Full text of Economic Policy Review (Federal Reserve Bank of New York) : Volume 3, Number 4

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

Bank Capital Requirements for
Market Risk: The Internal Models
Approach
Darryll Hendricks and Beverly Hirtle*

T

he increased prominence of trading activities
at many large banking companies has highlighted bank exposure to market risk—the
risk of loss from adverse movements in financial market rates and prices. Recognizing the importance
of trading operations, banks have sought ways to measure
and to manage the associated risks. At the same time, bank
supervisors in the United States and abroad have taken
steps to ensure that banks have adequate internal controls
and capital resources to address these risks.
Prominent among the steps taken by supervisors is
the development of formal capital requirements for the
market risk exposures arising from banks’ trading activities. These market risk capital requirements, which will
take full effect in January 1998, depart from earlier capital
rules in two notable ways. First, the capital charge is based
on the output of a bank’s internal risk measurement model

*Darryll Hendricks and Beverly Hirtle are vice presidents at the Federal
Reserve Bank of New York.

rather than on an externally imposed supervisory measure.
Second, the capital requirements incorporate qualitative
standards for a bank’s risk measurement system.
This paper presents an overview of the new capital
requirements. In the first section, we describe the structure
of the requirements and the considerations that went into
their design. In addition, we address some of the concerns
that have been raised about the methods of calculating capital charges under the new rules. The paper’s second section
considers the probable impact of the market risk capital
requirements. After performing a set of rough calculations
to show that the effect of the internal models approach on
required capital levels and capital ratios will probably be
modest, we identify some significant benefits of the new
approach. Most notably, the approach will lead to regulatory capital charges that conform more closely to banks’
true risk exposures. Moreover, the information generated
by the models will allow supervisors and financial market
participants to compare risk exposures over time and across
institutions.

FRBNY ECONOMIC POLICY REVIEW / DECEMBER 1997

1

THE STRUCTURE OF THE MARKET RISK
CAPITAL REQUIREMENTS
The new capital requirements for market risk have been
put forward as an amendment to existing capital rules. In
late 1990, banks and bank holding companies in the
United States became subject to a set of regulatory capital
guidelines that defined minimum amounts of capital to
be held against various categories of on- and off-balancesheet positions.1 The guidelines also specified which debt
and equity instruments on a bank’s balance sheet qualified
as regulatory capital. These guidelines were based on the
1988 Basle Accord adopted by the Basle Committee on
Banking Supervision, a group made up of bank supervisors
from the Group of Ten countries.
While the original Basle Accord and U.S. riskbased capital guidelines primarily addressed banks’
exposure to credit risk, the new requirements set minimum
capital standards for banks’ market risk exposure.2 Broadly
speaking, market risk is the risk of loss from adverse
movements in the market values of assets, liabilities, or
off-balance-sheet positions. Market risk generally arises
from movements in the underlying risk factors—interest
rates, exchange rates, equity prices, or commodity prices—
that affect the value of these on- and off-balance-sheet
positions. Thus, a bank’s market risk exposure is determined both by the volatility of underlying risk factors and
the sensitivity of the bank’s portfolio to movements in
those risk factors.
Banks face market risk from the full range of
positions held in their portfolios, but the capital standards focus largely on the market risks arising from
banks’ trading activities.3 This focus reflects the idea
that market risk is a major component of the risks arising from trading activities and, further, that market risk
exposures are more visible and more easily measured
within the trading portfolio because these positions are
marked to market daily. Thus, under the amended capital
standards, positions in a bank’s trading book are subject
to the market risk capital requirements but are exempt
from the original risk-based capital charges for credit risk
exposure.4 In addition, commodity and foreign exchange
positions held throughout the institution (both inside

2

FRBNY ECONOMIC POLICY REVIEW / DECEMBER 1997

and outside the trading account) are subject to the market
risk capital requirements.
Because the capital standards principally address
the market risk arising from trading activities, only those
U.S. banks and bank holding companies with significant
amounts of trading activity are subject to the market risk
requirements. In particular, the U.S. standards apply to
banks and bank holding companies with trading account
positions (assets plus liabilities) exceeding $1 billion
or 10 percent of total assets. The institutions meeting

By substituting banks’ internal risk
measurement models for broad, uniform
regulatory measures of risk exposure, [the
new rule] should lead to capital charges
that more accurately reflect individual
banks’ true risk exposures.

these criteria, while relatively few in number, account for
the vast majority of trading positions held by U.S. banks.5
Supervisors also have the discretion to impose the standards
on institutions that do not meet these criteria if such a step
appears necessary for safety and soundness reasons. The
rules become effective as of January 1998, although the
U.S. regulation also permits banks to elect early adoption
during 1997.

INNOVATIVE FEATURES
The market risk capital standards have drawn considerable
attention because they differ significantly in approach from
the risk-based capital rules for credit risk. The market risk
standards impose a quantitative minimum capital charge
that is calculated for each bank using the output of that
bank’s internal risk measurement model; they also establish
a set of qualitative standards for the measurement and
management of market risk. In both regards, the capital

standards break new ground. By substituting banks’ internal risk measurement models for broad, uniform regulatory
measures of risk exposure, this approach should lead to
capital charges that more accurately reflect individual banks’
true risk exposures. And by including qualitative standards,

Any bank or bank holding company subject to
the market risk capital requirements must be
able to demonstrate that it has a conceptually

underlying risk factors such as interest rates, exchange
rates, equity prices, and commodity prices. Specific risk
is defined as the risk of an adverse movement in the
price of an individual security resulting from factors
related to the security’s issuer. At one level, general and
specific market risk are analogous to systematic and
nonsystematic risk in a standard asset-pricing framework.
Specific risk, however, is intended to cover variation both
from day-to-day price fluctuations and from surprise
events, such as an unexpected bond default. The following
subsections provide an overview of the capital treatment
of the two types of risk.

