loan loss provision.

11

greater incentive to reduce the information asymmetry between management and

investors. In this setting, I do not expect that type of information asymmetry will affect

the incentive to mislead the bidder bank about the target’s underlying fundamentals. Both

the public and private targets have the same incentive, to mislead the bidder and increase

the purchase price. If there are differences between the two groups, I expect it will be

because of differences in the ability to manage earnings or to predict the timing of the

deal.

It is unclear if one or the other group would have an advantage in predicting the

timing of any potential merger. Public firms that are merger candidates are often

discussed in the trade press. Private firms generally do not have public attention paid to

them, but might have a greater amount of private communication with any potential

suitors.

The two groups’ ability to manage earnings might be different because of

constraints on the public banks. A difference also might arise because of differential

reliance on accounting information in forming price across the two groups. Beatty and

Harris (1999) determine that public banks are managing earnings to mitigate information

asymmetry. The earnings management that they find is bi-directional; that is, it is an

earnings-smoothing story. They test for earnings management in realized securities gains

and losses only, but if public banks have the incentive to smooth earnings with this tool

then they might be using other tools as well. If this ongoing management limits the public

12

banks’ ability to manage earnings in the specific period then the private banks will be less

constrained in managing earnings surrounding the merger event.

Another potential difference across the two groups might be audit quality. Both

groups are subject to regulatory review, but the public firms also must have a financial

audit by an independent external auditor. While the private banks are not required to have

a public audit firm perform an external audit, it is likely that most do. The quality of that

audit is likely to depend on the level of ownership concentration, and the level of

management ownership (Clarkson and Simunic, 1994). If private banks on average have

higher concentration of ownership and managerial control then they are likely to contract

for lower quality audits. This may allow a greater ability to manage earnings [Becker,

Defond, Jiambalvo and Subramanyam (1998) and Francis, Maydew and Sparks (1999)].

Although the timing issue is unclear, it seems that the private banks face lower

constraints in their ability to manage earnings. This leads me to the second hypothesis

concerning the target banks.

H1B: Privately held bank holding companies that are targets in a merger in a calendar year will have a negative unexpected loan loss provision compared to the publicly held targets.

I expect the split between intrastate and interstate mergers to find differences in the

ability to predict the timing of the merger. The trade press portrays many of the intrastate

mergers as a tactic to avoid the takeover of either by out-of-state rivals. Many of the press

articles that relate to intrastate mergers specifically discuss the need to combine in-state

forces to stave off the pending out-of-state invasion. While both of these types of mergers

13

are somewhat forced on the target, the intrastate type is to be more likely to be initiated

by the target. Those banks that are the target of an interstate acquisition are either less

likely to have predicted the event or they had less ability to find a suitable partner. In

either event, the target in an interstate merger seems to be not only predicting, but also

perhaps controlling the timing of the merger. This leads to the following hypothesis.

H1C: Bank holding companies that are interstate targets in a merger in a calendar year will have a positive unexpected loan loss provision compared to the intrastate targets.

The next prediction that I make concerns the ability of all types of targets to predict

the timing of the takeover. Erickson and Wang (1999) predict “similar to acquiring firms,

target firms may also have an incentive to increase pre-merger earnings in an attempt to

increase the transaction price”. They do not find that their sample bears this out, and

conclude; “the timing of the acquisition may explain the weak results”. During the period

of my sample the probability of any bank in my sample being a target in the next year

ranges between 1.95% in 1989 and 7.70% in 1997, Table 1 details these calculations.

There is a clear break in this probability, from 1987 to 1991, the average was 2.85%, but

between 1992 and 1997, the average probability jumped to 5.94%. If Erickson and Wang

(1999) are correct and the desire for targets exists, but they do not have the ability to

predict the timing, then targets will be more likely to consider themselves potential

targets during this higher probability period. This leads to the final hypothesis concerning

the targets.

14

H1D: Bank holding companies that are targets in a merger in calendar years when there is a lower average expected rate of takeover would have positive unexpected loan loss provision compared to the targets in other years.

The bidders in these mergers also have the incentive to manage their earnings

upwards. Many studies have shown a relation between current earnings and the firms’

market value [e.g., Ball and Brown (1968), Foster (1977), or Beaver et al. (1979)]. The

value of current earnings is particularly important in valuing the bidders stock

[DeAngelo (1986, 1990)].

Erickson and Wang (1999) make the case for managing earnings prior to a stock-

for-stock merger explicitly. In a stock-for-stock merger there is either a negotiated

exchange rate for the shares of the two firms or there is a dollar value of the bidders stock

that will be exchanged for the target’s stock. Their argument is that by increasing the

reported earnings prior to the negotiations for the transaction the stock price will increase,

and the bidder firm can use fewer shares of their stock to purchase the target. Even if a

bidder bank does not explicitly engage in a stock-for-stock merger, they still might have a

need to raise equity. This leads to my first hypothesis concerning the bidders.

H2A: Bank Holding Companies that are the bidders in a merger in a calendar year will have a negative unexpected loan loss provision at the end of the previous year, compared to non-bidders.

The next hypothesis is concerned with the differences between public and private

bidders. If bidders in general have the incentive and the ability to manage their earnings

prior to a takeover, are there differences across bank types. Erickson and Wang (1999)

15

looked exclusively at public bidders. Their results suggest that the bidder can manage

earnings and increase their stock price. This mechanism relies on information asymmetry

between managers and the investors of the bidder firm. It is unlikely that the same level

of asymmetry will exist in the privately held bank holding companies. There should be a

greater concentration of ownership, as well as a higher level of managerial ownership in

private firms. Any successful earnings management done by the private bidders must be

an attempt to mislead only the target bank and their representatives. The private banks

might have more ability to manage their earnings, because of a higher level of managerial

ownership.

Just as in the case of the targets, the private bidders might face fewer constraints

from their auditors. Additionally, the results from Beatty and Harris (1999) indicate that

the public banks might be constrained in their earnings management choices. If public

banks are engaging in an ongoing process of earnings management to mitigate the

information asymmetry, then they might be constrained in the actions that they could take

to manage earnings in a period prior to a merger action. This seems to indicate that the

private banks will have a greater ability to manage their earnings surrounding the merger

period.

Although the above arguments seem to indicate that private bidders might have a

greater ability to manage earnings, there are also reasons that they might not. The

incentive for the bidder in these transactions to manage earnings is to lower the cost of

acquisition by increasing the relative value of the medium of exchange, the stock of the

16

bidder. However, the shareholders of a target, that is being offered a private bidder’s

stock as an exchange, are likely to require a thorough investigation of that bidder's

valuation. This might result in the private bidder engaging a high quality auditor, and

perhaps revealing a greater amount of private information about the firm. These actions

could mitigate any earnings management that did take place. It is unclear to me which of

these effects would dominate, but if one of them does dominate the other, then the

discretionary loan loss provision for private bidders will differ from that of their public

counterparts. This leads me to my final hypothesis concerning the bidders.

H2B: Privately held bank holding companies that are bidders in a merger in a calendar year will have a difference in unexpected loan loss provision compared to the publicly held bidders.

If targets and bidders are managing their earnings upwards in one period, then that

will mean a reversing effect must happen in future periods. For a smaller set of bank

holding companies, I can jointly observe the accounts of the bidder and the target both

before and after the merger. Based on the above arguments it is a straightforward

argument to make the case that if targets and bidders manage their loan loss provision

downward prior to the merger, then they should manage it upward after the merger. This

leads to the final hypothesis.

H3: Bank holding companies that are either targets or bidders in a merger in a calendar year will experience negative unexpected loan loss provisions in the years prior to the merger and positive unexpected loan loss provisions in the years after the merger.

17

Overall, the set of hypothesis can be summed up into the following statements. The

managements of banks that are involved in mergers have incentives to manage earnings

upward prior to the completion of the merger. The extent that the bank can do this

depends on any other constraints that they might face and for target banks, the ability to

predict that they will be a target.

18

4. Empirical Methods

a. Modeling loan loss provisions

I assume that target and bidder bank managers are concerned about the relative,

rather than the absolute, levels of the provision. In the following model, I further assume

that the main approach to setting the provision is a balance sheet approach. That is, the

managers are using the provision (without discretion) to correctly state the value of their

allowance for loan losses. Under these assumptions it follows that what managers are

concerned with is the size of their provision relative to the ending balance of loans. The

following model and related tests implicitly assume that in the short run management has

greater discretion over the provision than they do over the level of loans. Of course,

management has complete control over both variables, but on December 31, the level of

loans is fixed, while the level of allowance has a discretionary component that is still

available to management. To control for the determinants of loan loss provision, I use

both a set of firm specific and a set of macro-economic variables. I describe both types

below. The ratio of the provision to loans (ProvRatio) is regressed on other bank specific

ratios that have been shown to influence both the discretionary and the non-discretionary

portions of the provision.

The control variables are a combination of flow variables (income statement and

changes in balance sheet items) and stock variables (balance sheet items). These variables

are all measured as ratios. In general, I scale the flow variables by a corresponding flow

variable and the stock variables by a corresponding stock variable. For instance, I scale

19

the change in non-performing loans (∆NPL) by the average loans for the period while I

scale the beginning of the period loan loss allowance (Allowance) with the total assets for

the same time2.

To control for both cross-sectional and temporal correlations, I use a pooled cross-

sectional model that incorporates fixed effects for both time and firm specific

correlations. Because these fixed effects are not the variables of interest, I do not report

them.

