View original document

The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.

Federal Reserve Bank
of Chicago
.reu"‘S St. LOU'S

Thirrl Onorf/^r


Determinants of supplier plant location:
Evidence from the auto industry


Switching primary federal regulators:
Is it beneficial for U.S. banks?


Financial constraints and entrepreneurship:
Evidence from the Thai financial crisis


Seasonal monetary policy



Michael H. Moskow
Senior Vice President and Director of Research
Charles Evans
Research Department
Financial Studies
Douglas Evanoff, Vice President
Macroeconomic Policy
David Marshall, Vice President

Microeconomic Policy
Daniel Sullivan, Vice President
Payment Studies
Richard Porter, Vice President
Regional Programs
William A. Testa, Vice President

Economics Editor
Craig Furfine, Economic Adviser
Helen O’D. Koshy

Associate Editors
Kathryn Moran
Han Y. Choi

Rita Molloy
Julia Baker
Economic Perspectives is published by the Research
Department of the Federal Reserve Bank of Chicago. The
views expressed are the authors’ and do not necessarily
reflect the views of the Federal Reserve Bank of Chicago
or the Federal Reserve System.

© 2005 Federal Reserve Bank of Chicago
Economic Perspectives articles may be reproduced in
whole or in part, provided the articles are not reproduced
or distributed for commercial gain and provided the source
is appropriately credited. Prior written permission must
be obtained for any other reproduction, distribution,
republication, or creation of derivative works of Economic
Perspectives articles. To request permission, please con­
tact Helen Koshy, senior editor, at 312-322-5830 or email
Economic Perspectives and other Bank
publications are available on the World Wide Web
at http:/

& chicagofed. org
ISSN 0164-0682


Third Quarter 2005, Volume XXIX, Issue 3


Determinants of supplier plant location:
Evidence from the auto industry
Thomas Klier
This article analyzes the geography of the auto parts sector in North America. Drawing on a large
plant-level data set it shows an industry that is very spatially concentrated. Formal models of plant
location highlight the role of transportation infrastructure as well as the importance of being within
a day’s drive of the assembly plant customer in the location choices of auto supplier plants.


Switching primary federal regulators:
Is it beneficial for U.S. banks?
Richard J. Rosen
This article examines the impact of switching primary federal regulators on banks’ return and risk,
using data from 1977 to 2003. Return increases and risk changes minimally for banks that switch
regulators from 1992 to 2003, while there is no significant change in either return or risk for banks
switching earlier. The improved performance at banks switching between 1992 and 2003 is evidence
for beneficial competition among regulators, and the absence of an increase in risk throughout the
sample period is inconsistent with a “race for the bottom” among regulators.


Financial constraints and entrepreneurship:
Evidence from the Thai financial crisis
Anna L. Paulson and Robert M. Townsend
Using their own data gathered in Thailand from 1997 to 2001, the authors show that ignoring labor
market conditions in empirical studies of financial constraints and entrepreneurial activity can lead
researchers to conclude that financial constraints are not important when in fact they are.


Seasonal monetary policy
Marcelo Veracierto
This article uses a dynamic general equilibrium cash-in-advance model to study the role of monetary
policy in U.S. seasonal cycles. The article finds that the seasonal monetary policy regime is largely
irrelevant: Smoothing interest rates across the seasons or following a constant growth rate of money
lead to basically the same real allocations. Only nominal interest rates are significantly affected.

Determinants of supplier plant location:
Evidence from the auto industry

Thomas Klier

Introduction and summary
The auto industry in the United States directly employs
over 1 million workers, and is so large that gross mo­
tor vehicle output alone represents more than 3 per­
cent of the U.S. economy. In discussing its fortunes,
however, we often focus on the assembly segment of
the industry. Assembly-related activities represent only
the most visible part of this industry, the tip of the ice­
berg, if you will. Below the waterline lies the entire
supply structure that ultimately feeds into the assembly
line, at the end of which rolls off a car or light truck.
That part of the industry, which encompasses every­
thing from inputs such as steel coils to the subassem­
bly of entire vehicle interiors, is larger, both by count
of plants and employment, than the assembly part of
the industry.1 Yet our understanding of the auto sup­
plier industry is quite limited, mostly due to the nois­
iness of the publicly available data for that sector.2
From numerous trade and business press stories,
we know that the way auto suppliers relate to their
assembly customers has fundamentally changed over
the last 20 years. The main driver was the arrival of
lean manufacturing, a production system aimed at the
elimination of waste in every area of production in­
cluding product design, supplier networks, and factory
management, in North America during the early 1980s.
Since then, lean manufacturing production techniques
have become standard practice for auto assembly as
well as the largest supplier companies. Some auto as­
semblers even operate “supplier support organizations”
in order to transfer technology and knowledge to im­
prove the efficiency of operations at their suppliers.
Furthermore, assemblers no longer interact directly
with most of their suppliers. The number of indepen­
dent supplier plants assembly companies work with
directly has fallen greatly during the last ten years to
15 years. In turn, many suppliers now supply prima­
rily other supplier plants. At the same time, the Big


Three automakers, notably Ford and General Motors
(GM), have increased the share of parts they procure
from outside their company. For example, both Ford
and GM spun off many of their own parts plants as
independent companies several years ago. In addition,
the remaining assembler-owned parts plants have ex­
perienced rather dramatic job reductions over the last
few years (Klier, 2005). Finally, this industry, like most
manufacturing industries, has become noticeably
more international. As producers of cars and light
trucks pursue a global manufacturing footprint, their
main suppliers need to be able to meet the needs of
the assemblers globally (Roland Berger, 2004).
In estimating models of supplier plant location,
this article contributes to the cunent discussion of the
changing geography in the U.S. auto industry. The
ongoing loss of market share by the domestically head­
quartered producers to foreign-headquartered producers
of vehicles, both through imports as well as production
in the U.S., raises important questions about the location
trends for the industry (Klier, 2005).3 Between the
first quarter of 2000 and the first quarter of 2005, the
U.S. share of light-vehicle sales by Big Three name­
plates has fallen from 67.9 percent to 57.8 percent.
While some of that market share loss is attributable
to a rise in imports, most of it is explained by increased
U.S. production of foreign-headquartered assembly
companies. This matters for the geography of this in­
dustry as most of these “new domestic” assembly plants
in North America tend to be located farther south
than the assembly plants of the traditional domestic
Thomas Klier is a senior economist in the Economic
Research Department at the Federal Reserve Bank of
Chicago. The author would like to thank Jeff Campbell,
Craig Furfine, and Dan McMillen for helpful comments
and Cole Bolton, Anna Gacia, Joanna Karasewicz, Paul
Ma, and Alexei Zelenev for excellent research assistance.

3Q/2005, Economic Perspectives

producers. In fact, the assembly plants opened most
recently, such as the Honda plant in Lincoln, Alabama,
and the Nissan plant in Canton, Mississippi, have been
situated in the most southern area of the auto region.
As the geography of the auto sector continues to change,
one wonders whether Detroit can continue to be the
hub of this industry over the medium-term horizon.4
The public policy issues of a changing location pat­
tern in the auto sector are considerable as the traditional
auto states are struggling with this southward shift of
auto production and related economic activities.5 For
example, Michigan is currently suffering from its heavy
exposure to the domestic auto and parts makers. In her
2005 State of the State address, Michigan Governor
Jennifer Granholm proposed a sizable bond issue to
attract and retain jobs in the state. The business press
reported recently that Michigan is heavily recruiting
Toyota to locate one of two currently proposed assem­
bly facilities in the state (Hakim, 2005).
This article utilizes detailed plant-based data on
the U.S. auto supplier industry. After describing the
spatial properties of this data, I estimate two simple
models of plant location.61 find the auto industry to
be strongly spatially concentrated. The core of the
auto region is densely packed with plants, reaching
from Michigan up into Ontario, west to Chicago, and
south to northern Alabama and into the Carolinas.
The states within the auto region show variations along
a number of dimensions. For example, the northern
half of the auto region is more densely populated by
domestic supplier plants7 whereas foreign plants are
more concentrated in the southern half. That pattern
is not surprising as it replicates the regional distribu­
tion of assembly facilities. Union plants are concen­
trated in Michigan, Indiana, and Ontario. Larger plants,
however, tend to be located farther away from Detroit.
A plant-level model of employment shows that plants
located farther from Detroit tend to have larger em­
ployment, as do tier 1 (discussed in detail later in the
text) and foreign-owned plants. In addition, I find plant
size to vary by type of part produced. Modeling plant
location choices of recently opened supplier plants at
the county level consistently finds the presence of an
interstate highway to be significantly related to plants
locating in such counties. In addition, the size of the
market, as measured by the number of assembly plants
within a day’s drive (approximately 450 miles) from
a county, is positively related to the number of recent­
ly opened plants in a county.

Literature review
Economic interest in agglomeration issues goes
back to at least Alfred Marshall (1920); for more

Federal Reserve Bank of Chicago

recent research, see Krugman (1991) and Ellison and
Glaeser (1997).
Regarding the question of what drives the geogra­
phy of the auto industry, a number of studies address
the reconcentration of assembly plants in the Mid­
west, a development which started in the mid-1970s.
Rubenstein (1992) attributes this to the demise of the
branch plant system, which was based on producing
identical models in plants located close to population
centers. The subsequent reconcentration of assembly
plants in the heart of the country was driven by an
increase in the choice of models available to the con­
sumer that far outpaced the growth of the market, re­
sulting in much reduced production runs per model.
As a result, individual models tend to support only a
single assembly plant. That plant is then best located
in the heart of the country, as the final product has to
be shipped all over the country from that one produc­
tion location.
Geographic trends in the supplier industry have
followed a different pattern. While this part of the auto
industry has remained remarkably concentrated in
the Midwest since the industry’s beginning over 100
years ago, it has experienced a migration of mostly
labor-intensive parts to the southern U.S. and Mexico
for some time. For example, in 2002, 73 percent of
all wiring harnesses—gatherings of electrical wires
terminating in a central plug that distribute electricity
in a car to operate the turn signals, brake lights, etc.—
“consumed” in the U.S. were imported, 82.7 percent
of which were produced in Mexico.
There is evidence that, within the auto region,
assembly and supplier plants want to locate in prox­
imity to one another (see Smith and Florida, 1994, for
a model for Japanese-affiliated manufacturing estab­
lishments in auto-related industries). State of the art
supply chain management requires most supplier plants
to be located within a day’s drive from the assembly
plant customer (see Klier, 1999, and 2005). And so,
supplier networks of individual assembly plants are
of a regional nature, as the existing transportation in­
frastructure allows for reliable on-time delivery of
products (see Woodward, 1992, and Smith and Florida,
1994, for the importance of highway transportation).
Yet, as the auto industry continues to be very
highly concentrated across space, the geographic
extension of its core region has changed. No longer
reaching eastward from Detroit to Pennsylvania and
New York, it now is defined in a marked north-south
direction, extending from Detroit to Kentucky and
Tennessee and beyond with fingers reaching north
into Canada and south into Mexico. In other words,
the core auto region has pivoted around Detroit over


several decades. During the last few years this devel­
opment has gained greater attention as the old-line auto
states have been losing production and employment
to the southern end of the auto corridor. The chang­
ing fortunes of domestic and foreign assembly plant
customers appear to be profoundly reshaping the re­
gional distribution of supplier employment (Klier, 2005).

How to measure the auto supplier industry?
Overview of the supplier industry
For the purpose of this article, auto suppliers are
companies that supply light-vehicle assembly com­
panies.8 Among them, one can distinguish the follow­
ing categories: suppliers that deal directly with the
assembly company and those that deal primarily with
other suppliers. The first category is commonly re­
ferred to as tier 1 suppliers, while the other category
is referred to as tier 2 suppliers. The number of tier 1
suppliers has been shrinking over the last decade, as
assemblers have been reducing the number of com­
panies they do business with directly. At the same time,
that segment of the supplier industry has been sub­
ject to a series of mergers and acquisitions. Finally,
there are a number of tier 1 parts operations that are
owned and operated by the assemblers themselves,
such as engine and stamping facilities. These are
generally referred to as captive suppliers. A number
of years ago the two largest U.S. assemblers decided

to spin off the majority of their captive parts opera­
tions. In 1999, GM spun off most of its captive plants
as Delphi, which instantly became the largest inde­
pendent tier 1 auto parts supplier. One year later, Ford
Motor Company divested a large number of its cap­
tive plants as a separate company called Visteon. It
then became the second largest independent parts
supplier in North America.9 Table 1 lists the 15 largest
auto supplier companies as ranked by the industry
weekly Automotive News in 2003 based on sales in
North America. The 50 largest suppliers on that list
each have global sales exceeding $1 billion, amounting
to a total of about $285 billion. If one classifies these
companies based on the location of their headquarters,
the following pattern emerges: 53 percent of the 150
largest suppliers represent companies based in one of
the NAFTA (North American Free Trade Agreement)
countries, 20 percent are from Japan, and the remain­
ing 27 percent are from Europe. This illustrates the
degree of global competition present in this industry.

Plant-level data
The analysis of auto supplier plants presented in
this article is based on data acquired from ELM Inter­
national, a Michigan-based vendor. While not designed
with research applications in mind, the ELM database
is intended to cover auto supplier companies and their
plants in North America.10 The database provides 3,542
plant-level records. Included is information on a plant’s


Largest auto supplier companies, 2003
OEM automotive parts sales ($ bn.)


Company name
Delphi Corp.
Visteon Corp.
Lear Corp.
Magna International
Johnson Controls Inc.
Dana Corp.
Robert Bosch Corp.
TRW Automotive Inc.
Denso International America Inc.
ThyssenKrupp Automotive AG
American Axle
Collins & Aikman
DuPont Automotive
Continental AG
Yazaki North America

HQ in





North America






Note: OEM is original equipment manufacturer; CDN is Canada; GER is Germany; and J is Japan.
Source: Automotive News, available at, by subscription.


3Q/2005, Economic Perspectives

address, employment, parts produced, customer(s),
union status, as well as square footage. In order to
clean up the data for research purposes, several oper­
ations were performed. First, records were cross-checked
with state manufacturing directories to obtain informa­
tion on the plant’s age.11 We also appended information
on the nationality of the company to the record of each
plant from the ELM company-level data.12 Plants of
supplier companies listed in the 2003 Automotive
News “top 150 automotive suppliers list” were coded
with the companies’ ranks in that listing. Information
on captive parts plants was also checked with Harbour
(2003). For all the Automotive News top 150 compa­
nies, the accuracy and completeness of ELM’s plant
listings—that is, the number of plants as well as their
location—was crosschecked with the companies’ web­
sites when possible.13 Overall, that resulted in a net
addition of 335 records. Finally, the accuracy of the
employment for the largest plants (employment greater
2,000) was also checked with company websites or
phone calls. After this preparation the data consists
of 3,877 observations of auto supplier plants located
in the U.S. and Canada (see table 2).14 To my knowl­
edge, this may well be the most accurate plant-level
description of the North American auto supplier in­
dustry currently available.
Table 2 summarizes the supplier plant data for the
U.S. and Canada along several dimensions. Of the
3,877 plants more than half are characterized as low­
er tier suppliers. That is, they primarily do business
with other supplier companies. These plants tend to be
smaller (their average employment is 241) than tier 1
suppliers (average employment of 388), which make
up 42 percent of all plants. Captive suppliers, while
small in numbers, represent by far the largest plants.
Their average employment is above 1,000. Of the

three groups, captive plants tend to be located closest
to Detroit. The union variable covers only 83 percent
of all plants; 25 percent are unionized, while 58 per­
cent are not. Unionized plants have larger employment
and are located closer to Detroit than nonunion plants.
As for ownership, just under 80 percent of supplier
plants are part of a company that has its headquarters
in the U.S., Canada, or Mexico. “Foreign” plants are
larger and are located farther away from Detroit than
“domestic” plants. Finally, a quarter of the plants ap­
pears to be single-establishment firms.15 These plants
show the lowest average employment of all groups
listed in table 2.

Spatial characteristics of the auto
supplier industry
This plant-level data allows a fairly detailed de­
scription of the spatial properties of the auto supplier
industry. Figure 1 shows the distribution of auto sup­
plier plants. It represents all 3,877 U.S. and Canadian
plants in the data set, aggregated to the zip code level
of detail. The symbols representing supplier plants are
scaled to convey the spatial density of plant locations.
The most interesting feature of the map is the
high degree of clustering exhibited by this industry.
It is self-evident that southern Michigan represents
the hub of the North American auto sector.16 The core
region of this industry extends from that area west to
Chicago, northeast to Toronto, and south to Tennessee
and arguably into northern Mississippi, Alabama,
Georgia, and the Carolinas.17 Pennsylvania represents
the link between the heart of the industry in the
Midwest and a cluster on the East Coast. West of the
Mississippi the country is mostly empty of auto sup­
plier activity except for a thinly populated band that
extends from eastern Texas and northern Louisiana


Supplier data summary, U.S. and Canada, 2003

Tier 1 suppliers
Captive suppliers
Lower tier suppliers
Single plant

% of

% of





Median distance
to Detroit (miles)

Note: Based upon 3,877 observations at auto supplier plants.

Federal Reserve Bank of Chicago



north to Nebraska and Iowa and into Minnesota. Other
than that, one can observe two clusters in California,
one in the Bay area and the other in the L.A. basin.
Finally, Utah, Colorado, Arizona, and New Mexico
are home to small localized clusters, and the border
between Texas and Mexico shows centers of activity
around El Paso and Laredo/Brownsville. These are
related to border crossings that link the Mexico-based
maquiladora plants to the U.S.-based suppliers.18
Table 3 provides further detail on the distribution
ofplants and employment in the auto supplier indus­
try. The information is first summarized by the four
Census regions plus Canada (see panel A). The bot­
tom panel of the table provides an alterantive break­
down of the data, focusing on the two halves of the
auto corridor. Column 2 shows that 90.1 percent of
all 3,877 plants are located in the Midwest, South, or
Canada. Michigan alone is home to 22.5 percent of
all auto supplier plants, followed by Ohio (11.6 per­
cent) and Ontario (10.7 percent). The auto corridor
as a group represents just under 79 percent of all auto
supplier plants in the U.S. and Canada. Columns 3-8
of table 3 provide three different breakdowns of the
location of auto supplier plants.
Grouping supplier plants by nationality of com­
pany, one can see that the auto corridor consists of two
halves: The northern end shows a higher concentration


of domestic plants (64.7 percent) and lower concen­
tration of foreign-owned plants (46.7 percent) than
overall. Likewise, the southern end shows a much
higher concentration of foreign-owned supplier plants
(33.7 percent) and a smaller share of domestics (13.8
percent). In addition, 21.5 percent of domestic auto­
motive supplier plants in the U.S. and Canada (and
19.6 percent of foreign ones) are located outside the
auto corridor. The share of foreign supplier plants lo­
cated at the southern end of the auto corridor is 2.4
times as large as the share of domestic plants. This
pattern suggests an influence of the location of the pri­
mary customer on the supplier plant location (Klier,
1999, and Smith and Florida, 1994). The median dis­
tance of foreign-owned supplier plants to Detroit is
309 miles, noticeably larger than the 210 miles for
domestic supplier plants (see table 2).19 One can argue
that in setting up operations in North America, foreign
suppliers choose locations close to foreign-owned as­
sembly plants, which presumably were their prime
customers at that time.
The tier status of a supplier plant is measured by
its inclusion in Automotive News’ top 150 supplier com­
panies list. That is a somewhat arbitrary yet plausible
way to define which plants are tier 1 plants. In essence,
it assumes that all of the large supplier companies’ plants
deal directly with assembly plants. Since captive

3Q/2005, Economic Perspectives

Fe de ra l R eserv e Ban k o f Ch icag o


Distribution of plants and employment by region, 2003
A. By Census region
Plant count






Tier 1




Employment count










Tier 1






























































































B. By auto corridor location

Auto corridor
Auto corridor
rest of US/CDN

Notes: Seventeen percent of plants have no information on their union status. Therefore, this comparison (columns 6, 7,13, and 14) only applies to 83 percent of the records.
States not listed do not have automotive supplier plants located in them.
Midwest: IA, IL, IN, KS, Ml, MN, MO, NE, OH, SD, Wl
Northeast: CT, MA, ME, NH, NJ, NY, PA, Rl, VT
South: AL, AR, DE, FL, GA, KY, LA, MD, MS, NC, OK, SC, TN, TX, VA, WV
West: AZ, CA, CO, NM, NV, OR, UT, WA
Auto corridor North: IL, IN, Ml, OH, Ontario, Wl
Auto corridor South: AL, GA, KY, MS, NC, SC, TN
Source: Automotive News, available at, by subscription.

suppliers tend to interact directly with assembly plants,
they are grouped with tier 1 plants in table 3. While
generally very similar in their regional distribution,
tier 1/captive plants are more prevalent in the South
and less so in the Northeast.
Table 3 also shows a disproportionate concentra­
tion of unionized supplier plants in the Midwest and
Ontario.20 Nonunionized plants, on the other hand, are
concentrated in the South where many states have right
to work laws. Within the auto corridor, this split shows
very strongly. Seventy-two percent of all union plants
are found in the northern end of the auto corridor.
Correspondingly, they are quite rare in the southern
end (7.8 percent of all unionized plants versus 21.4
percent of all nonunionized plants).
The location of employment, shown in columns
9-15, resembles the location of plants, column 2, very
closely in the aggregate. The auto corridor is home to
76.6 percent of the industry’s employment and 78.8
percent of its plants. At a more disaggregate level,
table 3 reveals a regional difference in the geography
of plants and employment, indicating that plants lo­
cated in the northern end of the auto corridor tend to
have, on average, fewer employees. For example, em­
ployment at foreign-owned plants is noticeably more
concentrated in the southern half of the auto corridor
than employment at domestic plants. The foreign-owned
plants located in the south also tend to be dispropor­
tionately large, as measured by employment. They
represent 33.7 percent of all plants, yet 36.2 percent
of all employment in the sector. In contrast, both do­
mestic and foreign-owned plants located in the north­
ern half are disproportionately smaller; that is, they
represent a smaller share of industry employment
than of plants. However, that pattern does not apply
to unionized plants. For example, Michigan is home
to 26.9 percent of unionized plants and 29.1 percent
of employment at unionized plants.

Formal analysis of employment and
plant distribution
This section reports on two formal models to es­
timate the location of employment as well as plant
distribution. The idea is to formally test what underlies
the observed agglomeration in the auto supplier industry.
The models utilize data on U.S. plant locations only.
Table 4 lists the summary statistics for both the plantlevel as well as the county-level models reported.
First, we regress p/u«Z-level employment on a
number of plant-level characteristics that the detailed
database allows us to draw on. The model also uses a
number of variables that are measured at the county
level, such as the presence of an interstate highway.