sound risk measurement system that is
implemented with integrity.

the approach is consistent with the shift in supervisory
interest from a focus on risk measurement to a more comprehensive evaluation of banks’ overall risk management.
The qualitative standards are designed to incorporate basic principles of sound risk management in the
capital requirements. Any bank or bank holding company subject to the market risk capital requirements
must be able to demonstrate that it has a conceptually
sound risk measurement system that is implemented
with integrity. The risk estimates produced must be
closely integrated with the risk management process: for
example, management could rely on daily reports from
the system to assess current strategy or could base its
limit structure on the risk estimates. In addition, the
bank must conduct periodic stress tests of its portfolio
to gauge the impact of extreme market conditions.
Further, the bank must have a risk control unit that is
fully independent of the business units that generate
market risk exposures. Finally, internal and/or external
auditors must conduct an independent review of the
bank’s risk management and measurement process.
The quantitative capital requirements distinguish between general market risk and specific risk. As
defined in the capital standards, general market risk is
the risk arising from movements in the general level of

CAPITAL REQUIREMENTS FOR GENERAL
MARKET RISK
The capital requirements for general market risk are based
on the output of a bank’s internal value-at-risk model, calibrated to a common supervisory standard. In brief, a valueat-risk model produces an estimate of the maximum
amount that the bank can lose on a particular portfolio over
a given holding period with a given degree of statistical
confidence.6 Although there are a variety of empirical
approaches to calculating value at risk, estimates are almost
always derived from the behavior of underlying risk factors
(such as interest rates and exchange rates) during a recent
historical observation period.
The general market risk capital requirement is
based on value-at-risk estimates calibrated to a ten-day,
99th percentile standard. That is, if the ten-day, 99th percentile value-at-risk estimate is equal to $100, then the
bank would expect to lose more than $100 on only 1 out
of 100 ten-day periods. The common supervisory standard
is imposed to ensure that the capital charge entails a
consistent prudential level across banks. The value-at-risk
estimates must be calculated on a daily basis using a minimum historical observation period of one year, or the
equivalent of one year if observations are weighted over
time. The capital charge for general market risk is equal to
the average value-at-risk estimate over the previous sixty
trading days (approximately one quarter of the trading
year) multiplied by a “scaling factor,” which is generally
equal to three.7

FRBNY ECONOMIC POLICY REVIEW / DECEMBER 1997

3

Several aspects of this calculation have generated
considerable discussion, and thus it is worth taking a
moment to consider them further. First, the ten-day holding period has been criticized as being overly conservative,
since under normal market conditions, many positions in a
bank’s trading portfolio could be liquidated in less than
this amount of time.8 The ten-day standard, however, also
reflects the need to address the risks posed by options and
other positions with nonlinear price characteristics.
Because options’ sensitivities to changes in market risk
factors can grow at a rate that is disproportionate to the
size of changes in the risk factors, a longer holding period
can reveal risk exposures that might not be evident with
the smaller risk factor movements associated with shorter
holding periods. Thus, the choice of a ten-day holding
period stems from the view that the value-at-risk estimates
used in the capital calculation should incorporate the impact
of instantaneous ten-day-sized price moves in the market
risk factors. In the language of options, the ten-day holding
period serves to calibrate the coverage of “gamma” risk.9
Second, the minimum historical observation
period has come under question. Critics characterize the
year-long minimum as intrusive and argue that longer
observation periods have not been shown to result in more
accurate value-at-risk estimates. In fact, however, the minimum historical observation period requirement primarily
reflects concerns about the variability of the capital
requirement across institutions, rather than a judgment by
supervisors about the historical observation period
likely to produce the most accurate value-at-risk estimates
for capital or risk management purposes.10
The basic idea behind this requirement is that
banks with similar risk exposures should face similar capital charges. In this regard, empirical evidence suggests that
shorter observation periods tend to generate value-at-risk
estimates that are more volatile over time (Hendricks
1996). Thus, for a set of banks with similar risk exposures,
this result implies that the dispersion of value-at-risk
estimates across banks will tend to be greater when some
of the banks are using short observation periods. The minimum one-year historical observation period is an attempt
to limit this disparity.

4

FRBNY ECONOMIC POLICY REVIEW / DECEMBER 1997

A third element of the new capital requirements
that has proved controversial—indeed, more controversial
than any other element—is the scaling factor. The scaling
factor has been criticized as an ad hoc supervisory adjustment that undercuts the benefits of basing a capital charge
on banks’ internal models. In this view, the key advantage
of using internal risk measurement models is that they

By itself, even a perfectly measured ten-day,
99th percentile value-at-risk figure may not
provide a sufficient degree of risk coverage
to serve as a prudent capital standard.