The general form of the model is as follows:

eteInterestRa*ßloymentStateUnemp*ßeStateIncom*ß

teInterestRa*ßloymentStateUnemp*ßeStateIncom*ßPrivate*ßSize*ßEBTP*ßCapRatio*ß

Assets*ßAllowance*ßNPL*ß

Bidder*ßTarget*ßProvRatio

1-t151-t141-t13

121110

9876

543

21

+∆+∆+∆+

∆+∆+∆+++++

∆++∆+

+=

In the actual estimation of the model, I allow several of the independent variables

to vary across types of banks, types of mergers and periods. I discuss these specific

modeling choices below. However, I present the basic framework and the variable

definitions for the readers benefit.

2 The one exception to this rule is the change in assets (∆Assets). At some point, I used the beginning assets, the average assets and the ending assets as scalars. All of these provided similar results; the results presented are with the change in assets scaled by the end of the year assets.

20

Dependent Variable: ProvRatio is the loan loss provision for the period scaled by the value of gross loans at the end of

the period. Test Variables:

Target is an indicator variable that is one if the bank is a merger target in the following calendar year.

Bidder - is an indicator variable that is one if the bank is an acquirer in the following calendar year. Firm Specific Control Variables:

∆NPL is the change in non-accrual loans scaled by the average total loans for the year. Allowance is the banks’ loan loss allowance at the beginning of the year scaled by total loans at

the beginning of the year. ∆Assets is the change in total assets for the year scaled by total assets at the end of the year. CapRatio is owners’ equity plus the tax adjusted loan loss provision scaled by total assets at the

end of the year. EBTP is the pretax net income plus the tax adjusted loan loss provision scaled by total assets at

the end of the year. Size is the log of total assets. Private is an indicator variable that is one if the bank is privately held.

Macro-Economic Control Variables: ∆StateIncome is the yearly percentage change in the fourth-quarter personal income in the state

that the bank is headquartered in. ∆StateUnemployment is the yearly percentage change in the fourth-quarter unemployment

payments in the state that the bank is headquartered in. ∆InterestRate is the one-year change in the average monthly quotes for the one-year treasury rate

in the fourth quarter.

b. Test Variables

I select the targets and bidders in two different manners. For H1 and H2 I use a set

of targets and bidders that I will refer to as the merger-file sample. For H3 I use a set of

targets and bidders that I will refer to as the status-change sample.

I base the merger-file sample on a data file of mergers maintained by the Federal

Reserve Bank of Chicago. This file lists the targets and bidders in 3002 mergers that have

ultimately resulted in the bidder fully incorporating the target. This file tracks mergers

21

going back to 1976, but the data for bank holding company performance (the Federal

Reserve’s Y-9 report) is only available in electronically readable form beginning in 1986.

If an observation is on the merger-file as a target and is a top-tier holding company

in the last year on the data files, then I code the last year the bank is on the data file as a

target observation. This results in 641 merger-file target observations. Table 4 describes

these observations.

To find the merger-file bidder observations I began with the list of targets

determined by the above process. I retained the bidders related to these targets. There are

fewer bidder banks than targets, because a bidder can be a bidder more than once. There

are fewer bidder observations than target observations because a bidder can acquire more

than one target in a year. Table 5 describes the distribution of the bidder observations

selected through this process.

Once identified, the target and bidder observations were interacted with other

dummy variables to obtain a marginal effect. To test for the differences between public

and private bidders and targets, hypotheses H1B and H2B, I use an indicator variable for

being a private bidder or private target. These variables take on the value one according

to the distribution in Table 4. To test for the differences between intrastate targets and

interstate targets, hypothesis H1C, I form an indicator variable if the target and bidder

banks are headquartered in different states. Finally, to test for the difference in the high

and low probability periods, hypothesis H1D, I use an indicator variable that is one for

22

targets during the years 1987 – 1991 and 1998. These years are those with lower

probability of being a takeover.

I identify the status-change sample in a different manner. Bank holding companies

that acquire other bank holding companies sometimes choose to report the two entities

separately. When the top-tier holding company of a subsidiary changes, the change shows

up in the data file. This change lets me identify both the new subsidiary and the parent

company. I find previously independent top-tier firms that are newly reported as a

subsidiary. I also include those subsidiary banks that are sold from one parent to another.

Because I wish to allow for a before and after comparison, I limit this sample to those

banks that have three years prior to the merger and two years after. This procedure results

in 116 observations of 107 banks. Table 13 lists these targets. I also pick up the bidders

that relate to the status-change targets. There are only 18 banks involved in 27 merger-

year observations. Table 13 also lists these bidders.

c. Firm-Specific Controls

Ahmed et al. (1999), Kim and Kross (1998), Collins et al. (1995) and Beatty et al.

(1995) all model the loan loss provision as a function of control variables and then use

measures of capital level and earnings as test variables. One of the common controls in

these papers is the quality of the loan portfolio. I include three variables in my model that

captures this property. First, I include the change in non-performing loans (∆NPL) for the

year scaled by average total assets. Ahmed et al. (1999) and Kim and Kross (1998) both

find that this measure is positively related to the loan loss provision, and I expect the

23

same to be true in my sample. Next, I include a measure of loans allowed for at the

beginning of the year. Collins et al. (1995) and Beatty et al. (1995) both include the

beginning balance of the loan loss allowance to control for loan quality. They both found

the beginning allowance to be negatively related to the loan loss provision at the end of

the year, indicating a reversion to the mean effect for the value of the loan portfolio. In

my model, I include the variable Allowance that is the loan loss allowance scaled by total

loans at the beginning of the year. Accordingly, I expect that this variable will be

negatively related to the loan loss provision. Finally, I include an explicit measure of

growth to capture the relative ages of the loan portfolio. I expect that those banks that are

growing faster are doing so by generating new loans, which are less likely to need an

allowance before they are a year old. Thus, I expect the relation between ∆Assets and the

provision to be negative.

I also include as control variables measures of earnings and capital. These variables

have been test variables in prior studies [see Moyer (1990), Collins et al. (1995), Beatty

et al. (1995), Kim and Kross (1998) and Ahmed et al. (1999)], and there is evidence that

they are significantly related to the provision. Earnings smoothing, increasing

(decreasing) the provision in times of high (low) earnings and the management of

regulatory capital, increasing the provision to increase the level of regulatory capital,

have both been thought to influence the valuation of the provision. The evidence on these

two influences on the provision is mixed. Moyer (1990), Beatty et al. (1995) and Ahmed

et al. (1999) all support the influence of capital management, while Collins et al. (1995)

24

does not. The earnings smoothing influence is supported by Collins et al. (1995) and not

by either Beatty et al. (1995) or Ahmed et al. (1999).

I proxy for the effect of capital constraints with a simple measure of bank equity

(CapRatio), adjusted for the current tax adjusted provision and scaled by assets. This

measure does not capture the fine nuances of all of the various capital regulations, but it

will capture broad differences across banks. In addition, the rules for the regulation of

capital level changed in 1991, and the influence of the provision in meeting minimum

capital standards was lessened3. Ahmed et al. (1999) tests for this change and finds a

significant difference between the two periods. Accordingly, I let the impact of the

capital level of the bank vary across the two periods. If my measure of capital is

capturing the regulatory cost of violating the minimum capital levels, than I would expect

to see a negative relationship in the pre-1991 period and a significantly less negative

relationship in the post 1991 era.

To control for the influence of non-provision earnings, I include the variable EBTP.

I form this by adjusting pre-tax net income by the current tax-adjusted provision4. The

earnings smoothing story suggests that there should be a positive relation between this

3 Ahmed et al. (1999) presents a thorough discussion of the capital regulation in both periods. 4 For EBTP I use an ad-hoc 40% tax rate. I also estimate EBTP using a derived tax rate based on the

current years tax expense. The two variables are quite similar, and I run the tests using both, the results are not different across the two methods.

25

variable and the provision. Further complicating the potential relation between the

provision and earnings, is the fact that my sample contains both public and private banks.

The motivation for earnings smoothing is to lower the cost of capital in an environment

of information asymmetry. This motivation should only exist in the public banks and not

in the private ones. Beatty, Petroni and Ke (2000) find that public banks are much less

likely than private banks to report small negative earnings changes, indicating that

meeting benchmarks is more important for the publicly owned banks than for private

ones. Accordingly, I allow the impact of earnings to vary across the two groups, and I

expect that the marginal effect of earnings on the provision, for the private banks will be

negative. The expectation is that the net effect is one of no relationship between earnings

and the provision for these banks.

I include two variables to control for broad differences across banks. I include the

log of assets to control for any differences in the loan portfolios that might vary with the

size of the bank. Finally, I include an indicator if the bank is a private bank.

d. Macro-Economic Controls

Ahmed et al. (1999) found that including a measure of the health of the economy in

the locations where the bank has loans is an important determinant of the loan loss

provision. They form an index of business failures weighted by the locations of the loans

a bank holds. To do this they have to restrict their sample to those banks where the

geographical distribution of loans is available. In an attempt to capture the same type of

macro-economic expectations without limiting the sample, I include three variables and

26

their lags. I use the fourth-quarter to fourth-quarter change in income in the state in which

the bank is headquartered as one component of this expectation. I expect that as state

income increases the expectation of future loan loss will decrease. Banks will see their

customer base growing, and their customers will have greater income, lowering their

default risk. I expect that this variable will be negatively related to the current provision.