The model incorporates that information only for coun­
ties in which plants are actually located. That explains
why the mean of the interstate highway variable is
0.78 in the plant-level model: 78 percent of plants
are located in counties that are reached by an inter­
state highway.
The geography of plants is measured by two dif­
ferent variables. DISTANCE measures the straightline distance between the centroid of the zip code in
which the supplier plant is located and the centroid
of the zip code for downtown Detroit.21 Detroit seems
an obvious spatial reference point as it is clearly the
hub of this industry. VDISTANCE measures distance
to Detroit only in the north-south direction. In addition,
the following set of plant characteristics is included
in the model. A set of dummy variables indicating if
the plant is part of a single plant company; if it is
part of one of the largest 150 supplier companies;22
if it is an assembler-owned supplier plant (CAPTIVE)',
if it is unionized;23 and if its headquarter operations
are located outside North America. In addition, a group
of dummy variables controls for what subsystem of
the car the plant’s output feeds into (table 5, p. 10).24
Finally, the model includes a control variable for
counties in right-to-work states as well as a couple
interactive terms of the plant control variables.
Table 6 (p. 11) reports the results of three differ­
ent specifications and the variables used in construct­
ing each of them. A simple model (specification 1)
can explain about 20 percent of the variation in the
dependent variable. In addition, the model identifies
a statistically significant relationship between the
plant-level employment and tier status as well as na­
tionality of headquarters: Plants of tier 1 supplier com­
panies as well as plants of foreign-headquartered
companies are found to have larger employment. The
presence of unions in a supplier plant is only related
to larger plant employment if the plant is either cap­
tive or part of a tier 1 supplier company. That is to
say, unionized plants are larger than others only if
they are either tier 1 or captive plants. Specification 2
controls for what the supplier plants are producing
by distinguishing 8 major subsystems of a car. Employ­
ment at plants producing parts for chassis (such as tires),
body, engine electrical (which includes the electron­
ics components suppliers), and engine attached (of­
ten referred to as air and fuel handling) is consistently
found to be larger than that of the control group, plants
that produce generic parts. Finally, specification 3
controls for a number of county-level characteristics
that might influence plant location decisions, such as
the degree of local work force education, transporta­
tion infrastructure, as well as the presence of other

3Q/2005, Economic Perspectives


Descriptive statistics
County-level model


All new

All new


Share of young supplier plants


Share of domestic young suppliers


Share of foreign young suppliers

Log employment
Distance to Detroit (miles)

Vertical distance to Detroit (miles)
Single plant company
Plant part of top 150 supplier
Plant is captive
Plant is unionized
Company headquarters outside North America
Right-to-work state
Interaction top 150 and unionized
Interaction captive and unionized
Parts for body (%)

Parts for chassis (%)
Parts for drivetrain (%)
Parts for engine attached (%)

Parts for engine electrical (%)
Parts for engine (%)
Parts for interior (%)
Generic parts (%)

Presence of interstate highway

Share of employment in manufacturing
High school education (%)
Population in 1990 (million)

No. of supplier plants in county













No. of domestic supplier plants in county
No. of foreign supplier plants in county
No. of assembly plants within 450 miles





No. of domestic assembly plants in county
No. of foreign assembly plants in county
No. of observations

All new

Note: Standard deviations are in parentheses for continuous variables.

Federal Reserve Bank of Chicago



Parts classification
Major subsystem

ELM subsystem

Engine proper
Engine electrical


of parts listed (%)


Ignition systems
Electronic supply


Engine cooling
Climate control
Fuel systems
Exhaust systems


Engine attached



Chassis electrical
Chassis systems
Wheels and tires




Interior body
Passenger restraints



Body glass
Body components






also includes a measure of how many sup­
pliers had previously located in a county
to account for agglomeration effects. Fi­
nally, the set of county-level controls used
in specification 3 of the plant-level model
(table 6) is included in the county-level
model as well. Table 7 reports the results
that utilize information for all counties east
of the Mississippi to capture the region
of the country most densely populated by
the auto industry.27
Across all specifications estimated,
the presence of an interstate highway in a
county is consistently associated with a
higher share of recently opened supplier
plants in that county. In addition, the size
of the market for suppliers, as measured
by the number of assembly plants within
a day’s drive from a county, is related to
suppliers choosing a county. Specifications
2 and 3 distinguish domestic and foreign
plants, both for the dependent as well as
the independent agglomeration variables.
It turns out that only the presence of foreign
assembly plants within a 450 mile radius
is significantly related to the incidence of
both domestic and foreign “young” sup­
plier plants locating in a county.

Source: ELM and author’s calculations.

Simulation of policy effects
supplier and assembly companies. However, the coun­
ty-level variables do not add to the plant-level model
of employment (table 6).
Next, I estimate a model of plant location at the
county level (table 7, p. 12). The dependent variable is
the share of supplier plants in a county that opened re­
cently.25 As the underlying data is cross-sectional in
nature, it seems prudent to focus on location decisions of
more recently established plants.26 Going back much fur­
ther in time could introduce survivor bias to the model.
The premise is that county characteristics matter in plant
location decisions. The model accounts for the presence
of existing assembly and supplier plants to capture pos­
sible agglomeration effects within the auto industry.
The number of assembly plants located within
450 miles of a county’s centroid measures the size of
the market available to a supplier locating in that
county. That is an important reference point as the
ability to deliver reliably within a day is a key require­
ment of the just-in-time production system. The dis­
tance of 450 miles corresponds to an industry rule of
being able to deliver within a day’s drive. The model


Based on the model results presented
in table 7,1 perform two simple simulation exercises.
The idea is to elicit from the model what the estimated
response in the distribution of supplier plants would
be to a simulated change in the location of an assem­
bly plant. First, assume that Tennessee has one less
light-vehicle assembly plant and Michigan has one
more. I assume Spring Hill as the location of the plant
in Tennessee, and Grand Rapids for the fictional plant
in Michigan. Subsequently, I re-calibrated the variable
that measures the number of assembly plants located
within a 450-mile radius of each county. To that re­
configured variable and all the others in the model,
the estimated coefficients as reported in table 7 were
subsequently applied. In doing so one performs what
is referred to as an “out-of-sample” forecast. In essence,
one can simulate what would happen to the distribu­
tion of young supplier plants if Grand Rapids had an
assembly plant and Spring Hill did not. Constraining
the estimation to result in a zero sum redistribution
of supplier plants, the following result emerges. The
three states of Michigan, Indiana, and Ohio would
increase their count of supplier plants that opened

3Q/2005, Economic Perspectives


Estimation of plant employment




Distance to Detroit




Vertical distance to Detroit




Single plant company




Top 150 supplier




Captive supplier




Unionized plant




Headquarters outside North America




Right-to-work state




Top 150 supplier and unionized




Captive supplier and unionized






Drivetrain %



Interior %



Body %



Engine %



Engine electrical %



Engine attached %



Chassis %

Presence of interstate highway


Manufacturing employment (%)


High school education (%)


Population in 1990


No. of supplier plants in county


No. of assembly plants within 450 miles


No. of observations

R squared










**Significant at the 5% level.
*Significant at the 10% level.
Note: Standard errors are in parentheses.

Federal Reserve Bank of Chicago



Supplier plant locations between 1994 and 2003
No. assembly plants w/450 miles

Foreign only

Domestic only


No. domestic assembly plants w/450 miles



No. foreign assembly plants w/450 miles



No. existing domestic suppliers



No. existing foreign suppliers



No. existing supplier plants


Interstate highway




Right to work state




Share of manuf. employment




Percent high school ed.




Population, 1990




Distance to Detroit















R squared
**Significant at the 5% level.
*Significant at the 10% level.

Notes: Standard errors are in parentheses. Observations: 1,607. Model is estimated for all counties east of the Mississippi.

between 1995 and 2003 by 42, from 122 to 164. The
three states of Kentucky, Tennessee, and Alabama would
see their count of young supplier plants fall by 37, from
65 to 28. The simulated redistribution represents about
14 percent of all young supplier plants opened during
the last 10 years. That represents a significant impact.28
A second experiment consisted allocating afor­
eign assembly plant in Michigan (again, Grand Rapids),
instead of Spartanburg, South Carolina, and estimating
the effect on the distribution of foreign-owned young
supplier plants (there were 107 of them that opened
between 1995 and 2003). Michigan, Indiana, and Ohio
would gain young foreign suppliers. The count for
the three states would increase by 27 from 30 to 57.
By the same token, South Carolina and the surround­
ing auto corridor states North Carolina, Kentucky,
Tennessee, Alabama, and Georgia would have received
fewer recently opened foreign suppliers: Their plant


count of foreign young would go down by 26 from 57
to 31.29 According to this simulation, placing one for­
eign assembly plant into Michigan instead of South
Carolina would affect the location of a quarter of all
foreign supplier plants opened between 1995 and 2003.

This study set out with the intent to shed more
light on the geography of the auto parts sector which
is far less understood than that of the auto assembly
sector of the auto industry. The analysis of a rich plantlevel data set with records of almost 3,800 auto sup­
plier plants located in the U.S. and Canada shows an
industry that is very spatially concentrated. Today
Detroit remains the center of a highly clustered auto
region that extends north-south from Michigan, reach­
ing up into Ontario, west to Chicago, and south to
northern Alabama and into the Carolinas. While the

3Q/2005, Economic Perspectives

analysis is purely cross-sectional, it reveals a surpris­
ing amount of variation in the location pattern exhib­
ited along a number of dimensions. The study confirms
the north-south split within the auto region by nation­
ality of plant: Plants of domestically headquartered
suppliers are concentrated in the northern end of the
auto corridor and plants of foreign-headquartered sup­
pliers are concentrated in the southern end. Overall,
employment and plants are distributed quite similarly.
A plant-level model of employment shows that
plants located farther from Detroit tend to have greater
employment, as do tier 1 and foreign-owned plants.
In addition, we find plant size to vary by type of part
produced. A simple model of recent supplier plant
openings at the county-level points out the importance
of regional transportation infrastructure. The presence
of interstate highway access in a county is consistently

related to a higher share of recently located supplier
plants. Furthermore, the number of assembly plant
customers reachable within a day’s drive is also related
to supplier location choices. This finding points to the
continued importance of agglomeration in this industry.
A policy simulation asks what the effect of a change
in the location of one assembly plant would be on
the geography of recent supplier plant openings. Two
different simulations are presented, one moving an
assembly plant from Tennessee to Michigan, the other
moving a foreign assembly plant from South Carolina
to Michigan. Both suggest a sizable regional effect
on the location of supplier plants. A number of them
would have located closer to the “new” location of
the assembly plant as they need to be within 450 miles
of their assembly plant customers.

'U.S. motor vehicle parts employment is about four times as large
as employment in motor vehicle assembly.

9See White (2005) on the recent restructuring of the original agree­
ment between Ford and Visteon.

2Many different manufacturing sectors contribute to the production
of vehicles and at the same time supply non-automotive custom­
ers. Furthermore, the census data on shipments do not distinguish
between producers of parts for the aftermarket and the original
equipment market. The 2002 Census of Manufacturing, however,
reports the cost of materials used in U.S. light-vehicle assembly
plants at $152.5 billion. That measure includes imported parts.

10Data are available at the plant and company level. However, plants
producing primarily for the aftermarket are not part of database,
nor are plants that produce raw materials, such as steel and paint.
The ELM data were purchased at the end of 2003. The database
is continuously updated by the vendor.

3In addition, factors such as the continuing consolidation and in­
ternationalization within the supplier industry also affect its spa­
tial structure.
4The northern end of the auto corridor is home to over half of all
light-vehicle assembly plants in the U.S., 81 percent of these are
Big Three facilities. Conversely, the southern end of the auto
region is home to about 20 percent of all light-vehicle assembly
plants; half of these are foreign producer facilities. Testa, Klier,
and Mattoon (2005) identify such a regional shift as the most
likely structural threat to the Midwest’s economy.
5See the speech of Michigan’s Governor Granholm from August 4,
2004, in which she outlines a framework on how Michigan should
respond to the current challenges facing its most important manu­
facturing sector. See also McAlinden and Hill (2003).

6The role of the border is not addressed in this article. Post 9/11,
elevated national security concerns have exacerbated demands on
the already strained border infrastructure between the U.S. and
Canada, potentially affecting plant location decisions in an indus­
try that continues to be very tightly integrated and has straddled
both sides of the border for many years (see Simon, 2004, and Klier
and Testa, 2002).
7“Domestic” refers to supplier companies which are headquar­
tered in either the U.S., Canada, or Mexico, “foreign” to compa­
nies headquartered elsewhere.
8The term light vehicles refers to passenger cars and light trucks,
which include minivans and sport utility vehicles.

Federal Reserve Bank of Chicago

“Plants for which no matching records were found were contacted
by phone.

12Based on the location of company headquarters, the article dis­
tinguishes North American (U.S.-, Canadian-, or Mexican-owned
plants), Japanese, as well as other foreign-owned plants.
13Thanks to my colleague Jim Rubenstein who shared his plant-level
data for the 150 largest supplier companies.

14Mexican data are available for 601 plants, but have not yet been
scrutinized to the same extent.
15I construct that variable from the database, utilizing plant names
and company information. It is possible that some of these single­
plant companies have plants that are not included in the database.
16A map of employment, instead of plant count, looks virtually

17Based on the shape of the core auto region, I define the “auto
corridor” to be the states and Canadian provinces that represent
the contiguous north-south cluster visible in figure 1. They are
Alabama, Georgia, Illinois, Indiana, Kentucky, Michigan, Missis­
sippi, North Carolina, Ohio, Ontario, South Carolina, Tennessee,
and Wisconsin. Mississippi and Alabama are included as they re­
cently received new assembly plants.
18Maquiladora plants in northern Mexico were established by the
1965 Border Industrialization Program. This program allowed U.S.
companies to assemble products in Mexico destined for export
elsewhere. Later companies from other countries also established
such plants near the northern Mexico border.


19Of all domestic assembly plants operating in the U.S, 38 percent
are located within 100 miles of Detroit. The corresponding figure
for foreign-owned assembly plants is only 7 percent.
20Note that 17 percent of plants have no information on their union
status. Therefore, this comparison (see columns 6 and 7) only ap­
plies to 83 percent of the records.

21The geographic coordinates for the zip code centroids come from
the Maptitude GIS program. The distance between the two sets of
coordinates is given by the following formula: acos(sin(Zal)*sin(/as')
+ cos(/tfl)*cos(/tf2)*cos(/o2 — Zol))*6370*.62, where lal and lol
are the latitude and longitude (in radians) of the zip code centroid
of the supplier plant and la2 and lo2 are the coordinates for the
zip code centroid of downtown Detroit.
22As explained earlier, tier 1 suppliers are the ones that interact directly
with the assembler. One would have to know the identity of a
supplier’s customer plants in order to identify that group. The top 150
variable tries to proxy for that relationship in the absence of such
detailed customer information. The underlying assumption is that
the vast majority of tier 1 suppliers happen to be large companies.

23In the estimation we treat plants with unknown union status as
not unionized. Based on size and location these plants are very
similar to plants identified as nonunion.
24The ELM data provide information on what parts an individual
plant produces in a very detailed way. Unfortunately, it does not
provide the distribution of actual output across the various parts.
The ELM parts classification system distinguishes 20 subsystems
in a car (table 5). Altogether, it identifies 492 individual parts. Uti­
lizing the relative frequency of the detailed parts listed for each plant,


we converted this information on what each plant produces into a
more aggregate system that distinguishes only 8 subsystems. They
are body, chassis, drivetrain, engine attached (such as the exhaust
system), engine electrical, engine proper, generic parts, as well as
interior parts. The subsystem variables measure the share of indi­
vidual parts codes in each of these by plant.
25A small downside of utilizing the information on plant age is
that it is missing for 19 percent of the data. Elowever, there seems
to be no relation between that and the location of plants. For a
slightly different treatment of such an estimation, see Klier, Ma,
and McMillen (2004).

26Table 7 reports results for supplier pants that were not older
than 10 years in 2003 (1994-2003). Estimating the model for a
smaller set of “young” plants, the ones that opened between 1999
and 2003, yields robust results.
27Estimating the county-level model for the auto corridor only as
well as for the entire U.S. produces robust results.
28To test for robustness of this exercise, I performed the same ex­
periment on the model that estimates the location determinants
for all supplier plants that opened between 1999 and 2003. The
resulting redistribution of suppliers, while different in absolute
numbers, represents a relative change of a similar order of mag­
nitude as described above.
29That result is found to be robust when basing it on the locations
of foreign supplier plants that opened since 1999 instead.

3Q/2005, Economic Perspectives


Ellison, Glenn, and Edward L. Glaeser, 1997,
“Geographic concentration in U.S. manufacturing in­
dustries: A dartboard approach,” Journal ofPolitical
Economy. Vol. 105, No. 5, pp. 889-927.
Granholm, Jennifer M., 2004, “Automotive futures:
Made in Michigan,” speech given at a conference in
Traverse City, MI, August 4.

Hakim, Danny, 2005, “Taking down the ‘No foreign
cars’ signs in Michigan,” New York Times, March 3.

Harbour and Associates, 2003, The Harbour Report—
North America 2003, Troy, MI.
Klier, Thomas, 2005, “Caution ahead—Challenges to
the Midwest’s role in the auto industry,” Chicago Fed
Letter, Federal Reserve Bank of Chicago, February,
No. 211.

__________ , 1999, “Agglomeration in the U.S. auto
supplier industry,” Economic Perspectives, Federal
Reserve Bank of Chicago, Vol. 23, No. 1, pp. 18-34.
Klier, Thomas, Paul Ma, and Dan McMillen, 2004,
“Comparing location decisions of domestic and for­
eign auto supplier plants,” Federal Reserve Bank of
Chicago, working paper, No. WP-2004-27.
Klier, Thomas, and William Testa, 2002, “Linkages
across the border—The Great Lakes economy,” Chicago
Fed Letter, Federal Reserve Bank of Chicago, July,
No. 179b.
Krugman, Paul, 1991, Geography and Trade, Cam­
bridge, MA: MIT Press.

Marshall, Alfred, 1920, Principles ofEconomics,
London: McMillan.

Federal Reserve Bank of Chicago

McAlinden, Sean P., and Kim Hill, 2003, The Mar­
ket Renewal ofMajor Automotive Manufacturing Fa­
cilities in Traditional Automotive Communities, Ann
Arbor, MI: Automotive Communities Program, Cen­
ter for Automotive Research, August.
Roland Berger Strategy Consultants, 2004, The
Odyssey of the Auto Industry’—Suppliers’ Changing
Manufacturing Footprint, Munich, Germany.
Rubenstein, James M., 1992, The Changing U.S.
Auto Industry’—A Geographical Analysis, London:
Shirouzu, Norihiko, 2004, “Chain reaction—Big
Three’s outsourcing plan: Make parts suppliers do
it,” Wall Street Journal, June 10, p. Al.

Simon, Bernard, 2004, “Wheels of trade seize up at the
world’s busiest border,” Financial Times, August 3, p. 3.

Smith, Donald, and Richard Florida, 1994, “Ag­
glomeration and industrial location: An econometric
analysis of Japanese-affiliated manufacturing estab­
lishments in automotive-related industries,” Journal
of Urban Economics, Vol. 36, No. 1, pp. 23-41.
Testa, William, Thomas Klier, and Richard Mattoon,
2005, “Challenges and prospects for Midwest manu­
facturing,” Chicago Fed Letter, Federal Reserve Bank
of Chicago, March, No. 212b.

White, Joseph B., 2005, “Ford to pay up to $1.8 bil­
lion on Visteon,” Wall Street Journal, May 26, p. A3.
Woodward, Douglas, 1992, “Locational determinants
of Japanese manufacturing start-ups in the United
States,” Southern Economic Journal, Vol. 58, No. 3,
pp. 690-708.


Switching primary federal regulators: Is it beneficial
for U.S. banks?

Richard J. Rosen

Introduction and summary
In the United States, commercial banks can select among
three primary federal regulators. A bank chooses a
chartering agency and decides whether it will be a
Federal Reserve System (Fed) member, thereby select­
ing its regulatory authority. A nationally chartered bank
is regulated by the Office of the Comptroller of the
Currency (OCC). If it is a Fed member, a state-chartered
bank has the Fed as its primary federal regulator; other­
wise, it is overseen by the Federal Deposit Insurance
Company (FDIC).1 By choosing its charter and deciding
whether to be a Fed member, a bank effectively selects
its regulator.
This article explores how banks use their option
to select a regulator. Specifically, I examine banks that
switch from one regulator to another. Is the ability to
switch regulators a positive aspect of our current sys­
tem? I offer some insight into this issue by examining
whether banks benefit from switching and how switch­
ing affects social welfare. This study helps shed light
on the behavior of regulators and the efficacy of the
current system of multiple regulators. There has been
debate about whether regulators, when setting policies,
act in the public interest or not. This article builds on
Rosen (2003), where I focused on whether the regula­
tory competition was beneficial or destructive. Com­
petition could spur useful innovation or regulatory
flexibility, thereby allowing banks to benefit without
reducing social welfare. It could also be a “race for
the bottom”—or a “competition for laxity,” to use
former Federal Reserve Chairman Arthur Burns’s
term—if regulators try to attract banks by easing re­
strictions on unsafe or unsound practices. The evidence
presented here is not consistent with a race for the
bottom, while there is some evidence of beneficial
competition. In general, a bank’s return either stays
the same or increases after it switches regulators, while
its risk of failure does not rise.


While most banks never switch regulators, the ag­
gregate number of switchers is not small. Over 10 per­
cent of banks switched regulators at least once during
the period 1977-2003.1 compare banks that switch
with others that do not in an attempt to learn why banks
switch. I find that, prior to changing regulators, switchers
have approximately the same return on assets as oth­
er banks, and switchers are somewhat riskier. Small
banks are less likely than large banks to switch regu­
lators, but this is largely due to the fact that small banks
are less likely to be in a bank holding company. Non­
lead banks that are in a holding company are more
likely to switch than either lead (largest) banks in a
holding company or banks not in a holding company.
The effect of a switch on return and risk can in­
dicate whether switches are beneficial. I find that banks
generally increase their return when they switch reg­
ulators. There is little significant impact of a switch
on risk. Banks tend to reduce their equity-to-asset ra­
tio following a switch, but more inclusive measures
of risk, such as the bank failure rate, point toward no
increase in risk. An increase in return with no signifi­
cant increase in risk is evidence consistent with bene­
ficial competition. However, the aggregate results hide
differences over the sample period in the performance
of banks that switch.
The percentage of banks switching varies through­
out my sample period—rising in the late 1970s, then
falling to a lower rate in the 1980s, before rising again
in the 1990s (see figure 1). There are many reasons
why banks switch regulators, some of which may ex­
plain part of the pattern of switching over time. A switch
Richard J. Rosen is a senior economist and economic
advisor at the Federal Reserve Bank of Chicago. The
author would like to thank Tara Rice, Craig Furfine, and
the participants in a workshop at the Chicago Fedfor
their helpful comments.