provide more accurate measures of an individual bank’s
risk exposure than do broad supervisory measures. Accordingly, some have argued that a bank that can demonstrate
convincingly that its model is accurate should be subject
to a scaling factor of one.
In considering this argument, however, it is important to recognize that the overall purpose of the scaling
factor is to produce the desired degree of coverage for the
market risk capital charge. The market risk capital requirements are intended to ensure that banks hold sufficient
capital to withstand the consequences of prolonged and/or
severe adverse movements in the market rates and prices
affecting the value of their trading portfolios. The key
assumption behind the internal models approach is that a
value-at-risk estimate calibrated to a ten-day, 99th percentile standard is well correlated with the degree of such risk
inherent in the portfolio, and thus is a reasonable base for a
minimum capital standard.
Nonetheless, by itself, even a perfectly measured
ten-day, 99th percentile value-at-risk figure may not provide a sufficient degree of risk coverage to serve as a
prudent capital standard. For one, such a standard implies
that a bank is expected to have trading portfolio losses that
exceed its required capital in one ten-day period out of a
hundred, or about once every four years. An environment

in which banks depleted their market risk capital so
frequently could be highly unstable, particularly if such
events happened to many banks at the same time (which
could occur if banks adopted similar trading strategies).
Further, value-at-risk estimates based only on recent historical market data may not incorporate the possibility of
severe market events. Thus, a capital standard based on
unadjusted value-at-risk estimates might not provide sufficient capital for a bank to withstand the effects of market
breaks or unanticipated regime shifts.
The role of the scaling factor is to translate the
value-at-risk estimates into an appropriate minimum
capital requirement, reflecting considerations both about
the accuracy of a bank’s value-at-risk model and about
prudent capital coverage. The capital cushion should
cover possible losses due to market risk over a reasonable
capital planning horizon—which is generally seen to
reflect a period between one quarter and one year—while
at the same time reflecting the fact that banks’ trading
positions change rapidly over time. As an alternative to
the scaling factor, supervisors could have based the capital
charge on value-at-risk estimates calibrated to a very
stringent prudential standard (for example, a one-year
holding period or a 99.999th percentile standard). In
practice, however, it is very difficult to derive reliable and
verifiable value-at-risk estimates for such extreme parameter values. Actual observations of such “tail events” are
few, greatly complicating the task of verifying that any
model is accurately measuring the probability of these
occurrences. Thus, instead of representing a more “scientific” alternative to the scaling factor, a requirement of this
kind would simply introduce a false sense of precision into
the capital standards.
By contrast, the scaling factor has the advantage of
being simple and easy to implement. It does not require
banks to make (or supervisors to evaluate) complex calculations intended to model rare or as yet unobserved events,
such as regime shifts or market breaks. At the same time,
however, it does seek to provide a capital cushion against
such incidents. In addition, it is similar to the techniques
used by some banks for internal capital allocation, in which
one-day value-at-risk estimates are extrapolated to a much

longer holding period (for example, six months or one year)
by multiplying by the square root of time (in the case
of ten-day value-at-risk estimates, this calculation for a
one-year holding period implies a multiplication factor of
five). Moreover, comparisons of ten-day, 99th percentile
value-at-risk estimates with banks’ actual daily trading
results suggest that the scaling factor of three provides an
adequate level of capital coverage. The results of bank
stress-testing programs were also a key input in the
decision to use a scaling factor of three.
For additional protection, the market risk capital
requirements incorporate a feature intended to ensure that
models that systematically underestimate risk exposures
are subject to a higher multiplication factor. This feature is
the so-called backtesting requirement. Backtesting is a
process of confirming the accuracy of value-at-risk models
by comparing value-at-risk estimates with subsequent
trading outcomes. For instance, an accurate model will
produce one-day, 99th percentile value-at-risk estimates
that are exceeded by actual trading losses only 1 percent of
the time.
The backtesting procedures in the market risk
capital requirements use a very simple statistical test based
on the number of times during a year that trading losses
exceed value-at-risk estimates. For purposes of the backtest, banks will compare daily end-of-day value-at-risk
estimates calibrated to a one-day, 99th percentile standard
with the next day’s trading outcome. Each instance in
which a trading loss exceeds the value-at-risk estimate is
termed an exception. Since it is unlikely that an accurate model would produce a large number of exceptions,
banks with five or more exceptions over a one-year period
are subject to a higher scaling factor. The increase in the
scaling factor is as large as 33 percent (from three to four)
for banks with a very large number of exceptions.
The introduction of the higher scaling factor for
banks experiencing five or more exceptions is based on a
simple statistical technique that calculates the probability
that an accurate value-at-risk model would generate a
given number of exceptions during a year of trading days.
In theory, these probabilities are independent of the design
of any particular model, so the same number of exceptions

FRBNY ECONOMIC POLICY REVIEW / DECEMBER 1997

5

is used as the starting point for the higher scaling factor
across all banks. Overall, the backtest is calibrated to
ensure that a bank with an accurate value-at-risk model is
very unlikely to face an increased scaling factor. The relationship between the number of exceptions and the scaling
factor is reported in Table 1.11
For technical reasons, the backtests conducted by
banks may deviate from the ideal conditions assumed in
the statistical derivation. For one, the trading gains and
losses used in the backtest calculation may be based on the
actual trading outcomes booked by the bank, and in that
case will include fee income and the profits and losses from
intraday trading. This means that the profit and loss figures used in the backtest could reflect influences not
incorporated into the value-at-risk model, potentially
introducing bias into the backtest results. The direction of
the bias is not clear, however. On the one hand, including
fee income in the profit and loss figures will tend to reduce
the number of exceptions identified. On the other hand,
the impact of intraday trading will likely increase the volatility of the daily profit and loss figures relative to the
value-at-risk estimates, increasing the probability of an
exception.
One possible response would be to require banks
to calculate hypothetical profit and loss figures by holding
end-of-day positions constant and excluding fee income.
This calculation could become quite burdensome, however.

