ArticlePDF Available

Estimating the Effect of Hierarchies on Information Use

Authors:

Abstract and Figures

Theory suggests that greater hierarchical distance between a subordinate and his boss makes it more difficult to share abstract and subjective information in decision making. A novel dataset put together from credit dossiers of large corporate loan applicants enables us to observe the information collected by loan officers, and how it is used by the ultimate loan approving officer. We find that greater hierarchical/geographical distance between the information collecting agent and the loan approving officer leads to less reliance on subjective information and more on objective information. By exploiting nonlinearities in the “assignment rules” that determine an applicant's hierarchical distance, and using information collecting agent fixed effects, we show that our result cannot be driven by endogenous assignment of applicants. We also find that higher frequency of interactions between the information collecting agent and loan approving officer, both over time and through geographical proximity, helps mitigate the effects of hierarchical distance on information use. Our results show that hierarchical distance influences information use, and highlights the importance of “human touch” in communication.
Content may be subject to copyright.
OXFORD UNIVERSITY PRESS LTD JOURNAL 00 (2008), 1–26
doi:10.1093/OUP Journal/XXX000
Estimating the Effect of Hierarchies on Information Use
Jose M. Libertiaand Atif R. Mianb
aKellstadt Graduate School of Business, DePaul University. Email: jliberti@depaul.edu
bGraduate School of Business, University of Chicago and NBER. Email: atif@chicagogsb.edu
Abstract: Theory suggests that greater hierarchical distance between a subordinate and his boss makes it more
difficult to share abstract and subjective information in decision making. A novel data set put together from credit
dossiers of large corporate loan applicants enables us to observe the information collected by loan officers and also how
it is used by the ultimate loan approving officer. We find that greater hierarchical / geographical distance between the
information collecting agent and the loan approving officer leads to less reliance on subjective information and more on
objective information. By exploiting non-linearities in the “assignment rules” that determine an applicant’s hierarchical
distance, and using information collecting agent fixed effects, we show that our result cannot be driven by endogenous
assignment of applicants. We also find that higher frequency of interactions between the information collecting agent
and loan approving officer, both over time and through geographical proximity, helps mitigate the effects of hierarchical
distance on information use. Our results show that hierarchical distance influences information use, and highlights the
importance of “human touch” in communication.
Copyright c
°2008 Oxford University Press
Prepared using oupau.cls [Version: 2007/02/05 v1.00]
2 J.M. LIBERTI AND A.R. MIAN
Acknowledgement
We thank George Baker, Francesca Cornelli, Wouter Dessein, Luis Garicano, Bob Gibbons, Andrew Hertzberg,
Steven Ongena, Francisco P´erez Gonz´alez, Mitchell Petersen, Patricia Pierotti, Raghuram Rajan, Jeremy
Stein, Tano Santos, Steve Tadelis, Eric Van den Steen, and Luigi Zingales, as well as seminar participants
at the 2006 American Finance Association Meetings, Board of Governors of the Federal Reserve System,
Columbia Business School, DePaul University (Kellstadt), Federal Reserve Bank (Chicago), Harvard-MIT
Organizational Economic Seminar, London Business School, Loyola University Chicago, NBER Corporate
Finance Summer Institute (2006), Northwestern University (Kellogg), Purdue University (Krannert), Tilburg
University (CentER), University of California, Berkeley (Haas), Dartmouth (Tuck), University of Chicago GSB,
University of Illinois (Chicago), University of Illinois (Urbana-Champaign), University of Wisconsin (Madison),
Ohio State University (Fisher), and World Bank for helpful comments and suggestions. The superb research
assistance of Paula Canavese, Stuart Currey, and Michael Niessner is greatly appreciated. A special thanks to
all members of the financial institution who kindly provided their time and effort in answering questions that
enabled assembling the database. All errors are our own.
ESTIMATING THE EFFECT OF HIERARCHIES ON INFORMATION USE 3
Do hierarchies effect the sharing of information within a firm? The evolution of firms from family businesses
to large hierarchical organizations sparked an extensive debate on how the design of hierarchies impacts
information sharing and allocation of tasks [Radner (1993); Bolton and Dewatripont (1994); Aghion and Tirole
(1997); Garicano (2000); Stein (2002); Dewatripont and Tirole (2005); and Garicano and Rossi-Hansberg (2006)].
A central idea in this literature argues that information sharing —particularly when information is subjective
and more nuanced in nature— becomes harder in more hierarchical production processes. However, despite
the practical relevance of understanding how hierarchical design impacts information flows, empirical testing
remains elusive.
There are several difficulties in empirically testing informational theories relating to hierarchical design.
First, theoretical constructs such as “subjective information,” “authority,” and “hierarchical design” often lack
a formal empirical counterpart and are thus difficult to measure. Second, theories of organizational design are
often based on intra-firm dynamics, such as the sharing of information between employees within a firm. Such
intra-firm transactions are seldom recorded in a form that can be analyzed. Third, even if the necessary data
were available, there remains the challenge of finding plausibly exogenous variation in hierarchical design. For
example, differences in how information is used across various hierarchical designs may be primarily driven by
omitted factors that happen to be correlated with hierarchical design.
This paper addresses these empirical hurdles to estimate the effect of hierarchical distance, i.e. distance
between a decision-making officer and his subordinate who collects information, on the type of information
used in the decision-making process. A number of theories suggest that private non-verifiable information,
which we classify as “subjective information,” is difficult to use across organizational layers.1The precise
channels vary from ex-ante incentives for information collection [Aghion and Tirole (1997); and Stein (2002)],
to strategic manipulation of information [Crawford and Sobel (1982)] and ex-post communication costs [Sah
and Stiglitz (1986); Radner (1993); and Bolton and Dewatripont (1994)]. However, regardless of the underlying
channels, these papers share the common hypothesis that hierarchical distance makes it harder to use subjective
information, hence increasing reliance on objective information.
We test the above prediction using data from the loan approval process of a large multi-national bank. The
data is constructed using credit dossiers of large corporate loan applicants, and offers a natural environment
for testing the impact of hierarchical distance on information use. First, the loan approval process is heavily
dependent on information regarding an applicant’s quality and future outlook. Second, the credit dossiers
enable us to observe the flow of information (both objective and subjective) between the loan officer collecting
information on an applicant, and the ultimate credit approval officer. Third, there is significant variation across
applicants in the hierarchical distance travelled by a loan application. For example, some loan applications
are approved by credit approval officers who sit very close (in terms of hierarchical distance) to the loan
officers who collect information. While others are approved by credit approval officers who sit further away
from information collecting officers in the hierarchy. An important feature of the hierarchical distance travelled
4 J.M. LIBERTI AND A.R. MIAN
by a loan application is that it is determined ex ante by “bank rules.” This can be exploited to generate plausibly
exogenous variation in hierarchical distance across applicants.
Our empirical methodology in this paper can be better understood from the following representative
example. Consider a loan officer who receives a firm’s loan application, and then collects a variety of information
about the firm. For simplicity, summarize this information into a subjective signal (S) that represents the
loan officer’s personal assessment of the firm (for example, an “A” in “Management Professionalism”), and an
objective signal (H) representing firm performance [for example, “10%” in Return on Assets (ROA)].
Suppose for now that the loan officer also has the authority to decide how much to lend to the applicant.
He will then infer the firm’s inherent quality given the two signals and decide on the loan limit to approve. All
else equal, higher firm quality should lead to higher loan approvals. Furthermore, define the informativeness of a
signal as its covariance with the underlying firm quality of interest. Then the loan officer will put weights βSand
βHon the two signals in his loan approval decision, where the weights are proportional to the informativeness
of the respective signals.
Next suppose that for some firms the loan officer cannot make the loan approval decision. Instead, after
collecting signals Sand Hfor these firms, he must send the loan application “upwards” to his boss (the manager)
for approval. Will the manager give the same importance to the two signals as the loan officer? The theoretical
literature cited earlier suggests this is not the case. For example, knowing that he no longer has discretion over
the final decision, the loan officer might put less effort in collecting S, hence reducing its degree of informativeness
[Aghion and Tirole (1997); and Stein (2002)]. Alternatively he might strategically add noise to S[Crawford and
Sobel (1982)]. Thus when signals Sand Hreach the manager, Swould have lost part of its informativeness.
The same is not true of H(i.e., audited financials) since this signal is verifiable. Hence weights β0
Sand β0
Hused
by the manager will be such that β0
S< βSand β0
H> βH.In other words, credit approved by the manager will
be less sensitive to subjective and more sensitive to objective information compared to credit approved by the
loan officer (see Section 1 for full details).
We test this prediction using data from the credit dossiers of a large multi-national bank in Argentina. The
data tracks the entire loan approval process for all 424 corporate loan applicants in the year 1998. It contains
all of the information collected by a loan officer regarding an applicant, including subjective information such
as the loan officer’s impression about an applicant’s management quality, as well as objective information
including audited firm financials. Moreover, there is variation in the hierarchical distance travelled by different
loan applications. While some are approved at a low hierarchical level (including the loan officer himself), others
have to go higher up for approval.
We find strong support for the theoretical and empirical predictions highlighted earlier. Sensitivity of
approved loan amount to objective information is much higher at higher levels of approval, while sensitivity
to subjective information is significantly lower at higher levels of approval. However, a key concern with this
finding is that the differences in sensitivity to information might be spuriously driven by endogenous allocation
ESTIMATING THE EFFECT OF HIERARCHIES ON INFORMATION USE 5
of applicants to approval levels, or alternatively the endogenous allocation of applicants to loan officers collecting
information.
Under the endogenous bank assignment concern, applicants may be assigned to different hierarchical levels
for approval in a way that systematically affects the sensitivity of credit to information. For example, suppose
larger firms have naturally more informative objective information and are also more likely to be sent higher up
for approval by the bank. Then credit sensitivity to objective information will be higher for firms approved at
higher levels because of the type of firms being sent higher up, not because of hierarchical distance. In general,
this concern can be difficult to address as bank assignment may be based on unobserved firm characteristics.
Fortunately however, the assignment principle chosen by our bank to assign firms to approval levels is
codified in its credit manuals, and based on observable applicant characteristics. The bank’s credit manual
prescribes a pre-specified set of rules that are a non-linear function of some observable firm characteristics,
such as applicant size, industry, and other firm-specific verifiable characteristics. We can thus exploit these non-
linearities to provide a plausibly exogenous source of variation in hierarchical distance. We do so by controlling
for linear and other higher powered functions of applicant characteristics that the bank uses in its allocation
rules.
The second endogenous loan-officer assignment concern is that firms that are approved at low levels (e.g., by
loan-officers themselves) are assigned to better or more experienced loan officers that generate more informative
subjective information due to their higher ability. If this were true, then the differential sensitivity across approval
levels will be driven by differences in the ability of loan officers collecting information rather than any direct
effect of hierarchical distance. However, we can completely account for this concern non-parametrically by using
loan officer fixed effects appropriately. The fixed-effects strategy forces comparison across firms that are approved
at different hierarchical levels but whose information is collected by the same loan officer.
Our result remains robust to controlling for the endogeneity concerns above. Credit sensitivity to subjective
information remains smaller, and credit sensitivity to objective information remains larger for firms approved
at higher levels. Additional tests further bolster the case that our result is driven by features of organizational
design rather than any spurious correlation.
In particular, we find that the change in information sensitivity at higher levels is not gradual. Loan approval
process within our bank can have up to five hierarchical layers, and the change in information sensitivity (for
both subjective and objective information) occurs suddenly between Levels 2 and 3. Exploring this further, we
find that these sharp changes in information sensitivity are driven by differences in the geographical location of
bank officers. The change in credit sensitivity to information occurs only when the loan approving officer sits in
a different geographical region than the loan officer.
The co-location result suggests that close proximity with the loan officer (who collects information) helps in
communicating subjective information. The importance of repeated contacts is further strengthened as we find
that the decline in sensitivity to subjective information at higher levels is smaller when information is generated
6 J.M. LIBERTI AND A.R. MIAN
by a more experienced loan officer. Higher level bank officers might be better able to understand, trust and
“decode” subjective information from more experienced loan officers as a result of repeated interactions with
them.
Finally, we decompose the aggregate index of subjective information into its constituent parts. The decline
in subjective information sensitivity is larger for more subjective sub-components, reaffirming the interpretation
that it is the subjectivity of a piece of information that makes it more difficult to use at higher hierarchical
distances.
There is a vast theoretical literature related to many of the issues our paper touches upon, but a review is
not feasible here. Overall our results are in line with the view that greater hierarchical distance discourages the
use of subjective and more abstract information. Although we discuss possible interpretations at the end, we
want to emphasize that our primary purpose is not to discriminate between various theories that might lead to
this reduced reliance on subjective information. In contrast, Liberti (2004) provides support for the loan officers’
incentives view in Stein (2002) by showing that loan officers who receive relatively more formal authority as
opposed to real authority put more effort into collecting soft information from their large corporate borrowers.