I also include the changes in state unemployment payments as another indicator of

macro-economic expectation. I see this as an opposite indicator to the income indicator

above. As economic conditions degrade in a state, the prospects for a banks' customer

also degrade, increasing their default risk. I expect this variable to have a positive relation

with the provision.

Finally, I include the change in the one-year interest rate to control for changes that

affect the borrower's ability to repay loans.

e. Test for reversing behavior.

I use the model described above to test the merger-file sample in the period just

prior to the merger year. However, the status-change sample offers a unique opportunity

to examine the behavior of the provision over time for both the targets and bidders

involved in mergers. Because of the smaller sample size, I do not try to sub-divide the

targets and bidders into any other groups. I estimate the model using the control variables

listed above and form a set of ten indicator variables that measure the unexpected

provision in each of five years for both the targets and bidders. The Targets and Bidders

variable is the year before the merger, the same as in the merger-file tests. I then add two

27

years before and two years after for each group. Consistent with H3, my expectation is

that for both targets and bidders the unexpected provision will be negative in the first

three periods, and positive in the last two.

28

5. Sample Selection and Descriptive Statistics

The data used in this study come from three publicly available sources. I obtain the

bank-specific data from the Y-9 report filed with federal bank regulators. These data are

available in downloadable files from the Federal Reserve Bank of Chicago at

www.frbchi.org. The data files are available for all quarters in the period between the

second quarter 1986 and the second quarter 1999. In this study, I use the yearly data that

is available, the files for the period ending 12/31. The data on bank holding company

mergers is also available from the Federal Reserve Bank of Chicago. They have available

a list of 3,002 mergers between bank holding companies that occur between 1976 and

1999. This merger-file contains the date that the non-surviving bank holding company

consolidates into the survivor.

The state income data are available from the Bureau of Economic Analysis.

Machine-readable data for all state quarters from 1969 through 1998 is available at

www.bea.doc.gov. The interest rate data is located at FRED, the Federal Reserve

Economic Data, which is at www.stls.frb.org

The main data files for the bank holding companies during this period list 3,127

banks with 19,392 observations that report positive assets. I eliminate 26 banks that are

not headquartered in the 50 states or D.C. I next remove 187 banks that have

interruptions in their time series of data. I lose 2,914 observations and another 409 banks

due to the use of beginning of the period and change variables. Finally, I eliminate 473

observations of 91 banks that have missing or anomalous data. Table 2 outlines the

29

sample selection procedure. The final sample consists of 14,574 observations from 2,414

bank holding companies over 12 years. This sample represents 96.8% of all the available

observations of bank holding companies during this period.

To determine the cut off points for anomalous data, I examined the 50 extreme

observations for all variables. For the ProvRatio I eliminated those greater than 0.5 and

those less than –0.07. The ∆NPL I limited to be between -0.3 and 0.3. CapRatio was

limited to be greater than 0.01 and less than 0.5. EBTP was constrained to range between

–0.2 and 0.25. All of these cut-off points were based on an observation of the raw data,

and the decision rule was based on the gap between data points. For all of the data

eliminated, if any of the observations were anomalous, all of the observations for the

bank were eliminated.

Table 3 describes the distribution of observations across years and the public and

private distinction. The public or private designation is determined in the following

fashion. The Y-9 reports indicate if the bank was a SEC filer during the period covered

by the Y-9. I code as public banks those that are SEC filers and those that switch from

being a non-filer to being a filer at any time during the sample period. I code as private

all those that are non-filers, or switch from filer to non-filer status. Being a SEC filer does

not insure that the firm’s stock is publicly traded, but being a SEC non-filer insures that it

is not. Therefore, in my coding scheme I may code as public some private banks. This

works to bias against my finding results that support H1B and H2B.

30

Table 4 and Table 5 describe the merger-file target and bidder observations

respectively. The public, private, intrastate, and interstate columns describe these subsets

of each of the bidder and target samples. The target sample is split rather evenly between

public and private, while the bidder sample is 72% public banks. Both types of targets

and bidders are about 60% intrastate mergers.

Table 6 presents the descriptive statistics for the total sample. The loan loss

provision has a mean (median) of 0.66% (0.37%) of the gross outstanding loans at the

end of the year. This level is similar to studies that limit their sample to larger public

banks [see for example Ahmed et al. (1999)]. Although the provision is an expense for

financial reporting purposes, it can be negative in the regulatory reports5. There are 356

observations, approximately 2.4% of the sample, which report a negative loan loss

provision. To better explain the effect the screening process had on the data, I present the

following details the process on low end of the ProvRatio distribution. I screened this

variable at -0.07 on the low end. Prior to the screen, there were five observations less

than that, between -0.21 and –114.0 and there were five observations between –0.03 and

–0.21. After the screen, and the removal of all of the observations of the banks that had

observations less than –0.07 there were only three observations less than –0.03.

5 Ahmed et al. also report a negative minimum value for their provision variable.

31

The range in the size of the sample is larger than most prior research that looks

only at public banks. The sample includes a minimum observation of a bank with just 8.2

million dollars in assets all the way to an observation of a bank with 617 billion dollars in

assets.

Tables 7 and 8 present the same set of descriptive statistics for the target and bidder

observations. As expected, the bidder banks are on average much larger than the targets,

although there is considerable overlap in the distributions. The other major differences

between the two groups are their growth rate and the ProvRatio. The bidders in the year

prior to the merger are growing at an average of almost 11% per year, while the targets

are just over 6% per year. The targets are providing for loan loss at a significantly higher

level than the bidders, at a rate of about 0.2% of total loans.

Tables 13, 14 and 15 describe the smaller, status-change sample. The difference

between the targets and bidders in this sample is extreme. The status-change targets are

on average smaller than their bidder counterparts by a factor of 26. These targets are

smaller than the merger-file targets by about 40%, while the bidders are nearly 3 times

the size of the merger-file sample bidders.

32

6. Results

I divide this chapter into two sections. The first discusses the tests on the merger-

file sample. The second discusses the test of the status-change sample.

a. The merger-file sample

Table 9 presents the regression results of the scaled loan loss provision on the

control and test variables. To account for the effects of time and firms I use a full fixed

effect model. A test of the first and second moments of the model results in the rejection

of the null of no heteroscedasticity at less than the .0001 level. Accordingly, the t-stats

presented are adjusted with White’s (1980) method to correct for heteroscedasticity.

(1) Test Variables

Table 9 and 10 present the results of the main tests. In the regression estimate

presented in table 9, the coefficient on the Targets variable is negative and significant,

indicating that the public intrastate targets in the high probability period are exhibiting

negative unexpected ProvRatio. The level of the coefficient, -0.00114, equates to a

difference of about 5% of pre-tax, pre-provision income for the targets.

The coefficient on the private targets is not significantly different from zero. This

variable is measuring the difference in the average ProvRatio between the public targets

and the private targets. This indicates that there was not a significant difference between

the public targets and the private targets in all settings. This result is after controlling for

a bank being private. Hypothesis H1B proposed that private banks would manage their

33

provision downward in a greater amount than the public banks, because they had fewer

constraints on their earnings management ability. This hypothesis can be rejected.

The coefficient on the Interstate Targets variable is positive and significant. This

variable is measuring the difference in the average ProvRatio between the intrastate

targets and the targets of interstate mergers. Hypothesis H1C suggested that the interstate

targets would not have the ability to manage their provision because of an inability to

predict the timing of the merger. The prediction was that all targets would have a

negative loan loss provision and interstate targets would be positively offset from that

negative position. The result on the Interstate Targets variable supports that hypothesis.

The positive and significant coefficient on the Low Probability Targets indicates

that the time-period matters. This variable is measuring the difference in the average

ProvRatio between the targets in the high probability period (1992-1997) and those same

banks in the low probability period (1987-1991, 1998). This result lends support to the

idea that the targets will engage in earnings management if they are in a position to

predict the timing of the takeover. However, the size of the coefficient on this variable is

large enough to overwhelm any earnings management effect during the low probability

time-period. In other words, during the low probability period, the targets had a positive

unexpected provision, and in the high probability period, they had a negative unexpected

provision. To insure that the low/high probability distinction was in fact capturing the

ability of the targets to predict, and not some other time specific anomaly, I estimated the

34

model letting the bidders vary across the two periods. The results (not presented) indicate

that the bidders were not affected by the time-period.

As an additional check to address the issue of targets managing earnings during

particular periods when the probability of takeover is high, I also estimate the model

allowing the coefficient on all targets and all bidders to vary across all years. Table 11

reports just the estimation of the unexpected provision for all targets and bidders, by year.

The model is estimated with the same control variables as in Table 9. The estimates for

the controls are virtually identical and are not reported. Then I correlate the estimated

coefficients and the yearly probabilities of takeover from Figure 1. Table 12 presents

these correlations. If the targets’ level of earnings management is related with the

probability of takeover in the manner I expect, then the correlation between the target

coefficient and the yearly probabilities should be negative. The results support this

expectation; the Pearson correlation is negative, but not statistically significant, while the

Spearman rank-correlation is negative and significant at the 0.07 level. Additionally, the

estimates for the unexpected provisions for the bidders are negative 2 out of 12 years, and

none of the 12 is positive. In addition, there is no observable correlation between the

bidders’ estimated coefficients and the yearly probabilities.

Table 10 presents the total effect for each of the different sets of targets. I split the

targets into the high and low-probability time-periods. In each period I report the total

estimate of the unexpected provision for the public/private and intrastate/interstate splits

of the targets. Any negative provision management is only apparent during the high-

35

probability period. It is strongest for the private intrastate targets, and progressively

weaker for public intrastate targets and private interstate targets. It is non-existent for the

public interstate targets.

