Do Bank-affiliated Analysts Benefit from Lending Relationships?
By Xiumin Martin
Abstract
This paper investigates whether private information from lending activities improves the
forecast accuracy of bank-affiliated analysts. Using a matched sample design, matching
by affiliated bank or borrower, we demonstrate that the forecast accuracy of bankaffiliated analysts increases after the followed firm borrows from the affiliated bank. We
also find that the increase in forecast accuracy is more pronounced for borrowers with
greater information asymmetry and bad news, and for deals with financial covenants.
Last, we find that the informational advantage of bank-affiliated analysts exists only
when the affiliated banks serve as lead arrangers, not merely as participating lenders.
Overall our evidence suggests that information flows from commercial banking to equity
research divisions within financial conglomerates.
JEL: G14, G21, G24, G28
Key Words: Bank-affiliated analyst, conglomerate forecast, information sharing
“Do Bank Affiliated Analysts Benefit from Lending Relationships?”
By Xiumin Martin
Amid the financial crisis that started in 2007, large US investment banks such as
Bear Sterns and Lehman Brothers have completely disappeared from the banking scene.
The universal banking model, which allows financial conglomerates to combine a wide
range of financial activities has emerged over 1990s particularly after the Gramm-LeachBlieley Act of 1999 that formally repealed the Glass-Steagall Act of 1933. Such system is
argued to be a more desirable structure for a financial institution from viewpoint of policy
makers due to its resilience to adverse shocks. This drastic change in the landscaping of
the financial industry has many implications for corporations in the U.S. In this paper,
we focus on the informational impact.
Traditionally banks play a role of financial intermediary, who collect money from
depositors and lend to other businesses. In doing so, banks as lenders have unique
information advantage and incentives to monitor borrowers (Fama, 1985; Chemmanur
and Fulghieri, 1994). For example, borrowers usually have a much closer relationship
with their banks than with investors in their public securities such as stocks and bonds. In
particular, borrowers often provide material and price-sensitive information such as
revenue projection updates or acquisition and divestiture plans, to their banks well in
advance of release to the public. In the absence of perfect “Chinese Wall” that separates
the public from the private domain within a financial conglomerate, the private
information possessed by loan officers about a borrower can migrate to the public domain
-the equity analysts along with the public trading and sales. Consequently, security
analysts can incorporate this information into their earnings forecasts and stock
recommendations, which is eventually transmitted to the market before the borrower
makes any public announcement.
Based on syndicated loans obtained from Dealscan database, analyst forecasts
from First Call and companies’ financial information from Compustat, four key findings
are demonstrated in this paper. First, as shown in Table 1 the accuracy of earnings
forecasts from a bank-affiliated analyst for a borrower increases after the loan origination
compared to the forecasts made by the same analyst for non-borrowing firms and
compared to the forecasts made by non bank-affiliated analysts for the same borrower.
Relative to the benchmark forecasts, bank-affiliated analysts reduce annual EPS forecast
error by seven cents, which is about one sixth of the average EPS forecast error in our
sample. Second, the results presented in Table 2 indicate that the increase in forecast
accuracy of bank-affiliated analysts is more pronounced for borrowers with greater
information asymmetry as measured by size and the standard deviation of analyst
forecasts. For example, these borrowers are usually smaller in size and the forecasts of
their security analysts are more diverged. Third, the increase in forecast accuracy of
bank-affiliated analysts concentrates in instances when borrowers experience bad news,
when borrowers have high credit risk such as lower credit ratings or no credit ratings and
higher leverage ratio, and when loans contain financial covenants. Fourth, we find an
informational advantage for conglomerate analysts only when conglomerates serve as
lead arrangers but not as participating lenders. Taken together, our results provide a
consistent picture that there is information spillover from the commercial lending
division to the equity research division of a financial conglomerate and bank affiliated
analysts benefit from this information spillover via more accurate forecasts.
Although information sharing is beneficial to financial conglomerates, it is not
without controversy, particularly when much of the superior information comes from
ongoing correspondence between borrowers and banks. In recent years, regulators and
market participants have expressed concerns that the spillover of private information into
the public domain might breach confidentiality agreements between lenders and issuers
and, more importantly, could lead to illegal trading (Standard & Poor’s, 2008). Banks
have tried to address this concern by establishing limits to the flow of information among
different parts of a financial conglomerate: i.e., erecting Chinese Walls. Analysts, along
with public trading and sales desks that they are associated with, belong to the public side
of the wall and are therefore not supposed to receive private information. Our findings
suggest that despite the presumed existence of Chinese Walls, financial analysts still have
access to superior information from lending relationships and take advantage of this
access in improving their forecast accuracy. As a consequence, information spillover
among different divisions within a financial conglomerate is likely to be of greater
concerns.