Despite almost an explosion of work in the theory of organizations, empirical work has far lagged behind.
Ours is one of the first papers that uses intra-firm data to directly test a key prediction of organizational
theory. While there is some empirical literature that associates specialization of certain bank types to their
organizational design [Berger et al. (2005); and Mian (2006)], the evidence that links organizational design to
information use in these papers is indirect. In contrast, our paper provides a more direct test of the effect of
hierarchical distance on information use.
1 Information and Hierarchies
A number of papers investigate how hierarchies affect the acquisition, transmission, and usage of information
within an organization. A common theme that runs through this literature is that separation of tasks across
organizational layers, such that employees in one layer rely on information generated by another, makes it more
difficult to share information. We find it useful to categorize this literature into three classes:
(1) Incentive-Based Theories - Aghion and Tirole (1997) and Stein (2002) argue that large hierarchical systems
inhibit the ex-ante incentives to collect and use information, particularly soft information. The drop in incentives
occurs because employees in charge of collecting information cannot act on it and instead have to send
information upwards for final decision. Given the “soft” nature of information, there is always a chance that it
may be overruled or disregarded. Anticipation of such overrules reduces the incentives for investing in information
collection effort.
(2) Strategic Manipulation of Information - The seminal work by Crawford and Sobel (1982) showed that senders
of information will deliberately coarsify their information and make it noisier if their preferences are not perfectly
aligned with those of the “receivers,” who again have the authority to take final action.
ESTIMATING THE EFFECT OF HIERARCHIES ON INFORMATION USE 7
(3) Ex-Post Communication Costs - Work such as, Becker and Murphy (1992), Radner (1993), and Bolton
and Dewatripont (1994) focuses on the ex-post costs of communication, and argues that while hierarchies
provide advantages such as specialization and parallel processing, they also bring trade-offs in the form of costly
communication across hierarchical levels. Such costs are likely to be larger for subjective information that is
harder to verify by a third party.
While the work cited above differs in its foundations, it shares a common theme. The literature predicts that
introducing layers between employees generating information and those taking decisions, leads to difficulties in
generating and transmitting information, particularly subjective information that is softer in nature. It is this
particular prediction that we take to data, and remain mostly agnostic as to which of the three classes of theories
might generate the observed empirical relationships.
1.1 Empirical specification
We motivate our empirical specification through an example that closely mirrors how our data is generated.
Consider a bank trying to decide how much to lend to a given firm. The bank is arranged as a hierarchy of two
layers as shown in Figure 1. A loan officer sits at the lower level and his manager at the higher level. The loan
officer is responsible for receiving and reviewing each loan application. The review process involves collecting a
variety of information about the firm. We summarize this information into two types: an objective signal Hand
a subjective signal S. The objective signal consists of easily quantifiable information such as size, profitability
(e.g., 10% ROA), and other audited financial ratios. The subjective signal on the other hand is qualitative in
nature and includes information such as the loan officer’s assessment (e.g., a grade of “A”) of firm’s management
quality and project strength.
Insert Figure 1 Approximately Here
Once necessary information has been collected by the loan officer, there are two possible scenarios.
Depending on the firm, either the loan officer has discretion to make the final credit approval decision, or
he refers the case to his manager who then makes the final decision taking into account information collected
by the loan officer.
Thus while information is always collected by the loan officer, the final authority to decide what amount,
L, should be given to a firm can rest with either the loan officer or his manager. Both loan officer and manager
decide on Ldepending on firm quality Q, with higher quality firms getting larger loan amounts. However, loan
officer and manager differ in terms of how they determine Qgiven informational signals Hand S. In-line with
the theoretical work, suppose that Sloses part of its informativeness when it is used by someone higher up in
the hierarchy (manager in our example) who did not collect this information. Hon the other hand is based on
objective information that everyone can interpret and verify and hence does not lose informativeness (to the
same extent at least) when used across hierarchical layers.
8 J.M. LIBERTI AND A.R. MIAN
For example, Hmay include a firm’s ROA during the last three years as recorded in audited firm financials.
Son the other hand may include the subjective score given by the loan officer regarding the quality of the firm’s
new management. If the loan officer has to communicate these two signals to the manager, an ROA of say 10%
can be communicated without much loss of information. However, a subjective management quality grade of
say “A” is much harder to interpret for a bank manager, and the quality of this information depends on loan
officer’s incentives.
Formally if one defines “informativess” of a signal as its covariance with the underlying metric of interest
Q, then theory predicts that the covariance drops faster for subjective information when communicated across
hierarchical levels. This gives us the following empirical prediction2:
Lij =α+β1×MGRij +βH×Hij +βM
H×(Hij ×MGRij ) + βS×Sij +βM
S×(Sij ×MGRij ) + εij (1)
Lij is the log of approved credit for firm iwhose information is collected by loan officer j. M GRiis an
indicator variable equal to 1 if firm iis approved by the manager, and 0 if firm iis approved by the loan officer.
The main prediction is that βM
H>0 and βM
S<0, i.e., sensitivity of the credit approval decision to subjective
information is smaller for managers than loan officers, and vice versa for objective information. There is no
particular prediction on the level of sensitivity to subjective and objective information [i.e., coefficients βHand
βSin (1)].
We have deliberately used the term “subjective” rather than “soft” to denote the more nuanced information.
A strict definition of soft information makes it impossible to be codified and hence (by assumption) soft
information cannot be observed by an econometrician. Subjective information, on the other hand, can be
categorized into grades. However, such grades do not have a well-specified objective metric like height, weight,
or profitability, and hence cannot be objectively “verified” by a third party.
2 Data Description
We estimate Equation (1) using data from a bank whose organizational structure closely mirrors the description
in Section 1. The data covers information contained in the credit folders of all of the 424 corporate clients of a
large multi-national bank in Argentina in 1998. A firm is classified as corporate by the bank if its annual net
sales exceed 50 million pesos.3The advantage of having full access to these credit folders is that we observe the
entire life cycle of loan origination. Our data set contains all of the information collected by a loan officer as
part of the loan review process. We also observe the hierarchical level at which a given loan is approved, the
geographical location of the final approving officer, and the approved and requested loan amounts.
The timing of a typical loan review at the bank is as follows. Once a firm requests credit from the bank, it is
assigned a loan officer who is in charge of developing the firm-bank relationship. At the same time, given the basic
ESTIMATING THE EFFECT OF HIERARCHIES ON INFORMATION USE 9
verifiable information provided by the firm in its application, the bank’s credit policy manuals determine the
ultimate hierarchical level of approval. Two points are important to emphasize here. First, the final hierarchical
level of approval is determined before the loan officer collects his firm-specific information. This ex-ante knowledge
of who has the final discretion over the approval process is likely to effect incentives of the loan officer collecting
information. Second, the final hierarchical level of approval is determined by a set of observable objective firm
attributes that do not depend on the loan officer’s subjective assessment. These attributes, which we refer to
as approval level rule variables, are collected as part of the initial loan application (i.e., before the loan officer
collects more detailed information in the loan review process). Given these rule variables, a set of pre-specified
rules in the credit manual determine which hierarchical level within the bank the loan application must go for
final approval.
The pre-specified set of rules in the credit manual guarantee that the loan officer has no discretion in
determining the final level of credit approval for a firm. This is rational for a profit maximizing bank. If the
bank believes that the loan officer does not have sufficient capability to approve a loan for certain firms, then
it would not want the loan officer to decide what those firms are.4There are five different levels of approval in
the hierarchical design of our bank, with the loan officer sitting at the lowest level (see Figure 2).
Insert Figure 2 Approximately Here
Once the final level of credit approval is determined, a loan officer collects detailed information regarding
the firm’s financials, as well as subjective information through interviews and plant visits. The content, type
and quality of information is consistent across credit folders, with all credit folders containing the same type of
information. Bank credit manuals specify exactly what kind of questions and information each loan officer must
seek for a given loan application.
After a loan officer has completed the information required for a given loan application, the application
travels sequentially through all hierarchical levels until it reaches its final level of credit approval. The final level
of approval can of course be the loan officer himself.
We divide variables constructed from the credit folders into approval level rule variables collected at the
time of initial loan application, informational variables collected by the loan officer as part of the loan review
process, and credit approval variables determined by the final approving authority. These variables are described
in detail below.
2.1 Approval level rule variables
Given the five hierarchical levels in the bank, Table 1 shows how firms are distributed across these levels for
credit approval. The loan officer himself approves at Level 1 26.7% of loans.5Another 37.3% are approved at
Level 2, and the remaining are approximately equally divided among Levels 3, 4, and 5.
10 J.M. LIBERTI AND A.R. MIAN
Firms are sent to one of the five hierarchical levels as determined by the rule variables. Although theoretically
there are 19 rule variables, many of these are “exceptions” that are used very rarely. In particular there are 11
such variables that taken together influence the approval level of only 48 firms in sample.6For brevity we do
not report their summary statistics, although they will be included in the regression analysis later on.
The eight primary variables responsible for assigning applicants to different approval levels are described
in Appendix A, and their averages by the five approval levels are given in Panel A of Table 1. These variables
include applicant characteristics such as: loan maturity and level of collateralization, applicant loan size, central
bank credit score, foreign bank branch guarantee, family company indicator, and indicator variables for whether
a firm belongs to a pre-specified industry or fails to pass an industry “threshold level.”
It should be kept in mind that credit manual guidelines that map rule variables to approval levels cannot
be expressed in a single closed form function. There are a number of discontinuities and trigger points built
into the credit manual guidelines. For example, larger applicants are more likely to be sent to higher levels for
approval. However this relationship is not smooth, and by necessity there are cut-off points deciding the level of
firms. Similarly a number of other reasons, such as maturity structure, firm industry, and credit score can send
a firm to higher levels for approval even if the firm falls in a lower level according to applicant size. It is thus a
combination of several non-linear rules that decides the ultimate approval level for a firm.
General principles underlying assignment rules can be understood from Panel A of Table 1 that provides
means of all rule variables broken down by the five approval levels. The means shows that firms requesting larger
loans are more likely to be sent to higher levels for approval. Since bigger firms have larger and more complex
funding requirements, the bank is more inclined to send such firms to officers higher up in the hierarchy as
they have more experience and expertise. Similarly, firms belonging to volatile industries, poor credit history,
long-term loans and unsecured loan applications are more likely to be sent to higher levels for approval. On the
other hand, firms with guarantees from foreign affiliates of the bank are unlikely to be sent up for approval.
These patterns again reflect the belief that more senior officers are better able to evaluate more complex loans.
Insert Table 1 Approximately Here
Table 2 formally investigates the relationship between the approval level and rule variables used by the
bank’s credit manual to allocate firms across levels. Column (1) includes all of the (19) rule variables on the
right-hand side, and reaffirms that larger applicants, applicants with worse credit scores, firms with more complex
loan requests, and firms belonging to volatile or nascent industries are more likely to be sent to higher levels for
approval. These results are very much in line with the “management by exception” criteria of Garicano (2000),
where the role of a hierarchy is to conserve the time of experts so that they only intervene when no one else can
solve a problem. Although Column (1) includes all of the rule variables used by the bank, the R2is still only
0.42. These low R2reflect the non-linear nature of the assignment procedure followed by the bank. It is neither
due to the bank ignoring assignment rules at times, nor is it due to missing rule variables. For example, we can
ESTIMATING THE EFFECT OF HIERARCHIES ON INFORMATION USE 11
get an R2of 1 if we manually apply the credit manual procedure to the rule variables associated with each firm.
The “predicted” approval level from doing this exercise matches the actual approval level in all of the 424 firms
in our sample.
Insert Table 2 Approximately Here
Column (2) of Table 2 includes all pair-wise interactions of the top four rule variables to allow for some
non-linearities. The top four rule variables are Maturity/Collateral Score, Applicant Size, Central Bank Credit
Score, and Foreign Bank Branch Guarantee. The R2increases to 0.54. Column (3) adds higher powers of the
rule variables by including functions of powers 2 and 3 for all the 19 rule variables. The R2increases slightly to
0.55 as a result. Column (4) shows that most of the variation in approval levels is explained by the top four rule
variables as these four variables alone give us an R2of 0.34 compared to 0.42 in Column (1).
Since approval levels only take integer values, OLS may not be an appropriate estimation technique.
Correspondingly we experiment with ordered probit and ordered logit specification in Columns (5) and (6).
However, even with such non-normal estimation techniques Pseudo R2is not very high.