Table 1

BACKTESTING AND THE SCALING FACTOR
Number of Exceptions
(Out of 250 Trading Days)
0 to 4
5
6
7
8
9
10 or more

Scaling
Factor
3.00
3.40
3.50
3.65
3.75
3.85
4.00

Cumulative
Probability
(Percent)
10.78
4.12
1.37
0.40
0.11
0.03
<0.01

Note: The “cumulative probability” column reports the probability that an
accurate model would generate more than the number of exceptions reported in
the first column. These figures are generated using a binomial distribution,
assuming a sample size of 250 trading days. For the purpose of the backtest, an
accurate model is one that produces an accurate estimate of the 99th percentile
of the distribution of one-day trading gains and losses. Thus, an accurate
value-at-risk model will produce more than five exceptions over a 250-day
trading period 4.12 percent of the time.

6

FRBNY ECONOMIC POLICY REVIEW / DECEMBER 1997

For this reason, and because the use of actual profit and loss
figures in the backtest does not produce a clear bias in the
test, banks are allowed to use the profit and loss information already at hand.
Finally, the backtest is calibrated to a one-day
standard, whereas the value-at-risk estimates used for
capital purposes are calibrated to a ten-day standard.
Many commentators have pointed out that this difference introduces a discrepancy between the value-at-risk
estimates validated in the backtest and the estimate
actually used for capital purposes. Once again, the reasoning behind this specification reflects the practical
limitations of testing value-at-risk estimates calibrated
to a ten-day standard: backtesting such estimates would
require a significant amount of historical data to generate
a series of independent ten-day profit and loss figures.
With only a limited number of such observations—just
twenty-six over a one-year horizon—the power of the
backtest to distinguish between accurate and inaccurate
models is very limited. Thus, the supervisory backtest is
calibrated to a one-day standard to strike a balance
between the need to have a sufficient amount of data
to give the backtest statistical power and the desire to
determine the accuracy of the value-at-risk model used
in the capital calculations.

CAPITAL REQUIREMENTS FOR SPECIFIC RISK
As noted earlier, the capital requirements for specific risk
are intended to cover the risk of adverse price movements
stemming from factors related to the issuer of an individual
security. Thus, debt and equity positions in bank trading
portfolios are assumed to be subject to specific risk. Under
the original risk-based capital guidelines put forth in
1988, long debt and equity positions in a trading portfolio were subject to capital charges ranging from 0 percent
(for government securities) to 8 percent (for corporate debt
and equity) of the book value of the positions. Under the
amended guidelines, both long and short debt and equity
positions are covered by the market risk capital requirement for specific risk.
Banks whose value-at-risk models incorporate
specific risk can use the specific risk estimates generated

by their models.12 Under the most recent announcement
by the Basle Committee on Banking Supervision (1997),
these model-based specific risk estimates are subject to a
scaling factor of four until market practice evolves and
banks can demonstrate that their models of specific risk
adequately address both idiosyncratic risks and “event risks”

Banks whose value-at-risk models
incorporate specific risk can use the specific
risk estimates generated by their models. . . .
These model-based specific risk estimates are
subject to a scaling factor of four.

that might not be captured in a value-at-risk model. 13
This provision holds out the prospect of harmonizing the
specific risk capital requirements fully with the general
market risk requirements as market practices with respect
to positions subject to significant event risks become
clearer. This approach is consistent with the view that
there is no compelling conceptual reason to separate market risk into a general and a specific portion in a value-atrisk model, or to apply different standards to one portion
than to another.

IMPACT OF THE CAPITAL REQUIREMENTS
EFFECT ON CAPITAL LEVELS AND CAPITAL RATIOS
How the market risk requirements will affect banks’
required capital ratios is difficult to calculate precisely with
the data currently in the public domain. Such calculations
require both information on banks’ value-at-risk estimates—
calibrated to the ten-day, 99th percentile supervisory
standard—and information about the distribution of
trading assets and liabilities among various specific risk categories. Despite the lack of such data, however, it is possible
to make a rough estimate of the impact of the capital charge
by using information reported in banks’ annual reports.

Table 2 reports 1996 average value-at-risk estimates for a sample of large bank holding companies that
presented annual average value-at-risk estimates in their
1996 annual reports along with sufficient descriptive detail
to identify the holding period and percentile underlying
the estimate.14 As indicated in Table 2, all of the estimates
were based on a one-day holding period, with percentiles
ranging from the 95th to the 99th. The divergence in these
parameters, as well as in other aspects of the estimates such
as correlation assumptions, makes direct comparisons of
these figures across institutions difficult.
Nevertheless, these figures suggest that the
impact of the market risk capital charge on required capital
levels and capital ratios is likely to be quite small. Using
these numbers, we calculate that the estimated increase in
the level of required capital from the general market risk
component of the new capital charge ranges roughly
between 1.5 and 7.5 percent for these banking companies.
We find that the impact on the capital ratios is also fairly
modest, with an average decline of about 30 basis points
and 40 basis points in the tier 1 and total capital ratios,
respectively. These calculations are at best rough estimates,
however, and could differ significantly from the actual
impact of the capital charge at the time it becomes effective. Such differences would reflect both estimation error in
translating the reported figures to the supervisory stan-

Table 2

1996 ANNUAL AVERAGE VALUE-AT-RISK ESTIMATES
FOR SELECTED U.S. BANK HOLDING COMPANIES
Bank Holding
Company
BankAmerica
Bankers Trust
Chase Manhattan
Citicorp
J.P. Morgan

1996 Average
Daily VAR
(Millions of Dollars)
42a
39
24b
45c
21

Percentile
Basis
97.5
99.0
95.0
2σ
95.0

Holding
Period
1 day
1 day
1 day
1 day
1 day

Note: The average 1996 value-at-risk (VAR) figures are drawn from the
companies’ 1996 annual reports.
a
Figure assumes a correlation of one between broad risk categories. The
comparable figure assuming a correlation of zero is $18 million.
b
Figure is based on the volatility of actual daily trading results, as reported in
the 1996 annual report.
c
The 2 σ VAR figure is equivalent to the 97.7th percentile under a normal
distribution.