3Q/2005, Economic Perspectives


Banks that switch primary federal regulators, 1977-2003

may be prompted by changes in the structure of a
banking organization, issues relating to the interaction
between a banking organization and its regulators, or
a shift in strategy sought by a banking organization.
The sample period I explore—1977-2003—was one
of major changes in banking, both in the structure of
the industry and in the regulatory framework under
which it operated. I explore whether the characteris­
tics of banks that switch regulators vary over time,
perhaps indicating changing motivations for switch­
ing. I find that prior to 1992, switching regulators
has little impact on overall risk and return. However,
switches in the latter part of the sample, specifically
1992-2003, have a significant impact on performance.
Banks that switch in this period show an increase in
return without a commensurate increase in risk, as
would be expected if there is beneficial competition.
Note that the post-1991 period is also when the rate
of switching is at its highest.
The plan of the article is as follows. First, I pro­
vide an overview of when banks switch primary reg­
ulators. Next, I review the arguments for and against
a system in which banks can choose among multiple
regulators. Then, I examine the characteristics of banks
that switch primary regulators and present an analysis
of how switches affect performance, including failure

Federal Reserve Bank of Chicago

The pattern of banks switching primary
Banks have been switching primary regulatory
agencies for many years (Scott, 1977, documents
switches from 1950 to 1974). I examine switches that
occurred from 1977 to 2003, a period that covers ma­
jor changes in banking and bank regulation. I identi­
fy the year a bank changes primary regulators from
the Reports ofIncome and Condition (call reports).
Table 1 gives an overview of the banks that switched
primary regulators. As table 1 shows, there were 2,298
switches during the sample period, an average of 85 per
year. Over the sample period, 10.8 percent of banks
left their respective regulators at least once (0.7 percent
of banks switched more than once). Table 1 also pro­
vides a breakdown of switches based on the size of the
bank. The smallest banks were the least likely to switch.
The pattern of banks switching regulators can be
partially explained by regulatory changes. In 1980,
the Depository Institutions Deregulation and Monetary
Control Act (DIDMCA) was passed. Prior to DIDMCA,
there were important differences among regulators.
For example, reserve requirements (the funds a bank
must hold against specified deposit liabilities) depended
on whether a bank was a member of the Federal Re­
serve System. DIDMCA leveled the playing field for all
banks, regardless of membership in the Federal Reserve
System. It is possible that many of the regulatory



Banks that switch primary federal regulators, 1977-2003
All switching

Total assets less
than $1 billion


of banks

of banks

of banks





of banks


Total assets
between $1 billion
and $10 billion

of banks


of banks

Total assets greater
than $10 billion
of banks



of banks

Note: Size classes are based on total assets in 2003 dollars.
Source: Data from Federal Deposit Insurance Corporation, 1977-2003, Reports of Income and Condition, Washington, DC.

switches that occurred prior to and immediately after
passage of DIDMCA were related to the changes insti­
tuted by the act rather than any actions of the regulators.
During the 1980s, states gradually reduced their
restrictions on interstate and intrastate expansion (Amel,
1991; Amel and Starr-McCluer, 2002). This may have
prompted the merger wave in the 1980s and could
have led to some of the switches of primary regulators
during that decade. In addition, the Riegle-Neal Act
of 1994 removed the restrictions on interstate branching.
This act was phased in over the next few years as states
gradually adopted its provisions (Dick, 2006).
Merger activity, and switches associated with
mergers, varied significantly over the sample period.
Figure 2 gives the number of merger-related and other
switches by year. I define a bank as having switched
because of a merger if it switches its primary regula­
tor in the year of its merger or the following year. If


banks with different primary regulators merge, the
newly formed bank will have to choose one of the two
regulators. Following a merger, if the acquiring bank
changes from its pre-merger regulator to the target
bank’s regulator, then I record this as a switch of pri­
mary regulators for the newly formed bank. A total of
779 banks switched regulators following a merger,
one-third of all switches.
In the main analysis of the article that follows, I do
not include banks that have recently merged. An ob­
jective of this article is to examine whether the abili­
ty of banks to switch regulators is a valuable option.
This is difficult to do with switches following mergers
for at least two reasons. The first is that, as noted above,
banks with different primary regulators are forced to
choose one. This leads to a different—and likely, low­
er—threshold for switching regulators than for non­
merging banks. It is possible that the inclusion of

3Q/2005, Economic Perspectives


Merger-related and other switches

Are multiple regulators

There has been a debate over the best
regulatory structure for a long time (see
Rosen, 2003, for some examples). This
section briefly explores why banks switch
regulators and discusses some concerns
about the current regulatory system, as
well as some of its benefits.
When bank managers are asked why
they change primary regulators, they gen­
erally respond in one of three ways. These
managers claim that a bank switches be­
cause it can gain additional powers (as
Chase Manhattan Bank did when it changed
the primary regulator of its Delaware bank
1977 ’79 ’81 ’83 ’85 ’87 ’89 '91 ’93 ’95 ’97 ’99 2001’03
in 1990); save on regulatory compliance
costs (as Chase Manhattan Bank did after
Source: Data from Federal Deposit Insurance Corporation, 1977-2003, Reports of
Income and Condition, Washington, DC.
its merger with Chemical Bank in 1995);
or expand more easily nationwide (as HSBC
USA did when it changed its charter in
2004). Broadly speaking, regulation at the three agen­
banks that switch regulators concurrent with a merg­
er will bias the results toward finding no impact from
cies—the OCC, the Fed, and the FDIC—and among
switching. The second reason for dropping mergerthe states (for state-chartered banks) is similar.2 But
related switches is perhaps more important. A merger
for some banks, the differences among regulators
can significantly affect the reported return and risk
might be important enough to induce a switch. Dur­
for a bank. Costs related to the integration of the merg­
ing part of the sample period, for example, the insur­
ing banks can depress the return for several years.
ance powers granted to banks varied among regulators.
Also, the decision to participate in a merger may be
To conduct certain insurance activities, a bank might
related to return and risk. Banks may be more likely
have needed to switch regulators. Thus, Chase Man­
to be merger targets when their return has been de­
hattan switched the regulator of its Delaware bank to
clining or when their risk has been increasing. This
allow it to sell insurance.
may bias the before-and-after comparison of return
Some switches might be prompted because of
and risk for banks that merge.
the costs of regulation, which are both indirect and
Bank mergers affect not just the merging banks,
direct. The indirect costs include managerial and le­
but also other banks. The bank merger waves in the
gal costs involved in meeting with bank examiners
1980s and the 1990s increased the average size of a bank
and making required reports. Indirect costs also in­
dramatically. However, these waves had a much smaller
volve the opportunity costs of restrictions on portfo­
effect on local market competition, with the average
lio choices imposed on banks by regulators, such as
market concentration index essentially unchanged. These
reserve requirements and expedited funds availability.3
changes affected competition in local markets (see, for
There is no reason to believe that there are systemat­
example, Berger, Udell, and Rosen, 2005). Some switch­
ic differences in the indirect costs that banks would
es of regulators may have been partially in response to
face at the different agencies. However, there are dif­
these repercussions of bank consolidation. These switch­
ferences in direct costs. Both the OCC and the FDIC
es are included in the sample, since they do not suffer
charge for bank exams, but the Fed does not. This may
from the drawbacks noted in the previous paragraph.
seem to give the Fed a cost advantage, but examina­
DIDMCA and the merger waves may have induced
tion of state-chartered banks is shared with state reg­
some of the changes in my sample period. However,
ulators, who charge for their exams. Still, there can
there are many switches that cannot be explained
be cost differences among regulators. This may induce
purely by regulatory changes or industry consolida­
some switches if the OCC, the FDIC, or some states
tion. In the next section, I examine additional possible
change the cost of exams (or if, because of competition
explanations for switches of primary regulators.
in banking, a bank feels it has to squeeze out additional

Federal Reserve Bank of Chicago


cost savings). Cost considerations may have also
prompted some holding companies to simplify their
regulatory structures.
From a social perspective, some question whether
having multiple regulators is a good idea. There are
several potential drawbacks to the current regulatory
system. At minimum, having multiple regulators in­
troduces complications. For instance, when J. P.
Morgan Chase merged with Bank One in 2004, J. P.
Morgan Chase Bank had a state charter and Bank One
had a national charter. As part of the merger process,
J. P. Morgan Chase had to decide which charter to
adopt (and if it chose a state charter, whether to become
a Fed member). This took time and resources that
would not have been necessary if there were only a
single bank regulator. Moreover, when it selected a
national charter, the bank’s former state regulator had
to shift its personnel and pricing to account for the
loss of a major bank. These costs may not be large,
but they are certainly present.4
A potentially more serious issue is that regula­
tors might not always act in the social interest. Stigler
(1971) points out that regulators can be captured by
the firms they cover because those inside a particular
industry care a lot more about the regulators’ decisions
than outsiders do. As a result, they may choose poli­
cies that benefit banks rather than the public.
Related to this concern, the literature on regula­
tory structure explores a “race for the bottom” among
regulatory agencies. In the 1970s, then-Fed Chairman
Arthur Burns commented that he feared destructive
competition among regulators for banks (their custom­
ers, in a sense). He brought up the possibility of what
he called a “competition for laxity,” a scenario in which
banks would relax regulation to capture market share
(see Scott, 1977). Since the budget of an agency de­
pends in part on the number and size of the firms it
regulates, regulators might compete against each other
by offering lenient treatment in order to attract firms.
When Chase Manhattan Bank elected to have a state
rather than a national charter, subsequent to its merger
with Chemical Bank in 1995, the OCC lost fees amount­
ing to 2 percent of its budget. Similarly, when its suc­
cessor, J. P. Morgan Chase Bank, returned to a national
charter after its merger with Bank One in 2004, the
New York Banking Department (the state regulatory
agency) lost 27 percent of its revenues. If either agency
was concerned with maximizing its budget, it would
have an incentive to remove burdens on banks to keep
them from switching.5
A race for the bottom could allow banks to manip­
ulate the system. That is, banks might choose their
primary federal regulator (and potentially, their state


of incorporation, and thereby, their state regulator) to
take actions that benefit the bank but are not in the public
interest. An example of this would be a bank that
switched regulators in order to adopt a new, risky strate­
gy (or to hide risks it was already taking). The risk
could increase the exposure of the deposit insurance
fund. It is important to note that a bank can only switch
to a new regulator if that regulator approves. Thus, reg­
ulators have the ability to block switches of this kind.
On the other hand, having multiple regulators of­
fers potential benefits. A single regulator might have
less incentive to allow banks to undertake new powers
or to use new products. There is a natural tendency
for regulators to be risk averse, since they are assigned
blame for anything that goes wrong, but may not be
recognized for permitting beneficial changes. Poten­
tially beneficial changes that one regulator views as
too risky might be adopted by another regulator. In
addition, having multiple regulators allows for some
specialization. Tiebout (1956) presents a model of
public goods provision by local communities that has
often been modified to examine other regulatory is­
sues. The Tiebout framework can be used to show
that under certain conditions (including when there
are no externalities and there is costless mobility),
regulatory competition leads to optimal standards
setting. Different localities can offer distinct menus
of public goods, with each individual choosing the
menu best suited for that individual (referred to as
Tiebout sorting). This model underlies the arguments
for local control of securities regulation (Romano,
1998), antitrust enforcement (Easterbrook and Fischel,
1991), and environmental policy (Revesz, 2000). These
papers also claim that the benefits of competition among
local agencies eliminate (or should eliminate) a race
for the bottom.
Connected to Tiebout sorting, another reason
that banks might switch is that regulatory enforcement
may differ among agencies. There may be an explicit
policy shift at a particular agency. For example, in 1991,
Federal Reserve Chairman Alan Greenspan was worried
that examiners were contributing to a “credit crunch”
by requiring banks to hold too much capital against
loans. This was interpreted by some as a signal for ex­
aminers to relax enforcement. This could have encour­
aged banks to switch to the Fed from the other agencies.6
An additional complication to this analysis is that
a bank regulatory agency is essentially a collection of
examiners. Unlike regulators in many other areas, ex­
aminers in banking frequently make subjective deci­
sions about the banks they visit.7 Berger, Kyle, and
Scalise (2000) review examiner and regulatory agen­
cy discretion when monitoring banks. Examiners go

3Q/2005, Economic Perspectives

into a bank to evaluate its risk. Based on this assessment,
the examiners decide whether the bank’s reserve for
loan losses is sufficient, and then they assign a strength
rating—the CAMELS (capital, asset quality, manage­
ment, earnings, liquidity, and sensitivity) rating—to
the bank. If a bank wants to change its portfolio, its
examiners must decide how to react. The examiners
can either accede to the change or make it costly for
the bank by requesting a higher loan loss reserve
(resulting in a charge against income) or by giving
the bank a lower CAMELS rating (resulting in greater
regulatory costs for the bank). Thus, to an extent, ex­
aminers can decide how costly it is for a bank to add
risk. Having multiple regulators and the ability to
switch among them allows the bank to escape exam­
iners that the bank feels are out of line.
One potential problem that a bank might have is
that its examiners can exploit the discretion they have
when assessing the bank to serve their own ends. Some
examiners may be interested in leading a “quiet life”
(Rosen, 2003).8 That is, they may want to get by with
as little work and as little career risk as possible. To
get a quiet life, some examiners might prefer to regu­
late banks with portfolios that are as simple as possi­
ble to evaluate.
There is another reason why examiners may put
up roadblocks to change by banks. Regulatory be­
havior may be influenced by a desire to avoid criti­
cism from groups other than the firms that examiners
assess. Importantly, Congress and public interest
groups may criticize ex post actions that were proper
ex ante (as Kane, 1989, argues they did early in the
savings and loan crisis in the 1980s). This gives reg­
ulatory agencies and, by extension, examiners an in­
centive to avoid actions that could increase the risk
of bank failure. Fear of criticism may induce risk aver­
sion on the part of examiners who want a quiet life.
Whether having the ability to switch regulators
leads to beneficial competition or a race for the bot­
tom can be tested by examining which banks switch
and how switching affects the performance of these
banks. The key here is to decide which switches are
“beneficial” and which are not. A beneficial switch
allows a bank to move to a better risk-return trade­
off without increasing societal risk. I used the risk of
bank failure as a proxy for societal risk. Banks are
overseen by government agencies for many reasons.
For instance, banks are regulated in order to maintain
a smoothly operating payments system and to con­
firm that their deposits are insured. Both of these ob­
jectives imply that regulators want to limit excessive
risk-taking by banks, which should limit bank fail­
ures. A race for the bottom might work this way:

Federal Reserve Bank of Chicago

Regulators could allow banks that switch to increase
societal risk without a compensatory increase in re­
turn. Bank managers or shareholders could profit from
this, but only by taking advantage of the deposit in­
surance system. Beneficial competition among regu­
lators, on the other hand, would allow banks to move
to a better risk-return trade-off without increasing
failure probabilities.9 Note that these tests are sufficient
to indicate beneficial competition or a race for the
bottom, but there are other factors that may not be
figured in. Beneficial competition can help all banks,
not just those that switch. I cannot directly test for
this, but the increase in bank profits and decrease in
bank failures over the past 15 years are consistent with
beneficial competition—and not a race for the bot­
tom. Still, since these trends are also a function of
macroeconomic factors, this is at best weak evidence.

Characteristics of banks that switch
To evaluate banks that switch regulators, I need
measures of return and risk. Return is easy to mea­
sure. I use the return on assets (ROH), but its results
are similar to other measures, such as the return on
equity. Unfortunately, there is no simple inclusive
measure of risk. I use direct and indirect risk evalua­
tions. The direct measure of risk I use is a failure pre­
diction model. As noted above, bank failures can
reduce the smooth operating of the payments system
and increase losses to the deposit insurance fund.
Thus, if a regulator allows banks that switch to take
actions that increase their failure probabilities, this
suggests a race for the bottom. To attain a second es­
timate of failure probability and to determine how
risk changes relative to return, I use four accounting
ratios that capture different aspects of risk. The most
direct is the Sharpe ratio, which is the ratio of ROA
to the standard deviation of the ROA (again, the re­
sults are similar to the return on equity). To construct
this measure for year t, I use the ROA for year / as
the numerator. The denominator is the standard devi­
ation of the semiannual ROA (expressed as an annual
return) for all the periods from year / - 4 to year / for
which return data exists. I keep all observations with
at least two years of return data as of year t. Even
with ten semiannual periods, I do not have a very
precise measure of risk. Still, while noisy, the ratio
of ROA to its standard deviation does give a picture
of the risk-return trade-off.
I also use other accounting measures of risk tra­
ditionally used to evaluate banks. The equity-to-asset
ratio (EQUITY/ASSET) is a measure of leverage,
with higher values indicating lower risk, since equity



Performance of banks prior to switching regulators
Banks that switch









Banks that never switch





Test of difference
of means (p value)




aThe Sharpe ratio only includes banks with at least two years of data.
“Significant at 5 percent level.
‘“Significant at 1 percent level.
Notes: Banks that switch regulators include all banks that switch regulators, except those that switch in the year of or year following a merger.
Variable definitions are given in the text. The data are year-end (except for ROA, which is for the full year) for the period 1977-2003. Data for
switchers are from the year prior to a switch. The variables ROA, EQUITY/ASSET, LOAN/ASSET, CHRG/LOAN, and DEP/LIAB are expressed as
percentages. All other variables, except LOG ASSETS, are expressed as ratios.
Source: Data from Federal Deposit Insurance Corporation, 1977-2003, Reports of Income and Condition, Washington, DC.

offers a cushion against failure. The loan-to-asset
ratio (LOAN/ASSET) is likely to be correlated with
risk as well. Loans are among the riskiest assets on
bank balance sheets. A bank with more loans, all else
being equal, is more likely to fail. However, loans
can vary significantly in risk. To measure the riskiness
of a loan portfolio, I use the charge-off-to-loan ratio
(CHRG/LOAN).10 This ratio reflects expected losses
on loans made in the past. A riskier loan portfolio, all
else being equal, has higher charge-offs. Charge-offs
can also reflect bad luck, poor management, or invest­
ments in risky but predictable loans (for example, some
credit card loans). To capture risk differences on the
liability side, I use the ratio of deposits to liabilities
(DEP/LIAB). Deposits are a more stable source of
funding than other liabilities, such as loans from other
banks. Results based on these ratios should be viewed
with caution, since they may be associated with changes
in productivity as well as risk.
The loan-to-asset ratio and the charge-off-to-loan
ratio can also be viewed as proxies for the workloads
of bank examiners. Examiners have to spend more
effort when reviewing loans than other assets, and
they have to spend even more effort when reviewing
nonperforming loans than other loans. If examiners


desire a quiet life, they prefer banks to have nonloan
assets, such as cash and government securities, and
they are inclined toward banks that do not issue loans
with a high probability of becoming nonperforming.
To assess whether a switch of primary federal
regulators is beneficial, it is useful to know what leads
a bank to switch regulators. To do this, I use a simple
model to predict which banks will switch regulators
as a function of the return and risk characteristics of
the banks. The dependent variable, SWITCH, is a dum­
my that takes the value 1 in year t if a bank switches
regulators in the year t + 1. The model is:

When analyzing the data, I drop banks in any year
that they are in the top or bottom 1 percent of ROA,
To examine whether banks that switch regulators
are different from other banks, it is important to con­
trol for reasons unrelated to return and risk that might
lead a bank to shift its primary regulator. Table 1

3Q/2005, Economic Perspectives

shows that small banks are disproportionately less
likely than larger banks to switch regulators. For this
reason, I control for bank size using the log of total
assets (LOG ASSETS). Structural considerations may
play a role in the decision to switch. I control for holding
company status, using dummies for whether the bank is
the lead bank in a holding company (LEADBANK HC),
or whether it is a non-lead bank within a holding com­
pany that has the same (NONLEAD SREG) or differ­
ent (NONLEAD DREG) charter than the lead bank.
Banks not in a holding company (NONHC BANK)
compose the excluded category. This allows us to test
for switches that reduce the number of regulators to
which a holding company reports. There may also be
other differences across primary regulators. To con­
trol for this, I include dummies for whether a bank is
regulated by the Federal Reserve or the FDIC at the
end of year t- 1 (the OCC is the excluded category).
Finally, I include year dummies to control for systemic
changes, such as changes in overall levels of return
and risk in the industry as a whole.
Table 2 reports summary statistics for the return,
risk, and control variables. Banks that switch regula­
tors have a similar return and Sharpe ratio to other banks.
There are differences between the two groups in the
other risk measures. The equity-to-asset, loan-to-asset,
and deposit-to-liability ratios all indicate higher risk
for switchers than for other banks that have not switched,
but switchers have a lower charge-off-to-loan ratio. How­
ever, I need to account for correlations among these vari­
ables and patterns in the proportion and type of banks
that switch. I do this using a regression framework.
Equation 1 is estimated using a logistic regres­
sion. The results of the regression are reported in the
first column of table 3. Consistent with the univariate
statistics, the coefficient on ROA is not statistically
significantly different from zero. So, I cannot use a
bank’s return to predict whether it will switch regula­
tors. Most of the risk variables, on the other hand, are
significant and can help predict which banks will switch.
Banks with a lower Sharpe ratio, more leverage, and
a lower deposit-to-liability ratio—all indicators of
higher risk—are more likely to switch. Pointing to a
trend in the other direction, banks with fewer chargeoffs, signaling less risk, are also more likely to switch.11
As figure 1 and table 1 show, the proportion of
banks that switch regulators varies over time. It is
possible that the strength of banks varies along with
switching intensity. To test this, I divide my sample
period into two smaller periods. The early period in­
cludes all switches from 1977 to 1991. This covers
the implementation of DIDMCA and the lesser wave

Federal Reserve Bank of Chicago

of mergers in the 1980s.12 The late period includes all
switches from 1992 to 2003. This includes the peak
of bank consolidation. This is also the time when the
proportion of banks that switch regulators is largest.
The second and third columns of table 3 present
regression results for the two smaller periods. There
are differences across the two periods in the magni­
tude and statistical significance of the return and risk
variables. For example, the coefficients on the Sharpe
ratio and the charge-off-to-loan ratio are larger and
statistically significant only in the late period. Still,
the overall pattern is similar. Return is not a predictor
of switching in either period, and banks that switch
look somewhat riskier in every dimension except
their level of charge-offs.
The control variables differ in important ways
across the two periods. In the early period, 1977-91,
small banks, all else being equal, are more likely to
switch regulators. This is reversed in the late period,
1992-2003, when large banks are more likely to switch.
Overall, banks that are not the lead bank in a holding
company are more likely to switch than either lead
banks or banks not in a holding company. Consistent
with a desire to simplify the regulatory structure of
their respective holding companies, non-lead banks
that have different charters than their lead banks switch
more often in both periods. Non-lead banks with the
same regulator as their lead banks are only more likely
to switch in the early period. This may reflect banks
switching to exploit differences among regulators in
the types of investments allowed, such as insurance
activities. These differences tended to be larger in the
early period than in the late period, especially once
the Financial Modernization Act (also known as the
Gramm-Leach-Bliley Act) was passed in 1999. The
regulatory dummies also provide some interesting in­
sights. Banks are more likely to switch from the Fed
than either the OCC (the omitted regulator) or the FDIC
in the early period. In the late period, however, banks
under the Fed are less likely to switch than those un­
der the OCC—and as likely as those under the FDIC.
Over the entire period, banks under the FDIC are the
least likely to switch.
The results suggest that banks that switch are
different from those banks that do not. They also
suggest that these differences depend on when the
banks switch. These findings do not help determine
whether there is a race for the bottom or beneficial
competition, but they point out the importance of
controlling for why and when banks switch, as well
as other bank characteristics.