The opposite seems to be true during the low probability time-period. The large

positive effect of the time-period makes the public interstate targets a strongly positive

unexpected provision during this period. The private interstate targets and the public

intrastate targets are progressively weaker in the positive direction. The private intrastate

targets are not significantly different from zero.

The results for the Bidders are inconclusive. The coefficient on the bidder variable

in table 9 is negative but not significantly different from zero. The results from the

additional analysis in Table 11 also indicate that the bidders have generally negative (and

never positive) unexpected provisions over the sample period, and there is no relation

between the level of unexpected provision by bidders and the probability of a takeover.

(2) Control Variables

The inferences drawn from the results on the test variables necessarily depend on

the model's ability to control for the expected value of the loan loss provision. The three

variables intended to capture the quality of the loan portfolio work as they are intended.

The positive and significant coefficient on ∆NPL indicates that as banks’ non-performing

loans increase so does the provision for the total portfolio. The negative and significant

coefficient on Allowance indicates that banks on average miss the true value of their net

portfolio at the end of any given period. They decrease the loan provision to make up for

36

an estimate of the loan loss allowance that was high at the beginning of the year. The

significant negative coefficient on ∆Assets indicates that faster growing banks have a

smaller provision, perhaps because the average age of their loans is lower. I also ran the

model substituting a loan growth variable for the change in assets variable. The results

(not presented) are virtually identical.

The results on CapRatio are unexpected. Prior studies have generally documented a

negative relation between capital and the provision. The coefficient on CapRatio is

positive and significant in the period 1987 to 1990, and becomes less positive, but

remains significant in the period 1991 and after.

There is a significant negative coefficient on EBTP for public banks. The

coefficient is less negative, but still significant for private banks. This indicates that

banks that have higher income before their provision have even higher income after the

provision. This indicates that income smoothing is not occurring in this sample. This

result is troubling; perhaps it indicates that the model is still missing a variable that

adequately captures the quality of the loan portfolio. I discuss the results on both the

CapRatio and the EBTP in detail in the sensitivity analysis section that follows.

The size variable is positive and significant, indicating that the larger the bank the

greater the loan loss provision. The Private variable is negative, but not significantly

different from zero. This indicates that the public/private distinction is not a determinant

of loan loss provision.

37

The control variables for changes in state income seem to work as intended. When

the income in the state which the bank is headquartered in increases, the bank's provision

decreases. This seems to indicate that this variable is capturing the expectation for future

default as intended. The effect of a change in state unemployment in the state that the

bank is headquartered in is insignificant. This could be because of differential effects

across consumer and commercial loans. If an increase in unemployment increases the

probability of an increase in consumer loan default, then this should be positive.

However, an increase in unemployment might also signal lower labor costs for employers

and decrease the probability of commercial loan defaults. The results on the change in

interest rates are unexpected. I had anticipated that increases in interest rates would signal

higher default probabilities and result in higher provisions. The significant negative

coefficient on both the change and the lag of the change in rates indicates that as rates go

up, the provision goes down. Perhaps this is an indication that the effect of borrowers

refinancing their fixed rate loans as rates decrease lowers the overall probability of

default.

b. The status-change sample

Table 16 presents the results of the test run on the status-change sample. All of the

control variables are virtually identical to those in the previous tests. As for the target-test

variables, there is no conclusive evidence to support H3. In the target sample the

unexpected provision is negative for yearst-2 through yeart+1. It is significant in yeart-1, in

the correct direction, and in yeart+1, in the incorrect direction. Yeart+2 is positive, but

38

insignificant. The results indicate that the hypothesis of earnings management prior to the

merger and then a reverse of that management might be correct, but the timing is not as

expected. The unexpected provision is negative more than a year before the merger, and

in the year of the merger. While the positive unexpected provision does not occur as

expected, there is at least a non-negative unexpected provision two years after the

merger.

The results for the bidders are supportive of H3. The two years just prior to the

merger are negative. This is consistent with the bidders managing earnings upward prior

to the merger. The expected reversing behavior is not found. While the two years after

the merger are not positive, they are not negative either. This result could indicate that

any reversing behavior either takes longer than two years, or is masked by continued

growth

39

7. Sensitivity Analysis

a. Capital and Earnings

The unexpected results on my capital and earnings variables could be the result of a

number of factors. Prior studies looking at the influence of these two factors have used a

much smaller subset of the population of banks. However, even when I limit my sample

to only large public banks (results not presented) the signs on these two variables are the

same. The time period that I use in this study is also different, and includes more years in

the period after the introduction of risk based capital standards. However, even when I

limit the sample to large banks and the period between 1987 and 1994 the signs on these

variables are still the same. I also estimated the model letting capital vary by year. The

basic result is the same, significant positive coefficients for the early years of the sample,

insignificant coefficients in the middle years, and significant, though smaller in absolute

value than the early years, negative coefficients in the later years. I also estimated the

model letting capital vary across bank size peer groups. All of the groups were positive,

with the largest group being insignificant. None of these tests significantly affected the

test variables.

The prior work that looked at the effect of regulatory capital has some conflicting

findings. Moyer (1990), Beatty et al. (1995) and Ahmed et al. (1999) all support the

influence of capital management, while Collins et al. (1995) does not. The prior studies

set in the period before the 1990 change in capital requirements used Primary capital as

the measure capital. The CapRatio measure I use does not capture the effect of the

40

allowance, perpetual preferred stock or mandatory convertible debt. Ahmed et al. uses a

measure of Primary capital before 1990 and Tier I capital after 1990. Again, there are

similar differences between these measures and my CapRatio variable. These subtle

differences could account for the finding that is contrary to prior research.

The anomalous earnings result is even stronger than the capital, and I tried a

number of model designs to try to determine why. I let the earnings vary across both

years and the public and private distinction. Every year public banks were negative and

significant, and in every year, private banks were positive and significant. The R2 on this

model climbed to 0.4016, but the test variables remained essentially the same. The

earnings smoothing result is not as well supported by prior research as the capital

management result is. Only Collins et al. (1995) finds significant results that indicate the

relationship between the provision and earnings is positive. While Beatty et al. (1995) do

not find a significant positive relationship in their simultaneous equations model, they do

have a finding similar to mine in an ordinary least squares estimate.

Ahmed et al. (1999) claim that the inclusion of their proxy for business failures is

an important determinant of their model, and a prime reason why they do not find support

for the smoothing hypothesis in contrast to Collins et al. (1995). I attempt to control for

the macro-economic factors that should impact the level of the provision with state-by-

state variables that are less precise indicators of each banks exposure to changes in the

economic environment. It is possible that the difference in selection of control variables

impacts the model quite significantly.

41

There are most likely a number of factors that all contribute to the differences

between the findings of prior research and my current findings. Across all of the

aforementioned research the sample and period studied are all different sets of banks in

different time-periods. Each study uses a slightly different model design to test their

hypotheses. Beatty et al. (1995) uses a five-equation simultaneous equation model

estimated using both 2 and 3-stage least squares. Collins et al. (1995) uses a series of

firm-specific regressions. Ahmed et al. (1999) uses a pooled cross-sectional regression

without the use of fixed effects for firm and year. All of these different models have led

to slightly different inferences, and my use of a pooled cross-sectional full fixed-effects

model leads to slightly different results. The final difference among all of these papers is

the selection of and the measurement of the variables meant to control for the non-

discretionary portion of the loan loss provision. Only Ahmed et al. (1999) and the current

study attempt to control for macro-economic effects on the determination of the loan loss

provision. The variables used for this control in the two studies are very different because

of data constraints. Undoubtedly all of these research design differences have led to the

findings of the current study being different from the prior research

b. Sample choices

I decided to test the merger-file sample separately from the status-change sample

because the two groups must have had some fundamental differences. It is costly to

report a subsidiary separately; if a parent could consolidate, it would make sense to do so.

Both regulatory reporting costs as well as the loss of flexibility having to maintain two

42

entity’s capital levels instead of just one are obvious differences. The relative sizes of the

targets and bidders are very different as well. In the merger-file sample, the bidders are

on average 8 times larger than the targets. While the status-change sample, the bidders

are 26 times larger than the targets. All of these reasons convinced me that I should

perform two complementary tests. However, when I include the status-change sample in

the table 9 tests, the results for the targets are similar to those in table 9, and the results

for the bidders become significant. When the 27 bidder observations from the status-

change sample are included in the test in table 9, it results in a significant coefficient for

the bidders in table 9.

43

8. Contributions and Extensions

a. Contributions

This study seeks to provide evidence on the earnings management behavior of

bidders and targets in bank holding company mergers. The principal finding of this study

is that some of the targets of takeovers manage their earnings upwards with the use of the

loan loss provision. They do this only in periods when the expectation of being a target is

high. This indicates that some of the targets of mergers do have the incentive to manage

earnings, but they can only do so in select periods, or if they have some other special

insight into the coming deal.

I also find results concerning bidders that are supportive of those of Erickson and

Wang (1999) even though the firms studied and the methods used are very different. The

results on the bidders from the merger-file sample are marginally insignificant, but in the

proper direction, and the results from the status-change sample are supportive of bidders

managing their earnings upwards using the loan loss provision prior to a merger.

These results should be of interest to accounting researchers, capital market

participants, and regulators. All of these groups have an interest in understanding, and/or

undoing earnings management. The findings from this study, combined with those of

Erickson and Wang (1999), because of the very different nature of the studies, should

lend credibility to both studies and methods.

