Sei sulla pagina 1di 9

CFA Institute

Components of Earnings and the Structure of Bank Share Prices


Author(s): Mary E. Barth, William H. Beaver and Mark A. Wolfson
Source: Financial Analysts Journal, Vol. 46, No. 3 (May - Jun., 1990), pp. 53-60
Published by: CFA Institute
Stable URL: http://www.jstor.org/stable/4479330 .
Accessed: 18/06/2014 12:10
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp

.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.

CFA Institute is collaborating with JSTOR to digitize, preserve and extend access to Financial Analysts
Journal.

http://www.jstor.org

This content downloaded from 188.72.126.25 on Wed, 18 Jun 2014 12:10:30 PM


All use subject to JSTOR Terms and Conditions

by Mary E. Barth, WilliamH. Beaver and MarkA. Wolfson

Components
of Earnings and the
Structure of Bank Share
Prices
Empiricalexaminationof the relation betweencommonstock prices and two major
componentsof bankearningsshowsthatearningsbeforesecuritiesgains and lossesplayan
importantrole in explainingbankstockprices.Themarketappearsto assign a significant
multipleto this componentof earnings,judgingfromregressionresultsover the 1968-87
period.
Themarketappearsto pricerealizedsecuritiesgains and lossesdifferently.Themultiple
on securitiesgains and lossesdid not differsignificantlyfromzerooverthe periodstudied.
Theevidencealso suggests that securitiesgains and lossesbehavein a mannerconsistent
with smoothingearnings.That is, investorsapparentlyperceivethat reportedgains and
losses in banks'investmentsecuritiesare timedby bankmanagementsto offsetlossesand
gains in otherearnings.
Thesefindingsholdwhetherearningscomponents
areusedto explainpricelevelsor price
changes.Furthermore,the movementof the coefficienton earningsbeforesecuritiesgains
and lossescorrespondsto majoreconomicchangesthat might be expectedto affectbanks.
Over the 1972-74 period,for example,as financialconditionsgenerallydeclined,the
impliedmultiplefor earningsbeforesecuritiesgains and lossesalso declineddramatically.
C

ONVENTIONAL
VALUATIONMOD- nancial claims, in the form of loans and invest-

ELS often characterize stock price as a


function of earnings, linking the two
through an "earnings multiplier."But investors
may not want to apply the same multiple to
each component of earnings, because they may
not assign equal importance to every earnings
component.1 This article examines whether
stock price multiples differacross earnings components in the banking industry.2
We chose to look at the banking industry for
two reasons. First, the asset and liability structures of firms in the industry are relatively
homogeneous. The assets consist largely of fi1. Footnotes appear at the end of article..

Mary Barthis AssistantProfessorof Accountingat the


Schoolof Business,HarvardUniversity.WilliamBeaveris
theJoanE. HorngrenProfessorof Accountingat Stanford
UniversityGraduateSchoolof Business.MarkWolfsonis
theJosephMcDonaldProfessorof Accountingat Stanford
UniversityGraduateSchoolof Business.

ment securities, while liabilities consist largely


of deposits and other monetary liabilities. This
homogeneity is important,because our analysis
assumes that the multiple assigned to a given
earnings component is constant across firmsin a
given year. Second, conventional wisdom holds
that the marketassigns differentmultiples to the
components of banks' earnings. Securitiesgains
and losses, for example, are believed to be
assigned a lower multiple than earnings before
securities gains and losses. In fact, with banks,
security analysts often focus on net income
before securities gains and losses.
We examined the relation between banks'
common stock prices and their earnings to see
whether investors do indeed place different
emphases on differentcomponents of earnings.
As the dependent variablein a multiple regression analysis, we used bank stock price. We
looked at both the level of stock prices (the
marketvalue of common equity deflated by the

FINANCIAL ANALYSTS JOURNAL / MAY-JUNE 1990 O 53

This content downloaded from 188.72.126.25 on Wed, 18 Jun 2014 12:10:30 PM


All use subject to JSTOR Terms and Conditions

book value of common equity) and the percentage changes in marketvalue per share (the first
differences). For the explanatory variables, we
used two basic components of earningssecurities gains and losses (SGL)and earnings
before securities gains and losses (EB). The appendix describes the equations used. The resulting regression coefficients can be viewed as the
weights, or multiples, assigned by investors to
the earnings component.