2.2 Informational variables
Once a credit application is filed and its ultimate approval level is known, the credit folder is given to a loan
officer (LO) who collects all the firm level information. A typical loan officer manages around 20-25 firms (on
average) that are mostly clustered in a single or related industries. The collected information includes objective
information from audited firm financials, as well as subjective assessment of firm quality by the loan officer. The
subjective assessment is based upon visits to firm premises and interviews with firm management.
Our data set includes all of the objective and subjective pieces of information collected by the loan officer as
per bank rules. The bank pre-specifies what pieces of information have to be collected by a loan officer. In order
to avoid concerns of “data mining,” we desist from picking and choosing any particular set of informational
variables. Instead, we use all of the informational variables collected by a loan officer. These variables are
naturally classified by the bank into two categories: objective cardinal and subjective ordinal measures.
The first category of variables measure some cardinal firm characteristic. This category includes firm
financials from audited records, and we classify it as objective information. Appendix B provides the full list of
objective variables, which include leverage ratios, profitability, cash flows, and size measures. We classify these
variables as objective since they are quantifiable, easy to collect and transmit, and are verified by a third party
(the auditor). Therefore,, a loan officer collecting this information does not have any discretion in how to report
it, and also does not need much effort or expertise in collecting such information.
Since the objective variables (in particular leverage ratios) can have large variance, the bank translates these
ratios according to a pre-specified formula into a rating that goes from 0 to 22 for all financial ratios, and 1 to
6 for firm size. The ratings are a monotonic categorization of the financial ratios. The bank also constructs an
12 J.M. LIBERTI AND A.R. MIAN
overall index of these financial ratios and size information that we define as objective index. We standardize this
index by subtracting the sample mean and dividing by the sample standard deviation. We also divide objective
index into two (standardized) sub-indices: a performance index that averages all of the leverage, profitability
and current financial ratios, and a size index composed of firm size.
The second category of informational variables collected by the bank are subjective ordinal rankings
provided by the loan officer. These variables, which we classify as subjective information, are personal assessments
of the loan officer on various firm and management attributes. A differentiating feature of subjective information
is that it involves discretion on the part of the loan officer, and requires him to invest effort and expertise in order
to collect reliable information. As with objective information, the bank pre-specifies what pieces of subjective
information a loan officer must collect. These variables are described in Appendix C, and include loan officer’s
assessment of management quality, accounting practices, firm’s risk management policies, firm’s overall market
positioning, industry outlook, and firm’s access to external capital markets. The loan officer assigns an ordinal
(subjective) score of 1 through 7 to each subjective firm attribute, with larger scores signifying higher firm
quality. For any given subjective variable, a particular score corresponds to a pre-defined criteria. For example,
a 3 in professionalism corresponds to “At some Key Positions,” while a 5 corresponds to “At All Key Positions
In Operations and Management.” Appendix D provides a mapping of subjective categories into their respective
definitions for all variables.
The bank also aggregates its subjective information into an overall index, which is standardized to provide
us our subjective index. Although all variables with ordinal rankings are initially combined into one subjective
index, they differ in the degree of their subjectivity. For example, when a loan officer is asked to report on a
firm’s ability to access outside funds, he may use some objective verifiable information such as existing firm
lenders to arrive at an answer. However, a question regarding a firm’s “professionalism” is considerably more
subjective. We therefore also construct two sub-indices of the overall subjective index into a strong subjective
index and a weak subjective index. The strong subjective index is a standardized average of management and
competitive position variables that we think involve more subjectivity, a priori, than other variables. Variables
in industry risk assessment, risk management policies, and access to capital categories on the other hand are
classified as weakly subjective since they are partially based on objective information such as lending by other
banks, industry sales trends, and leverage and liquidity policies that can be inferred from audited financial
statements.
Although we stick with the bank’s construction of objective and subjective indices (to avoid concerns of
data mining), our results are completely robust to alternative definitions of objective and subjective indices
as we shall discuss in the robustness section. Panel B of Table 1 provides the mean and standard deviation
of information indices by the level of approval. The variation in information indices is similar across the five
levels of approval. This is useful since we estimate the sensitivity of loan approval to information separately at
different approval levels. Since some of our regression specifications use loan-officer fixed effects, we also report
ESTIMATING THE EFFECT OF HIERARCHIES ON INFORMATION USE 13
standard deviation in subjective and objective indices after demeaning these variables at the loan officer (i.e.,
information collecting agent) level. As Panel B shows, there is significant variation across the five levels even
within the same loan officer.
Table 3 provides the correlation matrix for the various sub-components of subjective and objective indices.
The sub-components are positively correlated as expected. However, the correlation is not perfect, signifying the
independent component that each sub-component brings to the overall indices. We shall explore the variation
in sub-components in some of the analysis as well.
Insert Table 3 Approximately Here
2.3 Credit approval variables
Once a loan officer collects all required information, credit is approved and authorized by the loan officer himself
if he has the authority to do so. Else, otherwise the credit file is sent up the hierarchy towards the bank officer
with the approving authority. The average credit facility provided by the bank in 1998 was $16.6 million and
there is significant variation in this amount across firms as shown in Panel A of Table 1. The approved credit
line aggregates all the short-, medium-, and long-term financing provided by the bank. Once a credit line is
approved, a firm does not have to utilize all of it. In fact, the average outstanding loan for a given firm is $10.7
million. The difference between approved and outstanding amounts partly reflects liquidity management on part
of the firms as their short-term credit demand fluctuates.
Other variables collected by the bank include credit risk rating of the firm, an indicator as to whether the
firm is in financial distress, maturity of all existing facilities over three years, percentage of unsecured existing
facilities, legal history of default and covenant violations, years in industry, ownership type, and access to other
financial institutions. We also have some specific information such as the time (in days) taken by the credit
analyst and LO to prepare the credit recommendation form and whether additional information was requested
by the loan officer along the process. Our final data set includes all clients with approved credit lines in 1998.
However, if a credit application were rejected by the bank, we do not have it in our data.
3 Empirical Methodology
3.1 Identification
We can estimate Equation (1) using data on subjective and objective information indices (Sand H) for loan
applicants approved at different hierarchical levels of approval. However, the concern is that differences in
sensitivity of credit to information may be driven by the endogenous assignment of loan applicants to various
approval levels, rather than a direct effect of hierarchical distance. We outline these concerns in more detail
below.
14 J.M. LIBERTI AND A.R. MIAN
3.1.1 Endogenous bank assignment
As we have outlined, the bank follows a systematic set of rule to assign applicants to various approval levels.
A concern therefore is that firms sent to higher levels for approval are inherently different in terms of how
important objective and subjective information is in evaluating them. For example, perhaps firms with less
reliable subjective information are deliberately sent further up in the hierarchy because more senior bank officers
are better able to tackle complicated loans with poor subjective information. If this were the case, managers will
put less weight on subjective information compared to loan officers even if there was no effect of hierarchical
distance on information use.
More generally, let Zbe a firm characteristic that the bank uses to assign firms to higher levels of approval.
For simplicity, assume that there is only one such variable, say firm size. The bank chooses a cutoff size Zsuch
that firms above this threshold are sent to the manager for approval while others are sent to the loan officer.
Figure 3 shows the function mapping Zto approval level. The endogeneity concern is that larger firms might
have less relevant subjective information, and/or more relevant objective information. This can be represented
statistically by plotting how the “informativeness” of subjective and objective information varies with Z.
Insert Figure 3 Approximately Here
Let σ2
qs and σ2
qh denote the covariance of subjective and objective information respectively with firm
quality Q. Furthermore, suppose σ2
qh and σ2
qs denote the maximum possible informativeness for a firm, i.e.,
the informativeness that the best loan officer can generate if he works efficiently. Then the general concern is
that any bank assignment criteria Zmight be positively correlated with σ2
qh and/or negatively correlated with
σ2
qs.Figure 3 plots some possible relationships between Zand σ2
qs,and Zand σ2
qh that can bias βM
Sdownwards
and βM
Hupwards respectively.
The endogenous bank assignment concern highlighted in Figure 3 is almost impossible to address if Zis
unknown or not observable. However, as pointed out, the bank has a pre-specified list of rule variables (i.e.,Z’s)
that determine which level a firm gets sent to. Moreover, these rule variables are based on third-party verifiable
objective criteria and not subject to the loan officer’s discretion. We can therefore control for endogenous bank
assignment concerns by including Z, and its interactions with subjective and objective information (Z×S) and
(Z×H) as controls in (1). We can also include higher powers of Z(such as Z2) and their interactions with H
and Sto allow for greater functional form flexibility.
The inclusion of linear and quadratic bank selection controls implies that the identification of βM
Hand βM
S
is coming from the non-linear and discontinuous part of the relationship between rule variables Zand approval
levels. For example, by necessity approval levels have to be partly a non-linear and discontinuous function
of the ex-ante firm selection variables. Once we control for linear and quadratic components of Z, it is these
non-linearities and “jumps” in the residual variance that are used to identify βM
Hand βM
S.
ESTIMATING THE EFFECT OF HIERARCHIES ON INFORMATION USE 15
3.1.2 Endogenous loan officer assignment
A separate concern in estimating (1) is the endogenous assignment of loan officers to firms. Since information
for all types of firms is collected by the loan officers, it might be the case that firms approved by loan officers
themselves are given to loan officers with better ability and expertise in collecting subjective information. If this
were the case, then firms approved by loan officers will get higher weight on subjective information not because
of the lower level of approval, but because the loan officer was better at collecting subjective information.
Since we know the identity of the loan officer, j, collecting information for each firm, i, we can fully address
the loan officer selection concern by including loan officer fixed effects, and interacting these fixed effects with
Hand S. This non-parametric approach ensures that we only compare firms at different approval levels whose
information was collected by the same loan officer. Bank selection controls and loan officer fixed effects (αj)
update (1) to:
Lij =βM
H(Hij ×MGRij ) + βM
S(Sij ×MGRij ) + αj+αjHij +αjSij +β1M GRij +β2Zi+β3(Hij Zi) + β4(Sij Zi) + εij
(2)
4 Results
4.1 Effect of hierarchy on information use
We estimate Equation (1) using the data and methodology described above. We begin by collapsing the five
approval levels into “high” and “low” around the median. This classifies approval Levels 1 and 2 as “low,” and
Levels 3, 4, and 5 as “high.” Column (1) of Table 4 estimates Equation (1) using the log of approved credit
line as the dependent variable. Coefficients on interaction between the information indices and the high level
dummy show that the sensitivity of credit approval to subjective information dramatically goes down for loans
approved higher up in the hierarchy, while sensitivity to objective information increases for loans approved at a
high level. These results are consistent with theoretical predictions.
Insert Table 4 Approximately Here
However, as Section 3 explained, the result may also be driven by endogenous assignment of firms and/or
loan officers. Column (2) of Table 4 includes loan officer fixed effects and their interactions with objective and
subjective indices. There are a total of 26 loan officers. The fixed effects non-parametrically control for the
person generating subjective and objective information, and force comparison between firms whose objective
and subjective information is generated by the same loan officer.7The results are very similar to those of Column
(1).
16 J.M. LIBERTI AND A.R. MIAN
Column (3) of Table 4 controls for endogenous bank assignment concern by including variables used by the
bank to assign firms to different levels as controls. We include the 19 rule variables described in Appendix A, as
well as their interactions with objective and subjective information indices as controls on the right-hand side.
In other words, Column (3) tests the full blown specification in equation (2) that exploits the non-linearities
in bank assignment rules to identify our coefficients of interest. Our coefficients of interest remain qualitatively
unchanged. It is worth emphasizing that the amount of loan requested by an applicant is one of the controls
in Column (3). In other words, both right- and left-hand side variables are conditioned on the amount of loan
requested by an applicant. Since we are exploiting non-linearities in rule approval to identify our coefficient of
interest in Column (3), the increase in objective information sensitivity and decrease in subjective information
sensitivity at higher levels is unlikely to be driven by spurious bank assignment criteria.
The magnitude of the effect of hierarchical distance is large. As all informational variables have already been
normalized, the coefficients can be interpreted as the effect of a one standard deviation change in information
variables. Then for a firm with a 1% higher objective information score (in s.d. units), getting approved at the
higher (more distant) hierarchical level increases its approved credit limit by about 0.8%. Of course if the same
firm also had 1% higher subjective information score, then it would lose out by about 0.7% if approved at the
higher level. Thus the net effect really depends on the correlation of objective and subjective scores across firms.
The information in Table 3, Panel C is instructive here. It shows that subjective and objective information
indices are positively correlated, and the correlation magnitude is strong.
The coefficient on objective information is essentially zero at low level of approval (Column (1) of Table 4).
However, if we take out subjective information from Column (1), then objective information is positive at the
low levels as well. In other words, it is the component of objective information that is orthogonal to subjective
information that is not given any weight in the credit making decision by officers at low levels. It is a bit puzzling
that objective information caries no significant (independent) weight in the evaluation of loans at low level. One
possible explanation is that when loan officers know that they themselves (or close associates) are approving
the loan, they incorporate all necessary information (in their view) into the subjective information grades. Thus
once this subjective information is taken into account, there is no residual power in the objective information
index.