The results of this study should be of particular interest to regulators. The SEC, the

FDIC, the Federal reserve Board, the Office of the Comptroller of the Currency and the

44

Office of Thrift Supervision have recently come together to form a joint interagency task

force with the purpose of examining the reporting of loan loss allowances in commercial

banks. This task force is concerned with the consistent application of GAAP in assessing

the level of the allowance. To the extent that a bank might step outside the bounds of

GAAP in practicing earnings management, then these results will be of interest to this set

of regulators.

b. Extensions

One major extension that is in progress is meant to determine how, if at all, does

this earnings management affect the outcome of the valuation. I am collecting pricing

data on as many of the merger targets as possible. By estimating a loan loss provision

model without any indicator variables, I can then use the residuals for the targets as a

continuous measure of earnings management. This measure will be regressed on the price

multiples from the targets’ pricing data. A negative relation would indicate that the bidder

was able to undo the earnings management, while a positive relation might indicate that

the targets with the greatest incentive to manage earnings did so the most.

45

46

References

Ahmed, Anwer, S., Carolyn Takeda and Shawn Thomas, “Bank Loan Loss Provisions: A Reexamination of Capital Management, Earnings Management and Signaling Effects,” Journal of Accounting & Economics (v28 n1 November 1999), 1-25.

Allen, Linda and Anthony Saunders, “Bank Window Dressing: Theory and Evidence,” Journal of Banking and Finance (v18 1992), 585-623.

Amel, Dean F., “State Laws Affecting the Geographic Expansion of Commercial Banks,” Unpublished Working Paper, Board of Governors of the Federal Reserve System, 1993.

Ball, Ray and Philip Brown, “An Empirical Evaluation of Accounting Income numbers” Journal of Accounting Research (n6, 1968) 159-178.

Beatty, Anne and David Harris “The Effects of Taxes, Agency Costs and Information Asymmetry on Earnings Management: A Comparison of Public and Private Firms” Review of Accounting Studies, Vol. 4, Issues 3 & 4, forthcoming.

Beatty, Anne, Bin Ke and Kathy Petroni “Differential Earnings Management to Avoid Earnings Declines and Losses across Publicly and Privately-Held Banks” Working Paper, Michigan State University, November 2000.

Beatty, Anne, Sandra L. Chamberlain and Joseph Magliolo, “Managing Financial Reports of Commercial Banks: The Influence of Taxes, Regulatory Capital, and Earnings,” Journal of Accounting Research (Autumn 1995), 231-261.

Beaver, William H and Ellen E. Engel “Discretionary behavior with respect to allowances for loan losses and the behavior of security prices” Journal of Accounting & Economics (August-December 1996) 177-206.

Beaver, William, Roger Clarke and William Wright “The Association Between Unsystematic Security Returns and the Magnitude of Earnings Forecast Errors” Journal of Accounting Research (Autumn 1979) 316-340.

Becker, Connie L., Mark L. Defond, James Jiambalvo and K. R. Subramanyam, “The Effect of Audit Quality on Earnings Management.” Contemporary Accounting Research (v15, n1, Spring 1998) 1-24.

Bhat, Vasanthakumar, N., “Banks and Income Smoothing: An Empirical Analysis.” Applied Financial Economics (6, 1996), 505-510.

Cahan, Steven F., “The effect of Antitrust Investigations on Discretionary Accruals: A Refined Test of the Political-Cost Hypothesis,” The Accounting Review (January 1992), 77-95.

47

Chamberlain, Sandra, “The Effect of Mergers on The Operating Performance and Discretionary Earnings Components of Acquired Banks,” PhD. Thesis, University of Chicago 1991.

Clarkson, Peter M and Dan Simunic, “The Association Between Audit Quality, Retained Ownership, and Firm-Specific Risk in U. S. versus Canadian IPO Markets,” Journal of Accounting & Economics (v17 n1 January 1994), 207-228.

Collins, Julie H., Douglas A. Shackelford, and James M. Wahlen, “Bank Differences in the Coordination of Regulatory Capital, Earnings, and Taxes,” Journal of Accounting Research (Autumn 1995), 263-291.

DeAngelo, Linda, “Accounting Numbers as Market Valuation Substitutes: A Study of Management Buyouts of Public Stockholders” The Accounting Review; (July 1986) 400-422.

DeAngelo, Linda, “Discussion of Evidence of Earnings Management from the Provision for Bad Debts,” Journal of Accounting Research (Supplement 1988), 32-40.

DeAngelo, Linda, Managerial Competition, Information Costs, and Corporate Governance: The Use of Accounting Performance Measures in Proxy Contests Journal of Accounting & Economics (Jan 1988) 3-37.

DeAngelo, Linda, “Equity Valuation and Corporate Control” The Accounting Review (Jan. 1990) 93-113.

Dechow, Patricia M. and Douglas J. Skinner, “Earnings Management: Reconciling the Views of Accounting Academics, Practitioners, and Regulators.” Accounting Horizons (June 2000) 235-250.

Dechow, Patricia M., Richard G. Sloan and Amy Sweeney, “ Detecting Earnings Management,” The Accounting Review (April 1995), 194-225.

Easterwood, Cintia M., “Takeovers And Incentives For Earnings Management: An Empirical Analysis,” Journal of Applied Business Research (v14, n1 1997/1998), 29-47.

Erickson, Merle and Shiing-wu Wang, “Earnings Management by Acquiring Firms in Stock for Stock Mergers,” Journal of Accounting and Economics (v27 1999), 149-176.

Foster, George, “Quarterly Accounting Data - Time Series Properties and Predictive - Ability Results” The Accounting Review (January 1977).

Francis, Jere, Edward Maydew and H. Charles Sparks, “The Role of Big 6 Auditors in the Credible Reporting of Accruals,” Auditing (v18, n2, Fall 1999), 17-34.

Gaver, J. K. Gaver, and J. Austin “Additional Evidence on Bonus Plans and Income Management,” Journal of Accounting and Economics (1995), 3-28.

48

Greenawalt, Mary Brady and Joseph F. Sinkey, Jr., “Bank Loan-Loss Provisions and the Income-Smoothing Hypothesis: An Empirical Analysis, 1976-1984,” Journal of Financial Services Research, (1, 1988), 301-318.

Greene, William H., “Econometric Analysis, 4th edition,” 2000 Prentice-Hall, Upper Saddle River, NJ.

Griffin, Paul A., “Further Evidence on the Economic Effects of Changes in Loan Loss Provisions on Bank Stock Returns,” Abacus (34(2), 1998), 188-203.

Gup, Benton E., David C. Cheng, Larry D. Wall, and Kartono Liano, “Regional Differences in Bank Merger Pricing,” In Bank Mergers: Current Issues and Perspectives, edited by Benton E. Gup, (Boston: Kluwer Academic Publishers 1989), 69-86.

Hansen, Glen, “Bias and Measurement Error in Discretionary Accrual Models” Working Paper, Penn State University, 1999.

Healy, Paul, “The Effect of Bonus Schemes on Accounting Decisions.” Journal of Accounting and Economics (7, 1985) 73-109.

Healy, Paul M. and James M. Wahlen, “A Review of The Earnings Management Literature and Its Implications for Standard Setting,” Accounting Horizons (December 1999) 365-383.

Holthausen, R., D. Larker and R. Sloan, “Annual Bonus Schemes and the Manipulation of Earnings” 1995 journal of Accounting and Economics (19) 29-74.

Hunter, William C. and Larry D. Wall, “Bank Merger Motivations: A Review of the Evidence and an examination of Key Target Bank Characteristics,” Economic Review (September/October 1989), 2-19.

Jayaratne, Jith and Philip E. Strahan, “Entry Restrictions, Industry Evolution, and Dynamic Efficiency: Evidence From Commercial Banking,” The Journal of Law and Economics (April 1998), 239-273.

Jones, Jennifer J., “Earnings Management During Import Relief Investigations,” Journal of Accounting Research (Autumn 1991), 193-228.

Key, Kimberly G., “Political Cost Incentives for Earnings Management in the Cable Television Industry,” Journal of Accounting and Economics (23, 1997), 309-337.

Kim, Myung-Sun and William Kross “The Impact of the 1989 Change in Bank Capital Standards on Loan Loss Provisions and Loan Write-Offs” Journal of Accounting and Economics (n25, 1998) 69-99.

Kroszner, Randall S. and Philip E. Strahan, “What drives deregulation? Economics and politics of the relaxation of bank branching restrictions,” The Quarterly Journal of Economics (November 1999), 1437-1467.

49

Levitt, A. “The Numbers Game,” Speech Delivered at the NYU Center for Law and Business, New York, NY September 28, 1998.

Padgett, Tania, “Look for Increase in Hostile Takeovers,” American Banker (October 16, 1996) Pg. 22.

Rangan, Srinivasan, “Earnings management and the performance of seasoned equity offerings,” Journal of Financial Economics (October 1998) 101-122.

Teoh, Siew Hong, Ivo Welch and T.J. Wong. “Earnings Management and the Long-Run Market Performance of Initial Public Offerings.” Journal of Finance (December 1998a) 1935-1974.

Teoh, Siew Hong, Ivo Welch and T.J. Wong. “Earnings Management and the Underperformance of Seasoned Public Offerings.” Journal of Financial Economics (October 1998b) 63-99.

Teoh, Siew Hong, S.H., T.J. Wong and G. Rao “Are Accruals During Initial Public Offerings Opportunistic?” Review of Accounting Studies (1998) 175-208.