SGL vs. EB
If the market places a greater emphasis on EB
than on SGLin valuing a firm's common equity,
then our regression analyses should yield a
lower coefficient for SGL than for EB. There are
several reasons why we might expect such a
result.
Earnings before securities gains and losses
(EB)arise primarilyfrom the deposit and lending activities of the bank. The EBcomponent of
earnings varies with the levelsof interest rates
(or interest rate spreads), which are highly correlated over time. Under historicalcost accounting, EBdoes not reflect changes in the market
value of bank assets and liabilities(except with a
time lag that may be considerable).3In particular, EB contains no explicit revaluations of the
loan portfolio or investment securities in the
form of unrealized gains and losses.4
In contrast, securities gains and losses (SGL)
essentially consist of changesin the marketvalue
of investment securities since their purchase
date. Income from investment securities (for
banks, mostly municipal, state and federal government securities) is likely to be more volatile
and transitory than other components of earnings, such as net interest income.
A very simple valuation, then, might consider
EB as consisting entirely of a permanent (i.e.,
recurring)component, carryinga multiple equal
to the reciprocalof the cost of equity capital. By
contrast, if SGL is viewed as nonrecurring, it
would have a dollar-for-dollarimpact on the
market value of common equity and carry a
coefficientof one, reflectingits one-period effect
on earnings.
In a more complex valuation, the importance
of SGL is somewhat more problematic. If all
information relevant to the valuation of securities is fully reflected in their prices, for example,
then future changes in their marketvalue would
be completely anticipated. Thus, under an appropriate amortization schedule, the expected

value of SGL would be zero. If banks possess


superior portfolio management skills, however,
SGLwould be expected to contain a permanent
component, with a regression coefficient in excess of one. In principle, the coefficientcould be
comparablein magnitude to that on EB. This is
one of the propositions we test.
The argument set forth above is strictly true
only when SGL reflects contemporaneous
changes in market value. Under historical cost
accounting rules, gains and losses are reported
in the period in which they are "realized"(i.e.,
when securitiesare sold). This impairsthe timeliness of the SGLnumber itself. Of course, with
appropriate supplemental disclosure, the bank
share price will reactwhen the change in market
value of the investment securities occurs, rather
than when the subsequent sale takes place.
Indeed, the market value of such securities is
reported parentheticallyin the annual report. If
there is a sufficient time lag between change in
value and sale, there may be no relation between bank share price (or changes in price) and
SGL in the period in which SGL is actually
reported(i.e., a zero coefficientmay be assigned
to SGL).
In other words, SGL represents realizedgains
and losses but does not include unrealizedgains
and losses. Realized gains and losses can thus
be viewed as measuring total gains and losses
(the sum of realized plus unrealized)with error.
In the simple case of a single explanatory variable that is measured with error, "white noise"
causes its coefficient to be biased toward zero.5
In the limit, the errorcould be sufficientlylarge
to lead to a coefficient that is empiricallyindistinguishable from zero.
Another factor that may affect the SGL coefficient is management's control over the timing
of securitiessales. Managementsmay, for example, time the sale of securities so that SGL will
"smooth" earnings. In that case, SGLwill tend
to be relatively high (low) when earnings from
other sources (i.e., EB)are relatively low (high).
If investors believe that SGL is being used to
smooth earnings, the coefficient for SGL will
tend to be lower-even negative. In other
words, a relatively high SGL figure would be
"bad news" because it would tend to occur
when earnings from other sources are relatively
poor.
These explanations for why the coefficienton
SGLmay differfrom that on EBare not mutually
exclusive, but they all lead to a prediction that

FINANCIAL ANALYSTS JOURNAL / MAY-JUNE 1990 O 54

This content downloaded from 188.72.126.25 on Wed, 18 Jun 2014 12:10:30 PM


All use subject to JSTOR Terms and Conditions

Table I Average Values of Variablesin Regressionsof Price Levels, 1968-1987*

Year

MarketValueof Equity
Mean
Median

EarningsBeforeSGL
Mean
Median

SecuritiesGains& Losses
Mean
Median

Obs.