Probability that bank will switch regulators
in the next year
Late period

Full sample

Early period
























the risk of bank failure, then that is evi­
dence consistent with a race for the bottom.
The section is divided into two parts. I
examine the accounting measures of perfor­
mance in the first part and then a failure
prediction model in the second part.
Accounting measures ofperformance
To examine how performance changes
preceding and following a switch of reg­
ulators, I use the following model:


Performance =fPre-change
indicators, Post-change indicators,
Control variables),

where performance is measured using
our return and risk variables. The model
is estimated for the entire sample of banks,
not just banks that switched. This allows
us to compare changes in performance at
banks that switch with otherwise similar
banks that have not.
A priori, there is no reason to believe
that the changes induced by a switch of
regulators should be immediately reflected
in the performance. For this reason, I look
over five-year periods before and after a
switch. This allows a long enough time
before a switch to see whether there was
some change in a bank’s performance that
prompt a switch. It also allows a
time after a switch to ensure
that all the changes that result from it are
‘Significant at 10 percent level.
reflected in the accounting data I examine.
“Significant at 5 percent level.
For banks that switch regulators, I use
“‘Significant at 1 percent level.
Notes: The data are from 1977 to 2003, with year dummies not shown.
dummy variables for pre- and post-switch
The dependent variable is a dummy for whether a bank switches regulators
as well as a trend variable. Let
in the next calendar year. Other variable definitions are given in the text.
Robust p values adjusted for cluster effects are in parentheses.
DUMMY PRE, the pre-switch dummy,
Source: Data from Federal Deposit Insurance Corporation, 1977-2003,
equal 1 for each of the five years prior to
Reports of Income and Condition, Washington, DC.
a switch (year t - 5 to t - 1 for a switch
in year t) and equal 0 otherwise. Similarly,
Performance of banks that switch
let DUMMY POST, the post-switch dummy, equal 1
for each of the five years following a switch (year t + 1
1 + 5 for a switch in year t) and equal 0 otherwise.
In this section, I examine the change in performance
banks that never switch, both DUMMY PRE and
at banks that switch regulators, comparing return and
POST equal 0.1 set the trend variables so
risk before and after a switch. This allows us to address
increasing in time. For banks that switch
two issues. The first is whether switching is good for
PRE take the value 1 in year / - 5,
banks and the second is whether it is good for society.
on until it has the value 5 in
Beneficial competition implies that banks can benefit
and other banks, it equals 0.
from switching while the probability of bank failure
POST as taking the value 1
(our proxy for social welfare) does not increase. If
2, and so on until it has the
switching allows banks to take actions that increase


3Q/2005, Economic Perspectives

value 5 in year t + 5 for switchers, and the value 0
The control variables are similar to those in the
prediction model in the previous section. The risk
choices a bank makes affect its return and vice versa.
Thus, I include risk and return variables as controls
(excluding the performance measure being estimated).
Results are qualitatively similar without these con­
trols. I also include the structural controls from the
prediction model. These cover the holding company
status and regulator of a bank. Finally, I use the log
of total assets as a control, since larger banks are more
diversified, all else being equal, and year dummies to
control for systemic changes.
Table 4 presents the results of regressions using
equation 2 for the risk and return measures. Panel A
gives the regression coefficients. Some but not all of
the trend and dummy variables are significant. What
I am most concerned with is the net change following
a switch. For example, the post-switch trend is signifi­
cant for the ROA, but the post-switch is not. What does
this say about the net change in ROA? To get an idea
of how important these changes are, it is necessary to
combine the trend and dummy variables. For example,
five years prior to a switch, the average bank has an
ROA that is 0.016 percentage points below that of an
otherwise similar bank that never switches (0.016 =
-0.018 + 0.002 x 1). By the year before the switch, ROA
is 0.008 percentage points below that of an otherwise
similar bank that never switches (0.008 = -0.018 +
0.002 x 5 with rounding), indicating an increase of
0.008 percentage points in ROA in the four years be­
fore a bank switches. The increasing return after a switch
is such that five years after a switch, the average bank
has an ROA that is 0.083 percentage points above
that of an otherwise similar bank that never switches
(0.083 = 0.008 + 0.015 x 5 with rounding). Panel B
of table 4 presents the estimated changes for the years
before and after a switch.
The results for the period prior to a switch indi­
cate that banks are changing their balance sheets sig­
nificantly prior to a switch. Leverage increases as the
equity-to-asset ratio falls. Banks are also shedding loans.
If examiners want a quiet life, then changes such as
these may make them unhappy. This may, in turn, make
it more probable that a bank will switch regulators.
The results in panel B of table 4 show that return
rises significantly in the five years after a switch. They
also provide evidence on the accounting risk measures.
Overall, the picture on risk changes before and after
a switch is mixed. The key factors are that the Sharpe
ratio is unchanged but the equity-to-asset ratio decreases
heading into and following a switch.

Federal Reserve Bank of Chicago

To get an idea of how the risk and return chang­
es compare after a switch, I use the data in panel B of
table 4 to compare the percentage change in the equity-to-asset ratio (the risk measure that increases) to the
percentage change in ROA. The percentage change is
measured by dividing the change by the pre-switch
mean and is given in the final row of the table. From
the year prior to a switch to five years after the switch,
ROA is estimated to increase by 0.091 percentage
points, 9.8 percent of the average ROA prior to the
switch. Over a similar period, the equity-to-asset ra­
tio is estimated to decrease by 0.463 percentage points,
5.3 percent of the average ratio prior to a switch. Thus,
return increases by a larger fraction than the risk (as
measured by the accounting variables) increases. This,
in combination with no significant change in the Sharpe
ratio, suggests that banks do better following a switch
and provides no evidence that social risk increases.
Recall that the regression results in table 3 show
that the factors that lead banks to switch regulators have
changed over time. It makes sense, then, to see whether
the performance of banks before and after a switch
differs over time. To do this, let EARLYbe a dummy
variable that takes the value 1 if a bank switches reg­
ulators between 1977 and 1991, and let LA TE be a
dummy variable that takes the value 1 if a bank switches
regulators between 1992 and 2003.1 create a series
of eight interaction variables using these dummies.
Each interaction variable is the product of one of the
period dummies and either TREND PRE, DUMMY
variables, I estimate equation 2. Rather than present­
ing the entire regression results, table 5 gives the es­
timated changes relative to otherwise similar banks
that have not switched in the accounting return and
risk measures for the three periods, mirroring panel B
of table 4.13 It is clear from the table that the effect of
switching on return appears only in the late period.
Return increases significantly in the late period but
changes little in the early period. The effect of a switch
on risk is mixed in both periods but is driven by dif­
ferent factors in each. In the early period, the Sharpe
ratio is unchanged and charge-offs decrease, signaling
no change or a decrease in risk. But, the equity-toasset ratio increases, indicating higher risk. In the late
period, on the other hand, the Sharpe ratio increases
(in large part due to the increase in ROA), while
charge-offs also increase. The only constant is that
banks add leverage following a switch in both periods.
Also, while the coefficient on the change alter a switch
in the late period is positive for the charge-off-to-loan
ratio regression, a deeper examination of the data (not
shown) indicates that the positive coefficient reflects



Performance regressions
A. Regression coefficients



Return on


Equity-toasset ratio





















Charge-off-toloan ratio


Deposit-toliability ratio






Charge-off-toloan ratio

Deposit-toliability ratio

B. Estimated changes in accounting variables
Return on


Equity-toasset ratio


Change from 5 years
prior to switch to
1 year prior to switch







Change from 1 year
prior to switch
to 1 year after switch







Change from 1 year
prior to 5 years
after switch











Change from 1 year
prior to 5 years
after switch divided
by sample mean



‘Significant at 10 percent level.
“Significant at 5 percent level.
‘“Significant at 1 percent level.
Notes: The data are from 1977 to 2003, with year dummies not shown. Variable definitions are given in the text. For both panels A and B,
robust p values adjusted for cluster effects are in parentheses.
Source: Data from Federal Deposit Insurance Corporation, 1977-2003, Reports of Income and Condition, Washington, DC.


3Q/2005, Economic Perspectives


Performance by periods, early (1977-91) and late (1992-2003)
Return on assets

Sharpe ratio

Equity-to-asset ratio





Change from 5 years prior to switch
to 1 year prior to switch







Change from 1 year prior to switch
to 1 year after switch







Change from 1 year prior
to 5 years after switch







Change from 1 year prior to 5 years
after switch divided by sample mean



Loan-to-asset ratio



Charge-off-to-loan ratio





Deposit-to-liability ratio






Change from 5 years prior to switch
to 1 year prior to switch







Change from 1 year prior to switch
to 1 year after switch







Change from 1 year prior
to 5 years after switch











Change from 1 year prior to 5 years



after switch divided by sample mean
‘Significant at 10 percent level.
“Significant at 5 percent level.
“‘Significant at 1 percent level.
Notes: The results are based on regressions of equation 1 with interaction terms between the period dummies and the pre- and post-switch
dummies and trend variables. Each regression has 249,988 observations. The change variables are calculated based on the coefficients on
the interaction involving the pre- and post-switch dummies and trend variables. Variable definitions are given in the text. Robust p values
adjusted for cluster effects are given in parentheses.
Source: Data from Federal Deposit Insurance Corporation, 1977-2003, Reports of Income and Condition, Washington, DC.

a large decline in charge-offs at banks that do not switch
rather than an increase at banks that switch.
The results in table 5 point out characteristics of
switchers and switching in the two periods, 1977-91
and 1992-2003. Prior to a switch, there is not much dif­
ference between the banks that switch in the early and
late periods. Return is flat, and the equity-to-asset and
loan-to-asset ratios are decreasing, although the change
in the two ratios is only significant in the early period.
The major differences between the two periods
are in the change in return and risk at banks that switch.
The most important is that return increases after a
switch only in the late period. In the early period, the
average change in return is statistically and economi­
cally insignificant. The findings for risk are mixed in
both periods. In the early period, there is no change in
the Sharpe ratio following a switch. This may reflect

Federal Reserve Bank of Chicago

the balancing of higher leverage and lower charge-offs.
In the late period, on the other hand, the Sharpe ratio
is increasing following a switch, indicating a reduction
in risk. However, leverage and charge-offs are increas­
ing, signifying higher risk.
The evidence using the accounting data is consis­
tent with beneficial competition, but only in the post1991 period. Return increases after a switch in the late
period, but not in the early period. The results for risk
are mixed, but there is no strong indication of higher
risk. In the late period, the best measure of risk—the
Sharpe ratio—signals a reduction in risk after a switch.
These findings are indicative of beneficial competi­
tion among regulatory agencies. However, before
drawing stronger conclusions, I need to examine the
direct measure of failure probabilities.


Failure probability model
The accounting data present a mixed picture of
how switching primary federal regulators affects risk.
From a social perspective, the critical issue is whether
the changes in risk promote bank failure. To directly
examine this, I use a failure prediction model.
I use two approaches to determine whether switch­
ing regulators makes a bank more likely to fail than
if it had not switched. First, I estimate a failure pre­
diction model with a dummy for whether a bank has
recently switched regulators. Since so few banks fail
in any given year (approximately 0.5 percent per year),
I look at three- and five-year horizons to minimize
noise in the model. Let FAIL DUMMY X be a variable
that takes the value 1 in year t if a bank fails prior to
the end of year / + x, where x is either 3 years or 5 years.
In my sample, an average of 1.5 percent of banks fail
over a three-year horizon and 2.3 percent fail over a
five-year horizon. For banks that switch regulators,
I include just data for the years following the switch
because a bank’s decision to switch is only observed
if it survives long enough to complete the switch. To
include switches in the failure prediction model, let
SWITCH be a dummy variable that takes the value 1
if a bank has switched regulators within the past three
years, and the value 0 otherwise.141 also interact
SWITCH with the period dummies.
For the prediction model, failure is assumed to
depend on the accounting return and risk measures used
earlier, as well as the log of total assets (since larger
banks are more diversified) and year dummies to
capture systemic movements in failure probabilities:

year dummies).
I estimate the model two ways. First, to establish
a baseline, I only include observations for banks that
never switch. Then, I include all banks. The model is
estimated over the years 1977-2001 to allow at least
three years after a switch for banks to potentially fail
(since I have failure data through 2004).
In the analysis of the accounting data, I dropped
outliers because they often have a disproportionate ef­
fect on regression results. In this section, on the other
hand, all observations are included except banks with
negative equity (since these have effectively failed al­
ready). This is because it is precisely the outlier banks,
at least those in the lower tail, that are most likely to
fail in the near term. Excluding the outliers pushes the
results more toward switches reducing the probability


of failure, although, for the most part, the differences
are not statistically significant.
Table 6 presents the results of estimating equa­
tion 3 using a logistic regression. The signs of the co­
efficients on the control variables are consistent with
expectations. Increasing either size or return decreases
failure probability, while increasing risk has the op­
posite effect. The coefficient on SWITCH is statistically
insignificant for both the three- and five-year failure
windows. This is not consistent with the hypothesis
that, all else being equal, a bank that has recently
switched regulators is more likely to fail than an other­
wise similar bank that has never switched.
The final column of table 6 includes the interac­
tion terms between SWITCH and the time dummies.
In this regression, the coefficient on SWITCH LATE
is positive and significant. The positive coefficient
on SWITCH LATE is consistent with the hypothesis
that, all else being equal, a bank that switched regula­
tors in the late period is more likely to fail than an
otherwise similar bank that has never switched.
The careful wording in the last sentences of the
previous two paragraphs reflects an assumption im­
plicit in the failure prediction model (equation 3). The
model assumes that a switching bank would have the
same risk-return profile whether or not it had switched.
In essence, it rules out the possibility that a bank is
able to, or chooses to, change its portfolio precisely
because it has switched regulators. For example, a
regulator involved in a race for the bottom might at­
tract new banks by allowing those banks to greatly
increase leverage (that is, decrease their equity-toasset ratio) after they switched to its oversight. If banks
that switched increased leverage, they would be more
likely to fail. However, if these banks failed at the
rate that otherwise similar banks with their new level
of leverage failed, then the coefficients on the switch
dummies in equation 3 would not be significantly
positive. Related to this proposition, if regulatory
specialization allows banks that switch regulators to
increase return and reduce their failure rate, but the
failure rate is still above that at otherwise similar banks
uy'/A their new ROA, then the coefficients on the switch
dummies in equation 3 would be significantly posi­
tive. Since ROA increases for banks that switch reg­
ulators in the late period, this means that the
significant positive coefficient on SWITCH LATE
does not necessarily imply that there is a race for the
bottom in that period.
A second approach is to assume that a bank would
have kept its pre-switch risk-return profile had it not
changed regulators. By taking this approach, I can
then examine whether a switching bank has a higher

3Q/2005, Economic Perspectives


Predicted failure probabilities




























































‘Significant at 10 percent level.
“Significant at 5 percent level.
‘“Significant at 1 percent level.
Notes: The regression is estimated for 1977 to 2001, with year dummies not shown. The logistic regressions in columns 1 and 3 include
all banks that never switch primary federal regulators. The logistic regressions in the other columns include banks that never switch plus
banks that have switched regulators in the previous six years (excluding the year of the switch). Variable definitions are given in the text.
Robust p values adjusted for cluster effects are given in parentheses.
Source: Data from Federal Deposit Insurance Corporation, 1977-2003, Reports of Income and Condition, Washington, DC.

failure rate after its change than its steadfast counter­
parts with similar pre-switching profiles. To do this,
I compare the predicted failure probability of the bank
in the year it switches with the actual failure rate. To
get the predicted failure probability, I use the five-year
failure rate model estimated over banks that never
switch regulators (that is, the model with coefficients
reported in column 3 of Table 6). Table 7 gives the
predicted and actual failure rates for all switches,
broken down by the time of the switch and the type
of switch (both merger-related and otherwise). There
is no statistically or economically significant difference
between the predicted and actual failure rates. Specifi­
cally, the failure rate is not higher for banks that switch
regulators in the late period, even if the switches do
not occur after a merger. This is consistent with the
positive coefficient on SWITCH LATE in table 6 arising

Federal Reserve Bank of Chicago

because banks that switch in the late period have
lower failure rates than if they had not switched, but
not as low as do banks with their new level of return.
Switches do not appear to increase failure risk.
Using a simple failure prediction model, I have shown
that for most switching banks, their post-switch fail­
ure rate is the same as that of otherwise similar banks.
The one exception is found among banks that switch
regulators after 1991. These banks fail at a higher rate
than otherwise similar banks. However, the failure
prediction model does not compare switchers to banks
that are otherwise similar to the switchers prior to their
changing regulators. In particular, in the late period,
return increases for banks after a switch. Thus, the
“higher failure rate” may be above that for banks with
the new, high ROA, but it is lower than for banks with
the pre-switch ROA. To test this, I have compared



Predicted and actual failure rates for banks that switch regulators
Predicted failure rate
over the next five
years using equation 3

Actual failures over
the next five years

p value for test of
difference between
predicted and actual
failure rate

Both periods (1977-2001)




Early period (1977-91)




Late period (1992-2001)




Notes: Failure rates over the next five years for banks that switch regulators as of the end of the year of the switch. The predicted failure
rate is based on the coefficient for regression reported in column 3 of table 6. The standard deviations of the predicted and actual failure
rates are in parentheses.
Source: Data from Federal Deposit Insurance Corporation, 1977-2003, Reports of Income and Condition, Washington, DC.

the actual failure rate to the level predicted in the
year of a switch. I have found that the actual failure
rate is no higher than the predicted rate, even for
switches after 1991. This implies that switches in
regulators do not increase the level of bank failures.

The focus of this article is on changes of primary
federal regulators. There are two potential alternative
approaches to analyzing changes among banks that
I address here. The first one involves an approach in
which the choice of a national versus state charter is
emphasized, without regard to the further choice of
taking membership in the Federal Reserve System
(for banks that elect state charters).15 Using a switch
of charters rather than a switch of primary federal reg­
ulators in the analysis does not change the qualitative
results. When I replicate the performance regressions
in table 4 or the failure prediction model in table 6 for
changes of charters rather than changes of primary
federal regulators, the same coefficients are significant
at the 5 percent confidence level.
A second approach takes into account that for
state-chartered banks, regulation is shared between
federal regulators and state regulators. To control for
the effect of state regulators, I add state dummies for
banks with a state charter. The qualitative results are
unchanged. Examining results on a state-by-state basis,
there are not enough switches to obtain meaningful
results, even for the largest states.
The choice of periods is motivated by changes in
regulation and the pattern of banks that switch. To test
the impact of the division, I run the regression (equa­
tion 1) with a separate set of switching trends and


dummies for each year in which a bank might switch.
I focus on the change in return between the year prior
to a switch and five years after a switch. This analysis
shows a distinct break between 1991 and 1992, with
the change in performance mixed for changes prior
to 1992, but consistently positive thereafter. This
suggests that the break between the early and late
periods is set correctly and is important.
In the main analysis, I exclude switches that might
be related to a merger. As discussed earlier, roughly
one-third of all switches are in the year of a merger
or the following year. Because the threshold for switch­
ing following a merger is different than for switching
at other times (and due to accounting issues), I dropped
merger-related switches from the main sample. When
I examine merger-related switches, the post-switch
changes are qualitatively similar to those for switches
at banks that did not merge in the period before and
after the switch. There is an increase in return, but only
in the late period, 1992-2003, and there is no unam­
biguous indication of an increase in either accounting
or failure risk. Prior to the switch, however, there are
differences in performance for merger-related and
other switches. Heading into a merger-related switch,
return is decreasing. This may be related to reasons
behind the merger (including accounting issues)—
and not to reasons behind the switch. Still, for the pur­
poses of this article, the key is that the post-switch
performance is similar for the two types of switchers.

This article has attempted to shed some light on
the effects of having multiple regulatory agencies in
commercial banking. I have studied the performance

3Q/2005, Economic Perspectives

of banks that switch their primary federal regulators
as an indication of whether there is beneficial compe­
tition or a race for the bottom among agencies. Whether
banks are able to increase return without increasing
risk following a switch constitutes my test for benefi­
cial competition. A race for the bottom would be evi­
denced by an increase in the failure rate of banks that
switch, especially if there is no compensatory increase
in return. Overall, I find evidence of beneficial com­
petition instead of a race for the bottom, since return
rises and failure rates remain effectively unchanged.
However, this masks important differences over time.
The reasons for switching regulators may have
changed over time. My sample includes banks that
switched between 1977 and 2003, a period of massive
changes in banking and bank regulation. I divide the
sample into two smaller periods. The early period,
1977-1991, combines two time spans—one marked
by the passage of the Depository Institutions Deregu­
lation and Monetary Control Act (DIDMCA) in 1980,
the other notable for the initial lessening of prohibi­
tions on interstate banking in the 1980s. Switches in
the late 1970s and early 1980s may be a response to
DIDMCA or to pre-DIDMCA differences among regu­
lators. Switches in the 1980s through 1991 may reflect
banks adjusting to their new competitive environment,
although the rate of switching during this period was
the lowest in my sample. Finally, in the late period,

Federal Reserve Bank of Chicago

1992-2003, prohibitions on interstate banking and on
mergers between banks and other financial firms were
essentially eliminated. Perhaps because of these chang­
es, there was again a major merger wave in banking.
I find that switches in the early part of my sam­
ple—those prior to 1992—had little impact on bank
performance. Return did not change significantly fol­
lowing a switch, and there was no unambiguous ef­
fect on accounting risk. Moreover, the evidence suggests
that bank failure rates did not increase as the result of
My results imply that banks switching regulators
in the late part of my sample, 1992-2003, increased
return without a rise in bank failures. This is evidence
of beneficial competition among regulators, and sup­
ports the hypothesis that there is specialization among
them. Interestingly, starting in 1992, there was an in­
crease in the rate of regulatory switching that lasted
through at least 2003. It is possible that the increase
in switches was associated with the onset of this type
of beneficial competition.
Finally, note that this analysis is intrinsically limit­
ed to looking at one aspect of regulatory competition.
While I find evidence of beneficial competition only
in the post-1991 period, that should not be taken to
imply that other types of beneficial competition did
not exist throughout my sample period.


’Regulatory authority for state-chartered banks is shared with the
appropriate state chartering agencies. Unless otherwise stated, when
I refer to a bank’s “regulator” (or “primary regulator”), I mean its
primary federal regulator.

2Butler and Macey (1988) point out that differences among regu­
lators are not very large due in part to the use of federal supremacy
laws. In essence, federal regulators impose their rules on state-char­
tered banks through direct regulation or by making federal deposit
insurance conditional on accepting certain rules.

8Berger and Hannan (1998) talk about the desire of bankers for a
quiet life.
9It is also possible to test the source of beneficial competition, but
this is beyond the scope of this article. See Rosen (2003).

10Results using the ratio of nonperforming loans to total loans are
more likely to indicate a reduction of risk after a switch than those
using the charge-off-to-loan ratio are. Nonetheless, I use chargeoffs rather than nonperforming loans since data on nonperforming
loans are not available for the entire sample period.

3Elliehausen (1998) gives estimates of the cost of regulation that
range between 5 percent and 15 percent of non-interest expense,
or between 2 percentage points and 6 percentage points of return
on equity.

1’Whalen (2002) finds lower return and higher risk at banks that
change charters.

4It typically takes between 15 days and 30 days to change primary
regulators. This time is necessary to get approval from the new
regulator. The approval process can be longer if the new regulator
chooses to do an exam prior to approving a new applicant; how­
ever, this is not generally done for banks that are financially strong
and well managed.

12The early period actually comprises two different subperiods,
one marked by the passage of the DIDMCA, the other notable for
the1980s merger wave. Switching activity in the DIDMCA subperiod
was higher than during the bulk of the 1980s. However, there was
no economically important difference in the relative performance
of banks that switched in either subperiod. Thus, to simplify the
exposition, I combined my findings from the two subperiods.

5Another potential drawback of having multiple regulatory agen­
cies is that the agencies may respond to their constituencies but
ignore externalities. When externalities are important, control by
local agencies may lead to too little regulation (Baumol and Oates,
1988; Stewart, 1992). As an example, for many years Britain did
not control sulfur emissions from its power plants because pre­
vailing winds blew them offshore, with most of the damage being
felt in continental Europe (Lomas, 1988). I do not examine this
here, since this sort of externality is not a big problem in banking.

13Recall that these are calculated by considering changes to the
pre- and post-switch trend and dummy variables only.

6Greenspan spoke in October 1991. Later that year, Treasury
Secretary Nicholas Brady made similar remarks. The OCC is
part of the Treasury Department.

15Whalen (2002) also examines banks that change their charters; how­
ever, that paper does not examine post-change performance indicators.

14I use the three years following a switch as the base years (and
thus, look at failures for either the first six or eight years after a
switch). The reason to restrict how long after a switch I examine
is that, eventually, one cannot attribute a failure to be the direct
result of a switch. However, looking out further after a switch
does not change the qualitative results.

7In other industries, interpretation of regulations most frequently
occurs at the agency level. There is literature that studies whether
regulatory agencies act as Congress wants them to (see, for example,
Libecap, 1996).


3Q/2005, Economic Perspectives


Amel, Dean F., 1991, “State laws affecting commer­
cial bank branching, multibank holding company ex­
pansion, and interstate banking,” Board of Governors
of the Federal Reserve System, working paper.

Kane, Edward J., 1989, The S&L Insurance Mess:
How Did It Happen?, Washington, DC: Urban Insti­
tute Press.

Amel, Dean F., and Martha Starr-McCluer, 2002,
“Market definition in banking: Recent evidence,”
Antitrust Bulletin, Vol. 47, No. 1, Spring, pp. 63-89.