References
FAMA, E.. “What’s Different about Banks?” Journal of Monetary Economics 15 (1985):
29-39.
CHEMMANUR, T, AND P. FULGHIERI. “Investment Bank Reputation, Information
Pro- duction and Financial Intermediation.” Journal of Finance 49 (1994): 57-79.
Appendix: Variable Definition
Dependent variables:
=
analyst forecast error, calculated as |annual EPS forecast – actual
EPS forecast|/price, where price is firm stock price at the beginning
of forecast month.
CONGLOMERATE
=
1 if an EPS forecast is issued by a bank-affiliated analyst,
and 0 otherwise.
POST
=
1 if an EPS forecast is issued after loan initiation, and 0
otherwise.
MKT
=
market value of equity (Compustat #199 * #25) measured at the
end of the fiscal year prior to an EPS forecast.
SMALLFI RM
=
dummy variable coded as 1 if LOGMKT is below the sample
median, and 0 otherwise .
ASSET
=
total assets (Compustat #6).
MB
=
market value of equity divided by book value of equity (Compustat
#60) measured at the end of the fiscal year prior to an EPS
forecast.
PLOSS
=
EPSUP
=
the probability of a firm experiencing loss in the fiscal year prior to
an EPS forecast is issued.
1 if a firm experiences an increase in EPS compared with previous
fiscal year, and 0 otherwise measured at the end of the fiscal year
prior to an EPS forecast.
NUMANALYST
=
average number of analysts following a firm over the course of a
fiscal year measured during the fiscal year prior to an EPS forecast.
STD DEV
=
the average of standard deviation of analyst annual EPS forecast
over the course of a fiscal year measured during the fiscal year
prior to an EPS forecast.
HORIZON
=
number of days between the time when an EPS forecast is issued
and the actual EPS announcement date.
EXPERIENCE
=
number of months between the time when an analyst started to
follow a firm and the time when a specific forecast is issued in
year t.
ERROR
Independent variables:
Table 1 The Association between Conglomerate Forecast Accuracy and Loan Initiation
ERROR = β0 +β1POST+β2CONGLOMERATE*POST
+ γCONTROLS +ε
Variable
INTERCEPT
POST
CONGLOMERATE * POST
LOGMKT
MB
PLOSS
EPSUP
LOGNUMANALYST
HORIZON
LOGEXP
(LOGEXP)2
Firm fixed effect
Broker fixed effect
Year fixed effect
N
Adj - R2 (%)
Predicted
Broker Constant Sample
Sign
Coeff. Estimate (1) p-value (2)
?
-0.022
<.001
?
0.000
0.560
-0.002
0.012
-0.006
<.001
+
0.000
0.047
+
0.065
<.001
-0.004
<.001
?
0.004
<.001
+
0.010
<.001
0.002
<.001
+
0.000
0.007
Firm Constant Sample
Coeff. Estimate (3) p-value (4)
0.014
<.001
0.000
0.619
-0.002
0.017
-0.004
<.001
-0.001
<.001
0.041
<.001
-0.006
<.001
0.003
<.001
0.006
<.001
-0.001
0.054
0.000
<.001
Yes
Yes
23872
Yes
Yes
20668
53.83
19.75
Table 1 presents results of testing the relation between conglomerate forecast accuracy
and loan initiation based on the broker-constant sample and the firm-constant sample,
respectively. In the broker-constant sample, a loan deal for the borrower is assigned for
their matching non-borrowers hypothetically, which assume the same loan initiation date.
Firm and year fixed effects are included in the model for the broker-constant sample and
broker and year fixed effects are included in the model for the firm-constant sample.
Robust standard errors clustered at the firm level are used to derive p values. Variable
definition is provided in Appendix.
Table 2 The Association of Conglomerate Forecast Accuracy with Loan Initiation and Borrower Information Asymmetry
Panel A:
ERROR = β0 +β1POST+β2CONGLOMERATE *POST+ γCONTROLS +ε
Broker Constant Sample
Firm Constant Sample
Predixted
SMALL FIRM=1
SMALL FIRM=0
SMALL FIRM=1
SMALL FIRM=0
Variable
Sign Coeff. Est. (1) p-value (2) Coeff. Est. (3) p-value (4) Coeff. Est. (5) p-value (6) Coeff. Est. (7) p-value (8)
INTERCEPT
?