FRBNY ECONOMIC POLICY REVIEW / DECEMBER 1997

7

dards and changes in the bank holding companies’ portfolios over time.
Once we account for the capital treatment of
specific risk, the overall impact of the market risk capital
charge is likely to be even smaller than our calculations
suggest. As noted earlier, many traded debt and equity
positions subject to the credit risk capital requirements
under the original capital guidelines are now subject to
specific risk capital requirements based on the output of
banks’ internal models. This “specific risk carve-out” will
offset the impact of the additional general market risk
capital charge, possibly to a considerable degree. Unfortunately, the data needed to make reasonably precise estimates
of this effect are not currently available. However, given the
significant positions that some institutions hold in instruments that will become subject to the specific risk capital
requirements, this carve-out may well result in a net
reduction in required capital levels for some institutions.

ADVANTAGES OF THE INTERNAL MODELS
APPROACH
Whatever the effect of the new standards on the level of
overall required capital, capital requirements based on
internal models should produce minimum regulatory
capital charges that more closely match banks’ true risk
exposures. This closer relationship is important not only
for determining the risk facing an institution at a particular moment in time, but also for tracing the evolution of
risk over time. That is, while the value-at-risk estimates
underlying the market risk capital charge are useful for
assessing the level of risk undertaken by a bank or bank
holding company at a given moment, they are potentially
even more beneficial for understanding changes in risk
exposure over time. By extension, the key benefit of the
market risk capital charge is that the required capital levels
will evolve with risk exposures over time.
In addition to tightening the link between risk
exposures and capital requirements, a capital charge based
on internal models may provide supervisors and the
financial markets with a consistent framework for making
comparisons across institutions. As the information in
Table 2 makes clear, the value-at-risk figures presented in

8

FRBNY ECONOMIC POLICY REVIEW / DECEMBER 1997

the annual reports of various bank holding companies are
calculated using different parameters, especially the
percentile of the loss distribution. These differences make
comparisons across institutions difficult without additional

In addition to tightening the link between risk
exposures and capital requirements, a capital
charge based on internal models may provide
supervisors and the financial markets with a
consistent framework for making comparisons
across institutions.

calculations to convert the figures to a common basis.
Typically, these calculations require assumptions that may
be only approximately correct, introducing additional
noise in the comparisons.
By contrast, the market risk capital charge provides a common standard for value-at-risk estimates that
makes comparisons across institutions easier and more
reliable. The value-at-risk estimates underlying banks’
capital charges will be based on a uniform set of prudential
parameters and will accurately reflect the assumptions and
specifications of each bank’s internal model (rather than an
external approximation). Further, the financial markets
may gain information about the performance and accuracy
of these models over time if banks make public the results
of their backtests. While disclosure of the details of these
results is purely discretionary, this backtesting information
is consistent with the type of disclosures about market
risks advocated in several recent discussion papers (see
Bank for International Settlements [1994] and Federal
Reserve Bank of New York [1994] for two examples).

CHALLENGES FOR SUPERVISORS
The actual benefits to be derived from the value-at-risk
estimates depend crucially on the quality and accuracy of

the models on which the estimates are based. To the extent
that these models are inaccurate and misstate banks’ true
risk exposures, then the quality of the information derived
from any public disclosure will be degraded. More important, inaccurate value-at-risk models or models that do not
produce consistent estimates over time will undercut the
main benefit of a models-based capital requirement: the
closer tie between capital requirements and true risk
exposures. Thus, assessment of the accuracy of these
models is a key concern for supervisors.
The discussion of value-at-risk models in this
paper might suggest that supervisory evaluation of banks’
internal models is a daunting task, necessitating the hiring
of large numbers of new staff with the same degree of
technical and market expertise as the bank personnel
responsible for developing and implementing the models.
This interpretation is somewhat mistaken, however.
Although the task of assessing value-at-risk models
requires supervisors to maintain staff with a high degree of
technical skill and experience in reviewing banks’ trading
operations, it is largely an extension of the activities
routinely performed by U.S. bank supervisors in overseeing
the trading operations of major banks. These activities have
typically entailed review and assessment of the accuracy
and appropriateness of the models used by banks for
pricing, risk management, and general ledger profit and
loss calculations. Thus, the basic procedures for evaluating
value-at-risk models are similar to those that have been
used by U.S. supervisors for some time in reviewing banks’
trading activities. The procedures followed by examiners
are also quite similar in spirit to the techniques used by
auditors and accountants to assess the accuracy of the books
and records of a banking institution.
As a first step, supervisors can turn to the internal
auditing and certification processes used by the banks to
validate the accuracy and performance of their models. The
qualitative standards imposed by the market risk capital
guidelines require independent validation of any models
used to value positions or to measure the sensitivity of
portfolios to market risk. As we have seen, the standards
also call for an independent risk management unit and an
independent internal or external audit of a bank’s risk man-

agement processes. The results of these internal reviews
provide supervisors with a valuable starting point for
their own evaluation. The standards also mandate that
the models be used as an integral part of a bank’s risk management process—for instance, as part of daily management reports or as the basis of the bank’s limit system.
Because the models are used for purposes that go well
beyond calculating regulatory capital levels, the interests of bank management in obtaining accurate value-atrisk estimates may be more closely aligned with the interests of supervisors.
Backtesting results—both those generated as
required for supervisory capital purposes and additional