Libecap, Gary D. (ed.), 1996, Advances in the Study
ofEntrepreneurship, Innovation, and Economic
Growth: Reinventing Government and the Problem
ofBureaucracy, Vol. 7, Greenwich, CT: JAI Press.

Baumol, William J., and Wallace E. Oates, 1988,
The Theory’ ofEnvironmental Policy, Cambridge,
UK: Cambridge University Press.

Lomas, Owen, 1988, “Environmental protection,
economic conflict, and the European Community,”
McGill Law Journal, Vol. 33, No. 3, pp. 506-539.

Berger, Allen N., and Timothy H. Hannan, 1998,
“The efficiency cost of market power in the banking
industry: A test of the ‘quiet life’ and related hypoth­
eses,” Review ofEconomics and Statistics, Vol. 80,
No. 3, August, pp. 454-465.

Revesz, Richard L., 2000, “Federalism and regula­
tion: Extrapolating from the analysis of environmental
regulation in the United States,” Journal ofInterna­
tional Economic Law, Vol. 3, No. 2, June, pp. 219-233.

Berger, Allen N., Margaret K. Kyle, and Joseph
M. Scalise, 2000, “Did U.S. bank supervisors get
tougher during the credit crunch? Did they get easier
during the lending boom? Did it matter to bank
lending?,” in Prudential Supervision: What Works
and What Doesn’t, Frederic M. Mishkin (ed.),
Chicago: University of Chicago Press, pp. 301-349.
Berger, Allen N., Gregory F. Udell, and Richard
J. Rosen, 2005, “Does market size structure affect
competition? The case of small business lending,”
Indiana University, working paper.

Butler, Henry N., and Jonathan R. Macey, 1988,
“The myth of competition in the dual banking system,”
Cornell Law Review, Vol. 73, May, pp. 677-718.
Dick, Astrid A., 2006, “Nationwide branching and
its impact on market structure, quality, and bank
performance,” Journal ofBusiness, Vol. 79, No. 2,
Easterbrook, Frank H., and Daniel R. Fischel,
1991, The Economic Structure of Corporate Law,
Cambridge, MA: Harvard University Press.
Elliehausen, Gregory, 1998, “The cost of bank reg­
ulation: A review of the evidence,” Board of Governors
of the Federal Reserve System, staff study, No. 171.

Federal Reserve Bank of Chicago

Romano, Roberta, 1998, “Empowering investors:
A market approach to securities regulation,” Ifr/e
Law Journal, Vol. 107, No. 8, June, pp. 2359-2430.

Rosen, Richard J., 2003, “Is three a crowd? Com­
petition among regulators in banking,” Journal of
Money, Credit, and Banking, Vol. 35, No. 6, part 1,
December, pp. 967-998.

Scott, Kenneth E., 1977, “The dual banking system:
Model of competition in regulation,” Stanford Law
Review, Vol. 30, pp. 1-49.
Stewart, Richard B., 1992, “Environmental law
in the United States and the European Community:
Spillovers, cooperation, rivalry, institutions,” Univer­
sity of Chicago Legal Forum, pp. 41-80.
Stigler, George J., 1971, “The economics theory of
regulation,” Bell Journal ofEconomics, Vol. 2, No. 1,
pp. 3-21.
Tiebout, Charles, 1956, “A pure theory of local ex­
penditures,” Journal ofPolitical Economy, Vol. 64,
No. 5, October, pp. 416-424.

Whalen, Gary, 2002, “Charter flips by national
banks,” Office of the Comptroller of the Currency,
working paper, No. WP2002-1, June.


Financial constraints and entrepreneurship:
Evidence from the Thai financial crisis

Anna L. Paulson and Robert M. Townsend

Introduction and summary
Poorly functioning financial markets can limit entry of
new firms and lead to inefficient production in existing
firms. Small-scale entrepreneurs that have limited ac­
cess to formal financial markets may be particularly
affected by financial constraints. Despite this, small en­
trepreneurial firms are an important source of innovation,
jobs, and economic growth in both developed and de­
veloping countries. In the U.S., 44 percent of the private
work force is employed in small firms, which account for
approximately 50 percent of non-farm gross domestic
product (GDP).1 Striking similarities exist between small
firms in the U.S. and those in developing countries. In
Thailand, for example, small firms employ 60 percent
of the work force and account for approximately 50
percent of GDP.2 Investment from banks and other for­
mal financial institutions is typically limited in small
firms. Thus, in both the U.S. and Thailand, two-thirds
of the initial investment in small firms comes from
savings and funds from family and friends?
Outside investment in small firms may be limited
for a number of reasons, including the difficulty of
providing credible information to investors about the
expected profitability of a planned investment project
or the entrepreneurial skill of a potential borrower. This
type of problem is typically called asymmetric infor­
mation. In addition, the provision of a loan may reduce
the incentives for an entrepreneur to exert the neces­
sary effort to make a project successful, since the prof­
its of a successful project will have to be shared with
investors. This type of problem is called moral hazard.
Asymmetric information and moral hazard are concerns
in both developed and developing economies. However,
these problems are likely to be acute in developing
economies where financial markets are less efficient.
When financial markets are less developed, en­
trepreneurial activity may also be vulnerable to events
like the Asian Financial Crisis. This crisis began in


July 1997 when the Thai government abandoned its
policy of pegging the value of Thailand’s currency,
the baht, to a basket of developed countries’ currencies
heavily weighted to the U.S. dollar. The Asian Finan­
cial Crisis led to widespread turmoil in international
financial markets and to recessions in many Asian coun­
tries. In the wake of the crisis, the Thai economy en­
tered a period of marked contraction. In 1997 Thailand’s
GDP fell 1.5 percent, and in 1998 it fell 11 percent.4
At the same time, entrepreneurial activity in
Thailand increased. In the 12 months following the
onset of the crisis, data from a survey we conducted
reveal that the number of business households more
than doubled (see figure 1). In the spring of 1997, ap­
proximately 11 percent of survey households operated
a business. One year later, the percentage had tripled,
with more than 30 percent of the survey households
operating a business. By studying entrepreneurial activ­
ity in Thailand before, during, and after the financial
crisis, we can enhance our understanding of entrepre­
neurship and financial constraints generally, and im­
prove our understanding of the role of small businesses
during a period of economic contraction.
We use new longitudinal data from rural and semiurban Thailand to examine the factors that influence
entrepreneurial activity in the pre-crisis and crisis pe­
riods. The data cover an interval from the spring of 1997
to the spring of 2001, so we are also able to gain

Anna L. Paulson is a senior economist at the Federal
Reserve Bank of Chicago. Robert M. Townsend is the
Charles E. Merriam Distinguished Service Professor of
the Department of Economics at the University of Chicago
and a consultant to the Federal Reserve Bank of Chicago.
The authors wish to thank Kristin Butcher, Craig Furfine,
Xavi Gine, Ellen Rissman, and Alicia Williams for helpful
comments, and Shirley Chiufor excellent research assistance.
They are also grateful to Sombat Sakuntasathien for making
the data collection possible and to the National Science
Foundation and the National Institutes ofHealth forfunding.

3Q/2005, Economic Perspectives

some insight into the post-crisis period. We are par­
ticularly interested in entrepreneurial activity during
the crisis period.
Before the crisis, we find that wealthier house­
holds are more likely to start businesses and that they
invest more in these businesses than their less wealthy
counterparts (Paulson and Townsend, 2004). During
the crisis, however, the positive correlation between
entrepreneurial activity and wealth disappears. These
findings are robust to the inclusion of various control
variables, alternative functional form assumptions, and
various techniques for controlling for the endogeneity
of wealth. The traditional explanation of these findings
would be that financial markets were inefficient prior
to the Asian Financial Crisis, but effectively allocated
capital to entrepreneurial activities during the crisis.
However, this interpretation strains credulity,
given the major weaknesses of the Thai financial system
revealed by the crisis itself. Restricting our attention
to the operation of financial markets in rural and
semi-urban areas, where the survey takes place, we
find it difficult to imagine that imperfections in these
financial markets were somehow alleviated during
the crisis period.
Instead, we argue that rising unemployment and
falling real wages during the crisis led to changes in
the types of people who started businesses—and in
the types of businesses they started. For instance, busi­
nesses that were initiated at the height of the financial
crisis required only a median of 1,250 baht (approxi­
mately $50) in start-up capital.5 The median initial in­
vestment in businesses that were started prior to the
crisis was 36,750 baht (approximately $1,470). To put
these figures into context, note that median annual

Federal Reserve Bank of Chicago

income in Thailand in the year before the
crisis was 40,000 baht ($1,600) for non­
business households and 100,000 baht
($4,000) for business households.
In this article, we provide some in­
sights into how rural and semi-urban
households in Thailand coped with the
financial crisis. The results of this article
also underscore the importance of care­
fully controlling for changes in the re­
turns to non-entrepreneurial activities,
notably labor market conditions, in study­
ing the determinants of entrepreneurial
activity more generally. These findings
help us to understand, for example, in­
creases in self-employment observed in
the U.S. during the recession that ended
in November 2001.
The rest of this article is organized as
follows. First, we discuss some of the relevant related
literature. Then, we provide more background on the
impact of the Thai financial crisis, detail the financial
environment in the survey areas, and describe the lon­
gitudinal data that we analyze. Next, we use regres­
sion analysis to examine the role of financial
constraints in explaining patterns of entrepreneurship
before, during, and after the crisis. Finally, we con­
sider how to interpret these findings in the light of
other trends in entrepreneurial characteristics over the
1997-2001 period.

Related literature
If financial constraints were not important, then
potential entrepreneurs would make the decision to
start a business based solely on the expected profitability
of the planned endeavor. If necessary, they would be
able to get outside financing to start the project, and
their own wealth would not be a significant factor in
whether the business was started. When financial con­
straints are important, however, outside financing may
be unavailable or insufficient. Wealthier households
will be more likely to start a business than poorer ones
under these conditions.
Holtz-Eakin, Joulfaian, and Rosen (1994) use data
from tax records in the U.S. to examine the reducedform relationship between inheritance and entrepre­
neurship, and conclude that financial constraints are
important. Using U.S. data from the National Longi­
tudinal Survey of Youth (NLSY), Evans and Jovanovic
(1989) draw the same conclusion in their structural
study of the impact of wealth on career choices. On the
other hand, Hurst and Lusardi (2004) find no evidence
that entrepreneurial activity in the U.S. is affected by


financial constraints when they allow for a non-linear
relationship between wealth and entrepreneurship.
In work that is particularly relevant to this article,
Rissman (2003) and Aaronson, Rissman, and Sullivan
(2004) point to the importance of taking into account
labor market conditions when analyzing the decision
to be self-employed. Rissman (2003) models self-em­
ployment as an alternative to unemployment, suggesting
that self-employment is countercyclical. This conclu­
sion is supported by her analysis of U.S. data from the
NLSY. Aaronson, Rissman, and Sullivan (2004) also
find some evidence of countercyclical self-employ­
ment in the U.S. in their analysis of Current Population
Survey data. They find that higher rates of unemploy­
ment are associated with higher rates of self-employ­
ment. They attribute recent increases in self-employment
to weak labor market conditions during the recession
ending in November 2001.
The operation of existing businesses will also be
affected by the entrepreneur’s wealth when financial
constraints are present. In particular, financial constraints
may prevent entrepreneurs from investing the optimal
amount in their businesses. If financial constraints did
not exist, then entrepreneurs would be able to make
up the shortfall between their own funds and the prof­
it-maximizing level of investment by borrowing. In
this situation, entrepreneurial investment and entre­
preneurial wealth would be independent of one another.
When there are financial constraints, however, entre­
preneurs may be unable to borrow, or only be able to
borrow a limited amount. In this case, wealthier entre­
preneurs will be able to invest more in their own busi­
nesses, since they are less dependent on the availability
of outside financing.
Fazzari, Hubbard, and Petersen (1988) explore
this implication of financial constraints in a sample
of publicly traded manufacturing firms in the U.S. and
show that investment is sensitive to cash flows for some
firms. In their two studies, Petersen and Raj an (1994,
1995) hypothesize that banking relationships increase
small businesses’ access to credit by overcoming in­
formation problems that would otherwise constrain
the availability of credit to them. Their analysis of data
collected by the Small Business Administration (SBA)
suggests that banking relationships do indeed play this
role for small firms. In contrast, McKenzie and Woodruff
(2003) use semi-parametric techniques to show that
returns on investment do not increase with investment
in a sample of small Mexican firms, as one would
expect if financial constraints were important.
A number of other theoretical studies, relying on
a wide variety of assumptions about how financial mar­
kets operate, imply a positive relationship between


entrepreneurship and wealth and between investment
and wealth.6 Paulson, Townsend, and Karaivanov (2005)
show that moral hazard concerns limit entrepreneurial
activity in Thailand in the period leading up to the
Asian Financial Crisis.

Background and data
Thaifinancial crisis
The initial repercussions of the Thai financial crisis
were felt in large urban areas, especially in Bangkok,
where many construction workers were laid off. Total
unemployment increased from an annual rate of 1.1
percent in 1996 to 3.4 percent in 1998, and wages and
hours worked fell as well.7 By some measures, rural
areas were particularly hard hit. In these areas, unem­
ployment increased from 3 percent to 8 percent. In the
poor northeastern region, real earnings fell by 8 per­
cent.8 Workers with little education were particularly
vulnerable. Real earnings fell 13-20 percent among
those who had, at most, completed primary school.
Prices also rose during this period, with the Consumer
Price Index increasing by 14 percent from 1996 to
1998. From 1998 to 2001, annual inflation in Thailand
averaged 1.2 percent.9
The overall poverty rate in Thailand increased 24
percent from 1996 to 1999, from 17 percent to 21 per­
cent.10 However, increases in poverty were not uniform
across the country. In the Northeast, for example, ru­
ral poverty rates increased nearly 40 percent, going
from 28 percent to 39 percent. In the Central region,
rural poverty actually decreased from 13 percent to
12 percent from 1996 to 1999. However, urban pov­
erty in the Central region increased nearly 9 percent,
going from 6.96 percent to 7.59 percent.

Financial environment
The formal financial sector in Thailand provides
two main sources of funding for households in rural
and semi-urban areas: the Bank for Agriculture and
Agricultural Cooperatives (BAAC) and commercial
banks.11 Of these two, the BAAC is much more active
in rural areas. Ninety-five percent of northeastern Thai
villages and 89 percent of Central Thai villages had
at least one BAAC borrower in 1994. The BAAC of­
fers two types of loans. One is a standard collateral­
ized loan, and the other requires no formal collateral
and is secured instead through a joint liability agree­
ment with a group of farmers who all belong to a
BAAC group.
While the bulk of the BAAC’s loans are uncollat­
eralized, these loans tend to be small, and the majority
of funds are lent through collateralized loans. Com­
mercial banks are active lenders in 41 percent of Thai

3Q/2005, Economic Perspectives

villages. However, commercial bank borrowers tend
to be concentrated in the relatively prosperous Central
region, where 50 percent of villages have at least one
commercial bank borrower. In contrast, only 31 per­
cent of northeastern villages have a commercial bank
borrower. Commercial bank loans are almost always
secured with a land title. In addition to these formal
sector lenders, there are a number of quasi-formal in­
stitutions that offer savings and lending services to
villagers: village savings and lending institutions and
rice banks. It is also common for households to bor­
row from relatives and neighbors and moneylenders.
Often households will borrow from several sources
to finance one investment project.

Survey data
The data that we analyze were derived from our
own ongoing socioeconomic study in Thailand, which
is funded by the U.S. National Institutes of Health and
the National Science Foundation. The initial survey
of households, village financial institutions, and village
key informants was completed in May 1997. It covers
regions at the doorstep of Bangkok as well as in the
relatively poor Northeast. The data provide a wealth
of pre-financial crisis data from 2,880 households, 606
small businesses, 192 villages, 161 local financial in­
stitutions, 262 borrowing groups of the BAAC, and
soil samples from 1,880 agricultural plots. A subset
of these households was included in an ongoing longi­
tudinal survey, which takes place between March and
May of each year. The data we analyze cover the pe­
riod from 1997 to 2001 and include 960 households.
The study focuses on four Thai provinces that
were chosen because of the availability of retrospec­
tive data from the Thai Socio-Economic Survey (SES).
These provinces are emblematic of two distinct regions
of Thailand: rural and semi-urban households living
in the Central region, close to Bangkok, and more ob­
viously rural households living in the semi-arid and
much poorer northeastern region. The Central region
is wealthier and more developed than the Northeast.
In each province, four geographic areas, called
tambons, were chosen at random. Each tambon includes
approximately ten villages. In each sample tambon,
four villages were chosen at random.12 Fifteen house­
holds were randomly selected from each sample village.
Overall, the data include five years of information for
960 households (4 provinces x 4 tambons x 4 villages
x 15 households) from 64 Thai villages (4 provinces
x 4 tambons x 4 villages).
The data include survey year and retrospective in­
formation on wealth (household, agricultural, business,
and financial); occupational history (transitions to and

Federal Reserve Bank of Chicago

from farm work, wage work, and entrepreneurship); and
access to and use of a wide variety of formal and informal
financial institutions (commercial banks, agricultural
banks, village lending institutions, and moneylenders,
as well as friends, family, and business associates). The
data also provide detailed information on household
demographics, entrepreneurial activities, and education.
The retrospective data on wealth and interactions with
financial institutions help us to disentangle the effects
of running a business from the forces that make it
possible to start a business in the first place.
Because these data provide rich and detailed in­
formation about both the firm and the entrepreneurial
household, as well as information on financial interme­
diaries, they are particularly well designed for study­
ing the relationship between entrepreneurship and
the financial system. Economic theory emphasizes
that both firm and entrepreneurial characteristics are
important in determining the supply and demand for
credit. In many studies the available data force a fo­
cus on either the firm or the entrepreneur, but do not
allow both to be treated with equal thoroughness.13

Business characteristics
In this section we highlight some of the key fea­
tures of the data that are important for this article. The
businesses we study are quite varied and include shops
and restaurants, trading activities, raising shrimp or
livestock, and the provision of construction or trans­
portation services. We rely on household reports on
whether its members ran a business except in the case
of shrimp and fish farming. All of these activities are
treated as businesses. It is quite common for house­
holds to run a business in addition to working for wages
and farming, usually rice. Most business households
run only a single business and rely very heavily on
family workers. Only 10 percent of the businesses paid
anyone for work during the year prior to the survey.
While there are many different types of businesses,
shrimp and/or fish raising, shops, and trade account
for most of the businesses. These categories account
for 65 percent of businesses founded prior to the cri­
sis, 60 percent founded in the year of the crisis, and
39 percent founded in the immediate post-crisis period.
The distribution of business types within these cate­
gories changes substantially following the crisis. Trade
accounts for 17 percent of all businesses that were
started in the five years before the crisis. However, 47
percent of the businesses that were founded in the year
of the crisis were in trade. The trade category includes
retail and wholesale trading activities, ranging from
selling desserts in a local market to selling gasoline to
shops and gas stations.


There is substantial variation in initial investment
in new businesses over time, as we alluded to in the
introduction (see table 1). The median initial investment
in a business founded prior to the crisis, between 1992
and 1997, is 36,747 baht. The median initial investment
in a business that began at the height of the crisis in
1998 is 1,350 baht. The median initial investment in
a trading business was 52,533 baht prior to the crisis,
just 793 baht in the year of the crisis, and zero in the
three years following the crisis. For all the major busi­
ness types, median initial investment is substantially
lower for businesses founded during the first year of
the crisis and afterwards compared with businesses
founded between 1992 and 1997.
Households rely heavily on savings (either in the
form of cash or through asset sales) to fund initial in­
vestment in their businesses. Approximately 60 percent
of the total initial investment in household businesses
that were founded between 1992 and 1997 comes from
savings. Loans from commercial banks account for
about 9 percent of total business investment, and BAAC
loans account for another 7 percent. In the Northeast,
the BAAC plays a larger role compared with commer­
cial banks, and in the Central region, the opposite is
true. In the crisis and post-crisis periods, when invest­
ment is lower, the importance of credit for funding
initial investment in the business declines.
In some of the empirical work, we control for par­
ticipation in formal and informal financial markets
by business and non-business households. We group
formal and informal financial institutions into six
categories. The first, formal financial institutions, in­
cludes commercial banks, finance companies, insurance
companies, and national employee credit unions, such
as the Teachers Credit Union. The second, village in­
stitutions and organizations, is made up of production
credit groups (PCGs),14 rice and buffalo banks, and

village poor and elderly funds. Formal loans from the
BAAC, the Agricultural Cooperative, and local farm­
ers’ groups are included in the third group, agricultural
organizations. BAAC customers whose loans are se­
cured through joint liability arrangements make up
the fourth group. Moneylenders and rotating savings
and credit associations (ROSCAs) make up the fifth
and sixth groups, respectively. Households were asked
to report when they became a customer or member of
each organization. Hence, we are able to look at the
influence of participation in these organizations prior
to starting a business, as distinct from becoming a
client of an institution because of the business.
Because households were asked to report when
they acquired household and agricultural assets and
land, the data provide measures of past wealth as well
as current wealth. In the empirical work, which we
discuss in the next section, we examine the relation­
ship between past wealth (that is, wealth prior to
starting a business) and entrepreneurship. This al­
lows us to avoid some problems of endogeneity that
are likely to plague current wealth measures, since
current wealth reflects both the resources available to
start a business for potential entrepreneurs and the past
profitability of a business for current entrepreneurs.
Because we can measure wealth before a business
was founded, we can isolate the resources available
to start a business.
For the time being, however, our interest is in
current rather than past wealth. Panel A of figure 2
describes the trend in median wealth in real 1997 Thai
baht for business and non-business households over
the years 1997-2001. Business households are wealthier
than their non-business counterparts over the entire
span, and all households experience modest declines
in wealth during the crisis. Between 2000 and 2001,
median wealth increases for all households, with


Thai business types and median initial investment
Business types
Shrimp and/or fish
Retail and wholesale trade
Sample size

Median inv.







Median inv.




Median inv.




Notes: Pre-crisis refers to businesses that were started between 1992 and 1997 and were still in operation in 1997. Crisis refers
to businesses that were started in 1998 and were still in operation in 2001. Post-crisis refers to businesses that were started
between 1999 and 2001 and were still in operation in 2001. Median initial investment (median inv.) is in real 1997 Thai baht.