1968
1969
1970
1971

1.70
1.51
1.44
1.45

1.60
1.40
1.32
1.30

.120
.128
.132
.124

.121
.126
.130
.122

-.0079
-.0081
-.0038
.0030

-.0065
-.0045
-.0013
.0027

70
76
78
87

1972
1973
1974
1975
1976
1977
1978

1.67
1.27
0.66
0.70
0.86
0.78
0.75

1.47
1.14
0.60
0.62
0.79
0.72
0.73

.124
.128
.121
.112
.108
.114
.128

.123
.126
.122
.116
.111
.117
.132

.0017
-.0021
-.0022
-.0006
.0021
.0005
-.0028

.0009
-.0012
-.0009
.0006
.0005
.0001
-.0017

87
88
95
95
95
99
119

1979
1980
1981
1982

0.72
0.74
0.82
0.81

0.67
0.69
0.75
0.77

.137
.141
.140
.135

.142
.143
.143
.138

-.0041
-.0066
-.0081
-.0059

-.0025
-.0016
-.0042
-.0023

119
122
127
137

1983

0.99

0.96

.125

.129

-.0002

.0002

138

1984

1.00

0.97

.127

.133

.0001

.0003

139

1985

1.28

1.27

.123

.132

.0071

.0044

142

1986
1987

1.26
1.08

1.25
1.03

.119
.110

.128
.116

.0173
.0065

.0075
.0032

147
147

All data are expressed as percentagesof the book value of common equity.

the coefficient assigned to SGL will be lower


than that assigned to the complementarycomponent of earnings, EB. We test this hypothesis
empirically, relative to the "null" hypothesis of
equal coefficientsfor both components. The null
hypothesis may prove correctif, in a "myopic"
market, investors do not look beyond the "bottom line" and make no attempt to adjust for
smoothing behavior; in that case, the coefficients on the two earnings components would
be the same.

Average Values
Our sample consisted of 150banks whose financial statement data appear on the 1987 Compustat Bank Tape.6 These banks are the largest
publicly traded banks in the United States.
Table I reports the average values of the variables used in the regression of stock price levels.
(MV, EBand SGL are all divided by the book
value of common equity.)
If book value perfectly reflects the value of a
corporation'snet assets, then the ratioof market
value to book value (the market-to-bookratio)
should equal one. Deviations from one can be
viewed as the market's assessment of valuation
errors in net book assets (expressed as a percentage of book value). From1968through 1973,
the average market-to-bookratio exceeded one,
indicating that the capital market was placing a
higher economic value on the banks' common

equity than the historical cost amount reported


in the financial statements.
Several factorscould lead to a market-to-book
ratio above one. (1) Banking regulation may
have on balance been favorable to the banking
industry during this period, creating an offbalance-sheet intangible asset. (2) The default
rate on the loan portfolio may have been low
relative to the provisions for loan losses set
during the late 1960s. (3) The market rates of
interest at the balance sheet dates may have
been less than the loans' contractual interest
rates. Given factors(2) and (3), the book value of
the loan portfolio would have provided a "conservative"estimate of the economic value of the
loans.
The market-to-book ratio declined dramatically from 1972 through 1974. The mean ratio
declined from 1.67 in 1972 to 0.66 in 1974 and
remainedbelow one through 1983. In 1974, over
90 per cent of the banks had market-to-book
ratios below one, and the situation had only
improved slightly by 1982. The deterioration
could be due to three majorfactors-a decline in
the value of regulation;a decline in the value of
the loans, reflecting unanticipated interest rate
increases since their issuance; and increases in
default risk not fully reflected in the loan loss
provisions.7 The mean market-to-bookratio increasedafter 1982,reflectingthe generallyfavorable stock market conditions.