We also explore the robustness of our results to concerns that they might be driven spuriously by firm
attributes such as firm size, firm profitability, and industry fixed effects in Column (4). However, inclusion of
these variables as controls does not effect our coefficients of interest qualitatively.
Columns (5) and (6) of Table 4 open up these five levels to see how the sensitivity to information changes
at each level. Column (5) includes loan officer fixed effects and their interactions with information indices, while
Column (6) adds bank selection criteria variables and their interactions with information indices as well. The
results show that the change in credit sensitivity is not gradual across the five approval levels. The change in
sensitivity to subjective and objective information happens relatively sharply at Level 3 and then persists at
ESTIMATING THE EFFECT OF HIERARCHIES ON INFORMATION USE 17
higher approval levels. Furthermore, as before results are symmetric for subjective and objective information.
Sensitivity to subjective information declines at Level 3 and beyond, while sensitivity to objective information
increases at the same levels.
4.2 Does geographical location matter for information flow?
If changes in information sensitivity are truly driven by the level of approval, then why does the effect kick in
at Level 3? For example, why is the effect not more gradual from Level 1 through Level 5? If the information
sensitivity effect is coming from differences in the organizational structure of the loan approval process, then
how are approvals at Level 2 so much different from approval at Level 3, but not from approval at Level 1?
The geographical location of officers at different hierarchical levels presents a possible explanation. Our
data includes information on the location of each loan officer involved in the loan process. Panel A of Table 5
shows the joint distribution of the level of approval, and the geographical distance between a loan officer and the
officer approving a given loan. The variable, geographical distance, is defined as 0 if the loan officer who collects
information and the loan approving officer are located in the same branch. Otherwise it is coded as 1. The
joint distribution shows that loan officers collecting information and loan approval officers at Level 2 always are
located in the same bank branch. They can therefore interact and communicate on a daily basis with ease and
are likely to know each other quite well. Since there is equal sensitivity to objective and subjective information
among Level 1 and Level 2 approvals, it suggests that communicating subjective information among co-workers
who work in close geographical proximity is easy.
Officers above Level 2 on the other hand, are not always located in the same bank branch as the loan
officer. In fact, Level 4 and 5 officers are never located in the same branch as their loan officers. These officers
are found in the larger headquarter offices and sometimes even outside the country. Officers at Level 3 however
sometimes sit inside and sometimes outside the local branch where information is collected. Out of 54 firms that
are approved by officers at Level 3, 17 are approved by officers who are locateed at the same branch and 37 by
officers who are found at a different location.
Insert Table 5 Approximately Here
We exploit variation in location of the loan approving officer to formally test whether the results in Table
5 were driven by the loss in informativeness due to officers sitting at different geographical locations. Column
(1) of Table 5 of Panel B re-runs Column (1) of Table 4, but replaces hierarchical distance with geographical
distance. The results show that the change in sensitivity to information happens when the approving officer and
the loan officer collecting information are at different locations . However, as Panel A showed, geographical and
hierarchical distance are highly correlated. The only independent variation in geographical distance occurs for
loans approved at Level 3. Therefore, Column (2) restricts the sample to the set of 54 firms that are approved
at Level 3. Even though the number of observations is much smaller, coefficients on interaction terms support
18 J.M. LIBERTI AND A.R. MIAN
the hypothesis that differences in geographical location are an important factor in the loss of informativeness.
When a Level 3 officer is located in the same branch as the loan officer, his sensitivity to subjective information
is much higher than a Level 3 officer located outside the loan officer’s branch. Similarly, sensitivity to objective
information increases when the officer sits outside the branch of the loan officer.8Column (3) repeats Column (2)
on the full data, but includes all the approval level dummies and their interactions with informational indices. It
thus replicates Column (2), but is more efficient for computing standard errors. The results are almost identical.9
The fact that changes in sensitivity to information are not gradual, but happen suddenly in between levels
where the geographical location of approving officers is different from loan officers, further strengthens the
interpretation that differential sensitivity is driven by organizational differences in the loan approval process of
different firms.
4.3 Are more experienced loan officers better at communicating subjective information?
The usefulness of co-location for communicating subjective information suggests the importance of repeated
interactions. Geographical proximity facilitates repeated interactions that help in understanding and relying on
each other’s subjective information. While geographical proximity is useful, a substitute for proximity might
be repeated interactions over time. For example, a more experienced loan officer is likely to have interacted
with senior officers more often, which can make the interpretation of subjective information easier for high level
officers. An analogy may be drawn here with the academic job market where a recruitment committee might
give more weight to a recommendation if they have personally interacted with the recommending professor often
over time.
Since we have information on the experience of a loan officer within the bank, we can test whether this
experience facilitates subjective information communication. We do so through our loan officer fixed effects
specification and add triple interactions of subjective and objective information sensitivities with loan officers’
experience. The median loan officer has six years of experience in the bank. There is a break in the loan officer
tenure distribution at seven, and as such we use seven years’ experience as the cutoff point to create a dummy
for “experience.” The results in Columns (1) and (2) of Table 6 show that the decline in subjective information
sensitivity is much smaller for more experienced loan officers.10
Insert Table 6 Approximately Here
Since we use loan officers’ fixed effects and their interactions with objective and subjective variables as
well, our result cannot be driven by more experienced loan officers having better overall quality of subjective
information. A higher overall level of subjective information can explain an overall greater sensitivity to
subjective information for all bank officers, but it cannot explain why the sensitivity improves more for higher
level officers. Thus, experience of a loan officer likely improves the communication of subjective information
across hierarchies.
ESTIMATING THE EFFECT OF HIERARCHIES ON INFORMATION USE 19
4.4 Is the effect stronger for more subjective information?
So far we have used the objective and subjective indices constructed by the bank to measure credit sensitivity.
However, since we also have the underlying variables used to construct these indices, we can check for the
robustness of results to different ways of aggregating the underlying variables. We first explore variation in
subjective information variables. Appendix B provided details of all the subjective information variables used
to construct subjective information rating. There are a total of 18 primary subjective information variables,
divided across five subjective information categories. The bank uses its own formula to weigh these 18 variables
in coming up with an overall subjective ranking. While we are not at liberty to disclose the bank’s internal
rating construction, we can construct alternative indices of our own using these 18 variables.
We construct two different definitions of overall subjective information rank. (1) AVGsubjective: This is a
simple arithmetic mean of all the 18 subjective information variables, and (2) WAVGsubjective: This weighs the
five categories equally while giving equal weights to the subjective information variables within each category.
Columns (1) and (2) in Table 7 repeat the primary regression specification but replace subjective information
rating with AVGsubjective and WAVGsubjective respectively. The result on credit sensitivity to subjective
information is very similar in spirit to what we found earlier. As such our main result is not sensitive to the
definition of how subjective information index is constructed.
Insert Table 7 Approximately Here
Subjective information variables also differ in their “subjectiveness” or the extent of subjectivity involved
in computing them. If sensitivity to subjective information declines as a result of communication losses across
hierarchies, then one would expect such losses to be greater for more subjective variables. We therefore divide
subjective variables according to the degree of subjectivity involved in computing them and split the subjective
index into a strong subjective index, and a weak subjective index (Section 2 explained their construction).
Columns (3) through (5) of Table 7 test whether the drop in sensitivity to subjective information at higher
levels is stronger for more subjective information. The results indicate that the drop in sensitivity of subjective
information is stronger for the more subjective sub-index. This result is also in-line with our earlier results and
interpretation that it is the subjectivity of information that makes it difficult to communicate across hierarchies.
For example, consider the components of the weak subjective index. In coming up with industry outlook indices,
a loan officer may use publicly verifiable industry data such as recent growth and volatility. Rating a firm’s
leverage or liquidity policy can also be judged to a reasonable extent from its balance sheet numbers. Similarly
access to capital data is generally available in verifiable formats such as central credit registry data or knowing
the number of relationships the firm has access to.
On the other hand, components of the strong subjective index such as, variables linked to a firm’s
competitiveness and management quality are more subjective. For instance, ranking a firm’s “professionalism,”
“ability to act decisively,” or “technology advantage” is inherently a much more subjective exercise.
20 J.M. LIBERTI AND A.R. MIAN
Finally, we test for the robustness of our results to the definition of objective information index. As explained
earlier, the bank uses seven different financial ratios to arrive at its objective information rating that we have so
far used in our analysis. We also constructed our own index of objective information by taking the arithmetic
mean of these financial ratios. Results with our index of objective information are qualitatively very similar to
those obtained with the bank’s objective risk rating (regressions not shown).
5 Discussion of Results
Our results indicate that greater hierarchical distance makes it difficult to use subjective information and favors
the use of objective information instead. A number of tests, such as exploring non-linearities in assignment of
applicants to approval levels and loan officer fixed effects, showed that the results are not driven by spurious
correlations. Further results on the importance of the co-location of loan officer and loan approving officer, and
experience of the loan officer bolster the importance of organizational design on information use.
Section 1 explained that a number of different theories all suggest that hierarchical distance should favor
objective over subjective information. Given that our results confirm this common prediction, we now outline
which economic interpretation is more favorable in the light of our results.
5.1 Loss in communication
One interpretation of our results is based on theories of costly communication. In particular, subjective
information may be more costly to communicate across hierarchies, particularly when communicating parties
are geographically separated, and when the person generating information has been with the bank for a brief
period of time. Subjective information is harder to communicate between people who do not work together since
they are not fully aware of each other’s trust, competence, and judgement criteria. For example, it is easier for
coauthors to exchange (subjective) ideas if they work in the same building compared to coauthors working in
separate cities. This interpretation is consistent with our result that credit sensitivity to subjective information
declines at higher levels, that the decline is larger for more subjective information, that the drop in sensitivity
only kicks in when an officer in the higher hierarchy is located in a different branch, and the effect is strongest
when the loan officer has spent the least time in the bank.
5.2 Incentives to gather information
A slightly different interpretation of our results could be that when a loan officer has little control over the use
of his information, he has less incentives to gather and use quality information. The view that decision-making
authority increases a loan officer’s incentives to collect information has already been proposed in papers such
as Aghion and Tirole (1997) and Stein (2002) and tested in Liberti (2004). An incentive based explanation is
more likely to effect subjective information acquisition since this type of information requires more effort and
thinking on the part of the loan officer. For an incentive based story to explain all of our results, we will have
ESTIMATING THE EFFECT OF HIERARCHIES ON INFORMATION USE 21
to assume that the loss of incentives is not great when the person making the final credit decision works in
close geographical proximity to the loan officer. In other words, the loan officer must feel sufficiently part of
the decision-making process if the approving officer work close to him. Similarly, we have to assume that the
greater subjectivity of a variables increases the effort required from a loan officer. In such a case more subjective
information is more likely to be affected by an incentive effect.
5.3 Strategic manipulation of information
A loan officer might strategically manipulate and coarsify his information, as in Crawford and Sobel (1982), if he
does not have control over decision making. For example, this might be done in an effort to retain more control
by the loan officers themselves, or to make the decisions of other officers look worse. Since objective information
is more difficult to manipulate, loan officers are more likely to manipulate subjective information. Therefore, if
strategic manipulation exists in equilibrium, officers at higher approval levels will deliberately put less weight
on subjective information as they know the information has been tempered with.
However, we feel that strategic manipulation is unlikely to be a main explanation of our results. Loan
officers must also have an incentive to provide accurate and useful information to their superiors in order to
maximize their chances of promotion and career development. Such incentives should suppress the desires to
manipulate information. Similarly, the effect of strategic manipulation should have been seen when the Level 2
officer has discretion over credit approval. However the drop in sensitivity to subjective information is only seen
at Level 3 and beyond, and only when the decision-making officer is located in a separate branch. This evidence
also lowers the likelihood of strategic manipulation as a primary explanation of our results.
5.4 Different abilities or objectives
Officers at different levels may have different abilities to handle objective and subjective information variables.
Alternatively officers at different levels may have different tastes or objectives in terms of incorporating objective
and subjective information into their decisions. However, there is no particular theory to suggest why such
differences might exist. The bank also has identical lending guidelines for loan approval regardless of the
hierarchical level of approval.
Even if differences in objectives exist, there is no strong reason to suggest that officers at higher levels
should have a stronger bias against subjective information. Moreover, any theory based on differences in tastes
and abilities will have to argue that such differences do not exist between Levels 1 and 2, but do exist at higher
levels, and only kick in when officers at higher levels are located in a different location. As such it is difficult to
come up with a plausible explanation for our results based on differences in objectives alone.