Wahlen, James M., “The Nature of Information in Commercial Bank Loan Loss Disclosures,” The Accounting Review (July 1994), 455-478.

Weidner, David, “Deep-Sixing of Pooling Method May Spark Dash to Finish Line,” American Banker (February 7, 2000) p. 6A.

50

51

Appendix: Tables

Table 1 Probability of Being a Target

Year

Number of Bank Holding Companies

Number of Targets

Probability of being a target

1987 1,045 41 0.0392 1988 1,111 33 0.0297 1989 1,181 23 0.0195 1990 1,256 28 0.0223 1991 1,350 44 0.0326 1992 1,336 61 0.0457 1993 1,331 83 0.0624 1994 1,165 65 0.0558 1995 1,179 69 0.0585 1996 1,192 68 0.0570 1997 1,195 92 0.0770 1998 1,233 34 0.0276

The above table shows the probability of being a target in the following year for the bank holding companies that are included in my sample. The average of the low probability years (1987-1991 and 1998) is 0.0285 while the average for the high probability years (1992-1997) is 0.0594.

52

Table 2 Sample Selection Procedures

Number of Observations lost

Number of BHCs lost / remaining

Total Observations

Number of observations in the Y-9 reports with positive total assets for the periods ending on 12-31 for the years 1986-1998.

3,127 19,392

Less BHCs not headquartered in the 50 states or D.C. 114 26 / 3,101 19,278

Less those bank holding companies that have interruptions in their yearly reporting. 1,317 187 / 2,914 17,961

Less the first year to accommodate the use of a change variable and the beginning of the year allowance for loan losses

2,914 409 / 2,505 15,047

Less all observations of any BHC’s that have any observations of either negative capital, missing data or, extreme observations.

473 91 / 2,414 14,574

Final Sample 2,414 14,574

53

Table 3 Distribution of all sample observations

Distribution of observations across years and bank holding company types. Year Total Observations Public Private 1987 1,045 533 512 1988 1,111 555 556 1989 1,181 578 603 1990 1,256 591 665 1991 1,350 612 738 1992 1,336 605 731 1993 1,331 612 719 1994 1,165 583 582 1995 1,179 583 596 1996 1,192 572 620 1997 1,195 549 646 1998 1,233 538 695 Total Observations 14,574 6,911 7,663

Number of BHC’s 2,414 970 1,444

The Public designation indicates that the BHC was listed as a SEC filer in the last period that they appear in the sample. The private designation indicates that the BHC was listed as a SEC non-filer during the last period they appear in the sample.

54

Table 4 Distribution of target observations

Distribution of target observations across years and merger types.

Year Target Observations

Public Targets

Private Targets

Intrastate Targets

Interstate Targets

1987 41 19 22 31 10 1988 33 16 17 25 8 1989 23 11 12 19 4 1990 28 12 16 20 8 1991 44 21 23 28 16 1992 61 17 44 38 23 1993 83 35 48 48 35 1994 65 26 39 40 25 1995 69 39 30 35 34 1996 68 42 26 34 34 1997 92 48 44 38 54 1998 34 15 19 18 16 Total

Observations 641 301 340 374 267

Number of BHC’s 641 301 340 374 267

The merger date designation indicates that a bank was involved in a merger in the following year. For example, if a bank were a target at any time during 1991, it would be a target observation in 1990. The public and private designations are as defined in Table 3. The intrastate and interstate designation relates to the relationship between the headquarters of the bidder and target.

55

Table 5 Distribution of bidder observations

Distribution of bidder observations across years and merger types.

Year Bidder Observations

Public Bidders

Private Bidders

Intrastate Bidders

Interstate Bidders

1987 19 15 4 16 3 1988 20 13 7 17 3 1989 10 9 1 8 2 1990 12 8 4 9 3 1991 20 14 6 11 9 1992 29 19 10 18 11 1993 49 40 9 31 18 1994 33 25 8 19 14 1995 35 29 6 18 17 1996 38 30 8 23 15 1997 57 45 12 27 30 1998 24 17 7 11 13 Total

Observations 346 264 82 208 138

Number of BHC’s 200 144 56 148 84

The merger date designation indicates that a bank was involved in a merger in the following year. For example, if a bank were a bidder at any time during 1991, it would be a bidder observation in 1990. To ensure that the timing of the bidder observations is correct, the bidders are selected from the targets identified in Table 4. For the bidder observations, a BHC can be involved in multiple mergers, as long as they are involved in at least one merger during the year, then they will be listed as a bidder on the previous 12-31. Because of the nature of a takeover, a bidder type can have more than one bidder observation.

56

Table 6 Descriptive statistics for all observations

Descriptive statistics for the sample 14,574 yearly observations for 2,414 bank holding companies that file Y-9 reports with the federal reserve and meet the data requirements listed in Table 2. Variable Mean SD Min 25% 50% 75% Max ProvRatio 0.00662 0.01079 -0.06165 0.00191 0.00369 0.00709 0.23768 Provision 17.8131 123.180 -115.00 0.24800 0.73500 2.66900 4410.00 Loans 1881.49 9626.47 3.06500 105.863 182.378 502.151 365614 ASSETS 3050.45 15734.4 8.52700 192.219 309.509 827.318 617679 ∆NPL 0.00030 0.01221 -0.16221 -0.00280 0.0000 0.00294 0.20427 CapRatio 0.08423 0.02645 0.01010 0.06821 0.08155 0.09673 0.37421 EBTP 0.01531 0.00975 -0.15282 0.01162 0.01589 0.01986 0.17944 Allowance 0.01634 0.00914 0.00000 0.01109 0.01402 0.01853 0.20342 AssetGrowth 0.07411 0.12071 -1.96555 0.02092 0.06244 0.11551 0.89052 ∆INC 0.05657 0.02144 -0.20702 0.04391 0.05595 0.06756 0.30530 ∆UNP 0.04505 0.25665 -0.62794 -0.11475 0.00102 0.17632 1.68750 ∆RATE -0.18591 1.40447 -2.43660 -1.15340 -0.52340 0.03670 3.07000 Variable Definitions: ProvRatio - is the loan loss provision for the period scaled by the value of gross loans at the end of the period. Provision – is the dollar value of the loan loss provision in millions of dollars. Loans - is the dollar value of loans at the end of the year in millions of dollars. ASSETS - is the end of year total assets measured in millions of dollars. ∆NPL - is the yearly change in non-accrual loans scaled by average total assets. CapRatio - is owner’s equity plus 60% of the loan loss provision scaled by total assets. EBTP - is the pretax net income plus 60% of the loan loss provision scaled by average total assets. Allowance – is the loan loss allowance at the beginning of the year scaled by the gross loans at the beginning of the year. AssetGrowth - is the change in assets during the year scaled by total assets at the end of the year. ∆INC - is the yearly percentage change in the fourth-quarter personal income in the state that the bank is headquartered in. ∆UNP - is the yearly percentage change in the fourth-quarter unemployment payments in the state that the bank is headquartered in. ∆RATE - is the one-year change in the average monthly quotes for the one-year treasury rate in the fourth quarter. That is ((RateOct+RateNov+RateDec)t/3)-((RateOct+RateNov+RateDec)t-1/3).

57

Table 7 Descriptive statistics for target observations

Descriptive statistics for the sample 641 observations for target bank holding companies that are described in Table 4. Variable Mean SD Min 25% 50% 75% Max ProvRatio 0.00697 0.01093 -0.01469 0.00161 0.00379 0.00795 0.11502 Provision 18.8077 123.075 -20.0000 0.23400 0.86500 3.00000 2618.00 Loans 1968.31 9000.14 7.64600 109.938 200.319 684.350 168105 ASSETS 3162.82 14226.6 14.9030 203.395 335.039 1171.20 260159 ∆NPL -0.00137 0.01368 -0.07996 -0.00363 -0.00018 0.00266 0.16211 CapRatio 0.08407 0.02604 0.01041 0.06923 0.08181 0.09692 0.29329 EBTP 0.01382 0.00974 -0.05846 0.00973 0.01484 0.01933 0.05498 Allowance 0.01769 0.01004 0.00198 0.01173 0.01496 0.01978 0.07544 AssetGrowth 0.04665 0.11096 -0.62685 -0.00326 0.04087 0.09372 .061466 ∆INC 0.05587 0.02067 -0.20702 0.04373 0.05503 0.06731 0.11596 ∆UNP -0.00195 0.22870 -0.60317 -0.14286 -0.03882 0.12500 0.84848 ∆RATE -0.05192 1.40722 -2.43660 -1.15340 -0.03660 0.03670 3.07000 Variable Definitions: ProvRatio - is the loan loss provision for the period scaled by the value of gross loans at the end of the period. Provision – is the dollar value of the loan loss provision in millions of dollars. Loans - is the dollar value of loans at the end of the year in millions of dollars. ASSETS - is the end of year total assets measured in millions of dollars. ∆NPL - is the yearly change in non-accrual loans scaled by average total assets. CapRatio - is owner’s equity plus 60% of the loan loss provision scaled by total assets. EBTP - is the pretax net income plus 60% of the loan loss provision scaled by average total assets. Allowance – is the loan loss allowance at the beginning of the year scaled by the gross loans at the beginning of the year. AssetGrowth - is the change in assets during the year scaled by total assets at the end of the year. ∆INC - is the yearly percentage change in the fourth-quarter personal income in the state that the bank is headquartered in. ∆UNP - is the yearly percentage change in the fourth-quarter unemployment payments in the state that the bank is headquartered in. ∆RATE - is the one-year change in the average monthly quotes for the one-year treasury rate in the fourth quarter. That is: ((RateOct+RateNov+RateDec)t/3)-((RateOct+RateNov+RateDec)t-1/3).