FINANCIAL ANALYSTS JOURNAL / MAY-JUNE 1990 O 55

This content downloaded from 188.72.126.25 on Wed, 18 Jun 2014 12:10:30 PM


All use subject to JSTOR Terms and Conditions

Table II Average Values of Variablesin Regressionsof Price Changes, 1969-1987

Year
1969
1970

% Changein MarketValue*
Median
Mean
-.046
-.075
.040
.032

%Changein EarningsBeforeSGLt
Median
Mean
.111
.139
.107
.136

% Changein SGLt
Mean
Median
-.060
-.038
-.012
-.034

Obs.
70
76

1971

.066

.040

.025

.022

79

1972
1973

.246
-.188

.201
-.207

.078
.108

.078
.121

.015
-.017

.006
-.011

87
87

1974

-.401

-.391

.012

.072

-.021

-.009

88

1975
1976
1977

.174
.319
-.013

.143
.308
-.012

.048
.040
.174

.054
.058
.152

-.005
.020
.006

-.019
.004
.001

95
95
95

1978
1979

.055
.128

.044
.052

.286
.197

.221
.171

-.033
- .036

-.023
- .020

99
119

1980
1981
1982
1983
1984

.145
.171
.109
.336
.107

.077
.155
.124
.343
.119

.132
.103
.051
.007
.130

.120
.101
.087
.016
.113

-.066
-.060
-.044
.021
.030

-.012
-.038
-.020
.001
.003

119
122
128
138
139

1985
1986
1987

.409
.052
- .159

.408
.062
- .143

.086
.047
.004

.089
.079
.052

.064
.228
.054

.038
.063
.027

140
142
148

-.004

-.003

Change in marketprice per share expressed as a percentageof previous year's price per share.
t Expressedas a percentageof previous year's earnings(per share)before securitiesgains and losses.

Despite the dramatic decline in market-tobook ratios in 1972-74, the return on equity, as
reflected in EB/BV,remained remarkablystable
at 0.124, 0.128 and 0.121. The ratio of MV/BVto
EB/BVcan be interpreted as a crude estimate of
the average earnings multiplier the market assigns to EB. Because the average MV/BV declined, while the average EB/BVremained essentially constant, the implied multiplier
dropped dramatically from 13.5 times to 5.5
times, a reduction of approximately60 per cent.
(The regression results reported below provide
more direct estimates of the multiplier.)
While the average EB/BVwas positive in every
year, the average SGL/BV fluctuated around
zero. The median SGL was positive in 10 years
and negative in 10 years. This behavior is consistent with the earlier contention that the expected value of SGL is zero.8
Table II reports the average values of the
variables used in the regressions of percentage
changes in market value. The variables are the
percentage change in price per share, the
change in EBper share expressed as a percentage of the previous year's EB and the level of
SGL per share expressed as a percentage of the
previous year's EB.
Not surprisingly, the annual percentage
changes in common stock were quite volatile,
ranging from a low of -0.401 in 1974to a high of
0.409 in 1985. Percentage changes in EB, which

can be interpretedas growth in EB, ranged from


a low of -0.004 in 1971 to a high of 0.286 in
1978. SGLwas positive in nine years and negative in 10 years, similarto the results reportedin
Table I.

Evidence of Smoothing
Smoothingbehaviorcan be examined by comparing changes in EBwith changes in SGL over
time. The median value of changes in EBwas
0.086, while the median value of changes in SGL
was -0.005. The smoothing argument implies
that when changes in EB are relatively low
(high), SGLcan be expected to be relativelyhigh
(low). Excludingthe years in which SGLand EB
had median values (1975 and 1985, respectively), in 13 of the 17 years when the changes in EB
were below (above) the median, SGLwas above
(below) its median. The null hypothesis (i.e.,
changes in EBand SGLare independent) can be
rejected at the 2 per cent level of significance.9
This behavior accords with the smoothing hypothesis.

Levels of Prices and Earnings


One widely used approach to examining the
price-earningsrelationshipis to regress the level
of share prices on the level of earnings. This
regression assesses the extent to which differences in market values across firms can be
explained by differences in the firms' earnings.