Although the task of assessing value-at-risk
models requires supervisors to maintain staff
with a high degree of technical skill and
experience in reviewing banks’ trading
operations, it is largely an extension of the
activities routinely performed by U.S. bank
supervisors in overseeing the trading
operations of major banks.

results generated by institutions for internal validation and
calibration—also provide supervisors with important
information about the accuracy of value-at-risk models.
Although the backtesting procedures incorporated in the
market risk capital requirements are based on relatively
simple statistical tests, researchers at the banks and
elsewhere are actively investigating how to use ex post
trading results to draw inferences about the accuracy and
performance of value-at-risk models (see Kupiec [1995],
Crnkovic and Drachman [1996], and Lopez [1997]). This
work may lead to better and more powerful techniques for
using these data to assess the accuracy of value-at-risk
models.

FRBNY ECONOMIC POLICY REVIEW / DECEMBER 1997

9

In addition to drawing on these resources, supervisors rely on a dialogue with risk management staff at the
bank in question and on a process of evaluating key
assumptions and parameters of the models. Both the
dialogue with the banks and the evaluation of the model
parameters depend on having supervisory staff that can
assess the technical work performed by a bank’s risk management and trading staff. But while developing and
retaining examiners with these skills is a key challenge for
supervisors, the task is likely to become easier over time.
Basic information about the structure and theoretical
underpinnings of value-at-risk models is spreading, and
the models are quickly becoming commonplace at financial
(and nonfinancial) institutions. An understanding of these
models is also emerging as a standard part of the skills
acquired through academic and on-the-job training in
finance and risk management. Thus, value-at-risk modeling is becoming a significantly less arcane area of both risk
management and supervisory oversight.
Taken together, these factors suggest that supervisors have a broad arsenal of approaches to use in evaluating
value-at-risk models. While experience over time will
determine whether the information generated by these
models is consistent and reliable, there is good reason to
believe that the market risk capital requirements will yield
information that is useful to both supervisors and market
participants.

IMPLICATIONS FOR THE FUTURE
Market risk capital requirements based on internal models
have drawn considerable attention since the initial proposal
for these requirements was released in 1995. During this
time, supervisory interest in value-at-risk models has
encouraged banks in the United States and abroad to direct

10

FRBNY ECONOMIC POLICY REVIEW / DECEMBER 1997

resources and attention toward the further development of
these models and their fuller integration with the risk
management process.
In the coming years, some of the key issues facing
banks in value-at-risk modeling—and in risk management
more generally—will concern the extension of these models to cover a broader range of the risks facing banking
institutions. For example, can quantitative risk models be
applied to credit, operational, and legal risks? And if so,
should supervisors expand the use of their internal models
to derive capital charges for these exposures? Interestingly,
these issues have already surfaced in banks’ efforts to model
specific risk. Specific risk incorporates elements of both
market risk and credit risk. In measuring specific risk,
banks face a number of difficult technical and conceptual
problems—how to measure the probability and likely
impact of events that occur infrequently and how to quantify the effects of complex events that depend on the interrelated actions of many parties. These problems, which are
at the frontier of thinking about regulatory capital and
banks’ internal capital allocation, will need to be resolved
if quantitative risk models are to be used systematically to
gauge other forms of risk.
At present, banks and other financial institutions
are still in the early stages of developing methods for
quantifying other types of risk and for integrating these
risks into a unified capital allocation framework. Understanding the ways that risk models can and cannot be used
is clearly one of the most significant challenges facing
financial institutions and their supervisors today. The
market risk capital requirements may further this understanding by providing a test case for the supervisory use of
internal models.

ENDNOTES

The authors thank Christine Cumming, Arturo Estrella, Fred Herriman, Pat
Parkinson, and Brian Peters for many valuable conversations on the topics
addressed in this paper.
1. See Board of Governors of the Federal Reserve System (1994) for a
description of the risk-based capital standards that apply to state member
banks and bank holding companies. The standards for state nonmember
banks and for national banks (administered by the Federal Deposit
Insurance Corporation and the Office of the Comptroller of the Currency,
respectively) are essentially identical.
2. Readers interested in the details of the market risk capital
requirements should see Basle Committee on Banking Supervision
(1996a) and U.S. Department of the Treasury, Federal Reserve System,
and Federal Deposit Insurance Corporation (1996). The amended Basle
Accord contains a second method for calculating market risk capital
requirements that is not included in the U.S. guidelines. This second
approach—the “standardized approach”—requires an institution to
apply certain uniform techniques to calculate the capital charge for
market risk. It is also important to distinguish the internal models
approach contained in the U.S. guidelines from the so-called
precommitment approach, which has been released for discussion by the
Board of Governors of the Federal Reserve System and is being explored
in a pilot project by the New York Clearing House (see Board of
Governors of the Federal Reserve System [1995]). Under the
precommitment approach, banks would have latitude to specify the
amount of capital they wished to allocate to market risk, subject to
penalties if subsequent trading losses exceeded this precommitted
amount. This approach is one of several alternative methods that have
been suggested for determining banks’ capital requirements. For another,
see Estrella (1995), who proposes capital supervision based on banks’
internally determined “optimal” capital levels, in combination with a
simple supervisory minimum.
3. The U.S. capital standards have recently been amended to require
that a bank’s capital be adequate to cover its overall exposure to interest
rate risk. This determination is made as part of a bank’s supervisory examination, rather than through a formal minimum capital requirement.
4. The exceptions are derivative positions, which continue to be subject
to counterparty credit risk capital requirements.
5. As of the end of 1996, seventeen commercial banks met these
criteria. These seventeen banks held nearly 98 percent of the trading
positions (assets plus liabilities) held by all U.S. commercial banks. In
addition, seventeen bank holding companies met the criteria, including
the holding companies associated with fourteen of the seventeen banks.
The actual number of institutions that are ultimately subject to the
market risk capital requirements may differ from these figures, for two
reasons: supervisors can, at their own discretion, include or exclude