58

Table 8 Descriptive statistics for bidder observations

Descriptive statistics for the sample 346 yearly observations for 200 bidder bank holding companies that are described in Table 5. Variable Mean SD Min 25% 50% 75% Max ProvRatio 0.00503 0.00550 -0.00694 0.00228 0.00384 0.00616 0.05566 Provision 67.5508 165.374 -17.202 1.89100 7.39050 57.8590 1305.00 Loans 10391.9 21158.6 5.87700 812.187 2514.16 10184.4 264562 ASSETS 16359.5 33889.4 23.3080 1334.72 4297.83 15429.7 264562 ∆NPL -0.00049 0.00622 -0.3941 -0.00194 -0.00012 0.00169 0.02612 CapRatio 0.08526 0.01908 0.04301 0.07399 0.08253 0.09481 0.24326 EBTP 0.1916 0.00609 -0.02317 0.01632 0.01916 0.02236 0.04786 Allowance 0.01859 0.00872 0.00768 0.01310 0.01573 0.02085 0.07841 AssetGrowth 0.12461 0.13791 -0.50503 0.03992 0.09675 0.18271 0.70276 ∆INC 0.05334 0.02251 -0.20702 0.04150 0.05328 0.06510 0.10923 ∆UNP -0.01437 0.20836 -0.44578 -0.13699 -0.04321 0.08836 0.67532 ∆RATE -0.03515 1.35652 -2.43660 -1.09330 -0.03660 0.03670 3.07000 Variable Definitions: ProvRatio - is the loan loss provision for the period scaled by the value of gross loans at the end of the period. Provision – is the dollar value of the loan loss provision in millions of dollars. Loans - is the dollar value of loans at the end of the year in millions of dollars. ASSETS - is the end of year total assets measured in millions of dollars. ∆NPL - is the yearly change in non-accrual loans scaled by average total assets. CapRatio - is owner’s equity plus 60% of the loan loss provision scaled by total assets. EBTP - is the pretax net income plus 60% of the loan loss provision scaled by average total assets. Allowance – is the loan loss allowance at the beginning of the year scaled by the gross loans at the beginning of the year. AssetGrowth - is the change in assets during the year scaled by total assets at the end of the year. ∆INC - is the yearly percentage change in the fourth-quarter personal income in the state that the bank is headquartered in. ∆UNP - is the yearly percentage change in the fourth-quarter unemployment payments in the state that the bank is headquartered in. ∆RATE - is the one-year change in the average monthly quotes for the one-year treasury rate in the fourth quarter. That is: ((RateOct+RateNov+RateDec)t/3)-((RateOct+RateNov+RateDec)t-1/3).

59

Table 9 Regression results for merger-file sample

Variable Coefficient Estimate

Adjusted t value p-value Expected

Coefficient Intercept 0.0000 0.0000 1.0000 Test Variables Targets -0.00114 -1.8461 0.0324 - Private Targets -0.00059 -0.9309 0.1759 - Interstate Targets 0.00100 1.4606 0.0721 + Low Probability Targets 0.00235 2.9509 0.0016 + Bidders -0.00040 -1.1761 0.1198 - Private Bidders 0.00078 0.9487 0.8286 - Firm Specific Control Variables ∆NPL 0.10489 6.5668 <.0001 + Allowance -0.11994 3.8307 0.0001 - ∆Assets -0.01206 7.6440 <.0001 - CapRatio 0.07466 6.6450 >.9999 - CapRatio*Reg -0.03609 6.0833 >.9999 + EBTP -0.73091 10.5080 >.9999 + EBTP*PRIVATE 0.18184 2.0165 0.9563 - Size 0.00255 6.7766 <.0001 +/- Private -0.00253 0.6583 0.2552 +/- Macro-Economic Control Variables ∆State Income -0.02937 7.8960 <.0001 - ∆State unemployment -0.00007 0.1291 -0.4486 + ∆Interest Rate -0.00068 3.7035 <.0001 + Lag ∆State Income 0.02762 7.0850 <.0001 - Lag ∆State unemployment -0.00003 0.0000 >.9999 + Lag ∆Interest Rate -0.00100 5.7451 >.9999 + Number of Observations 14,574 Degrees of Freedom 12,126 Adjusted R-Squared 0.3555

60

Table 9 continued Variable Definitions: Dependent Variable: ProvRatio is the loan loss provision for the period scaled by the value of gross loans at the end of the period. Test Variables: Targets - is an indicator variable that is one if the bank is a merger target in the following calendar year. Private Targets - is an indicator variable that is one if Targets is one and if the bank is private. Interstate Targets - is an indicator variable that is one if Targets is one and if the merger is with an out-of-state acquirer. Low Probability Targets - is an indicator variable that is one if Targets is one and the year is 1987-1991 or 1998. Bidders - is an indicator variable that is one if the bank is an acquirer in the following calendar year. Private Bidder – is and indicator variable that is one if Bidders is one and the bank is private. Firm Specific Control Variables: Capital Ratio is owners equity plus the loan loss provision scaled by total assets. Reg is an indicator variable that is one for the period 1991-1998. Change in non-performing loans is the change in non-accrual loans scaled by the amount of loans. Size is the log of total assets. EBTP is the pretax net income plus the loan loss provision scaled by assets. Private is an indicator variable that is one if the bank is a privately-held firm. Macro-Economic Control Variables: Change in state Income is the yearly percentage change in the fourth-quarter personal income in the state that the bank is headquartered in. Change in state unemployment is the yearly percentage change in the fourth-quarter unemployment payments in the state that the bank is headquartered in. Change in interest rate is the one-year change in the average monthly quotes for the one-year treasury rate in the fourth quarter.

61

Table 10 Total effect of unexpected provision

The total effect of unexpected loan loss provision for different groups of targets. Targets High Probability Time Period Low Probability Time Period Public Private Public Private Intrastate -0.00114

(0.0324) -0.00173

(0.0004) 0.00121

(0.9403) 0.000618

(0.8041) Interstate -0.00019

(0.3810) -0.00078

(0.0809) 0.002162

(0.9881) 0.00157 (0.9531)

Total coefficients are the combination of the appropriate estimated coefficients on the test variables from table 9. For example, the total coefficient for a private target in the low-probability period would be the sum of the coefficients on targets, low probability targets, and private targets. The probabilities below the coefficients are based on a chi-square test using White’s adjusted covariance matrix and are adjusted for loss of the 2626 degrees of freedom due to the use of the mean adjustment process to control for fixed effects. They are a single -sided test against the null that the total coefficient is greater than or equal to zero. Panel B The total effect of unexpected loan loss provision for public and private bidders. Bidders Public Private -0.0004

(0.1198) 0.000383

(0.3028)

Total coefficients are the combination of the appropriate estimated coefficients on the test variables from table 9. For example, the total coefficient for a private bidder is the total of the two bid coefficients. The probabilities below the coefficients are based on a chi-square test using White’s adjusted covariance matrix and are adjusted for loss of the 2626 degrees of freedom due to the use of the mean adjustment process to control for fixed effects. They are a single-sided test against the null that the total coefficient is greater than or equal to zero.

62

Table 11 Regression results for bidders and targets by year

Coefficient estimates for all target and bidder observations by year. I form a separate indicator variable for each years’ targets and bidders that correspond with the data in tables 4 and 5. The control variables (not presented) are virtually identical to the model presented in Table 9.

All Targets By Year Coefficient Estimate

White’s Adjusted t-statistic

p-value Expected Direction

1987 0.01994 3.0303 0.0024 +/- 1988 0.00136 1.4112 0.1582 +/- 1989 0.00183 1.4142 0.1573 +/- 1990 -0.00038 0.2041 0.8383 +/- 1991 0.00306 1.8819 0.9701 - 1992 -0.00042 0.3535 0.3618 - 1993 0.00026 0.3535 0.6382 - 1994 -0.00139 1.3260 0.0924 - 1995 -0.00159 1.7511 0.0400 - 1996 -0.00250 2.9097 0.0018 - 1997 -0.00049 0.8366 0.2014 - 1998 -0.00128 0.9309 0.3519 +/- All Bidders By Year 1987 0.00146 0.9747 0.8351 - 1988 -0.00050 0.7071 0.2398 - 1989 -0.00394 2.8225 0.0024 - 1990 -0.00136 1.0287 0.1518 - 1991 0.00024 0.2041 0.5809 - 1992 -0.00015 0.1826 0.4276 - 1993 -0.00008 0.0913 0.4636 - 1994 -0.00061 0.5986 0.2747 - 1995 -0.00145 1.5652 0.0588 - 1996 -0.00067 0.8612 0.1946 - 1997 0.00034 0.5986 0.7253 - 1998 0.00177 1.8050 0.9645 - Number of Observations 15,574

Degrees of Freedom 12,107 Adjusted R2 0.3559

63

Table 12 Correlation between yearly coefficients and takeover probability

Correlation between coefficient estimates and yearly takeover probabilities. I present the Pearson correlation coefficients in the upper-right and Spearman rank-correlation coefficients in the lower-left. The probability shown below the correlation coefficient is the one-sided probability that the correlation is not negative. Targets Takeover

Probability Target

Coefficient

Takeover Probability 1 -0.45455 (0.0689)

Target Coefficient -0.20580 (0.2606) 1

Bidders Takeover

Probability Bidder

Coefficient

Takeover Probability 1 0.16783 (0.3011)

Bidder Coefficient 0.22641 (0.2396) 1

64

Table 13 Distribution of the status -change targets and bidders

Observations of those banks that have a change of top-tier status but remain on the data files as a separately reported subsidiary.