FINANCIAL ANALYSTS JOURNAL / MAY-JUNE 1990 O 56

This content downloaded from 188.72.126.25 on Wed, 18 Jun 2014 12:10:30 PM


All use subject to JSTOR Terms and Conditions

Table III RegressionResults:MarketValue as a Functionof Two EarningsComponents, 1968-1987*


Year
1968
1969
1970
1971
1972
1973
1974
1975
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987

13
17.63
12.65
13.79
15.96
21.32
12.59
4.92
5.58
6.39
5.92
3.80
4.00
4.73
5.53
3.74
3.28
5.25
6.06
6.68
5.54

tI
8.4
5.1
6.3
8.1
8.3
4.9
6.1
7.6
7.8
12.7
6.2
6.2
7.5
8.3
7.5
5.4
12.0
8.9
10.7
8.7

2
-1.85
-0.62
-1.14
4.42
-10.49
8.69
-1.55
-1.77
-5.22
2.09
3.49
3.35
1.48
-1.31
-1.99
-1.22
-3.44
1.53
4.29
-1.43

t2

-0.5
-0.2
-0.2
0.7
-0.8
0.8
-0.3
-0.3
-1.0
0.6
0.8
1.0
0.7
-0.8
-1.5
-0.4
-1.0
0.6
1.8
-0.4

R2
.51
.26
.34
.42
.45
.21
.78
.40
.43
.64
.24
.24
.35
.35
.36
.22
.57
.39
.47
.42

Obs.
69
75
78
87
87
87
94
93
92
98
119
119
121
125
135
133
134
137
130
110

+ 132(SGLit/BVit)
+ uit. t1 and t2 are t-valuesassociatedwith f31and R2,respectively.
MVit/BVit= at + 01t(EBit/BVit)

The market value of common equity (typically


scaled by some size variable) is used as the
dependent variable and the earnings available
for common (scaled by the same size variable)as
an explanatory variable.10
While additional explanatory variables, such
as risk and growth, may be included, previous
research has shown that estimates of the earnings coefficients are largely insensitive to the
inclusion of these additional variables.11Moreover, the contribution of earnings to variations
in market value is typically much larger than
that provided by other variables. For these reasons, we examined the relation using only the
earnings variable. We chose as the deflator a
measure prominently used both in prior research and in bank analysts' reports-the book
value of common equity.
Table III reports some summary statistics
from the cross-sectionalregressions. The coefficient on EB (,81) was positive and statistically
significant in all 20 years.12Moreover, the general movement of the coefficient was consistent
with the major economic factors that affected
banks during the period.
In 1972-74, for example, nominal interest
rates rose substantially. This rise would be expected to have an adverse effect on an industry
that takes long positions in financialassets, the
market values of which move inversely -with
unexpected changes in market interest rates.
There was also substantial weakness in the

economy, as reflected in a decline of 45 to 50 per


cent in most of the majorstock price indexes, as
well as a dramaticincrease in the default risk of
bank loans. Furthermore,the economic value of
regulation was apparently declining (possibly
turning negative), as interest rate ceilings on deposits and disintermediationaffected banks adversely.
During this same period, the coefficienton EB
declined from 21.32 in 1972 to 4.92 in 1974. In
other words, the multiple applied to earnings
before securities gains and losses (EB)experienced about a 75 per cent reduction. This decline is even sharper than the 60 per cent
reduction evident in Table I. The coefficienthas
remained at this lower level since 1974.
In contrast to the behavior of the coefficient
on EB, the coefficient on SGL (132)did not have
the same sign over time. Its sign was positive in
eight years and negative in 12 years. Moreover,
the coefficientwas not statisticallysignificantin
any of the 20 years.13In fact, the average coefficientvalue over the full 20 years was 0.01; even
a joint test over the 20 years would not come
close to rejecting a null hypothesis of zero for
the coefficientvalue. Finally, /32 was less than ,31
in all 20 years.14
Price Changes and Earnings Components
Time-series dependency can be a problem in
examining levels of prices and earnings. One
way to address this problem is to examine the

FINANCIALANALYSTSJOURNAL/ MAY-JUNE1990O 57

This content downloaded from 188.72.126.25 on Wed, 18 Jun 2014 12:10:30 PM


All use subject to JSTOR Terms and Conditions

Table IV RegressionResults:PercentageChange in MarketPriceas a Functionof Two EarningsComponents, 1969-1987*


Year
1969
1970
1971
1972
1973
1974
1975
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
* APit/Pit-=

t2

1.21
0.26
0.58
1.10
0.54
0.43
0.20
0.40
0.22
0.21
0.27
0.68
0.52
0.57
0.29
0.17
0.28
0.49
0.07