3Q/2005, Economic Perspectives

increases being more dramatic for business households
compared with non-business households.
In figure 2, we compare important characteristics
of business and non-business households from 1997 to
2001. Prior to the crisis, the heads of business house­
holds were more educated than the heads of non-business
households (see figure 2, panel B). Business household
heads had almost 4.8 years of schooling compared with
3.9 years for non-business household heads. Table 2
provides further details on the distribution of educa­
tion (and other variables) for business and non-busi­
ness households. While 61 percent of business and
non-business household heads had completed four years
of school in 1997, 23 percent of business household
heads had additional education compared with just
13 percent of non-business household heads.15 During
the crisis, the gap in education between business and
non-business households narrowed substantially, in­
dicating that individuals who started businesses during
the crisis were less educated than those who started
businesses prior to the crisis. Among households that
started businesses in 1999, for example, 35 percent
of household heads had less than four years of schooling
(see table 2, panel B).
We see a similar pattern with age (see figure 2,
panel C). The heads of business households tend to
be younger than the heads of non-business households.
Before the crisis, they are almost three years younger.
However, this gap virtually disappears during the cri­
sis. This indicates that the people who founded busi­
nesses during the crisis were significantly older than
the individuals who founded businesses prior to the
financial crisis.
In panel D of figure 2, we examine trends in house­
hold size for business and non-business households.
Here we see a different pattern. Business households
tend to be larger than non-business households, and the
difference increases between 1997 and 2001. There are
two potential explanations for this trend, both of them
related to urban migrants returning to rural and semiurban areas in the wake of the crisis. One possibility
is that existing business households were more likely
to be joined by family members who had migrated prior
to the crisis. Another possibility is that urban migrants
were more likely to rejoin households that did not
have a business prior to the crisis, and these migrants
spurred the creation of businesses during the crisis.
Panel E of figure 2 reports on trends in median
income (net of expenses for business and farm activi­
ties) for business and non-business households.16
Business households have higher median income
than the non-business households over the 1997-2001
period. However, while non-business income drops

Federal Reserve Bank of Chicago

modestly during the crisis, business income decreases
significantly with the onset of the crisis. In 1997 me­
dian business income is nearly 90,000 baht, and in
1998 it is just 65,000 baht. As before, there are two
potential factors that lie behind this decline. Busi­
nesses in operation prior to the crisis may have expe­
rienced a dramatic drop in income during the crisis.
In addition, businesses started during the crisis may
simply generate less income than those started before
the crisis. We return to which of these factors is like­
ly to be more important later in this article.
In panel F of figure 2, we examine trends in me­
dian expenditure for business and non-business house­
holds. Expenditure provides a measure of both current
welfare and also reflects expectations about future
economic conditions. Households that expect crisis
conditions to continue are likely to curtail their expen­
ditures more than households that expect the crisis to
be resolved relatively quickly. Median expenditure is
higher for business households compared with non­
business households throughout the 1997-2001 period,
and expenditure decreases from 1997 to 2000 and then
increases in 2001 for all households. However, business
households experience a sharper decline in expendi­
ture from 1997 to 1998 than non-business households,
potentially driven by the entry of new households into
this category. By 2001, median non-business house­
hold expenditure exceeds pre-crisis levels. For busi­
ness households, median expenditure in 2001 is still
lower than it was in 1997.
Before moving on to discuss the results of a more
formal analysis of the role of financial markets before,
during, and after the crisis period, it is useful to review
the observations that we would like to be able to ac­
count for:
■ The percentage of business households nearly tripled
during the crisis.
■ Businesses started during the crisis tend to have very
low or even no initial investment.
■ The heads of households who established businesses
during or after the crisis tend to be less educated
and older than the heads of households with busi­
nesses already in operation prior to the crisis.
■ Business households have higher wealth, net income,
and expenditure compared with non-business house­
holds, although the gap between business and non­
business households narrows during the crisis period.

Evidence of financial constraints
In this section, we consider the evidence that fi­
nancial market imperfections played a role in shaping
patterns of entrepreneurship before, during, and after



Characteristics of Thai business households vs. non-business households
A. Median household wealth

B. Average years of schooling, household head


C. Average age, household head

D. Average household size

E. Median net income

F. Median expenditure

Notes: In panels A, E, and F, amounts are in real 1997 Thai baht. In panel B, years refer to years of schooling completed by
head of household. In panel C, years refer to the age of head of household. In panel D, size refers to the number of individuals
that make up a household.


3Q/2005, Economic Perspectives


Thai household characteristics
A. Non-business households, by year

Age of head















Years of schooling— head











0-3 years (percent)






4 years (percent)






5-16 years (percent)






No. of adult males in household
No. of adult females in household
No. of children (< 18 years) in household









































Median past wealth (in 000s)






No. of observations






Mean past wealth (in 000s)

B. Business households, by year business was started

Age of head


















Years of schooling— head













0-3 years (percent)







4 years (percent)







5-16 years (percent)







No. of adult males in household
No. of adult females in household
No. of children (< 18 years) in household
Mean past wealth (in 000s)

















































Median past wealth (in 000s)







No. of observations







Notes: Standard deviations are in parentheses. For 1998 through 2001, two rows— mean past wealth and median past wealth— refer to wealth in real
1997 Thai baht in the year prior to the year the business started. For example, for the column headed 2000, past wealth is the value of wealth in 1999,
expressed in real 1997 Thai baht. However, for the column headed 1997 in panel A, past wealth is the value of wealth in 1991, expressed in real 1997
Thai baht. And for the column headed 1992-97 in panel B, past wealth is the value of wealth in 1991, expressed in real 1997 Thai baht. In panel B, for
1998 through 2001, the figures describe businesses that were started in that given year and were still in operation in 2001; the column headed 1992-97
describes businesses that were started between 1992 and 1997 and were still in operation in 1997.

Federal Reserve Bank of Chicago


the financial crisis. We examine the implications of
financial constraints for business start-ups and for
initial investment in new businesses.
In the analysis, we divide household businesses
into three groups:


Pre-crisis businesses: businesses founded between
1992 and 1997, still in operation in 1997;


Crisis businesses: businesses founded in 1998,
still in operation in 2001; and


Post-crisis businesses: businesses founded between
1999 and 2001, still in operation in 2001.

For ease of exposition, we label the third group
“post-crisis,” but we do not mean to imply that the
impact of the Thai financial crisis was limited to 1998.
We concentrate on businesses that survived for some
period because of the design of the 1997 survey. The
1997 survey identifies businesses that were in opera­
tion at the time of the survey—that is, businesses that
were started at some point in the past and were still in
operation in 1997. We restrict our attention to businesses
that were started in the five years prior to this survey.
To make sure that we are looking at roughly comparable
businesses after 1997, the analysis excludes businesses
that were started in 1998 but failed between 1998 and
2001 and businesses that were started between 1999
and 2001 and were not in operation in 2001. Of the
businesses that were founded at the height of the cri­
sis in 1998, 63 percent were still in operation in 2001.
To examine the importance of financial constraints,
we focus on two key relationships. The first is the re­
lationship between the likelihood that a household starts
a business and household wealth prior to the time that
the business was founded. The second is the relation­
ship between the initial investment in the business and
household wealth prior to the time that the business
was founded. If financial constraints are important, we
expect that business start-ups will be sensitive to the
wealth of potential entrepreneurs and that wealthier
entrepreneurs will invest more in their businesses.17
In order to evaluate the implications of financial
constraints, we need to come up with appropriate mea­
sures of entrepreneurial talent and wealth. The proxy
we use for entrepreneurial talent is education. While
education is certainly not a perfect indicator of entre­
preneurial talent, it is likely to be positively related to
business skill. In Paulson, Townsend, and Karaivanov
(2005), we show that, at least for Thailand, formal
education seems to be strongly associated with busi­
ness skill.
The appropriate wealth variable is wealth at the
time the decision is made to start a business. For the
pre-crisis analysis, we use wealth six years prior to the


1997 survey as an empirical counterpart to this variable.
We exclude households with businesses that were
founded prior to 1992 from the analysis. For the cri­
sis and post-crisis periods, we measure wealth in the
year before the business was started. The items that
are included in the wealth variable are: the value of
household and agricultural assets and land. We do not
include the value of any business assets that the house­
hold may have owned prior to starting a business.
By using past, rather than current wealth, and by
excluding business assets acquired before the business
was started, we hope to avoid issues of endogeneity:
Wealthier people are more likely to start businesses,
and business owners have higher earnings than wage
workers, which allow business owners to become
even richer. In this scenario, current wealth captures
both the cause and the effect of having been able to
start a business in the past.

Wealth and the likelihood of starting
a business
In table 3, we estimate probit models of who be­
comes an entrepreneur for the three periods. The first
set of results in this table reports on the pre-crisis find­
ings. The dependent variable is equal to one if the house­
hold runs a business in 1997 that was founded between
1992 and 1997 and zero if the household does not have
a business in 1997.18 The second set of results reports
on the crisis findings, where the dependent variable is
equal to one if the household starts a business in 1998
that survives until 2001, and it is equal to zero other­
wise. The post-crisis findings are found in the third
set of results, and the dependent variable in this re­
gression is equal to one if the household has a business
in operation in 2001, which was founded between 1999
and 2001, and it is equal to zero otherwise. The figures
reported in the table indicate the marginal effect of
an infinitesimal change in each continuous variable
on the probability of starting a business. For dummy
variables, we report the impact of changing the vari­
able in question from zero to one.
In addition to wealth prior to starting a business,
the explanatory variables include characteristics of
the household head that may be indicators of business
talent—age, age squared, and years of schooling. There
are also variables that control for the amount of house­
hold labor that is available—the number of adult males,
adult females, and children under the age of 18 living
in the household.19
We control for credit market availability by in­
cluding measures of whether the household was a mem­
ber or customer of various financial institutions in
the past. Like the labor supply variables, we include

3Q/2005, Economic Perspectives


Probit estimates of Thai business start-ups

Age of head
Age of head squared
Years of schooling— head
No. of adult males in household
No. of adult females in household
No. of children (< 18 years) in household
Past wealth
Past wealth squared
Past member or customer of
Formal financial institutions3
Village institutions/organizations3
Agricultural lenders3
BAAC groups"
Pseudo R-squared (%)
Log likelihood
No. of observations





















aDummy variables.
Notes: Pre-crisis refers to businesses that were started between 1992 and 1997 and were still in operation in 1997. Crisis refers to
businesses that were started in 1998 and were still in operation in 2001. Post-crisis refers to businesses that were started between
1999 and 2001 and were still in operation in 2001. For dummy variables, dF/dx represents the change in probability when the dummy
variable goes from zero to one. For all other variables, dF/dx is the change in probability from an infinitesimal change in the independent
variable in question. Past wealth is made up of the value of household assets, agricultural assets, and land. The coefficient on past wealth
in the table is the actual one x 106. The coefficient on past wealth squared is the actual one x 1012. Sixteen geographic controls are also
included (tambons).

these variables so that we can appropriately interpret
the coefficient of the wealth variable. In order to sep­
arate the impact of the availability of a particular
credit institution in the local area from the impact of
being a client of the institution, the estimates also in­
clude controls for each of the tambons that were
sampled. The tambon controls are meant to capture
geographic variations in the supply of credit along with
other important characteristics, such as infrastructure
and the size of the market. The inclusion of the tam­
bon controls means that the credit market variables
provide an indication of the average probability that
patrons of the various institutions will start business­
es, relative to the probability that households in a
particular tambon will start businesses.
During the pre-crisis period, the likelihood that a
household starts a business is positively related to pre­
existing wealth. In particular, the coefficients reported
in the first set of results imply that a 1,000,000 baht
($40,000) increase in wealth would be associated with
a 2.3 percentage point increase in the likelihood of
starting a business.20 This is an increase of 21 percent
above the observed percentage of households that have
started a business in the past five years. The coefficient
on wealth squared is significant, although very small,
suggesting that the impact of wealth on starting a
business decreases as wealth increases.

Federal Reserve Bank of Chicago

In contrast to the pre-crisis findings, during the
crisis and post-crisis periods, there is no statistically
significant relationship between wealth and the like­
lihood of starting a business. This suggests that the
importance of financial constraints declines during
the crisis and post-crisis periods.
Table 3 estimates also reflect trends in the differ­
ence between the characteristics of business and non­
business households over the crisis period, described
previously. Prior to the crisis, older household heads
are significantly less likely to start a business. During
and after the crisis, there is no significant relationship
between the age of the household head and the likeli­
hood of starting a business. More education is associ­
ated with a greater likelihood of starting a business
prior to the crisis, but has no significant impact on
business start-ups during the crisis. Larger households,
as captured by the number of adult males and females,
are more likely to start businesses during and after
the crisis. These variables have no significant impact
on the likelihood of starting a business prior to the
crisis. Business talent appears to have been more im­
portant prior to the crisis than during the crisis, and
the availability of household labor seems to be more
important during the crisis than before the crisis.


In general, access to credit, as measured by past
patronage of the various financial institutions, does not
seem to play an important role in business start-ups
before, during, or after the crisis. With one exception,
the variables that control for access to credit are in­
significant. During the crisis, however, households
that had a prior relationship with the BAAC, in the
form of a joint liability borrowing arrangement, are
7.5 percentage points more likely to start a business
than those without prior ties to the BAAC. This cor­
responds to nearly a 30 percent increase in the likeli­
hood of starting a business during the crisis period.

Wealth and initial business investment
In table 4, we examine the relationship between
initial business investment and preexisting household
wealth for pre-crisis, crisis, and post-crisis business­
es. In these regressions, the log of initial business

investment plus one is regressed on household wealth
prior to the period when the business was started. In
panel A, the sample includes only businesses with
positive initial investment. In panel B, the sample is
augmented with businesses that began with zero initial
investment. When we restrict the sample to business­
es with positive initial investment, as we do in panel
A, it makes it more difficult to find no relationship
between investment and wealth.
In addition to household wealth, these regressions
also include the same household controls discussed
earlier.21 For businesses with positive initial investment,
higher levels of wealth prior to starting a business are
associated with greater initial business investment
prior to the crisis and after the crisis but not during
the crisis (see table 4, panel A). An increase in past
wealth of 1,000,000 baht is associated with an increase
in investment of 46 percent prior to the crisis. These


Regression estimates of log initial Thai business investment
A. Businesses with initial investment greater than zero

Age of head
Age of head squared
Years of schooling— head
No. of adult males in household
No. of adult females in household
No. of children (< 18 years) in household
Lag wealth
Lag wealth squared
Adjusted R-squared (%)
No. of observations














B. All businesses

Age of head
Age of head squared
Years of schooling— head
No. of adult males in household
No. of adult females in household
No. of children (< 18 years) in household
Lag wealth
Lag wealth squared
Adjusted R-squared (%)
No. of observations















Notes: Pre-crisis refers to businesses that were started between 1992 and 1997 and were still n operation in 1997. Crisis refers to
businesses that were started in 1998 and were still in operation in 2001. Post-crisis refers to businesses that were started between
1999 and 2001 and were still in operation in 2001. Lag wealth is made up of the value of household assets, agricultural assets, and
land in the year prior to starting a business. The coefficient on lag wealth is the actual one x 106. The coefficient on lag wealth squared
is the actual one x 1012. The dependent variable is the natural log of initial investment plus one In panel A, only businesses with non-zero
initial investment are included. In panel B. all businesses, regardless of initial investment, are included.


3Q/2005, Economic Perspectives

findings suggest that financial market imperfections
restrict investment levels prior to the crisis and after
the crisis but not during the crisis itself.
After the height of the crisis in 1998, the impor­
tance of financial constraints on investment levels
appears to return, at least for businesses with positive
initial investment. For these businesses, an increase
in past wealth of 1,000,000 baht is associated with an
increase in investment of 26 percent. Interestingly, dur­
ing the crisis more educated business household heads
invest significantly more in their businesses. There is
some evidence that this is also the case prior to the
crisis, but the size and the significance of the coeffi­
cient on schooling is smaller.
When we include businesses that begin with zero
initial investment (see table 4, panel B), we find no
relationship between initial business investment and
past wealth before, during, or after the crisis.22 Edu­
cation is a strong predictor of initial business invest­
ment during the crisis and post-crisis periods according
to these estimates, although the magnitude of the
effect is fairly small. An additional year of schooling
is associated with an increase in initial investment of
1.3 to 1.4 baht. Keep in mind, however, that 37 per­
cent of the crisis businesses and 56 percent of the post­
crisis businesses had zero initial investment.
Overall, the relationship between investment and
past wealth suggests that financial constraints led to
underinvestment in existing businesses prior to the crisis,
and possibly after the crisis, but did not place important
restrictions on business investment during the crisis.


Performance of Thai business households
A. Median gross income

B. Median expenditure

Business performance
In figure 3, we examine the performance of the
three groups of business households from 1997 to
2001. We examine three indicators of business house­
hold success: gross income, expenditure, and profit
(panels A, B, and C, respectively). Figure 3 under­
scores the emerging picture that households that start
businesses during and after the crisis are different along
important dimensions from households that were run­
ning businesses when the crisis hit. Gross income,
expenditure, and profit are all much higher for house­
holds that were already running a business at the time
of the crisis compared with households that started a
business during or after the crisis. Businesses found­
ed in the post-crisis period have notably lower profits
(figure 3, panel C). One potential explanation for this
finding is that households with more entrepreneurial
talent started businesses earlier—either before the
crisis or during the crisis. The businesses that were
founded in the post-crisis period may be operated by

Federal Reserve Bank of Chicago

C. Median profit

Note: All are measured in real 1997 Thai baht


relatively untalented individuals, and hence have very
low profits.
These patterns suggest that the narrowing gap be­
tween business and non-business households—in terms
of wealth, net income, and expenditure (figure 2, panels
A, E, and F, respectively)—is primarily due to the entry
of new businesses with lower income and expenditure
during and after the crisis rather than a weakening of
the economic status of existing businesses. Note, in
particular, that the income of households that had busi­
nesses at the time of the crisis went up from 1997 to
1999 at the height of the crisis (figure 3, panel A).

Beginning with the observation that the number
of household businesses in rural and semi-urban
Thailand nearly tripled in the wake of the Thai finan­
cial crisis, we describe and analyze a number of im­
portant features of pre-crisis, crisis, and post-crisis
businesses. In particular, we show that businesses
started during and after the Thai financial crisis are
more similar to non-business households than house­
holds that started businesses prior to the crisis. Prior
to the crisis, business start-ups and initial investment
are significantly related to past household wealth.
However, during the crisis, business start-ups and
initial investment are unaffected by household wealth.
In addition, crisis and post-crisis businesses are char­
acterized by low initial investment.
During the post-crisis period, business start-ups
are unaffected by wealth, but initial business invest­
ment (for businesses with non-zero investment) is in­
creasing with wealth. Recall that the median business
founded during the post-crisis period has zero initial
investment. Profits are highest for businesses started
prior to the crisis and lowest for businesses started
during the post-crisis period. Compared with busi­
nesses started during and after the crisis, pre-crisis
businesses appear to recover faster and more sharply.
Financial market imperfections seem to restrict
business start-ups and investment prior to the crisis
but not during the crisis. What might account for this
finding? It seems plausible to rule out improvements
in financial markets as an explanation, since the cri­
sis itself suggests that Thai financial markets are


(or at least were) quite fragile. The key to understand­
ing the apparent lack of financial constraints during
the crisis period in Thailand—and how financial con­
straints have an impact on entrepreneurial activity
more generally—is to consider the alternative occu­
pations available to households.
The model of Evans and Jovanovic (1989) pro­
vides a useful framework for understanding the in­
crease in business activity during the Thai financial
crisis and over the business cycle. Their model im­
plies that when wages fall, more businesses will be
started as the returns to entrepreneurial activity ex­
ceed wages for more households. In addition, this
model implies that the new businesses will tend to be
capitalized at lower levels and be run by less talented
entrepreneurs. We see evidence of this in the data—
crisis and post-crisis investment levels are very low,
profits are also low, and the household heads that
founded crisis and post-crisis businesses are also less
educated than those that founded businesses prior to
the crisis. We can reconcile the facts we have de­
scribed above by understanding how falling wages
affect both who finds entrepreneurial activity profit­
able and how much they invest in business activity.
As alternatives to business employment wors­
ened during the Thai financial crisis, households be­
gan businesses because their wage employment
options deteriorated. Low capital business opportuni­
ties that were unattractive prior to the crisis looked
good during the crisis. Note that business investment
during the crisis period generated lower profits than
pre-crisis investment. Despite the finding that busi­
ness start-ups and investment are insensitive to
wealth during the crisis, there was no improvement
in financial markets during this period. Instead, typi­
cal business investment during the financial crisis
was so low that credit was not required.
This article’s findings underscore the general im­
portance of taking into account economic conditions
at the time a business is founded in order to account
for firm investment and profitability. This insight ex­
tends to both developed and developing countries,
and applies to dramatic events like the Thai financial
crisis, as well as to more modest business cycle type
variation in economic conditions.

3Q/2005, Economic Perspectives

1 Small Business Administration (SB A) statistics are drawn from
the U.S. Bureau of the Census and Current Population Survey
data. According to the SBA, small firms are defined as manufac­
turing firms with fewer than 500 employees and non-manufactur­
ing firms with less than $5 million in annual sales.

2APEC (Asia-Pacific Economic Cooperation) Center for Technol­
ogy Exchange and Training for Small and Medium Enterprises.
Small Thai firms include manufacturing and service firms with 50
or fewer employees; wholesale trade firms with 25 or fewer em­
ployees; and retail trading operations with 15 or fewer employ­
ees. Medium-sized firms may have up to 200, 50, and 30
employees in each of these categories, respectively.
3This is determined from Bitler, Robb, and Wolken (2001) and
calculations from the authors’ survey from Thailand.
4In the years leading up to the crisis, the Thai economy had grown
rapidly. From 1980 to 1995, real per capita GDP had grown 8 per­
cent per year. Following the crisis, the Thai economy recovered
somewhat, and real per capita GDP growth averaged 3 percent per
year from 1999 to 2001 (World Bank, World Development Indicators).

Throughout this article, monetary values are reported in real
1997 Thai baht. Prior to the devaluation in July 1997, 25 Thai
baht equaled 1 U.S. dollar (25 baht = $1).

Tor example, these implications are shared by a model where there
is no credit (Lloyd-Ellis and Bernhardt, 2000), a model where
credit is exogenously limited to be a fixed multiple of household
wealth (Evans and Jovanovic, 1989), and a model where credit is
allocated as the optimal solution to an information-constrained
moral hazard problem (Aghion and Bolton, 1997). They are also
consistent with the asymmetric information framework emphasized
by Fazzari, Hubbard, and Petersen (1988, 2000).
’Unemployed individuals are those who are currently not working
but are actively looking for work (World Bank, World Development
8See The World Bank Group (2000).

Trior to the crisis, inflation in Thailand was determined by infla­
tion in the currencies to which the Thai baht was pegged; this means
that price increases in Thailand largely mimicked those of the U.S.
1OA11 poverty rate figures are reported in Thailand Development
Research Institute (2003) and are based on calculations from the
Thai National Statistics Office, Socio-Economic Survey (SES) data.
The poverty rate is defined as the percentage of people in a given
region living below the poverty line for that region.

12Each village is a distinct political entity with an elected headman
or woman, very much like a mayor.
13For example, the National Longitudinal Survey of Youth, analyzed
by Evans and Jovanovic (1989), has detailed information on the
self-employed, but very sparse information on the businesses they
run. The Small Business Administration (SBA) data analyzed by
Petersen and Rajan (1994, 1995) provide a wealth of details about
the firm but very little information about the entrepreneur.

14These are village-run savings institutions where members pledge
to save a certain amount and interest earnings are determined by the
profitability of the whole institution for the year. A sizable frac­
tion of PCGs offer loans, which are secured by savings, as well.
15Four years of schooling was the statutory minimum at the time
most of the sample’s household heads were in school.
16In each survey year, households were asked to report on income
and expenditure for the 12 months prior to the survey. Thus, for
the survey year 1997, income and expenditure figures cover the
period from the spring of 1996 to the spring of 1997.
17We explain why financial constraints generate these predictions
in the Related literature section.
18Households with businesses that are operating in 1997 but were
founded prior to 1992 are eliminated from the analysis.
19In table 2, these variables are summarized in panel A for non-busi­
ness households, by year, and in panel B for business households,
by the year the business was started.
20A 1,000,000 baht increase in wealth corresponds to doubling the
current wealth of the median business household in 1997 and tri­
pling the wealth of the median non-business household.

21Because the sample sizes are smaller here, we do not control for
past use of financial institutions and geographic location.
22We have experimented with different statistical models and got­
ten qualitatively similar results. For example, we have estimated
probit models where 0 corresponds to zero initial investment and 1
corresponds to positive initial investment and ordered probit models
where 0 corresponds to zero initial investment, 1 corresponds to ini­
tial investment of less than 10,000 baht, and 2 corresponds to initial
investment greater than 10,000 baht.

nThis section is based on the authors’ observations and discussions
with BAAC officials as well as on data from the Community De­
velopment Department of the Thai Ministry of the Interior that cover
60,000 Thai villages every other year from 1988 through 1994.