5.5 Corruption or related lending
Since loan approvals at lower levels of the hierarchy rely more on subjective information, perhaps the evidence
reflects corruption or related lending by local branches. Corrupt lending refers to loans that are not based on
22 J.M. LIBERTI AND A.R. MIAN
any informational advantage, but rather loans that do not deserve to be made on financial grounds. However,
corruption is unlikely to be an explanation for our results. First, the bank we study is a multi-national bank
with assets all over the world. With so much reputational capital at stake, the bank is very unlikely to engage
in related lending in a small market. Even less likely is the scenario that the bank would only engage in such
related lending at lower levels of hierarchy and not at higher levels. Second, we have ownership information on
borrowers. None of the borrowing firms are “related” to loan officers, or loan approving officers inside the bank.
6 Concluding Remarks
Our main purpose was to test how hierarchical design impacts information sharing and use. Does the impact of
hierarchical distance on information sharing also affect the efficiency of financial intermediation? While it is an
important question, we are limited in how far we can answer it. Measuring efficiency of financial intermediation
is difficult since measures such as default and firm profitability can be misleading. For example, realized default
can be a very poor proxy for expected default, particularly in volatile and non-stationary environments like
Argentina. This is especially problematic for us since Argentina went through a massive economic crisis a couple
of years after our sample period in 2001. Nonetheless, using an outcome measure such as future firm default, and
future firm profitability, we find no systematic difference between applicants getting approvals at high versus
low hierarchical levels.
We should also point out that our analysis took the hierarchical design in our sample as given. As such
questions regarding whether the hierarchical design is optimal remain outside the scope of our paper. Optimal
organizational design involves not just concerns of information sharing, but also a host of other issues, such as
career concerns, task specialization, etc. A meaningful analysis of organizational optimality needs to take all of
these dimensions into account.
A lot has been written on how the design of organizations affects incentives, flow of information, and
ultimately the scope of firms. Yet our empirical understanding of these issues lags far behind. The reasons are
mostly obvious. Information at the intra-firm level is seldom collected, and firms are reluctant to share such
information. Even with available information, it is difficult to find exogenous variation in the organizational
attribute of interest for identification. Furthermore, several theoretical constructs such as “power” and “soft
information” are difficult to define empirically. The methodology adopted in this study aimed to address some
of these issues as we had a rare opportunity to peek inside the decision-making process of a large hierarchy.
ESTIMATING THE EFFECT OF HIERARCHIES ON INFORMATION USE 23
Appendix A: Description of Approval Level Rule Variables
There are eight primary approval level rule variables described below:
Maturity/Collateral Score: This is the bank’s numerical indicator on a scale of 1 to 10 (including sub-
grades) to identify the overall risk associated with each specific facility/loan of the firm. The bank’s scale
follows the following scheme: 1 (best), 2, 2, +2, 3, 3, +3, 4, 4, +4, 5, 5, +5, 6, 6, +6, 7, 7, +7,
8, 9, and 10 (worse). We scaled this score using a numerical indicator between 1 (worst) and 22 (best).
The bank’s pre-specified rules map the worse of the scores for each of the loans granted to the firm into
a corresponding approval level. The score is based on both the maturity of the facilities and the level of
collateralization. The score is lower if credit facilities are short term (relative to long-term) and secured
or guaranteed (relative to unsecured and subordinated). Maturity refers to the length of time from the
current date to the maturity or expiration of the loan contract.
Applicant Loan Size: This is the total amount of credit facility (in millions) requested by the firm.
Central Bank Credit Score: This is the credit score reported by the Argentinean Central Bank Public
Credit Registry (Central de Deudores del Sistema Financiero CDSF) for an applicant. The score is based
on the applicant’s past credit history from all the banks in Argentina. The numerical rating is expressed on
a scale of 1 to 5, where: 1= Current/Accrual Basis, 2 = Evidence of Weakness, 3 = Substandard/Timely
Repayment is at Risk, 4 = Doubtful/Timely Repayment is Improbable, and 5 = Uncollectable/Write-off.
Foreign Bank Branch Guarantee: This is a dummy variables that captures when the parent company
of a foreign subsidiary firm or “support provider” provides commercial and financial support to its
local subsidiary. The risks that the bank is asked to assume are only nationalization, expropriation, and
convertibility. Foreign guarantees from the affiliates cover the financial and commercial risk.
Family Company?: This dummy variable captures whether the company is private or in the case of a public
corporation, whether it is controlled by family members.
Product Market Industry Benchmark: This variable is 1 if a firm’s product scope and scale is below a
pre-specified industry benchmark.
Financial Risk Industry Benchmark: This variable is 1 is a firm’s past financial performance is below a
pre-specificed industry benchmark.
Declining Industry: This variable is 1 if a firm belongs to the set of industries pre-specified by the bank as
“declining.” Industries are classified into 27 categories and they are expressed at the 2-digit SIC (Standard
Industrial Classifications) Code.
In addition to the above eight variables, there are 11 additional rule variables (called “exceptions”) that
are used very rarely. In particular, these 11 variables taken together influence the approval level of only 48
firms in sample. They are all binary 0-1 variables and include: Requested Amount Above Pre-Approved Limits
According to Firm Size, Downgrade in Firm’s Credit Score, Increase In Total Facilities Requested, Reported
Risk Event At The Firm, Adverse Change In Risk Profile of The Firm, Adverse Change In Industry, Change
24 J.M. LIBERTI AND A.R. MIAN
in Type of Collateral and/or Degree of Support, Covenant Violations, Qualified Auditors, Loan Documentation
Completion, and Internal Bank Debt Rating Model Not Used.
Appendix B:
Objective information variables
This table reports summary statistics of the objective information variables used in the paper. All sub-
components are reported both in raw form and as an implied rating. The bank translates the ratios according
to a pre-specified formula into a rating that goes from 0 to 22 for the sub-components of the Performance
Index, and from 1 to 6 for the component of the Size Index. For 30 firms the sub-components of the
Performance Index measure are missing, hence the total number of firms is 394. All measures are in book values.
Pre-Tax Interest Coverage is the ratio of (Net Income from Operations – Gross Interest Expense – Income Tax
– Net Sale of Equity)/Gross Interest Expense. Pre-Tax Funds Flow Interest Coverage is the ratio of (Free Cash
Flow – Gross Interest Expense – Income Tax – Net Sale of Equity)/Gross Interest Expense. Funds from
Operations/Total Debt is the ratio of Net Income from Operations/Total Debt where Total Debt is composed
by Short- and Long-Term Debt. Free Operating Cash Flows/Total Debt is the ratio of Free Cash Flow/Total
Debt. Pre-Tax Return on Average Capital is the ratio of (Net Income from Operations – Gross Interest Expense
– Income Tax)/(Total Debt + Net Worth + Minority Interest). Total Debt/Capitalization is the ratio of Total
Debt/(Total Debt + Net Worth + Minority Interest). Current Ratio is the ratio of Total Current Assets over
Total Current Liabilities. Firm Size is the total capitalization of the firm.
Objective Information Variable Mean SD Min. Max. Obs.
Sub-Component of Objective Index (Raw Form)
Performance Index
Pre-tax Interest Coverage (dec.) 4.68 8.12 -8.47 23.62 394
Pre-tax Funds Flow Interest Coverage (dec.) 7.79 10.71 -7.02 30.90 394
Funds from operations/Total Debt (%) 12.18 35.55 -0.73 121.37 394
Free Oper Cash Flow/Total Debt (%) 5.12 14.92 -1.24 50.38 394
Pre-Tax Return on Avg Capital (%) 0.06 0.24 -0.59 0.54 394
Total Debt / Capitalization (%) 0.42 0.28 0.00 0.92 394
Current Ratio (dec.) 1.25 0.65 0.32 2.86 394
Size Index
Firm Size 411,250 4,102,483 -102,675 84,000,000 424
Sub-Component of Objective Index (Implied Rating)
Performance Index
Pre-tax Interest Coverage 10.98 8.01 0.00 22.00 394
Pre-tax Funds Flow Interest Coverage 11.39 7.71 0.00 22.00 394
Funds from operations/Total Debt 10.21 7.86 0.00 22.00 394
Free Oper Cash Flow/Total Debt 10.34 8.66 0.00 22.00 394
Pre-Tax Return on Avg Capital 9.40 8.68 0.00 22.00 394
Total Debt / Capitalization 14.17 6.23 0.00 22.00 394
Current Ratio 7.14 5.86 0.00 22.00 394
Size Index
Firm Size 2.26 1.43 1.00 6.00 424
Standardized Objective Indices
Objective Index 0.00 1.00 -1.90 3.22 424
Performance Index 0.00 1.00 -1.88 2.47 424
Size Index 0.00 1.00 -0.89 2.62 424
Appendix C:
Subjective information variables
This table reports summary statistics of the subjective information variables, the primary components and the
sub-components used in the paper reported in Appendix D. There are a total of 18 primary subjective
information variables, divided across five subjective information categories. The main categories of the
Subjective Index are Industry Risk Assessment, Competitive Position, Management Quality, Risk Management Policies, and
Access to Capital. All ratings are reported by the loan officer and are between 1 (worse) and 7 (best). The Mean
Category is the average across the fields for each of the categories. There are 15 firms that have the sub-
component level information missing, therefore reducing the sample to 409 firms. Furthermore, for three
firms, some Competitive Position variables are not relevant, therefore reducing the sample to 406 firms.
Subjective Information Variable Mean SD Min. Max. Obs.
Industry Risk Assessment
Trend in Output 3.51 0.80 1.00 7.00 409
Trend in Earnings 3.27 0.78 1.00 7.00 409
Cyclicality 3.35 0.81 1.00 7.00 409
External Risks 3.53 0.71 2.00 5.00 409
Mean Category 3.41 0.59
1.75
5.00 409
Competitive Position
Market Position 4.28 1.47 1.00 7.00 407
Product Line Diversity 3.88 1.12 1.00 7.00 408
Operating Cost Advantage 3.46 0.89 1.00 7.00 406
Technology Advantage 3.70 0.92 1.00 7.00 406
Key Success Factors 3.67 0.84 1.00 7.00 406
Mean Category 3.80 0.81 1.00 6.60 408
Management Quality
Professionalism 3.67 0.90 1.00 7.00 409
Systems and Controls 3.66 0.89 1.00 7.00 409
Financial Disclosure 3.72 0.85 1.00 7.00 409
Ability to Act Decisively 3.77 0.80 1.00 7.00 409
Mean Category 3.70 0.75 1.00 6.50 409
Risk Management Policies
Leverage Policy 3.34 0.85 1.00 7.00 409
Liquidity Policy 3.36 0.86 1.00 7.00 409
Hedging Policy 3.60 0.86 1.00 7.00 409
Mean Category 3.43 0.72 1.00 6.30 409
A
ccess to Capital
Capital Markets 3.47 1.11 1.00 7.00 409
Banks 3.77 1.01 1.00 7.00 409
Mean Category 3.62 0.98 1.00 7.00 409
Standardized Subjective Indices
Overall Sub
j
ective Index 0.00 1.00 -3.75 2.32 424
Stron
g
Sub
j
ective Index 0.00 1.00 -3.85 3.58 409
Weak Subjective Index 0.00 1.00 -3.35 3.56 409
Appendix D:
Assessment criteria of subjective information variables
This table reports the assessment criteria of the subjective information variables. These are a series of
qualitative questions answered by the loan officer when completing the loan request dossier. The qualitative
questions and responses are as follows:
BUSINESS RISK ASSESSMENT
1 Competitive Position RR1-2 RR3 RR4 RR5 RR6 RR7
Market Position Below 2% / Minor
Player; Declining
Share
2 to 3% / Minos
Player Over 5% / Known
Player or Established
Niche
Over 10% / Major
Player or Strong
Niche
Over 20% / Dominant Over 50% / Clearly
Dominant
Product Line Diversity Only 1 Declining Line Only 1 Stable Line At least 2 Stable Lines At least 2 Growing
Lines Over 3 Lines Over 3 Growing Lines
Operating Cost Advantage High Cost Producer No Cost Advantages Some Cost
Advantages Has Lowest Local
Costs Achieves Low Global
Costs Global Leader
Technology Advantage Predominantly
Outdated Technology Follower Mostly New;
Upgrading Old Leader in Local
Market Global Player in Some
Areas Global Leader in
Many Areas
Key Success Factors None Strong in Some;
Weak in Others Strong Locally in
Some Factors Strong Locally in All
Factors Global Capabilities in
Most Factors Global Capabilities in
All Factors
2 Management RR1-2 RR3 RR4 RR5 RR6 RR7
Professionalism In Few Positions At Some Key
Positions At Most Key Positions
& Most Levels At all Key Posi- tions
in Operations &
Mana
g
ement
At all Levels in
Operations &
Mana
g
ement
At all Levels With
Extensive Experience
Systems and Controls Largely Absent Unreliable Acceptable Very Reliable and
Strong Meets Highest Local
Standards Meets Highest Global
Standards
Financial Disclosure Unreliable Delayed, Inaccu-rate
or Incomplete Satisfactory Reporting Usually Timely and
Accurate Always Timely and
Accurate Meets Highest Global
Standards
Ability to Act Decisively Hopeless W eak Good, but Untested Good, but Untested Proven to be Strong Proven to be Very
Strong
3 Risk Management
Policies RR1-2 RR3 RR4 RR5 RR6 RR7
Leverage Policy Unlimited Appetite High Tolerance Some Tolerance Low Tolerance Very Conservative Extremely
Conservative
Liquidity Policy No Policy Low Liquidity
Acceptable Maintains Some
Cushion Some Cushion &
Sound Contingency
Plan
Conservative Cushion
& Contingency Plan Extremely Conser-
vative Cushion
Hedging Policy No Hedging Policy /
Speculative Policy Risks Understood but
Most Not Covered Risks Understood but
Not Always Covered Most Risks
Understood; Few
O
en Positions
Most Risks
Understood; No Open
Positions
All Risks Understood;
No Open Positions
4Access to Capital RR1-2 RR3 RR4 RR5 RR6 RR7
Capital Markets No access to Capital
markets Limited Largely to
Domestic Banking Primarily Domes-tic
Banking; Some
Capital Markets
Primarily Domestic;
Some International Wide Access;
Domestic &
International
Wide Access;
Domestic &
International
Banks Bank Cutting Lines;
Some Locked-in No Bank Strongly
Committed or Some
Banks Getting Out
At Least One Bank
Strongly Committed At Least One Bank
Strongly Committed Established Re-
lationships; Strong
Commitments
Established Re-
lationships; Strong
Commitments
5Industry RR1-2 RR3 RR4 RR5 RR6 RR7
Trend in Output Declining Uncertain / Declining Stable Growth Strong Growth Very Strong Growth
Trend in Earnings Declining Uncertain / Declining Stable Growth Strong Growth Very Strong Growth
Cyclicality (Fluctuations) Large & Unpredictable Large Moderate Small Very Limited Very Stable
External Risks Widespread Risks Numerous Critical
Risks Variuos Critical Risks Few Critical Risks Few Risks, Non
Cyclical No Risks
ESTIMATING THE EFFECT OF HIERARCHIES ON INFORMATION USE 25
References
Aghion, P. and J. Tirole. 1997. Formal and Real Authority in Organizations. Journal of Political Economy, 105:1–29.