19.5
4.5
6.0
6.1
4.7
6.0
4.0
3.7
3.3
4.0
5.7
7.4
5.8
7.9
4.0
3.1
3.9
7.1
2.0

-0.27
-0.21
0.20
-1.04
0.49
-0.19
-0.90
-0.66
0.16
0.21
-0.00
-0.37
-0.12
-0.17
-0.83
-0.26
-0.18
0.15
-0.36

t2

R2

-1.5
-1.2
0.7
-1.8
1.7
-0.4
-0.2
-2.0
0.5
0.7
-0.0
-1.3
-0.7
-0.9
-2.0
-0.7
-0.8
1.2
-1.8

.85
.21
.31
.33
.20
.29
.18
.15
.09
.13
.22
.38
.22
.35
.16
.09
.12
.29
.04

Obs.
69
74
78
87
87
87
93
91
97
97
118
118
118
122
132
128
i29
121
102

at + 31t(AEBit/EBit-1) + P2t(SGLjt/EBit-1)+ uit. t1 and t2 are t-valuesassociatedwith PI and /2, respectively.

relation between prices and earnings components- in first-difference form; percentage


changes in price are approximatelyuncorrelated
over time. We thus ran regressions using percentage changes in stock prices as the dependent variableand the two earnings components
as the explanatory variables.
Prior research has shown that price changes
are related to the unexpected
portion of earnings.
Thus a further reason for separating earnings
into more than one component is that the appropriate earnings expectations model may
differ between unexpected and expected earnings. In particular,the nature of EB is such that
a "randomwalk with a drift"model is probably
a good approximation to capture the market's
expectationof earnings from year to year, where
the drift is implicitly assumed to be a constant
percentage across banks in a given year. The
percentage change in EB will then represent the
unexpected component of EB. Given the evidence presented in Tables I and II, we assumed
that SGL has an expectation of zero.15
Table IV reports the coefficients and tstatisticsobtained from a model with percentage
change in price as the dependent variable and
percentage changes in EBand SGL (deflated by
the lagged value of total earnings) as the explanatory variables. The coefficient on EB (I1) was
positive and statistically significant in all 19
years. The average earnings coefficient for the
banks-0.40-is similar to the average coefficient of 0.37 found for nonfinancial firms.16

The coefficients on SGL (/2) behaved very

differently. The coefficient was positive in five


years and negative in 14 years, and statistically
significantin only two of the 19 years. g2 did not
exceed f31in any of the years. Assuming timeseries independence of the regression estimates,
the null hypothesis that /P2 equals ,31can be
rejectedat extremely low levels.17The probability that the results could have occurred by
chance is less than one in 500,000.
These findings are consistent with the earlier
contention that EBmay be expected to contain a
larger permanent component than SGL. The
predominantlynegative (although insignificant)
coefficient for SGL is consistent with a market
that perceives that securities gains and losses
are used to smooth earnings. Insofar as their
marginal impact on share prices is concerned,
greater than expected realizedsecurities gains
are "bad news" rather than "good news." U
Footnotes
1. Consistent with this view, the FinancialAccounting Standards Board's Statementof FinancialAccountingConceptsNo. 5 (1984) states, "The individual items, subtotals, or other parts of a
financial statement may often be more useful
than the aggregate to those who make investment, credit, and similar decisions."
2. R. Bowen ("Valuationof Earnings Components
in the ElectricUtility Industry," AccountingReview, January1981) and R. Lipe ("The Information Containedin the Components of Earnings,"
Journalof AccountingResearch,Supplement 1987)

FINANCIAL ANALYSTS JOURNAL / MAY-JUNE 1990 0 58

This content downloaded from 188.72.126.25 on Wed, 18 Jun 2014 12:10:30 PM


All use subject to JSTOR Terms and Conditions

3.

4.

5.