Federal Reserve Bank of Chicago



Aaronson, Daniel, Ellen R. Rissman, and Daniel
G. Sullivan, 2004, “Assessing the jobless recovery,”
Economic Perspectives, Federal Reserve Bank of
Chicago, Vol. 28, No. 2, pp. 2-20.

Aghion, Phillippe, and Patrick Bolton, 1997, “A
trickle-down theory of growth and development,”
Review ofEconomic Studies, Vol. 64, No. 2, April,
pp. 151-172.
Binford, Michael, Tae Jeong Lee, and Robert
Townsend, 2001, “Sampling design for an integrated
socioeconomic and ecologic survey using satellite re­
mote sensing and ordination,” University of Chicago,
Bitler, Marianne P., Alicia M. Robb, and John D.
Wolken, 2001, “Financial services used by small
businesses: Evidence from the 1998 Survey ofSmall
Business Finance f Federal Reserve Bulletin, Board
of Governors of the Federal Reserve System, April,
pp. 183-205.

Evans, David S., and Boyan Jovanovic, 1989, “An
estimated model of entrepreneurial choice under liquid­
ity constraints,” Journal ofPolitical Economy, Vol. 97,
No. 4, August, pp. 808-827.

Fazzari, Stephen M., R. Glenn Hubbard, and Bruce
C. Petersen, 2000, “Investment-cash flow sensitivi­
ties are useful: A comment on Kaplan and Zingales,”
Quarterly Journal ofEconomics, Vol. 115, No. 2,
pp. 695-705.
__________ , 1988, “Financing constraints and cor­
porate investment,” Brookings Papers on Economic
Activity,No\. l,pp. 141-195.
Greenwood, Jeremy, and Boyan Jovanovic, 1990,
“Financial development, growth, and the distribution
of income,” Journal ofPolitical Economy, Vol. 98,
No. 5, pp. 1076-1107.

Holtz-Eakin, Douglas, David Joulfaian, and Harvey
S. Rosen, 1994, “Sticking it out: Entrepreneurial sur­
vival and liquidity constraints,” Journal ofPolitical
Economy, Vol. 102, No. 1, February, pp. 53-75.
Hurst, Erik, and Annamaria Lusardi, 2004, “Liquid­
ity constraints, household wealth, and entrepreneurship,” Journal ofPolitical Economy, Vol. 112, No. 2,
April, pp. 319-347.


Kaboski, Joseph, and Robert Townsend, 1998,
“Borrowing and lending in semi-urban and rural
Thailand,” University of Chicago, manuscript.

Lerner, Josh, 1999, “The government as venture
capitalist: The long-run impact of the SBIR program,”
Journal ofBusiness, Vol. 72, No. 3, July, pp. 285-318.
Lloyd-Ellis, Huw, and Dan Bernhardt, 2000, “En­
terprise, inequality, and economic development,” Re­
view ofEconomic Studies,No\. 67, No. l,pp. 147-168.

McKenzie, David, and Christopher Woodruff,
2003, “Do entry costs provide an empirical basis for
poverty traps? Evidence from Mexican microenter­
prises,” Stanford University, manuscript.
Paulson, Anna L., and Robert M. Townsend,
2004, “Entrepreneurship and financial constraints
in Thailand,” Journal of Corporate Finance, Vol. 10,
No. 2, pp. 229-262.

Paulson, Anna L., Robert M. Townsend, and
Alexander K. Karaivanov, 2005, “Distinguishing
limited liability from moral hazard in a model of en­
trepreneurship,” Federal Reserve Bank of Chicago,
working paper, revised April 2005, No. WP-2003-06.
Petersen, Mitchell A., and Raghuram G. Rajan,
1995, “The effect of credit market competition on
lending relationships,” Quarterly Journal ofEconomics,
Vol. 110, No. 2, May, pp. 407-443.

__________ , 1994, “The benefits of lender relation­
ships: Evidence from small business data,” Journal
ofFinance, Vol. 49, No. 1, pp. 3-37.

Rissman, Ellen R., 2003, “Self-employment as an
alternative to unemployment,” Federal Reserve Bank
of Chicago, working paper, No. WP-2003-34.
Thailand Development Research Institute, 2003,
“Thailand Economic Information Kit: 2003,” Bangkok,
Townsend, Robert M., with Anna L. Paulson,
Sombat Sakuntasathien, Tae Jeong Lee, and
Michael Binford, 1997, “Questionnaire design and data
collection for NICHD grant ‘Risk, insurance, and the
family’ and NSF grants,” University of Chicago.

World Bank Group, The, 2000, “Thai workers and
the crisis,” Thailand Social Monitor, July.

3Q/2005, Economic Perspectives

Seasonal monetary policy

Marcelo Veracierto

Introduction and summary
It is widely known that economic activity does not
evolve smoothly over the course of a year, but that it
varies systematically across the different seasons. This
is not surprising: Weather is an important factor in
many sectors of production. While agriculture is an
obvious example, construction is another important
activity affected by weather: No doubt, it is much
harder to build a house in Chicago during the winter
months than during the rest of the year. Institutional
arrangements also lead to seasonal fluctuations in
economic activity. For instance, a disproportionate
fraction of American families take vacations during
the summer months partly because they coincide with
school recess. Another example is Christmas, which
sharply increases retail activity during the last month
of the year. While most modem discussion about
monetary policy centers on what is the best policy to
follow over booms and recessions, very little is said
about what is the best policy to follow across differ­
ent seasons. However, this has not always been the
case. The evolution of U.S. monetary institutions and,
in particular, the creation of the Federal Reserve Sys­
tem have been partly guided by this discussion.
Before the creation of the Federal Reserve System
in 1914, the U.S. monetary system was commonly
criticized for its alleged “inelasticity” in responding
to fluctuations in the demand for credit. While some
of these fluctuations were associated with business
cycles and bank panics, an important part of them were
the result of regular seasonal fluctuations in economic
activity. As a matter of fact, in those days it was com­
mon for the U.S. economy to go through recurrent
periods of monetary tightness during the fall crop-mov­
ing and Christmas seasons (September through De­
cember). To illustrate this, it suffices to consider the
seasonal pattern for short-term interest rates. The rea­
son is that, to the extent that the end-of-year increase

Federal Reserve Bank of Chicago

in the demand for credit was not matched by a com­
parable increase in money supply, the short-term in­
terest rates would have to increase. A classic source
for the seasonal behavior of interest rates is Kemmerer
(1910), who reported the seasonal weekly pattern for
average interest rates on call loans in the New York
Stock Exchange between 1890 and 1908. Indeed,
Kemmerer showed a strong seasonal pattern: He re­
ported that the call rate decreased quite rapidly from
7.38 percent during the last week of the year to 2.50
percent during the last week of January. Moreover,
after a long period of relative stability, the call rate
increased from 3.04 percent during the first week of
September to reach a peak of 7.38 percent during the
last week of the year.
To use the words of Friedman and Schwartz (1963,
p. 292): “That seasonal movement was very much in
the minds of the founders of the (Federal Reserve)
System and was an important source of their belief in
the need for an ‘elastic’ currency.” In fact, the creation
of the Federal Reserve System in 1914 changed the
seasonal behavior of interest rates quite dramatically.
Figure 1 shows the average call rate in New York City
during the periods 1890-1913 (before the creation of
the Fed) and 1915-28 (after the creation of the Fed,
but before the Great Depression). For the period be­
fore the creation of the Fed, we see the same season­
al pattern that Kemmerer reported in weekly data:
Interest rates rising steadily between September and
December, and dropping sharply in January. During
the period after the creation of the Fed, we see inter­
est rates behaving much more smoothly. We still
Marcelo Veracierto is a senior economist in the Research
Department at the Federal Reserve Bank of Chicago.
The author benefited from helpful discussions with Bruce
Smith while they were colleagues at Cornell University.
He also thanks David Kang and Tina Lam for research


observe a noticeable increase at the end of the year, but
it is small compared to the sharp increases that took
place before the creation of the Fed. This type of evi­
dence led Friedman and Schwartz (1963, p. 293) to
claim that “the System was almost entirely successful
in the stated objective of eliminating seasonal strain.”
In order to attain such a smooth path for interest
rates, the Federal Reserve had to meet the seasonal
variations in demand with accommodating expansions
and contractions in the supply of high-powered money.
Indeed, after presenting supporting evidence, Friedman
and Schwartz (1963, p. 294) stated that “the seasonal
variation in currency outside the Treasury and Feder­
al Reserve Banks and, we presume, in the total stock
of money were decidedly wider in the 1920s than in
the earlier periods.” In recent times the Federal Re­
serve has continued to generate large seasonal varia­
tions in the quantity of money. Figure 2 reports the
seasonally unadjusted monetary base growth rate be­
tween 1959:Q2 and 1988:Q2. We see that the mone­
tary base follows a strong seasonal pattern: Its growth
rate is relatively low in the first quarter of the year and
increases monotonically throughout the rest of the year.
The purpose of this article is to evaluate the con­
sequences of the Federal Reserve following this type
of seasonal policy. While smoothing interest rates
across seasons was one of the initial objectives of the
Federal Reserve System, it is surprising how little
work has been done to analyze the associated effects.
Would allocations and welfare be significantly differ­
ent if, instead of following an “elastic” monetary pol­
icy across seasons, the Fed would follow more of a
“lean against the wind” stance? More precisely, what
would be the consequences of following a constant
growth rate of money instead of smoothing interest
rates across seasons?


The main exercise in this article is to analyze what
would the effects be of switching from the seasonal
money growth rates that the Fed engineers to a con­
stant growth rate of money. The results in terms of
nominal interest rates are quite dramatic. Under a
constant money growth rate, the nominal interest rate
would be constant during the first three quarters, but
would more than double during the last quarter of the
year. That is, the pattern for nominal interest rates
would resemble the one corresponding to the period
before the creation of the Federal Reserve System.
On the contrary, under current Federal Reserve policy,
most of the seasonal variations in nominal interest
rates are eliminated. Despite this, the seasonal mone­
tary policy regime has no important consequences for
real allocations: The seasonal patterns for consump­
tion, output, hours worked, and real cash balances
are basically the same if the Fed smooths interest rates
or if it follows a constant rate of growth of money.
As a consequence, the welfare effects of both types
of policies are virtually the same.
Smoothing interest rates across the different sea­
sons would have more significant effects if the nomi­
nal interest rate targeted were equal to zero at every
quarter, that is, if the Federal Reserve followed the
celebrated “Friedman rule.” Output would increase by
1.1 percent in every season. However, the welfare bene­
fits of switching to the zero interest rates would still
be small: only 0.1 percent in terms of consumption.
The rest of the article is organized as follows.
The related literature is discussed in the next section.
I describe the environment in the third section. The
benchmark economy is calibrated to U.S. data in
the fourth section. I compare the effects of different

3Q/2005, Economic Perspectives

seasonal monetary rules in the fifth section. In the
sixth section, I investigate the main source of seasonal
fluctuations in the U.S. economy. Three appendices
provide all the technical details.

Related literature
This is not the first article to analyze the effects
of seasonal monetary policy.1 Miron (1986) analyzed
the problem of a large number of identical banks that
take the nominal interest rate as given and must de­
cide how to allocate their deposits into reserves and
loans. The banks face a cost function, which depends
on their reserve-deposit ratios and on the stochastic
realization of a variable called “withdrawals.” The model
is closed with an exogenous amount of deposits and
a demand function for loans that depends negatively
on the interest rate and an exogenous activity level.
The price level and inflation rate are treated as exog­
enous. Analyzing this framework, Miron finds that
if the Federal Reserve controls the demand for loans
(through open market operations) in such a way that
equilibrium rates are smoothed across different seasons,
banks respond by reducing their seasonal changes in
reserve-deposit ratios, which in turn lowers the aver­
age costs that the banks face (given the convexity of
the cost function). This result is interpreted as a re­
duction in the likelihood of bank panics. While the
paper illustrates that smoothing interest can decrease
bank panics, it is hard to assess how plausible the model
is, given its highly stylized nature and the lack of quan­
titative analysis.
Mankiw and Miron (1991) also provide an anal­
ysis of seasonal monetary policy, but using an IS-LM
framework. After parameterizing the equations to U.S.
observations, they use their model to evaluate the
benefits of smoothing nominal interest rates across
seasons, against the alternative of holding the stock of
money constant across seasons. They find, both under
“classical” and “Keynesian” assumptions, that holding
the stock of money constant would lead to extremely
seasonal interest rates: The seasonal amplitude would
be about 500 basis points. They also find that, even
under extreme Keynesian assumptions about the price
level, moving to a constant stock of money regime would
have small effects on the seasonal behavior of output.
This article is more closely related to Mankiw and
Miron (1991) than to Miron (1986), since it is completely
silent on “bank panics.” However, a big methodolog­
ical difference is that it follows a modern dynamic
general equilibrium approach instead of an IS-LM
analysis. An advantage of this approach is that it al­
lows us to evaluate any welfare benefit of changes in
monetary policy. Another advantage is the internal

Federal Reserve Bank of Chicago

consistency between microeconomic decisions and
macroeconomic outcomes. Despite these important
differences, this article obtains results that are quite
similar to Mankiw and Miron (1991): Switching to a
smooth money rule would lead to extremely seasonal
nominal interest rates but would have negligible effects
on real variables.

The model economy
This article uses a prototype model that has been
previously used to evaluate the effects of monetary
policy over the business cycle. The model is the one
studied by Cooley and Hansen (1995), which intro­
duces a cash-in-advance constraint similar to Lucas
and Stokey (1983) into the real business cycle model
analyzed by Hansen (1985). An important difference
with Cooley and Hansen (1995) is that, instead of hav­
ing stochastic shocks, this article introduces systematic
seasonal changes in preferences, technology, and
monetary policy.
The model has a representative agent that likes
consuming both a cash good and a credit good, but
dislikes working. The household rents labor and capital
to a representative firm, which uses them to produce
the two consumption goods and investment. The house­
hold uses the wage and rental income that it receives
from the firm, together with a lump-sum transfer of
cash that the agent receives from the government, to
purchase consumption goods, investment goods, cash,
and bonds. Consumption of the cash good is subject
to a cash-in-advance constraint. The cash transfers
that the household receives from the government are
completely financed by monetary injections.
In this economy the time discount rate, the weight
of the cash good in the utility function, the disutility of
work, total factor productivity, and the growth rate of
money vary deterministically across seasons. Parameter
values will be calibrated to reproduce the seasonal
fluctuations in consumption, investment, hours worked,
real cash balances, and money growth rate observed
in U.S. data. Once the model is calibrated to the U.S.
seasonal cycles, it will be used to assess the conse­
quences of Federal Reserve monetary policy.
Hereon, a season will be identified with a quar­
ter. For this reason, it will be important to keep track
of the year and quarter of the different variables in
the model economy. In what follows, x will denote
the value of variable x in year t and quarter s, for
,s =1..... 4. To simplify notation, x(0 will be understood
to be x(14. Similarly, x(5 will refer to x(+1 r A detailed
description of the model economy now follows.
The economy is populated by a large number of
identical agents. Each agent is endowed with one unit


of time every period and has preferences described by
the following utility function:






Zp'IXK111^ +(1-aJlnau -ya3

where 0 < B < 1 is the annual discount factor, c, is
consumption of a cash good, a(s is consumption of a
credit good, and hts are hours worked. Note that the
parameter introduces a seasonal pattern in quarterly
discount factors. Similarly, a introduces seasonal
variations in the desired mix between cash and credit
goods, and y, introduces variations in the disutility of
work effort (that is, on how much agents dislike
working as opposed to enjoying leisure).2
Output is given by the following production
z.s —XXA

where ()< 0<l, /cis capital, and bts is labor. Note that
total factor productivity z is assumed to vary across
the different seasons.
There is a standard capital accumulation technol­
ogy given by:




where 0 < § < 1 is the depreciation rate of capital, and
z, is investment.
Not only are the cash good, c(j, and the consump­
tion credit good, n(j, perfect substitutes in production,
but there also is a linear technology to transform con­
sumption goods into investment, z'(j. The feasibility
condition for output is given by

At the beginning of every period there is an asset
trading session. Agents enter this session with mis units
of cash brought from the previous period, principal
plus interest payments (1 + Rs l)bts on nominal bonds
purchased during the previous period, and current lump­
sum cash transfers Tts received from the government.
Agents then acquire nominal bonds bis+l (which ma­
ture during the following period) and cash balances
(which are required to purchase the cash good). Agents
do not have access to any further cash balances to pur­
chase the cash good once the asset trading session is
over. Therefore, their cash-in-advance constraint is
given by



Pt,sct,s <mt,s + (1+7?
)b +Tt,s -bt,.
t,s-Y t,s

where P s is the price of the cash good in terms of money.
This constraint will always hold with equality as long
as the nominal interest rate is positive in every season.
Aside from this cash-in-advance constraint, house­
holds are subject to the following budget constraint:

a,, + h +



+ r,, k,, +



where w s is the wage rate and rts is the rental rate of
capital. This constraint states that any cash that was
not used to purchase the consumption good or bonds,
plus the total earnings from renting labor and capital
to the firms, can be used to purchase credit consump­
tion good, a(s, investment goods, z's, and cash balanc­
es to carry into the following period, zn(j+1.
The representative firm behaves competitively,
taking the wage rate and rental rate of capital as giv­
en. The problem of the firm is to maximize profits,
which are given by


For simplicity, I will assume that government
expenditures are equal to zero and that the govern­
ment doesn’t issue bonds. The budget constraint of
the government is then given by
6) TIs = Mis+1-Mis,

where Mis is the aggregate stock of money in circula­
tion. The monetary policy rule is assumed to be as

7) Mts+l = e"'M,s.
Observe that the government follows a constant
annual growth rate of money rule, but allows the quar­
terly growth rate to vary in a systematic way across
the different seasons.
In a competitive equilibrium: 1) households maxi­
mize their utility function (equation 1) subject to the
cash-in-advance constraint, (equation 3), the budget
constraint (equation 4) and the capital accumulation
equation (equation 2); 2) firms maximize profits

3Q/2005, Economic Perspectives

(equation 5); 3) the government budget constraint
(equation 6) is satisfied; 4) the cash market clears


m =M ;

and 5) the bonds market clears


bt,s =0.

dummy variable indicating the quarter (season) of v,,.
Observe that the estimated coefficient \j>0 provides
the quarterly growth rate of the variable. Since all
real variables in the model economy are stationary
in levels, the seasonal components v can then be
defined as follows:

11) vs = gih+w, for.s= 1, ...,3, and

The formal conditions characterizing a competi­
tive equilibrium are described in appendix A.

The rest of the article focuses on stationary com­
petitive equilibria. That is, equilibria in which each
real variable (including real cash balances) may take
different values across the different seasons, but the
seasonal values must be the same across the different
years.3 The purpose of this section is to select policy,
preference, and technology parameter values such
that the associated stationary competitive equilibria
reproduce the seasonal fluctuations observed in the
U.S. economy.
The first step in calibrating the model economy
is to determine empirical counterparts for its variables.
The empirical counterpart for total consumption, c s
+ n(j, is chosen to be consumption of nondurable goods
and services. At equilibrium, consumption of the cash
good, c(j, is equal to real cash balances, Mts+i/Pts.
Consequently, it is chosen to be the ratio of the mon­
etary base to the Consumer Price Index. Investment,
z'(s, is in turn associated with fixed private investment
plus consumption of durable goods (which entail pur­
chases of capital goods by the households sector).
Output, y(s, is then defined as the sum of these con­
sumption and investment components. Finally, the
empirical counterpart for hours worked, /z(s, is given
by the efficiency equivalent hours series constructed
by Hansen (1993), which basically weighs the hours
worked by individuals by their earnings.
Having determined the empirical counterparts
for the different variables, statistical methods can be
used to calculate the corresponding seasonal compo­
nents. In particular, for each real variable, v s, the fol­
lowing regression was estimated using non-seasonally
adjusted time-series data:

10) lnr(s = ![/,.(4 x i + s) +

+ V/2 +


where y0,
MU MU an<3 V, are coefficients, e(s is
an i.i.d. (independently and identically distributed)
normally distributed error with zero mean, and J is a

Federal Reserve Bank of Chicago

where ys is the estimated value of ys, for s = 1, ...,4.
Money, Mts, is the only non-stationary variable
in the model. However, it is stationary in growth rates.
For this reason, the following regression was estimated:

12) “77^ =

+ MHX + MnX + Mf, + c, s,

where, again, yp y2, y3, and y4 are coefficients, e(s
is an i.i.d. normally distributed error with zero mean,
and J is a dummy variable indicating the quarter
(season) of AT s. The seasonal money growth rates ly
are then obtained as follows:
13) Li, = ij/s + y4, for5=1, ...,3, and
P4 = MU
where ys is the estimated value of ys, for s = 1, ...,4.
Table 1 reports the results of estimating equations
(equations 10 and 12) using U.S. data. Figure 3 depicts
the seasonal components obtained from equations 11
and 13 for the different variables, where the levels of
all variables with meaningless units of measurement
have been normalized to one during the fourth quar­
ter (Q4). We see that the seasonal fluctuations are ex­
tremely large in U.S. data. For instance, the output
level, ys, drops to 0.926 during the first quarter (Q1),
only to recover to 0.959 and 0.954 during the second
(Q2) and third quarters (Q3), respectively. A similar
pattern is followed by consumption, cs + as, and in­
vestment, z. The seasonal pattern for hours, h , is also
significant, but differs quite considerably from the
previous variables: Its lowest level takes place during
Q3, when it drops to 0.950. Real cash balances, on
the other hand, have a weak seasonal pattern: In Q4,
they are only 1 percent larger than during the rest of
the year. However, (as was evident from figure 2) the
growth rate of money, |i, has a strong seasonal pattern:
The growth rate is basically zero during Ql, jumps to
1.7 percent during Q2, and rises slowly thereafter



Regression coefficients































Cs + ds



Real cash balances






Money growth rate


Note: t-statistics are in parenthesis. N.A. indicates not applicable.

reaching 2.0 percent and 2.3 percent during Q3 and Q4,
Once the seasonal components of the different vari­
ables have been determined, parameter values can be
selected so that the model economy mimics them quite
closely. Appendix C describes this procedure in detail.
All calibrated parameter values are depicted in figure 4.

Seasonal monetary policy
While Friedman and Schwartz (1963) acknowl­
edged that “the [Federal Reserve] System was almost
entirely successful in the stated objective of eliminat­
ing seasonal strain,” they had some doubts about the
desirability of this type of policy. On page 295, they
give the following qualified statement: “Within the
year, there seems little harm and perhaps some merit
in permitting the stock of money to decline during
the summer months and rise in the fall and winter.”
At the end of the same paragraph they state “This
kind of ‘elasticity’ of the total money stock is perhaps
desirable.” Friedman (1959, p. 92) takes a much
stronger position: “My own tentative conclusion is
that it would be preferable to dispense with seasonal
adjustments and to adopt the rule that the actual stock
of money should grow month by month at the prede­
termined rate.”
The following question thus arises: Which policy
has more merit? Smoothing interest rates across sea­
sons, Friedman’s proposal of following a constant
growth rate of money, or some other alternative? The
rest of this section explores the different possibilities.