Baker, G., R. Gibbons, and K.J. Murphy. 1999. Informal Authority in Organizations. Journal of Law, Economics, and Organizations,
15:56–73.
Becker, G. and K.M. Murphy. 1992. The Division of Labor, Coordination Costs, and Knowledge. Quarterly Journal of Economics,
107:1137–1160.
Berger, A., N. Miller, M. Petersen, R. Rajan, and J. Stein. 2005. Does Function Follow Organizational Form? Evidence From the
Lending Practices of Large and Small Banks. Journal of Financial Economics, 76:237–269.
Bolton, P. and M. Dewatripont. 1994. The Firm as a Communication Network. Quarterly Journal of Economics, 109:809–40.
Crawford, D. and J. Sobel. 1982. Strategic Information Transmission. Econometrica, 50:1431–1452.
Credit Policiy Manuals and Credit Program Reports. 1998. Source: International Bank.
Dewatripont, M. and J. Tirole. 2005. Modes of Communication. Journal of Political Economy, 113:1217–1238.
Garicano, L. 2000. Hierarchies and the Organization of Knowledge in Production. Journal of Political Economy, 108:874–904.
Garicano, L. and E. Rossi-Hansberg. 2006. Organization and Inequality in a Knowledge Economy. Quarterly Journal of Economics,
121:1383–1435.
Hayek, F.A. 1945. The Use of Knowledge in Society. American Economic Review, 35:519–30.
Hao, L. and S.Wing. 2004. Delegating Decision to Experts. Journal of Political Economy, 112:s311–s335.
Liberti, J.M. 2004. Initiative, Incentives and Soft Information: How Does Delegation Impact the Role of Bank Relationship
Managers? London Business School, mimeo, IFA Working Paper 404.
MacLeod, B. 2003. Optimal Contracting with Subjective Evaluation. American Economic Review, 93:216–240.
Mian, A. 2006. Distance Constraints: The Limits of Foreign Lending in Poor Economies. Journal of Finance, 61:1465–1505.
Petersen, M. 2004. Objective and Subjective Information: Implications for Finance and Financial Research. Working Paper, mimeo,
Northwestern University.
Radner, R. 1993. The Organization of Decentralized Information Processing. Econometrica, 61:1109–1146.
Sah, R. and J. Stiglitz. 1986. The Architecture of Economic Systems: Hierarchies and Polyarchies. American Economic Review,
76:716–727.
Stein, J. 2002. Information Production and Capital Allocation: Decentralized vs. Hierarchical Firms. Journal of Finance, 57:1891–
1921.
26 J.M. LIBERTI AND A.R. MIAN
Notes
1Hayek (1945 p.524) was perhaps the first to formally emphasize the role that subjective information plays
in decision making: “[...] the sort of knowledge with which I have been concerned is knowledge of the kind which
by nature cannot enter into statistics and therefore cannot be conveyed to any central authority in statistical
form.”
2A formal proof was provided in an earlier version of the paper, which is available from the authors upon
request.
3In 1998 the bank was ranked third in terms of total assets and fifth in terms of net worth among all financial
institutions in Argentina. We have signed a non-disclosure agreement with the institution and therefore cannot
mention in any written document the name of the institution where the data comes from. As per Law 23,928
“Ley de Convertibilidad del Austral” from March 27, 1991, $1 Argentine Peso was equivalent to 1 US Dollar.
4There might still be some room for the loan officer to indirectly manipulate how firms are assigned to
different levels of hierarchy. We shall discuss these issues in greater detail in the next section.
5A loan is aggregated at the firm level.
6These variables are all binary 0-1 variables. They are: Requested Amount Above Pre-Approved Limits
According to Firm Size, Downgrade in Firm’s Credit Score, Increase In Total Facilities Requested, Reported
Risk Event At The Firm, Adverse Change In Risk Profile of The Firm, Adverse Change In Industry, Change
in Type of Collateral and/or Degree of Support, Covenant Violations, Qualified Auditors, Loan Documentation
Completion, and Internal Bank Debt Rating Model Not Used.
7We do not report coefficients on loan officer fixed effects, and their interactions with Subjective index and
Objective index for brevity sake. Also note that due to the inclusion of these interactions, the coefficient on the
subjective and objective indexes only reflects the omitted loan officer category and is hence not shown.
8We also compared basic descriptive statistics for Level 3 firms approved inside and outside the loan officer’s
branch. The firms are in general quite similar, showing that the geographical location of Level 3 officers is not
systematically biased in a particular direction so as to bias our coefficients of interest. Also note that since we
are only using variation from 54 observations to identify our coefficient of interest, we no longer have the power
to put in our usual set of control variables.
9Since we are only using the the 54 firms from Level 3 for identifying our coefficient of interest, we no
longer have sufficient power to include the loan officer fixed effects, rule variables, and their interactions with
information indices.
10Defining the “experience” dummy at the break in the distribution is important for our results. For example,
the triple interaction results lose significance if we define the “experience” above median.
Figure1:Anexampleofbankhierarchicalstructure
Bank Manager
Loan Officer
FIRM
Subjective Information (S)
e.g. “A” in Professionalism
Objective Information (H)
e.g. Audited Financials
High Level
Low Level
Credit approval done by manager
or loan officer.
Loan officer collects subjective and
objective information from the firm.
σ2
qs|L
σ2
qs|M< σ2
qs|L
σ2
qh|L
σ2
qh|M= σ2
qh|L
Quality of Firm, Q
Figure2:Hierarchicaldecisionmakingprocess
Level 1
Level 3
Level 2
Level 4
Level 5
Collects all the information.
Credit approval assigned to one of
the five levels according to credit
manual rules.
Figure3:Empiricalstrategy
Hierarchical
Level
Rule Variable (Z)
Z
-
σ2
qh
σ2
qs
TABLE 1
Summary statistics
The table presents summary statistics for the main variables used in the paper by a loan’s approval level.
Approval Level Rule Variables determine how firms are allocated across approval levels (Levels 1 through 5)
by the bank. Panel A shows the eight primary approval level rule variables. There are an additional 11 rule
variables that are used occasionally. The complete list (and definitions) of rule variables is given in Appendix
A. Panel B provides summary stats on Information Indices that are constructed using information collected
by a loan-officer on an applicant firm. The Objective Index is constructed using a firm’s audited financial ratios
and size (see Appendix B for details), while the Subjective Index is constructed using a subjective assessment of
various firm attributes (see Appendix C for details).
Level 1 Level 2 Level 3 Level 4 Level 5
Number of Firms 113 158 54 57 42
Percentage of Firms 26.7% 37.3% 12.7% 13.4% 9.9%
A
pproval Level Rule Variables
Maturity / Collateral Score 9.61 11.89 14.37 14.42 16.40
Applicant Size (in Million $) 6.06 17.15 17.20 39.05 36.98
Central Bank Credit Score 1.06 1.09 1.24 1.23 1.88
Foreign Bank Branch Guarantee 0.77 0.02 0.01 0.00 0.00
Family Company? 0.06 0.07 0.44 0.25 0.29
Product Market Industry Benchmark 0.13 0.10 0.50 0.61 0.50
Financial Risk Industry Benchmark 0.02 0.05 0.35 0.40 0.33
Declining Industry 0.04 0.02 0.24 0.09 0.00
Other Variables
Total Facilities (in Million $) 5.49 16.26 14.30 34.59 26.41
Total Outstanding (in Million $) 3.03 10.21 8.31 22.24 20.94
Net Sales (in Million $) 57.64 140.41 304.90 488.29 545.68
Net Income (in Million $) 0.56 1.12 14.66 14.24 55.50
Total Assets (in Million $) 64.83 145.87 293.21 862.43 1,333.80
Mean
Objective Index -0.37 -0.19 0.36 0.66 0.34
Performance Index 0.01 -0.08 0.35 -0.03 -0.23
Size Index -0.56 -0.21 0.17 1.00 0.72
Subjective Index -0.32 -0.09 0.02 0.57 0.41
Strong Subjective Index -0.32 -0.14 0.13 0.36 0.48
Weak Subjective Index -0.25 -0.15 0.14 0.40 0.44
Standard Deviation
Objective Index 0.85 1.04 0.78 0.76 1.12
Subjective Index 0.83 0.97 0.87 1.07 1.16
Ob
j
ective Index
(Demeaned at Loan Officer Level)
0.86 1.01 0.71 0.65 1.08
Sub
j
ective Index
(Demeaned at Loan Officer Level)
0.85 0.94 0.82 0.84 0.97
PANEL A: Means of Variables By Approval Level
PANEL B: Mean and Standard Deviation of Information Variables By Approval Level
TABLE 2
Mapping of rule variables to level of approval assignment
This table predicts approval level based on functions of rule variables used in the credit manuals to assign
loan applicant firms to approval levels. The dependent variable is the Approval Level, which varies from 1 to 5.
All regressions include the 19 rule variables described in Appendix A, but coefficients on only the eight
primary rule variables are shown for brevity. Columns (1) to (4) report Ordinary Least Squares (OLS)
estimates while Columns (5) and (6) report Ordered Probit estimates. Column (1) only includes the 19 rule
variables as controls. Column (2) adds 16 pair-wise interactions of the four top rule variables:
Maturity/Collateral Score, Applicant Size, Central Bank Credit Score, and Foreign Bank Branch Guarantee. Column (3)
further includes powers 2 and 3 of the 19 rule variables, i.e., a total of 38 additional controls. Column (4) only
includes the four top rule variables on the right hand side. In Columns (3) and (6), Foreign Bank Branch
Guarantee/Family Company?, and Family Company?, respectively, are dropped due to perfect collinearity.Robust
standard errors clustered at the holding company level are reported in parenthesis.