6.

which we are likely to have more observations


have investigated aspects of this issue in a nonwith positive transitory components in EB than
banking context.
negative transitorycomponents. This should exA minor exception is gains and losses on securiert a downward bias on the estimate of the
ties held for trading purposes. Since 1982, these
coefficientfor EB.
gains and losses are based on total gains (i.e.,
realized plus unrealized gains). Prior to 1982, 7. The decline in the market-to-bookratiocould also
be due to increases in book value. However,
gains were based exclusively on realized gains.
increases in book values were relatively small
Because securities held for trading purposes are
during this three-yearperiod.
typically held on a short-termbasis, this change
In a given year, however, the expected value
8.
have
had
a
to
in accounting method is unlikely
might be different from zero if gains and losses
significant impact on the reported gains and
are
timed strategically(e.g., to smooth earnings).
losses for trading securities.
While
totalsecurities gains and losses (i.e., realSuch revaluations may be implicitly reflected in
ized
plus
unrealized) are expected to be serially
the current yields on those assets. For example,
realized SGL appear to be posiuncorrelated,
when interest rates increase unexpectedly, a
In successive years, the median
correlated.
tively
fixed-rate asset financed by variable-rateliabiliSGL are of the same sign in 14 cases, but are of
ties will result in a lower interestrate spread. This
opposite sign in only five cases. Using a binomial
spread change persists for a durationequal to the
test, serial independence can be rejected at the 3
remaining term to maturity of the fixed-rateasper cent level of significance.
set. Note, however, that the change in the inter9. The correlationbetween the median AEB/EBand
est spread can be "captured"through the sale of
the median SGL/EBis -0.40, which is significant
the fixed-rate assets, leading to a realized loss.
at the 5 per cent level of significance(assuming a
Such gains and losses are not expected to persist
one-tailed test). The cross-sectional correlation
in the future. Collectionexperiencedifferentfrom
between these two variables is negative in 16
that expected at the time loans are originatedcan
years and positive in only three years. Based on a
have similar effects on interest spreads.
binomial test, the null hypothesis of zero correThe actualvalue of the coefficientswould depend
lation can be rejectedat a significancelevel of less
upon the nature of measurement error in SGL
than 1 per cent. The negative correlationis all the
and EB, as well as on the linear relationbetween
more impressive when one considers that in the
the two earnings components. EB also reflects
absence of strategic management of SGL, we
some measurement error because of the use of
might expect a positive dependency. The reason
historical cost, among other factors. In a regresis that, if banks hold more longer-duration,fixedsion that contains two explanatoryvariablesmearate assets than liabilities, EB should move insured with error, it is generallyambiguouswhat
versely with interest rates, as should the market
the sign of bias might be. However, given the
value (hence the implied gains and losses) of
natureof the linearrelationbetween the two earninvestment securities.
ings components, if both SGL and EB include 10. The problems induced by the use of the same
"white noise" measurementerror,it can be demdeflatorfor both the dependent and independent
onstratedthateach coefficientwill be biasedtoward
variables are discussed in W. Beaver and W.
zero and that an increasein measurementerrorin
Landsman, "IncrementalInformationContent of
Statement 33 Disclosures: Research Report" (FiSGL will have a more pronounced effect on the
nancial Accounting StandardsBoard, November
SGL coefficient than on the EB coefficient and
1983).
converselyfor an increasein measurementerrorin
EB. If EB is measured with sufficient error, its 11. Ibid.
estimatedcoefficientcould be empiricallyindistin- 12. The t-statistics should be viewed largely as descriptive. Assuming the regression residuals are
guishable from that on SGL, even if the null hyindependently and normally distributed, the tpothesis of equal coefficientsis false.
statistics can be used to infer levels of signifiObservations in a given year were eliminated if
cance. A conventional level of significance for
EB was negative that year, because the pricerejectingthe null hypothesis is 5 per cent. Under
earnings relation is not well defined when earna one-tailed test, t-values exceeding 1.65 to 1.67
ings are negative. With respect to the percentage
(depending upon the numberof observationsin a
change in EB, observations in a given year were
given year) representstatisticallysignificantcoefdeleted if EBwas negative in the previous year or
ficients. Under a two-tailed test, the criticalvalif the percentage change in earnings was greater
ues of the t-statistic are 1.98 to 2.00. Because a
than 300 per cent. Such deletion rules are similar
positive coefficientis expected for EB, a one-tailed
to those used in prior research. Note that the
test is appropriatehere, while a two-tailed test is
elimination of observations for which EB was
appropriate for SGL because the sign of the
negative means that we are left with a sample in
FINANCIAL ANALYSTS JOURNAL / MAY-JUNE 1990