Year Target Observations

Bidder Observations

1989 6 1 1990 41 2 1991 6 4 1992 21 6 1993 17 4 1994 13 3 1995 8 5 1996 4 2 Total

Observations 116 27

Number of BHC’s 107 18

65

Table 14 Descriptive statistics for the status -change targets

Descriptive statistics for the sample 116 yearly observations for 107 target bank holding companies that are described in Table 13.

Variable Mean SD Min 25% 50% 75% Max ProvRatio 0.00804 0.01201 -0.00373 0.00183 0.00420 0.00968 0.07028 Provision 17.7291 62.1583 -0.20000 0.15400 0.92400 7.27100 557.300 Loans 1097.03 2234.68 6.22000 85.8920 237.351 833.385 15428.1 ASSETS 1808.22 3296.89 22.8790 166.618 516.280 1821.31 20106.9 ∆NPL -0.00240 0.01384 -0.11082 -0.00394 -0.00045 0.00310 0.02425 CapRatio 0.08431 0.03932 0.03395 0.06194 0.07538 0.09064 0.31166 EBTP 0.01484 0.01528 -0.03686 0.01013 0.01509 0.01944 0.12429 Allowance 0.01962 0.01103 0.00443 0.01229 0.01603 0.02572 0.06673 AssetGrowth 0.07459 0.13750 -0.34613 0.03153 0.06503 0.11285 0.86268 ∆INC 0.06070 0.01563 0.01647 0.05161 0.06098 0.07414 0.08941 ∆UNP 0.09954 0.26876 -0.42047 -0.08247 0.02920 0.36492 0.75179 ∆RATE -0.32000 1.33482 -2.43660 -1.15340 -0.52340 -0.03660 3.0700 Variable Definitions: ProvRatio - is the loan loss provision for the period scaled by the value of gross loans at the end of the period. Provision – is the dollar value of the loan loss provision in millions of dollars. Loans - is the dollar value of loans at the end of the year in millions of dollars. ASSETS - is the end of year total assets measured in millions of dollars. ∆NPL - is the yearly change in non-accrual loans scaled by average total assets. CapRatio - is owner’s equity plus 60% of the loan loss provision scaled by total assets. EBTP - is the pretax net income plus 60% of the loan loss provision scaled by average total assets. Allowance – is the loan loss allowance at the beginning of the year scaled by the gross loans at the beginning of the year. AssetGrowth - is the change in assets during the year scaled by total assets at the end of the year. ∆INC - is the yearly percentage change in the fourth-quarter personal income in the state that the bank is headquartered in. ∆UNP - is the yearly percentage change in the fourth-quarter unemployment payments in the state that the bank is headquartered in. ∆RATE - is the one-year change in the average monthly quotes for the one-year treasury rate in the fourth quarter. That is ((RateOct+RateNov+RateDec)t/3)-((RateOct+RateNov+RateDec)t-

1/3).

66

Table 15 Descriptive statistics for the status -change bidders

Descriptive statistics for the sample of 27 yearly observations for 18 bidder bank holding companies that are described in Table 13.

Variable Mean SD Min 25% 50% 75% Max ProvRatio 0.00586 0.00490 -0.01006 0.00397 0.00614 0.00754 0.01573 Provision 190.666 173.202 -18.6000 45.5540 164.900 250.000 605.000 Loans 29497.5 27465.2 1806.08 9702.85 23217.3 37722.7 123845 ASSETS 47596.1 46718.6 2706.61 16745.9 32346.1 61331.5 185794 ∆NPL -0.00155 0.00503 -0.01931 -0.00422 -0.00067 0.00088 0.00854 CapRatio 0.07605 0.00985 0.04670 0.07081 0.07607 0.08455 0.09022 EBTP 0.01961 0.00381 0.01218 0.01763 0.01870 0.02215 0.02762 Allowance 0.02201 0.00799 0.01084 0.01720 0.02127 0.02393 0.04359 AssetGrowth 0.12478 0.11661 -0.01095 0.05552 0.09598 0.17770 0.54925 ∆INC 0.05372 0.01838 0.01946 0.03697 0.05720 0.06581 0.08941 ∆UNP 0.06328 0.25612 -0.38964 -0.09043 0.06926 0.19035 0.83333 ∆RATE -0.59050 1.52541 -2.43660 -1.30340 -1.15340 -0.03660 3.0700 Variable Definitions: ProvRatio - is the loan loss provision for the period scaled by the value of gross loans at the end of the period. Provision – is the dollar value of the loan loss provision in millions of dollars. Loans - is the dollar value of loans at the end of the year in millions of dollars. ASSETS - is the end of year total assets measured in millions of dollars. ∆NPL - is the yearly change in non-accrual loans scaled by average total assets. CapRatio - is owner’s equity plus 60% of the loan loss provision scaled by total assets. EBTP - is the pretax net income plus 60% of the loan loss provision scaled by average total assets. Allowance – is the loan loss allowance at the beginning of the year scaled by the gross loans at the beginning of the year. AssetGrowth - is the change in assets during the year scaled by total assets at the end of the year. ∆INC - is the yearly percentage change in the fourth-quarter personal income in the state that the bank is headquartered in. ∆UNP - is the yearly percentage change in the fourth-quarter unemployment payments in the state that the bank is headquartered in. ∆RATE - is the one-year change in the average monthly quotes for the one-year treasury rate in the fourth quarter. That is ((RateOct+RateNov+RateDec)t/3)-((RateOct+RateNov+RateDec)t-

1/3).

67

Table 16 Regression results for the status -change sample

Test of reversing accruals using the sample of 116 target observations for those acquired banks that continue to be reported as subsidiary banks on the data files.

Variable Coefficient Estimate

Adjusted t value p-value Expected

Coefficient Intercept 0.00000 0.00000 1.0000 +/- Test Variables Targets t-2 -0.00079 -1.1180 0.1318 - Targets t-1 -0.00152 -2.2545 0.0121 - Targets -0.00028 -0.3291 0.3710 - Targets t+1 -0.00121 -1.6456 0.9501 + Targets t+2 0.00030 0.4378 0.3308 + Bidders t-2 -0.00028 -0.2582 0.3981 - Bidders t-1 -0.00145 -1.7150 0.0432 - Bidders -0.00228 -2.2097 0.0136 - Bidders t+1 -0.00092 -0.9398 0.8263 + Bidders t+2 -0.00104 -0.9036 0.8169 + Firm Specific Control Variables ∆NPL 0.10522 6.5894 <.0001 + Allowance -0.12092 -3.8663 0.0001 - ∆Assets -0.01199 -7.5830 <.0001 - CapRatio 0.07484 6.6655 >.9999 - CapRatio*Reg -0.03699 -6.2420 >.9999 + EBTP -0.73146 -10.5343 >.9999 + EBTP*PRIVATE 0.18284 2.0309 0.9799 - Size 0.00251 6.6842 <.0001 + Private -0.00314 -0.8113 0.2086 - Macro-Economic Control Variables ∆State Income -0.02934 -7.8937 <.0001 - ∆State unemployment -0.00013 -0.2236 0.5885 + ∆Interest Rate -0.00073 -3.9389 <.0001 +/- Lag ∆State Income -0.02772 -7.0995 <.0001 - Lag ∆State unemployment -0.00004 -0.0913 0.5364 + Lag ∆Interest Rate -0.00101 -5.7869 <.0001 +/- Number of Observations 14,574 Degrees of Freedom 12,126 Adjusted R-Squared 0.3553

Vita Bruce Bettinghaus

Education:

Ph.D. Business Administration, Major Field: Accounting, The Pennsylvania State University, Graduation Date: Fall 2000

B.B.A. Grand Valley State University, 1994, Major: Accounting, Minor: Economics

Academic Experience: Assistant Professor, University of Missouri-Columbia, July 2000 –present Graduate Assistant, Penn State University, August 1994 – June 2000

Teaching Interests: Financial Accounting, Financial Statement Analysis, Strategic Reporting

Teaching Awards Graduate Assistant Teaching Award, The Graduate School, PSU, 1999 Ossian MacKenzie Teaching Award, Smeal College of Business, 1998

Professional Association: American Accounting Association

Research Interests: I am interested in empirical research that focuses on the banking and financial services industry. Issues that I am currently working on include earnings management, risk management techniques, and the disclosure of market risks

Current Working Papers: “Earnings Management by Merger Targets: Discretion over the Loan Loss Provision in Commercial Banks”. “Evidence on the Efficacy of Market Risk Disclosures By Commercial Banks” with Anwer Ahmed and Anne Beatty, currently under second review at Contemporary Accounting Research. "The Choice of Interest Rate Risk Disclosures by Bank Holding Companies: The role of Competitive Environment" “Interest Rate Risk Management of Bank Holding Companies: An Examination of Trade-offs in the Use of Investment Securities and Interest Rate Swaps” with Anwer Ahmed and Anne Beatty.

Refereed Presentations: “Hedging by Bank Holding Companies: An Examination of Trade-offs in the Use of Investment Securities and Interest Rate Swaps” with Anne Beatty, Annual Meeting of the American Accounting Association, 1997