De
p
endent Variable
(1) (2) (3) (4) (5) (6)
Maturity / Collateral Score 0.09 0.06 -0.39 0.12 0.12 0.51
(0.01) (0.06) (0.20) (0.01) (0.02) (0.14)
Applicant Size 0.01 0.03 0.04 0.01 0.01 0.05
(0.00) (0.01) (0.01) (0.00) (0.00) (0.01)
Central Bank Credit Score 0.31 1.36 6.71 0.29 0.29 1.29
(0.10) (0.51) (2.93) (0.10) (0.12) (0.70)
Foreign Bank Branch Guarantee -0.02 -0.29 -0.32 0.42
(0.12) (0.45) (0.11) (0.14)
Family Company? 0.32 0.25 0.44
(0.16) (0.15) (0.18)
Product Market Industry Benchmark 0.48 0.30 0.29 0.55 0.36
(0.15) (0.13) (0.13) (0.16) (0.17)
Financial Risk Industry Benchmark 0.60 0.41 0.35 0.62 0.45
(0.17) (0.15) (0.15) (0.18) (0.19)
Declining Industry -0.52 -0.67 -0.68 -0.51 -0.81
(0.23) (0.20) (0.20) (0.25) (0.25)
Pair-wise Interaction Of Top 4 Rule
Variables Yes Yes Yes
Powers 2 and 3 of Rule Variables
included? Yes Yes
No. of Obs. 424 424 424 424 424 424
Adj R-Sq / Pseudo R-Sq 0.42 0.54 0.55 0.34 0.19 0.29
OLS Ordered Probit
Approval Level
TABLE 3
Correlation of matrix of information indices and their sub-components
This table reports the correlation between information indices and their sub-components. See Appendix B
and Appendix C for variable description and summary statistics for these variables.
Pre-tax Interest
Coverage
Pre-tax Funds Flow
Interest Coverage
Funds From
Operations / Total
Debt
Free Oper. Cash
Flow/Total Debt
Pre-Tax Return on
Avg Capital
Total Debt /
Capitalization
Current Ratio
Firm Size
Pre-tax Interest Coverage 1.00
Pre-Tax Funds Flow Interest Coverage 0.91 1.00
Funds From Oper/Total Debt (%) 0.67 0.75 1.00
Free Oper. Cash Flow/Total Debt (%) 0.42 0.44 0.61 1.00
Pre-Tax Return on Avg Capital (%) 0.66 0.53 0.62 0.38 1.00
Total Debt / Capitalization (%) 0.40 0.44 0.68 0.47 0.27 1.00
Current Ratio (dec.) 0.20 0.19 0.24 0.14 0.09 0.33 1.00
Firm Size 0.13 0.12 0.04 0.03 0.03 0.03 -0.17 1.00
PANEL A: Correlation Matrix for Sub-Components of Objective Index
Industry Risk
Assessment
Competitive
Position
Management
Quality
Risk
Management
Policies
Access to
Capital
Industry Risk Assessment 1.00
Competitive Position 0.40 1.00
Management Quality 0.44 0.67 1.00
Risk Management Policies 0.41 0.54 0.61 1.00
Access to Capital 0.49 0.64 0.67 0.51 1.00
PANEL B: Correlation Matrix for Sub Components of Subjective Index
Objective Index
Performance
Index
Size Rating
Subjective Index
Strong
Subjective Index
Weak Subjective
Index
Objective Index 1.00
Performance Index 0.73 1.00
Size Rating 0.73 0.06 1.00
Subjective Index 0.46 0.23 0.44 1.00
Strong Subjective Index 0.42 0.20 0.42 0.78 1.00
Weak Subjective Index 0.43 0.22 0.40 0.79 0.77 1.00
PANEL C: Correlation Matrix for Information Indices
TABLE 4
Does reliance of information vary with hierarchical distance?
This table estimates the credit sensitivity to objective and subjective information variables for firms getting credit
approvals at various hierarchical levels within the bank. The dependent variable is the logarithm of the amount of credit
approved, Log(Approved Credit). High Level is an indicator variable that collapses the 5 approval levels into “high” and
“low” around the median. High Level takes a value of 1 for Levels 3, 4, and 5; and a value of 0 otherwise. Columns (2)
through (6) add loan office fixed effects and their interactions with the Subjective and Objective Information variables
(75 additional controls in all). Columns (3), (4), and (6) further add the 19 rule variables used by the bank to assign firms
to different levels and their interactions with Objective and Subjective Information measures (57 controls in all). The 19
rule variables are described in Appendix A. Columns (5) and (6) open up the five hierarchical levels to see how
sensitivity to information changes at each level. The omitted category in these columns is Level 1. Robust standard
errors clustered at the holding company level are reported in parenthesis.
Dependent Variable
(1) (2) (3) (4) (5) (6)
High Level 0.53 0.32 0.15 0.17
(0.15) (0.19) (0.26) (0.27)
Subjective Index 0.41 -- -- -- -- --
(0.08)
Objective Index -0.04 -- -- -- -- --
(0.06)
Subjective Index × High Level -0.43 -0.78 -0.68 -0.60
(0.13) (0.23) (0.31) (0.31)
Objective Index × High Level 0.84 0.92 0.82 0.93
(0.13) (0.29) (0.28) (0.31)
Subjective Index × Level 2 -0.12 0.24
(0.17) (0.33)
Subjective Index × Level 3 -0.87 -0.76
(0.31) (0.40)
Subjective Index × Level 4 -0.91 -0.62
(0.36) (0.42)
Subjective Index × Level 5 -0.88 -0.74
(0.48) (0.53)
Objective Index × Level 2 0.25 -0.25
(0.14) (0.38)
Objective Index × Level 3 1.13 0.92
(0.40) (0.42)
Objective Index × Level 4 1.37 0.70
(0.39) (0.46)
Objective Index × Level 5 1.14 0.57
(0.38) (0.37)
Constant 1.72 -- -- -- --
(0.07)
Rule Variables and Their Information
Interactions Yes Yes Yes
Loan Officer FE and Their
Information Interactions YesYesYesYes Yes
Industry FE, Firm Size, and ROA. Yes
Indicator Variables for each level Yes Yes
No. of Obs. 424 424 424 423 424 424
Adj R-sq 0.27 0.39 0.65 0.73 0.43 0.66
Log (Approved Credit)
TABLE 5
Geographical distance
Panel A reports the joint distribution of the level of approval, and geographical distance between a loan
officer and the ultimate officer approving a loan. Geographic Distance takes a value of 0 if the loan officer
who collects information and the ultimate approving officer are located in the same branch; otherwise it is
coded as 1. Panel B estimates the credit sensitivity to Objective and Subjective Information variables for firms
getting credit approvals from officers located at different facilities. The dependent variable is the logarithm of
the amount of approved credit, Log(Approved Credit).
Geographical Distance 12 3 4 5
0 113 158 17 0 0
100375742
Dependent Variable
Level 3
Onl
y
(1) (2) (3)
0.86 0.89 0.89
(0.13) (0.45) (0.42)
-0.44 -0.36 -0.36
(0.14) (0.38) (0.36)
∗∗
No. of Obs. 424 54 424
Adj R-sq 0.27 0.28 0.34
Other Variables included in regression but
coefficients not shown
Geographical Distance × Objective Index
Geographical Distance × Subjective Index
Log (Approved Credit)
Level of Approval (Hierarchical Distance)
Panel B: Information Use and Geographical Distance
Panel A: Joint Distribution of Hierarchical and Geographical Distance
* Geographical Distance, Objective Index, Subjective Index, and a Constant.
** Geographical Distance, Objective Index, Subjective Index, Approval Level indicator variables, and the interaction of
these indicator variables with the Objective Index and the Subjective Index.
TABLE 6
Does experience help the use of subjective information?
This table tests whether the experience of a loan officer who collects information helps in the use of
subjective information in the credit approval decision. Tenure is a dummy variable that takes a value of 1 when
the years of experience in the bank are seven or more years and 0 otherwise. The dependent variable is the
logarithm of the amount of approved credit, Log(Approved Credit). Column (1) adds loan office fixed effects
and their interactions with Subjective and Objective Information variables (75 additional controls in all).
Column (2) further adds the 19 rule variables used by the bank to assign firms to different levels and their
interactions with Objective and Subjective Information measures (57 controls in all). The 19 rule variables are
described in Appendix A. Robust standard errors clustered at the holding company level are reported in
parenthesis.
Dependent Variable
(1) (2)
Subjective Index × High Level × Tenure 2.16 2.83
(0.56) (0.47)
Objective Index × High Level × Tenure -1.16 -1.70
(0.65) (0.49)
Rule Variables and Their Information
Interactions Yes
Loan Officer FE and Their Information
Interactions Yes Yes
Other Variables included in regression but
coefficients not shown ∗∗
No. of Obs. 424 424
R-sq 0.40 0.66
Log (Approved Credit)
* Subjective Index, Objective Index, High Level, Subjective Index × High Level, Objective Index × High Level, Tenure,
Tenure × Subjective Index, Tenure × Objective Index, and Tenure × High Level.
TABLE 7
Decomposing subjective information
This table checks the robustness of the main results by using different measures of the Subjective Index. We
construct two different measures of overall subjective information rank. AVGsubjective is a simple arithmetic
mean of all the 18 subjective information variables. WAVGsubjective weighs the five categories that form the
Subjective Index equally while giving equal weights to the subjective information variables within each category.
Strong Subjective Index is a standardized average of the categories Management Quality and Competitive Position.
Weak Subjective Index is a standardized average of the categories Industry Risk Assessment, Risk Management
Policies, and Access to Capital. Columns (1), (2), (4), and (5) include loan office fixed effects and their
interactions with Subjective and Objective Information variables (75 additional controls in all). The
dependent variable is the logarithm of the amount of approved credit, Log(Approved Credit). Columns (1), (2),
and (5) further add the 19 rule variables used by the bank to assign firms to different levels and their
interactions with Objective and Subjective Information measures (57 controls in all). The 19 rule variables are
described in Appendix A. Robust standard errors clustered at the holding company level are reported in
parenthesis.
Dependent Variable
(1) (2) (3) (4) (5)
Sub
j
ective Index × Hi
g
h Level -0.68 -0.65
(0.19) (0.19)
Objective Index × High Level 0.85 0.84 0.90 0.87 0.85
(0.25) (0.25) (0.13) (0.27) (0.25)
Weak Subjective Index × High Level -0.18 -0.29 -0.23
(0.18) (0.21) (0.20)
Strong Subjective Index × High Level -0.40 -0.58 -0.52
(0.19) (0.20) (0.19)
Definition of Subjective Rating
Average Weighted
Rule Variables With Information
Interactions Yes Yes Yes
Loan Officer FE and Information
Interactions Yes Yes Yes Yes
Other Variables included in regression
but coefficients not shown ∗∗ ∗∗ ∗∗
No. of obs. 409 409 409 409 409
Adj R-sq 0.73 0.73 0.29 0.42 0.74
Log (Approved Credit)
* Subjective Index, Objective Index, and High Level.
** Objective Index, Weak Subjective Index, Strong Subjective Index, High Level, and a Constant.
... This study highlights the significance of soft information in loan issuance decisions. While existing literature emphasizes banks' reliance on hard information to evaluate borrowers' trustworthiness (Ball et al. 2015), our research diverges by focusing on the role of soft information in loan contracts, a topic of increasing interest in related literature (Agarwal & Ben-David 2018;Bozanic et al. 2018;Liberti & Mian 2009;Lin et al. 2011;Ye et al. 2022). By demonstrating that borrower trustworthiness, as proxied by CEOs' political preferences, mitigates the agency cost of debt, our empirical evidence supports the literature linking borrower trustworthiness to lending decisions (Wang et al. 2015;Hasan et al. 2017;Tang et al. 2018). ...
... Scholars have posited that bank loan managers often rely on qualitative information in their decision-making processes (e.g., Berger et al. 2005;Liberti & Mian 2009). We argue that a borrower's perceived trustworthiness is of considerable importance to banks. ...
... 7 Stein (2002) suggests that if banks' capital allocations are "decentralized, " which might prevail in multi-bank holding companies, then there is greater incentive for the bank's local agents to gather and use soft information. Liberti and Mian (2009) use data from a large multinational bank in Argentina and find that the bank uses soft information in its decisions to lend to large firms. They also find that the negative impact of hierarchical distance, e.g., between the agent collecting the soft information and the manager responsible for the loan approval, on the use of soft information is mitigated by a closer geographic relationship between the two and the manager's personal experience with the agent's information collecting skills. ...
Article
Full-text available
Banks rely on soft information when assessing loan applications, making it crucial to evaluate the trustworthiness of potential borrowers in order to set loan conditions, even in a legal environment where contractual rights are straightforwardly enforced. Given the common belief that certain personality traits—such as trustworthiness, reliability, loyalty, thriftiness, and stinginess—are more often linked to conservatives (Republicans) than to liberals (Democrats), we investigate whether companies with conservative chief executive officers (CEOs) secure more advantageous loan terms compared to others. Our findings indicate that firms with conservative CEOs are able to negotiate bank loans with lower interest spreads and upfront fees. While we do not observe a direct impact of CEO overconfidence on loan pricing, we reveal that the combined influence of CEO conservatism and overconfidence contributes to our primary findings. Additionally, we discovered that conservative CEOs tend to receive more favorable non-price conditions (fewer covenants) and are less inclined to offer collateral.