59

This content downloaded from 188.72.126.25 on Wed, 18 Jun 2014 12:10:30 PM


All use subject to JSTOR Terms and Conditions

coefficient would be predicted to be either positive or negative, depending upon how the market views SGL.
13. Concernhas been expressed about positive crosssectional dependency in the regression residuals
when the sample contains industry clustering,as
we have here. While cross-sectional dependencies are mitigated by permitting the intercept
term to vary year-by-year,any remainingpositive
dependence would tend to understate the standard errors and would bias the significancetests
in favor of rejectingthe null hypothesis. In spite
of this potential bias, the null hypothesis that 132
equals zero cannot be rejected at conventional
levels of significance. Departures from independence would affect our interpretationof the tstatistics reported for 131as well, although the
t-values for 31 average 7.8, or nearly five times
that required to achieve conventional significance, when the standard conditions are met.
Note that departures from independence do not
bias the coefficientestimates for ,1 and I32.
14. The ,I1and 132estimates are likely to be correlated
from year to year. This limits our ability to
conduct significance tests on the times-series
differences in 31 and f82. A score of 20 out of 20
would nevertheless appear to be impressive, notwithstanding the time-series dependence.
15. The assumption of a zero mean is not conditioned upon firm-specificor year-specific information other than AEB.The regressionscondition
the expected value of the SGLvariableupon AEB
and hence at least partiallyreflect the potentially
strategic nature of SGL. However, the serial
dependency discussed in footnote 8 would suggest that an alternativespecificationof the unexpected SGL might include prior years' SGL.
16. W. Beaver, R. Lambert and D. Morse, "The
Information Content of Security Prices," journal
of Accountingand Economics,March 1980.
17. Another approachto the time-seriesdependency
in the data is to estimate the regression model
coefficients using a technique known as "seemingly unrelated regressions." This technique incorporatesthe correlationbetween the regression
residuals in differentyears' regressions in a way
that provides more efficient coefficientestimates
and unbiased tests of the significance of those
coefficients. When the coefficients were reestimated using this technique, the results were
essentially the same as those reported here (i.e.,
I31is positive, 12 fluctuatesaround zero, and 12 iS
less than 131).

Appendix
To examine the contributions of different earnings components to bank stock prices, we used
the following equations:
MVit/BVit

= at + 1lt(EBit/BVit)

+ uit
+ I32t(SGLit/BVit)

(Al)

and
APit/Pit-

1 = at + I3t(zEBit/Nlit-

+ 0 2t(SGLit/Mit-1) + uit,

1)

(A2)

where
MVit = the market value of common equity
for firm i at the end of year t,
=
BVit the book value of common equity for
firm i at the end of year t,
APit

Pit

-Pit-,

Pit = common stock price for firm i at the


end of year t,
=
SGLit securities gains and losses for firm i
in year t,
=
earnings before securities gains and
EBit
losses for firm i in year t,
AEBit = EBit - EBit-1 and
NIit-l = EBit-1 + SGLit-1.

Glossary
Null Hypothesis: A statement that a certain relationship holds. In this study, the null hypothesis is that
a particularcoefficient is equal to zero. Statistical
tests are conducted to determine the probability
that statement is true, given the observed data. If
the probability(called the level of significance) is
sufficientlysmall, the null hypothesis is rejectedas
false.
Myopic Market: A market where prices are established based only on a portion or subset of all the
relevant informationavailable. An example would
be a market that focuses only on net income and
ignores other relevant information.
Random Walk with a Drift: A process that describes
how a variablechanges over time. For example, if
earnings move as a random walk with a drift, next
period's earnings are expected to be equal to the
most recently reported earnings plus a growth
factor, called a drift.

FINANCIAL ANALYSTS JOURNAL / MAY-JUNE 1990 O 60

This content downloaded from 188.72.126.25 on Wed, 18 Jun 2014 12:10:30 PM


All use subject to JSTOR Terms and Conditions

Potrebbero piacerti anche