-0.023
0.008
-0.007
0.379
0.047
<.001
0.015
<.001
POST
?
0.002
0.002
-0.001
0.002
0.002
0.005
-0.001
0.023
CONGLOMERATE * POST
-0.002
0.030
-0.001
0.425
-0.003
0.039
0.000
0.949
LOGMKT
-0.011
<.001
-0.005
<.001
-0.010
<.001
-0.004
<.001
MB
+
0.001
<.001
0.000
0.080
-0.002
<.001
0.000
<.001
PLOSS
+
0.049
<.001
0.043
0.001
0.038
<.001
-0.002
0.767
EPSUP
-0.006
<.001
-0.002
<.001
-0.009
<.001
-0.003
<.001
LOGNUMANALYST
0.004
0.013
0.002
0.049
0.003
0.009
0.003
<.001
HORIZON
+
0.014
<.001
0.007
<.001
0.009
<.001
0.004
<.001
LOGEXP
0.002
0.029
-0.001
0.134
-0.001
0.202
-0.002
0.006
(LOGEXP)2
Firm fixed effect
Broker fixed effect
Year fixed effect
N
Adj - R2 (%)
+
0.000
0.024
0.000
0.419
0.000
0.047
0.000
<.001
Yes
Yes
Yes
11928
Yes
11944
Yes
Yes
10271
Yes
Yes
10397
58.95
42.80
26.07
21.31
Table 2 (To be continued)
Panel B:
ERROR = β0 +β1POST+β2CONGLOMERATE *POST+ γCONTROLS +ε
Broker Constant Sample
Firm Constant Sample
Predixted
HIGH STDDEV=1
HIGH STDDEV=0
HIGH STDDEV=1
HIGH STDDEV=0
Variable
Sign Coeff. Est. (1) p-value (2) Coeff. Est. (3) p-value (4) Coeff. Est. (5) p-value (6) Coeff. Est. (7) p-value (8)
INTERCEPT
?
-0.034
0.003
-0.002
0.736
0.037
<.001
0.004
0.185
POST
?
0.000
0.706
0.000
0.652
0.002
0.072
0.000
0.694
CONGLOMERATE * POST
-0.005
<.001
0.001
0.202
-0.005
0.007
-0.001
0.266
LOGMKT
-0.008
<.001
-0.004
<.001
-0.007
<.001
-0.001
<.001
MB
+
0.000
0.967
0.000
0.039
-0.001
<.001
0.000
0.445
PLOSS
+
0.101
<.001
0.015
0.200
0.044
<.001
0.012
0.001
EPSUP
-0.006
<.001
-0.002
<.001
-0.010
<.001
-0.001
<.001
LOGNUMANALYST
0.005
0.013
0.002
0.173
0.004
<.001
0.000
0.981
HORIZON
+
0.015
<.001
0.005
<.001
0.010
<.001
0.004
<.001
LOGEXP
0.004
0.010
0.000
0.303
-0.002
0.355
0.000
0.311
(LOGEXP)2
Firm fixed effect
Broker fixed effect
Year fixed effect
N
Adj - R2 (%)
+
0.000
0.029
0.000
0.507
0.000
0.084
0.000
0.876
Yes
Yes
Yes
10237
Yes
10240
Yes
Yes
7911
Yes
Yes
9783
56.29
45.09
27.63
16.20
Table 2 presents results of testing the relation between the change in conglomerate analyst forecast accuracy after loan initiation and
borrower information asymmetry as measured by based on the broker-constant sample and the firm-constant sample, respectively.
SMALLFIRM is a dummy variable equal to 1 if total market value of equity measured at the fiscal year prior to loan initiation
(LOGMKT) is below the sample median, and equal to 0 otherwise; HIGH STDDEV is a dummy variable equal to 1 if the standard
deviation of analyst annual earnings forecasts made in the fiscal year prior to loan initiation (STD DEV) is above the sample median,
and equal to 0 otherwise. In the broker-constant sample, a loan deal for the borrower is assigned for their matching non-borrowers
hypothetically, which assume the same loan initiation date. Firm and year fixed effects are included in the model for the brokerconstant sample and broker and year fixed effects are included in the model for the firm-constant sample. Robust standard errors
clustered at the firm level are used to derive p values. Variable definition is provided in Appendix.
Download full paper.