... Therefore, in the case of diversified credit, commercial banks will spend more time and energy on customer tracking and management, collecting customer information, and accumulating customer and industry experience (A. N. Berger et al., 2005;Liberti & Mian, 2008). It will also result in in associated costs and fees. ...
Article
Full-text available
In recent years, there has been significant interest in economic policy uncertainty and its economic consequences. Fluctuations in economic policies can introduce instability into business operations, resulting in elevated credit default risks. It is important to investigate whether Chinese commercial banks take economic policy uncertainty into account and adopt a more diversified loan structure to mitigate the risks associated with policy changes. In this study, we investigate the impact of economic policy uncertainty on loan concentration within a sample of Chinese commercial banks from 2007 to 2020. Using a panel dataset of 311 banks, we employ fixed-effects regression models and conduct robustness checks with instrumental variables and dynamic panel regressions to address endogeneity concerns. Our findings reveal a significant negative correlation between economic policy uncertainty and the loan concentration of banks. These results hold strong even when subjected to rigorous testing for endogeneity using instrumental variables and dynamic panel regressions. Furthermore, we also find that the negative impact of economic policy uncertainty on loan concentration is more pronounced in regional banks, small-sized banks, and banks with lower market shares. Moreover, the mechanism analysis demonstrates that operational risk serves as a vital channel through which economic policy uncertainty affects loan concentration. By shedding light on how macroeconomic policies impact the financial behavior of commercial banks, we provide new empirical evidence and valuable insights into the dynamics of bank loan structures. In light of our findings, we propose policy recommendations to ensure economic policy stability, encourage the diversification of loan portfolios, expand the regulatory scope on loan concentration, and foster improvements in the regulatory system.
... The latter requires adequate and persistent "relationship banking" with the borrowing firm, its owners and the local community as well as ongoing monitoring of corporate performance across all types of financial relationships, including contracting, spending and financing. Liberti and Mian (2009) argue that the extent of geographical and hierarchical distance between the information-collecting officer and the loan-approving officer affects the extent of reliance on objective versus subjective knowledge. Uchida (2011) finds that, during the loan application screening process, smaller and provincial banks assign higher importance to relationshipbased knowledge and use third-party provided information often, but they also place greater emphasis on the borrowers' collateral value as insurance. ...
Article
Full-text available
Purpose-This study employs firm-level data to evaluate how the knowledge economy impacts the financing constraints of businesses across 106 low-and middle-income nations, focusing on the influence of technological transformation on corporate financing choices. Design/methodology/approach-The research centers on privately held, unlisted firms and examines the distinct effects of knowledge at both the within-country and between-country levels using a panel dataset. Rigorous sensitivity and endogeneity analyses are conducted to ensure the reliability of the findings. Findings-The findings indicate that greater levels of the knowledge economy correlate with reduced financing constraints for firms. However, this effect varies depending on the location within a country and across different geographical regions. Firms situated in larger urban centers and more innovative regions reap the most significant benefits from the knowledge economy when seeking external funding. Conversely, firms in smaller cities, rural areas and regions characterized by structural and institutional inefficiencies in knowledge generation experience fewer advantages. Originality/value-The impact of knowledge exhibits variability not only within and among countries but also between poor and affluent developing nations, as well as between larger and smaller countries. The knowledge effect on firms' access to external finance is influenced by factors such as financial openness and development, educational quality, technological absorption capabilities and agglomeration conditions within each country.
... Second, their relatively small size and horizontal organizational structure allow NCDFIs to delegate significant authority to loan officers. This is crucial for acquiring, transmitting, interpreting, and ultimately leveraging soft information (see, e.g., Stein 2002;Liberti and Mian 2009;Uchida et al. 2012;Ogura and Uchida 2014;Liberti 2018). ...
Article
Full-text available
Native Community Development Financial Institutions (NCDFIs) promote financial inclusion in financially underserved Native communities by adopting innovative lending strategies, including designing their own soft-information-based measures of borrower risk. Drawing on business loan data from one prominent NCDFI, a nonprofit loan fund, we examine to what extent the NCDFI-generated borrower risk measures help explain the NCDFI's loan performance and pricing above and beyond the effect of the credit score, a conventional credit-bureau-produced, hard-information-based metric. All else equal, both loan delinquency hazard and loan interest rate are robustly predicted by one of the NCDFI's two proprietary soft-information-based measures, the character score, but do not vary with the other one, commitment to business score. The credit score is an important determinant of loan delinquency hazard but, all else equal, does not exhibit a detectable relationship with the loan interest rate. We do not find evidence of noteworthy interactions among the three NCDFI-used borrower risk measures. Our study offers evidence in support of the unique underwriting practices and relationship-based lending operations that characterize the NCDFI industry.
Article
We study the broad question about the sources of FinTech lending growth by examining a specific representative product for which the technologies of both FinTech and the incumbent competitors can be identified and compared—small business lending. We test whether the presence of incumbents employing different technologies affects FinTech penetration, and find more FinTech lending where large/out-of-market banks are more prevalent. Using stress test exposures and Community Reinvestment Act examinations as instruments, we find that FinTech credit more often substitutes for loans by large/out-of-market banks than small/in-market banks. Results are consistent with FinTech advantages in processing hard information, rather than hardening soft information. (JEL G21, G23, O33)
Article
In this paper, we adopt a mixed‐methods approach that combines case studies and fuzzy‐set qualitative comparative analysis (fsQCA) to analyse the collaborative efforts of financial technology (fintech) and large banks in mitigating small and medium‐sized enterprise (SME) financing problems. The findings demonstrate that fintech has challenged the traditional lending structure of “large bank–large firms, small bank–small firms” by reducing information asymmetry between banks and enterprises and enhancing technological capabilities. The inherent advantages of large banks, including economies of scale, risk diversification, and talent attraction, can accelerate fintech's role in promoting inclusive finance for SMEs. Overall, our results underscore the importance of fintech applications and the active involvement of large banks in addressing SME financing challenges.
Article
When the Federal Reserve first paid interest on excess reserves (IOER) in October 2008, banks faced a choice to earn a “better than” risk‐free rate, or lend to earn a higher, riskier rate. Evidence suggests the “reserves‐lending puzzle” is not driven by endogeneity from reverse causality, flight to safety, or increased Treasury supply, but by the introduction of the “reserve premium” (IOER‐3MT), which is associated with a reduction of domestic bank‐level lending by ‐5.1% (‐$420.2B). Findings suggest the reserves risk channel can aid in restricting inflation. Additionally, recent Senior Financial Officer Surveys corroborate the conclusions presented in this paper.
Article
In a sample representing 40% of Peru’s formal economy, firms shocked by the death of their chief executive officer experience lower sales and a higher probability of exit, lower borrowing amounts and credit performance, and a reduced number of lenders compared with very similar firms. Existing lenders bear the negative consequences. Monthly risk assessments by lenders worsen immediately after the death events. The negative effects are observed across various firm sizes, while industry characteristics, incorporation, and family ties are relevant dimensions of heterogeneity. Lenders with a longer relationship history, and those less constrained for loan-making, protect themselves by introducing collateral clauses. (JEL D25, G30, L25).
Article
Full-text available
We present a model of delegation with self-interested and privately informed experts. A team of experts with extreme but opposite biases is acceptable to a wide range of decision makers with diverse preferences, but the value of expertise from such a team is low. A decision maker wants to appoint experts who are less partisan than he is in order to facilitate information pooling by the expert team. Selective delegation, either by controlling the decision-making process or by conditioning the delegation decision on his own information, is an effective way for the decision maker to safeguard own interests while making use of expert information.
Article
We assert that decision rights in organizations are not contractible: the boss can always overturn a subordinate's decision, so formal authority resides only at the top. Although decision rights cannot be formally delegated, they might be informally delegated through self-enforcing relational contracts. We examine the feasibility of informal authority in two informational environments. We show that different information structures produce different decisions not only because different information is brought to bear in the decision-making process, but also because different information creates different temptations to renege on relational contracts. In addition, we explore the implications of formal delegation achieved through divestitures. Copyright 1999 by Oxford University Press.
Article
How does organizational design shape the behavior of banks? Using a panel data of over 1,600 banks in 100 emerging economies, we identify the strengths and weaknesses of the three dominant organizational designs in emerging markets. Private domestic banks have an advantage in lending to "soft information" firms which allows them to lend more, and at higher rates without substantially higher default rates. On the other hand, foreign banks have the advantage of access to "external liquidity" from their parent banks which lowers their deposit cost. The external support however comes at the cost of foreign banks being limited (presumably by the parent banks) to only lend to "hard information" firms. Whereas foreign and private domestic banks have their own comparative advantages and disadvantages, we find that government banks perform uniformly poorly, and only survive due to strong government support.
Article
Theories based on incomplete contracting suggest that small organizations have a comparative advantage in activities that make extensive use of “soft” information. We provide evidence consistent with small banks being better able to collect and act on soft information than large banks. In particular, large banks are less willing to lend to informationally “difficult” credits, such as firms with no financial records. Moreover, after controlling for the endogeneity of bank-firm matching, we find that large banks lend at a greater distance, interact more impersonally with their borrowers, have shorter and less exclusive relationships, and do not alleviate credit constraints as effectively.
Article
This paper considers specialization and the division of labor. A more extensive division of labor raises productivity because returns to the time spent on tasks are usually greater to workers who concentrate on a narrower range of skills. The traditional discussion of the division of labor emphasizes the limitations to specialization imposed by the extent of the market. We claim that the degree of specialization is more often determined by other considerations. Especially emphasized are various costs of “coordinating” specialized workers who perform complementary tasks, and the amount of general knowledge available.
Article
What is the problem we wish to solve when we try to construct a rational economic order? On certain familiar assumptions the answer is simple enough. If we possess all the relevant information, if we can start out from a given system of preferences, and if we command complete knowledge of available means, the problem which remains is purely one of logic. That is, the answer to the question of what is the best use of the available means is implicit in our assumptions. The conditions which the solution of this optimum problem must satisfy have been fully worked out and can be stated best in mathematical form: put at their briefest, they are that the marginal rates of substitution between any two commodities or factors must be the same in all their different uses. This, however, is emphatically not the economic problem which society faces. And the economic calculus which we have developed to solve this logical problem, though an important step toward the solution of the economic problem of society, does not yet provide an answer to it. The reason for this is that the “data” from which the economic calculus starts are never for the whole society “given” to a single mind which could work out the implications, and can never be so given. The peculiar character of the problem of a rational economic order is determined precisely by the fact that the knowledge of the circumstances of which we must make use never exists in concentrated or integrated form but solely as the dispersed bits of incomplete and frequently contradictory knowledge which all the separate individuals possess.
Article
This paper develops a theory of the allocation of formal authority (the right to decide) and real authority (the effective control over decisions) within organizations, and it illustrates how a formally integrated structure can accommodate various degrees of "real" integration. Real authority is determined by the structure of information, which in turn depends on the allocation of formal authority. An increase in an agent's real authority promotes initiative but results in a loss of control for the principal. After spelling out (some of) the main determinants of the delegation of formal authority within organizations, the paper examines a number of factors that increase the subordinates' real authority in a formally integrated structure: overload, lenient rules, urgency of decision, reputation, performance measurement, and multiplicity of superiors. Finally, the amount of communication in an organization is shown to depend on the allocation of formal authority.
Article
The paper develops a theory of costly communication in which the sender's and receiver's motivations and abilities endogenously determine the communication mode and the transfer of knowledge. Communication is modeled as a problem of moral hazard in teams, in which the sender and receiver select persuasion and message elaboration efforts. The model is shown to provide a rich set of insights concerning (i) the impact of incentive alignment on communication strategies, (ii) the relative influence and the complementarity/substitutability between issue-relevant communication and cues (information that relates to the credibility of the sender rather than to the issue at stake), and (iii) the path dependency of communication.
Article
This paper studies how communication allows for the specialized acquisition of knowledge. It shows that a knowledge-based hierarchy is a natural way to organize the acquisition of knowledge when matching problems with those who know how to solve them is costly. In such an organization, production workers acquire knowledge about the most common or easiest problems confronted, and specialized problem solvers deal with the more exceptional or harder problems. The paper shows that the model is consistent with stylized facts in the theory of organizations and uses it to analyze the impact of changes in production and information technology on organizational design.
Article
This paper develops a theory of the allocation of formal and real authority within organizations. Real authority is determined by the structure of information, which in turn depends on the allocation of formal authority. An increase in an agent's real authority promotes initiative but results in a loss of control for the principal. The paper analyzes the allocation of formal authority as well as some determinants of the subordinates' real authority: overload, lenient rules, urgency of decision, reputation, performance measurement, and multiplicity of superiors. Finally, the amount of communication in an organization is shown to depend on the allocation of formal authority. Copyright 1997 by the University of Chicago.