Smooth nominal interest rates
The benchmark economy was calibrated under
the actual money growth rates that the U.S. imple­
ments across seasons. Figure 4, panel D shows that
this policy generates nominal interest rates that are
relatively smooth but are not perfectly constant. The
first policy question that concerns us is then: What would
be the consequences of the Fed changing its actual
policy to one of perfectly smoothing interest rates?
To answer this question, I perform the following
experiment. I replace the benchmark quarterly money
growth rates, ia", calibrated in the previous section
with a seasonal pattern that generates a constant nominal
interest rate. The constant interest rate is chosen so
that the annual interest rate is the same as in the bench­
mark economy.4 The effects of switching to this policy
are shown in figure 5. To ease comparisons, bench­
mark values (corresponding to the economy calibrat­
ed in the previous section) are also reported.
Figure 5, panel D, shows the change in interest
rates from the benchmark case to the constant inter­
est rate. Observe that the change in interest rates is so
small that an almost imperceptible change in mone­
tary growth rates is required to generate it (see figure 5,
panel A). With a higher interest rate in the first quar­
ter and a lower interest rate in the third quarter (rela­
tive to the benchmark economy), the constant interest
rate leads to real cash balances that are somewhat
smaller in the first quarter and somewhat larger in the
third quarter (figure 5, panel C). This in turn leads to
a higher inflation rate in the first quarter and a lower

3Q/2005, Economic Perspectives

inflation rate in the third quarter
(figure 5, panel B). Aside from
these changes, we see that the
rest of the real variables remain
mostly unaffected: The effects
on hours, total consumption, in­
vestment, and output are negli­
gible. The simulation results
thus suggest that the Federal
Reserve Bank policy has been
quite effective in terms of
smoothing interest rates across
seasons: Allocations would be
basically the same if it com­
pletely eliminated any seasonal
variations in interest rates.
Constant money growth rate
This section evaluates
Friedman’s recommendation of
switching to a constant growth
rate of money. To do this, I re­
place the benchmark quarterly
money growth rates calibrated
earlier with a constant money
growth rate that generates the
same annual money growth
rate.5 Figure 6 shows the results.
Figure 6, panel A depicts
the constant growth rate of mon­
ey. We see that, relative to the
benchmark case, the growth rate
of money is now higher in the
first quarter and lower in the
third and fourth quarters. The
more expansionary monetary
policy in the first quarter puts
upward pressure on the nominal
interest rate during the fourth
quarter of the year. Similarly,
the more contractionary policy during the third and
fourth quarters lower nominal interest rates in the
second and third quarter.6 As a result, the interest rate
becomes sharply more seasonal than in the benchmark
case. In particular, switching to a constant growth
rate of money would make the nominal interest rate
constant at about 1.54 percent during the first quarter
of the year, but would more than double during the
fourth quarter of the year, to 3.34 percent (see figure
6, panel D). Thus, a constant growth rate of money
would lead to the same type of increase in fourth
quarter nominal interest rates that were observed
previous to the creation of the Federal Reserve.

Federal Reserve Bank of Chicago

Note that the lower nominal interest rates during
the second and third quarters and the higher nominal
interest rate during the fourth quarter make real cash
balances increase during the second and third quarter
and decrease during the fourth. The reason is that the
nominal interest rate is the opportunity cost of hold­
ing money. The effects on the consumption of cash
goods (that is, real cash balances) translate into qual­
itatively similar effects for total consumption. However,
the effects are much smaller in magnitude. Figure 6,
panel F and panel I, show that the effects on hours
and output are also negligible.
Given the small effects on real allocations, the
welfare gains of moving to a constant growth rate of


money are equal to zero. We
conclude that perfectly smooth­
ing interest rates across seasons
or following a constant growth
rate of money is irrelevant from
a welfare point of view: Real
variables are hardly affected.

Calibrated values

The Friedman rule
In the two previous subsec­
tions, I found that smoothing in­
terest rates or the growth rate of
money gives rise to similar out­
comes, but this doesn’t mean
that money does not play a role
C. Rental rate of capital
in this economy. This section
shows that allocations can be
significantly affected by switch­
ing to a zero nominal interest
rate across seasons (that is, by
implementing the “Friedman
rule”). Figure 7, panel A depicts
the seasonal money growth rates
that are needed to implement the
zero nominal interest rule.7 Since
nominal interest rates are rather
smooth in the benchmark econ­
omy, but at a relatively high lev­
el, it is not surprising that this
path is basically a downward
shift of the benchmark path.
With the zero interest rates,
real cash balances increase dur­
ing each season. The reason is
that real cash balances have be­
come uniformly cheaper. This,
in turn, translates into an in­
crease in total consumption in
each quarter. To satisfy this uni­
form increase in consumption,
hours worked, output, and in­
vestment must also increase in
every season. The effects are
substantial: Output increases by
about 1.1 percent in every quarter.
Despite the significant ef­
fects on real allocations, the
welfare consequences of switch­
ing to the Friedman rule are
small.8 Agents should be com­
Friedman rule. The intuition for why the Friedman
pensated by having their consumption levels increase
by 0.1 percent at every date, to make them indifferent
rule increases welfare is quite straightforward. A pos­
with living in a world where the Fed follows the
itive nominal interest rate makes real cash balances


3Q/2005, Economic Perspectives


Interest rate smoothing
C. Real cash balances

F. Hours

I. Output

Circle markers: Interest rate smoothing
Triangle markers: Fed policy

costly, so agents substitute credit goods for cash
goods. However, the technological rate of transforma­
tion of cash goods to credit goods is equal to one.
That is, there are no technological costs for transform­
ing credit goods into cash goods. The only way to
make agents internalize that this transformation is re­
ally costless is by driving the nominal interest rate to
zero. With a zero nominal interest rate, agents are able
to choose the optimal mix of credit goods and cash
goods in the model economy.

Federal Reserve Bank of Chicago

The sources of seasonal fluctuations
The results so far indicate that monetary policy
plays a negligible role in seasonal fluctuations. How­
ever, I have shown earlier that seasonal fluctuations
in the U.S. are quite substantial. An important ques­
tion that therefore remains is: What is the most im­
portant source for U.S. seasonal fluctuations? Since
the model has used variations in different parameter
values to generate these cycles, it can be used to ex­
plore which of these parameters play the most pre­
dominant role. This section pursues such analysis.



Constant growth rate of money
C. Real cash balances

F. Hours

I. Output

Circle markers: Constant growth rate of money
Triangle markers: Fed policy

Preference weight on consumption
of cash goods (o.)
Figure 4, panel E shows that the benchmark
economy embodies a strong seasonal pattern for the
weight, a, of cash goods in the utility function. In
particular, cash goods are much more valued in the
first quarter of the year than in the last. To evaluate
what role this plays in U.S. seasonal cycles, I perform
the following experiment. I make these weights con­
stant and equal to the cross-seasons average for the
benchmark economy. Under the new constant weight,
I reset the money growth rates, ps, so that the model


generates the same seasonal pattern for nominal in­
terest rates as in the U.S. economy. Thus, the Fed’s
monetary policy together with the rest of the parame­
ter values are kept the same.
Figure 8 shows the results.9 Removing the sea­
sonal pattern for the a weights reduces real cash bal­
ances by 2.6 percent in the first quarter and increases
them by 3.5 percent in the fourth quarter. But aside
from that, the effects on the rest of the variables are
negligible. Thus, variations in the velocity of circula­
tion of money are found to play no important role in
U.S. seasonal cycles.

3Q/2005, Economic Perspectives


The Friedman rule
C. Real cash balances

F. Hours

H. Investment

I. Output

Circle markers: Friedman rule
Triangle markers: Fed policy

Disutility of work (y )
Figure 4, panel H shows that in the benchmark
economy there is a large spike in the disutility of
work, ys, during the third quarter of the year. To eval­
uate what role this plays in U.S. seasonal cycles, I
make the disutility of work constant and equal to the
cross-seasons average for the benchmark economy.
Similar to the previous subsection, I reset the money
growth rates, ps, so that the model generates the same
seasonal pattern for nominal interest rates as in the
U.S economy.

Federal Reserve Bank of Chicago

Figure 9 shows the results. With a constant dis­
utility of work, hours become 7.7 percent higher in
the third quarter and 5.2 percent lower in the fourth
quarter. The effects on hours worked are reflected on
output, which becomes 4.8 percent higher in the third
quarter and 3.3 percent lower in the fourth quarter.
Given the strong preference for consumption smooth­
ing, all the effects on output are translated into in­
vestment while consumption remains unaffected.



Smooth weights on consumption of cash goods

H. Investment

I. Output

Circle markers: Smooth weights on consumption of cash goods
Triangle markers: Fed policy

Discount factors <({)J
Figure 4, panel F shows that the discount factors
increase sharply throughout the year. This section
evaluates the effects of this exogenous increase by
analyzing how the economy would behave if the
agent discounted time equally across the seasons,
that is, if the discount factors were given by those
depicted in figure 4, panel G.10
The results are shows in figure 10. Absent the ex­
ogenous increase in discount factors throughout the
year, consumption would be 3.5 percent higher in the
first quarter and 3.7 percent lower in the fourth quar­


ter. This is not surprising since with the increase in
discount factors, consumption becomes more heavily
weighted in the utility function toward the end of the
year. Since nominal interest rates remain unchanged
(by construction), the ratio of cash goods to total
consumption remains the same as in the benchmark
economy. As a consequence, the effects on real cash
balances are a mirror of those on total consumption.
Note that the smooth discount factors also make
work more costly in the first quarter and less costly
in the last quarter. As a consequence, hours decrease
by 8.5 percent in the first quarter and increase by

3Q/2005, Economic Perspectives


Smooth disutility of labor

F. Hours

I. Output

Circle markers: Smooth disutility of labor
Triangle markers: Fed policy

10.9 percent in the last quarter. The qualitative ef­
fects on output are the same as for hours, but they
have a smaller magnitude. Investment has to de­
crease by 25.2 percent in the first quarter and in­
crease by 30.4 percent in the fourth quarter to be
consistent with the opposite effects on consumption
and output.
Thus, exogenous changes in discount factors
play a significant role in generating seasonal cycles
in the U.S. economy.

Federal Reserve Bank of Chicago

Total factor productivity (zs)
Figure 4, panel B shows that in the benchmark
economy, total factor productivity, z, is low in the
first quarter and increases continuously throughout
the year. This section analyzes the role that this plays
in U.S. seasonal cycles by comparing the benchmark
economy with one that has a constant total factor
The results are shown in figure 11. The strong
preference for smoothing consumption over time im­
plicit in the utility function (equation 1) means that



Smooth discount factors

E. Rental price of capital (%)

I. Output

Circle markers: Smooth discount factors
Triangle markers: Fed policy

the seasonal pattern for total consumption and con­
sumption of cash goods remains unaffected by the
switch to a constant total factor productivity. All the
effects are felt in hours, investment, and output. This
is not surprising: Since the productivity of capital is
constant (instead of increasing), investment does not
need to increase throughout the year. In fact, given
the strong seasonal pattern in other parameters (in
particular, in discount factors) investment would
sharply decrease throughout the year. Since hours en­
ter linearly in the utility function, there are no gains
in smoothing them over time. As a result, the sharp


decline in investment would be achieved by increas­
ing hours by 9.6 percent during the first quarter and
decreasing them by 7.1 percent during the fourth
quarter, allowing consumption to remain unchanged.
Thus, we see that seasonal variations in total fac­
tor productivity play a key role in offsetting the ef­
fects of seasonal variations in discount factors that
were analyzed in the previous subsection.

In this article, I have used a dynamic general
equilibrium cash-in-advance model to study the role

3Q/2005, Economic Perspectives


Smooth total factor productivity

E. Rental price of capital (%)

I. Output

Circle markers: Smooth total factor productivity
Triangle markers: Fed policy

of monetary policy in U.S. seasonal cycles. I have
found that the seasonal monetary policy is largely ir­
relevant in the model economy: Smoothing interest
rates across the seasons and following a constant
growth rate of money lead to basically the same real
allocations. Only nominal interest rates are signifi­
cantly affected.
Smoothing interest rates can play a significant
role if the level targeted is equal to zero. In particu­
lar, following the Friedman rule leads to considerable
effects: Output increases by 1.1 percent in every
quarter. However, the welfare effects are small: The

Federal Reserve Bank of Chicago

consumption equivalent benefit of switching to the
Friedman rule is only 0.1 percent. Not surprisingly
these results are in line with Cooley and Hansen
(1995), who evaluated the welfare costs of inflation
abstracting from seasonal fluctuations.
I also find that the most important source of sea­
sonal fluctuations in the U.S. economy is exogenous
changes in demand, that is in how much agents value
consumption over the different seasons. I find a large
spike in demand during the last quarter of the year,
suggesting that Christmas plays a key role, and a
large drop during the first quarter, indicating that


people tend to postpone consumption during cold
weather. However, seasonal variations in total factor
productivity play an important role in offsetting large
parts of these effects. Cold weather directly affects ac­
tivities like construction and agriculture, making to­
tal factor productivity hit its lowest values during the
first quarter of the year. However, this does not im­
pose much strain on the economy since demand is

also the lowest during the first months of the year.
After the first quarter, total factor productivity in­
creases steadily to reach its peak during the last quar­
ter of the year, just in time to meet the spike in
aggregate demand. In turn, an increase in the value
of leisure plays a significant role in flattening the
path for hours, output, and investment during the
third quarter of the year.

1The list of papers analyzing seasonal fluctuations is more exten­
sive than the one provided in this section, and includes Braun and
Evans (1998) and Krane and Wascher (1999). However, the focus
of these papers has been real activity and not monetary policy.

7These growth rates are obtained from equations B.8 and B.9
once the R (for 5=1, ...,4) are set to zero. Note that, given the
seasonal variations in 0 and a , these money growth rates associ­
ated with the Friedman rule in general will not be constant.

2The assumption of linear preferences with respect to labor can
be justified on theoretical grounds as in Hansen (1985) and
Rogerson (1988).

8Despite this, the Friedman rule can be shown to be the optimal
monetary policy in this environment (from a welfare standpoint).

3Appendix B describes the formal conditions that a stationary
competitive equilibrium must satisfy.
4In particular, let R* be the nominal interest rates corresponding
to the benchmark economy (depicted in figure 4, panel D). The
constant interest rate, R chosen, satisfies the following condition:

9Observe that the scale for figures 8-11 is different than the scale
for figures 5—7, since the effects are much larger in the former set
of figures.

10Formally, the smooth discount factors,

are given as follows:

(i+«)4=(i+X)(i+X)(i+X)(i+«4’)The money growth rates, p? that generate this constant interest
rate, R, can be obtained from equations B.8 and B.9.
5In particular, let p* be the growth rates of money corresponding
to the benchmark economy (depicted in figure 3, panel F). The
constant money growth rate p satisfies the following condition:

where p is the annual discount factor in the benchmark economy.

4fi = ft + ft + ft + ft.
6Observe from equations B.8 and B.9 that the nominal interest
rate Rs is directly related to the growth rate of money in the fol­
lowing quarter, p .


3Q/2005, Economic Perspectives


At year t quarter 5, the household must be indifferent to
two alternatives: 1) using one less unit of the cash avail­
able for purchasing the cash good and sacrificing 1/P
units of the cash good, which entails a loss in marginal
utility equal to Ot <|> /c per unit, and 2) purchasing one
more unit of the bond, obtaining 1 + R units of cash
the following period (as interest payment) that can be
used to purchase 1/P units of the cash good, entailing
a gain in marginal utility equal to as+1(t)J+/c,s+1 Per unitThus, the following conditions must hold:


1 «A








(1 + V) otiP4>i

-L to) =

for , = !.....3

1 t,s+\
i <h4(i-«4)


(l-a )
i------ — = yj,for5 = l, ...,4.


The household must also be indifferent to: 1) pur­
chasing one less unit of the credit good (a ), which en­
tails a loss in marginal utility equal to <t)( 1 - O- )lat , and
2) purchasing 1/P additional units of end-of-period
cash balances that next period can be used to purchase
1/P units of the cash good, which entails a gain in
marginal utility equal to 0s+ias+1/c(s+1 per unit. Thus the
following conditions must hold:


The conditions characterizing the optimal behavior
of the representative firm are much easier to describe.
The firm hires labor up to the point where the marginal
productivity of labor equals the wage rate



Finally, the household must be indifferent to:
1) working one less unit of time, losing w units of the
credit good that the wage rate could buy, which entail a
loss in marginal utility equal to 0 (1 - a )/a per unit,
and 2) obtaining one more unit of leisure, which entails
gain in marginal utility equal to 0 y. Thus, the follow­
ing conditions must hold:







=zX(1-0)/jM>for5 = 1>->4>

and hires capital up to the point where the marginal
productivity of capital equals its rental rate


rt,s = zs GZ-8
'Z;1 8, for 5 = 1,...,
t,s t,s ’

A competitive equilibrium is then a sequence
{ct,S, at,S , ht,S , kt,&, mt,&, bt,S , wt,&, rt,S . Pt,& Rt,!> Tt,S Mt,S7} for
t = 0, ..., and 5 = 1, ..., 4, such that equations 2, 3, 4,
6, 7, 8, 9, A.l, A.2, A.3, A.4,A.5, and A.6 hold.

The household must also be indifferent to: 1) pur­
chasing one less unit of the credit good
which en­
tails a loss in marginal utility equal to <t)( 1 - a,)/a(v, and
2) purchasing one unit of capital (A- + ), and renting it to
the firm and selling-off the undepreciated portion, ob­
taining r + 1 - § units of the credit good the follow­
ing period, which entails a gain in marginal utility
equal to 0J+1(1 - a )/a per unit. Thus, the following
conditions must hold:
1 _ gi^+lQ Ti|j

A. 3)

for 5=1,



Federal Reserve Bank of Chicago


A stationary equilibrium is a vector (c , a , i ,y , k, h , r ,w , R \ for s = I, ..., 4, such that the following equations
are satisfied:


c +a + i =v,


^+1 = (l-§X + z;;




rs = X,






(l + Aj^^ = z;+l-5,

+ «,_ 1 , (1~aJ




(1+ Rj, (except for s-= 4),


i _‘I’s+i


1 = 3-^-^-—(1 + R.), and




for 5=1, ..., 4.


3Q/2005, Economic Perspectives

This appendix describes the procedure used to calibrate
parameter values.
The depreciation rate of capital, §, is chosen to be
0.025, which is a standard value in the real business cy­
cle literature. The seasonal pattern for the stock of capi­
tal, k is then chosen to reproduce the seasonal pattern
for investment, z when § = 0.025. The result is depict­
ed in figure 4, panel A, which shows no significant sea­
sonal variations for the stock of capital, k . This result
is obtained, despite the strong seasonal pattern in in­
vestment, because investment is small relative to the
size of capital.
The share of capital in national income is given, at
equilibrium, by the curvature parameter 0 in the pro­
duction function. For this reason, 0 is chosen to be
0.36, which is the share of capital implicit in the Na­
tional Income and Product Accounts. Given 0, and the
seasonal components for capital, k hours, h, and out­
put, y the seasonal pattern for total factor productivity,
z, can be obtained as a residual from the production
function. The result is depicted in figure 4, panel B,
which shows a strong seasonal pattern: Total factor pro­
ductivity drops to 0.938 during Q1 and slowly recovers
thereafter, reaching 0.966 and 0.986 during Q2 and Q3,
Given the capital share, 0, the capital-output ra­
tios, kJy have direct implications for the rental rate of
capital, r , in the model economy. Figure 4, panel C
shows that this rental rate has a significant seasonal
pattern, taking the lowest value during Ql.
The rental rate of capital and the depreciation rate
determine the seasonal pattern for the real interest rate
in the economy. Considering the seasonal inflation rate
pattern implied by real cash balances, c, and the money
growth rate, |_l, the nominal interest rates, R. can be
obtained from a version of the Fisher equation. Figure
4, panel D, shows that the nominal interest rate goes
through significant seasonal variations: It ranges from
1.67 percent during Ql to 2.36 percent during Q3.
The weight of cash goods in the utility function,
tt is a key determinant of the relation between the
nominal interest rate, A’, and the velocity of circulation
of money, cJ(c + ay that is, of the demand for money.
As a consequence, it was chosen to be consistent with
the values for the nominal interest rate, Rs , real cash
balances, c , and total consumption, c + a , obtained
above. The weights, cy, thus obtained are reported in

Federal Reserve Bank of Chicago

figure 4, panel E. We see that they have a strong sea­
sonal pattern, the desirability of cash goods being the
highest during Ql and decreasing smoothly throughout
the rest of the year.
Given these weights a the discount factors (3, (()
02, 0 , and (ty were selected to be consistent with the
nominal interest rates, R and money growth rates, Ly,
reported above. Figure 4, panel F reports that these dis­
count factors have a strong seasonal pattern. To make
this clear, figure 4, panel G reports the discount factors
that the representative agent should have if it discount­
ed time equally across the seasons. We see that both
paths differ quite substantially. In particular, the sea­
sonal pattern for the calibrated values of (() (() 0 and
(b4 indicate a monotone increase in demand throughout
the year, which becomes particularly sharp during Q4.
Finally, the disutility of work parameters, y are
selected to reproduce the seasonal pattern for total
hours worked, /? The resulting values of y in figure 4,
panel H indicate a large increase in the disutility of
work during Q3 and a sharp reversal during Q4.
The rest of the appendix describes in detail which
equations were used in each stage of the calibration
The following variables are directly obtained from
the data (as described in the model economy section):
i ,c ,a ,h , and Lt . Given these variables, model parameters are selected as follows.

Set § = 0.025.


Given i (for s = 1, ..., 4), choose seasonal pattern
for k, that is consistent with equation B.2.


Set 0 = 0.36.


Given c , a , and z, obtain v from equation B. 1.


Given y ,k ,h and 0, obtain z from equation B.3.


Given v k and 0, obtain r from equation B.4.


Given c ,


Given R , c , and a , obtain a from equation B.7.


Given cy, u , and A, set (() (this is just a normaliza­
tion) and obtain 0, for s = 2, ..., 4 and (3 from
equations B.8 and B.9.


r, and §, obtain R from equation B.6.

10)' Given as’, 0,’ as’, vs’, and hs’, c>get y*s from B. 10.



Braun, R. A. and C. L. Evans, 1998, “Seasonal
Solow residuals and Christmas: A case for labor
hoarding and increasing returns,” Journal ofMoney,
Credit, and Banking, Vol. 30, No. 3, pp. 306-330.

Cooley T. F., and G. D. Hansen, 1995, “Money and
the business cycle,” in Frontiers ofBusiness Cycle
Research, T. F. Cooley (ed.), Princeton, NJ: Prince­
ton University Press.
Friedman, M., 1959, A Program for Monetary’ Sta­
bility, New York City: Fordham University Press.

Friedman, M., and A. J. Schwartz, 1963, A Monetary’
History of the United States, 1867-1960, Princeton,
NJ: Princeton University Press.
Hansen, G., 1993, “The cyclical and secular behavior
of the labor input: Comparing efficiency units and
hours worked,” Journal ofApplied Econometrics,
Vol. 8, No. l,pp. 71-80.

Krane, S., and W. Wascher, 1999, “The cyclical
sensitivity of seasonality in U.S. employment,” Journal
ofMonetary Economics, Vol. 44, No. 3, pp. 523-553.

Lucas, R. E., and N. L. Stokey, 1983, “Optimal fiscal
and monetary policy in an economy without capital,”
Journal ofMonetary Economics, Vol. 12, pp. 55-93.

Mankiw, N. G., and J. A. Miron, 1991, “Should the
Fed smooth interest rates? The case of seasonal mon­
etary policy,” Carnegie-Rochester Conference Series
on Public Policy, Vol. 34, pp. 41-69

Miron, J. A., 1986, “Financial panics, the seasonality
of the nominal interest rate, and the founding of the
Fed,” American Economic Review, Vol. 76, No 1,
pp. 125-140.
Rogerson, R., 1988, “Indivisible labor, lotteries, and
equilibrium,” Journal ofMonetary’ Economics, Vol. 21,
No. l,pp. 3-16.

__________ , 1985, “Indivisible labor and the business
cycle,” Journal ofMonetary’ Economics, Vol. 16, No.
3, pp. 309-327. '

Kemmerer, E. W., 1910, “Seasonal variations in the
relative demand for money and capital in the United
States,” National Monetary Commission, report.


3Q/2005, Economic Perspectives