NOTES

particular institutions, and institutions have the option to become
subject to the capital requirements with supervisory approval.
6. See Jorion (1996) for a more detailed discussion of value-at-risk
models. Hendricks (1996) compares the performance of several types of
value-at-risk models.
7. To be precise, the capital charge for general market risk is equal to
the greater of the sixty-day average value-at-risk estimate times the
scaling factor or the previous day’s value-at-risk estimate. As a practical
matter, the previous day’s value-at-risk estimate should rarely, if ever,
exceed the sixty-day average times three.
8. Of course, some positions could take longer than ten days to liquidate.
The extent to which a ten-day holding period is a suitable average would
obviously depend on the characteristics of an individual portfolio.
9. Gamma risk arises from the fact that the sensitivity of an option’s
value to changes in the value of the option’s underlying instrument will
vary as the value of the underlying instrument changes.
10. Note, however, that the existing empirical evidence does not
suggest substantial differences in the performance of value-at-risk models
with varying observations periods.
11. For a full discussion of the use of backtesting in the market risk
capital requirements, see Basle Committee on Banking Supervision
(1996b). For a discussion of the statistical properties of backtesting and
other methods of evaluating the accuracy of value-at-risk models, see
Kupiec (1995) and Lopez (1997).
12. For banks whose value-at-risk models do not adequately incorporate
specific risk, debt and equity positions in the trading portfolio are subject
to a set of standardized specific risk charges, which apply to both long
and short positions. These charges are added to the value-at-risk-based
general market risk charge. The standardized charges are in many cases
significantly lower than the original credit risk capital charges. For
instance, an investment-grade corporate bond, which would have been
subject to an 8 percent credit risk capital charge under the earlier
guidelines, is now subject to a 1.6 percent specific risk charge.
13. There is a concern that measures of recent price variability may not
provide a complete guide to the potential risk inherent in some
positions—for example, illiquid positions that trade infrequently. This
concern, together with the existence of differing market practices in this
regard, has been a factor in shaping the interim approach to specific risk.
14. The institutions cited in Table 2 are used for illustrative purposes
only. They do not represent an exhaustive list of the bank holding
companies that reported value-at-risk estimates in their 1996 annual
reports.

FRBNY ECONOMIC POLICY REVIEW / DECEMBER 1997

11

REFERENCES

Bank for International Settlements. 1994. “A Discussion Paper on Public
Disclosure of Market and Credit Risks by Financial Intermediaries,”
September.

Federal Reserve Bank of New York. 1994. “Public Disclosure of Risks
Related to Market Activity: A Discussion Paper,” November.

Basle Committee on Banking Supervision. 1996a. “Amendment to the
Capital Accord to Incorporate Market Risks.”

Hendricks, Darryll. 1996. “Evaluation of Value-at-Risk Models Using
Historical Data.” Federal Reserve Bank of New York ECONOMIC
POLICY REVIEW 2, no. 1: 39-69.

———. 1996b. “Supervisory Framework for the Use of ‘Backtesting’ in
Conjunction with the Internal Models Approach to Market Risk
Capital Requirements.”

Jorion, Phillipe. 1996. V ALUE AT R ISK : T HE N EW B ENCHMARK
FOR CONTROLLING MARKET RISK. Chicago: Irwin Professional
Publishing.

———. 1997. “Explanatory Note: Modification of the Basle Capital
Accord of July 1988, as amended in January 1996.”

Kupiec, Paul H. 1995. “Techniques for Verifying the Accuracy of Risk
Measurement Models.” JOURNAL OF DERIVATIVES 3: 73-84.

Board of Governors of the Federal Reserve System. 1994. CAPITAL ADEQUACY
GUIDELINES.

Lopez, Jose A. 1997. “Regulatory Evaluation of Value-at-Risk Models.”
Federal Reserve Bank of New York Research Paper no. 9710.

———. 1995. “Capital Requirements for Market Risk.” FEDERAL
REGISTER 60, no. 142: 38142-4.

U.S. Department of the Treasury (Office of the Comptroller of the Currency),
Federal Reserve System, and Federal Deposit Insurance Corporation. 1996.
“Risk-Based Capital Standards: Market Risk.” FEDERAL REGISTER 61,
no. 174: 47357-78.

Crnkovic, C., and J. Drachman. 1996. “Quality Control.” RISK MAGAZINE
9: 139-43.
Estrella, Arturo. 1995. “A Prolegomenon to Future Capital Requirements.” Federal Reserve Bank of New York E CONOMIC P OLICY
R EVIEW 1, no. 2: 1-12.

The views expressed in this article are those of the authors and do not necessarily reflect the position of the Federal Reserve
Bank of New York or the Federal Reserve System. The Federal Reserve Bank of New York provides no warranty, express or
implied, as to the accuracy, timeliness, completeness, merchantability, or fitness for any particular purpose of any information
contained in documents produced and provided by the Federal Reserve Bank of New York in any form or manner whatsoever.

12

FRBNY ECONOMIC POLICY REVIEW / DECEMBER 1997

NOTES

The Benefits of Branching
Deregulation
Jith Jayaratne and Philip E. Strahan*

T

he Riegle-Neal Interstate Banking and
Branching Efficiency Act, implemented in
June 1997, enables banks to establish branches
and buy other banks across the country. This
legislation is the final stage of a quarter-century-long effort
to relax geographic limits on banks. As recently as 1975, no
state allowed out-of-state bank holding companies (BHCs) to
buy in-state banks, and only fourteen states permitted
statewide branching. By 1990, all states but Hawaii
allowed out-of-state BHCs to buy in-state banks, and all
but three states allowed statewide branching. The RiegleNeal Act removes the remaining restrictions by permitting
banks and BHCs to cross state lines freely.1
Although the effects of the recent federal legislation
will be known only over time, we can study the impact of
geographic restrictions on the banking industry by
examining an earlier stage of the deregulatory process.

*Jith Jayaratne and Philip E. Strahan are senior economists at the Federal
Reserve Bank of New York.

The states were most active in removing geographic limits
on banks in the fifteen years from 1978 to 1992. By
observing the changes in banking that followed the state
initiatives, we can learn much about the impact of these
limits.2 Previous research has suggested that geographic
restrictions destabilized the banking system by creating
small, poorly diversified banks that were vulnerable to
bank runs and portfolio shocks (Calomiris 1993). In this
article, we focus instead on the effect of the restrictions on
the efficiency of the banking system.
We find that bank efficiency improved greatly
once branching restrictions were lifted. Loan losses and
operating costs fell sharply, and the reduction in banks’
costs was largely passed along to bank borrowers in the form
of lower loan rates. The relaxation of state limits on interstate banking was also followed by improvements in bank
performance, but the gains were smaller and the evidence
of a causal relationship less robust.
Our analysis suggests that much of the efficiency
improvement brought about by branching was attributable

FRBNY ECONOMIC POLICY REVIEW / DECEMBER 1997

13

to a selection process whereby better performing banks
expanded at the expense of poorer performers. It appears that
the branching restrictions acted as a ceiling on the size of
well-managed banks, preventing their expansion and
retarding a process of industry evolution in which less efficient
firms routinely lose ground to more efficient ones.
While the improvements to the banking system
following deregulation helped bank customers directly, we
also find important benefits to the rest of the economy. In particular, state economies grew significantly faster once branching was allowed—in part, we suggest, because deregulation
permitted the expansion of those banks that were best able to
route savings to the most productive uses. Although it is
uncertain whether the observed acceleration in economic
growth will last beyond ten years, the stimulative effect of
branching deregulation on the economy has been considerable.

A BRIEF HISTORY OF GEOGRAPHIC
RESTRICTIONS ON BANKING
States began imposing limits on branch office locations in
the nineteenth century. Such limits were intended in part
to prevent unscrupulous bankers from “choosing inaccessible
office sites to deter customers from redeeming . . . circulating

As late as 1975 only fourteen states allowed
statewide branching. Twelve states prohibited
branching altogether, and the remainder
imposed restrictions of varying severity.

banknotes” (Kane 1996, p. 142). Geographic limits were
also justified by the political argument that allowing banks
to expand their operations freely could lead to an excessive
concentration of financial power. Appearing before Congress
in 1939, the Secretary of the Independent Bankers Association
warned that branch banking would “destroy a banking
system that is distinctively American and replace it
with a foreign system . . . a system that is monopolistic,

14

FRBNY ECONOMIC POLICY REVIEW / DECEMBER 1997

undemocratic and with tinges of fascism” (Chapman and
Westerfield 1942, p. 238).
Inefficient banks probably supported these restrictions because they prevented competition from other
banks. Economides, Hubbard, and Palia (1995) show that
states with many weakly capitalized small banks favored
the 1927 McFadden Act, which gave states the authority to
regulate national banks’ branching powers. The states
themselves often benefited from exercising control over the
supply of bank charters and the expansion of branch banking. Massachusetts and Delaware, for instance, received a
majority of their state revenues from bank regulation in the
early nineteenth century (Sylla, Legler, and Wallis 1987).
Geographic restrictions may not have seriously
constrained the banking industry before the appearance
of large corporations that required large-scale, multistate banking services. Rapid industrialization and the
growth of transcontinental railroads after the Civil War,
however, created firms whose need for comprehensive
corporate financial services could not be met adequately
by the existing system of fragmented unit banks. In
response, banks formed “chain banks”—an alliance of
several banks whose principal ownership rested with the
same group of investors—after 1890. A few years later,
“banking groups”—banks owned directly by a holding
company—were created in an effort to get around
branching restrictions (Calomiris 1993).
Nevertheless, branching restrictions persisted, and
as late as 1975 only fourteen states allowed statewide
branching. Twelve states prohibited branching altogether,
and the remainder imposed restrictions of varying severity.
Pennsylvania was representative of a partially restrictive
state. Until 1982, Pennsylvania banks were allowed to
branch only in the county where their head offices were
located and in contiguous counties.
In addition to facing restrictions on in-state
branching, banks have traditionally been limited in
their ability to cross state lines. The Douglas Amendment to the 1956 Bank Holding Company Act prohibited
a BHC from acquiring banks outside the state where it
was headquartered unless the target bank’s state permitted such acquisitions. Since no state allowed such

transactions in 1956, the amendment effectively barred
interstate banking organizations. Although states had
the option to allow out-of-state BHCs to enter, none
exercised that right until 1978, when Maine permitted
such transactions. Even then, however, little changed:
the Maine statute allowed an out-of-state BHC to buy a
Maine bank only if the home state of the acquiring BHC
permitted Maine-based BHCs the reciprocal right to
buy banks there; since no other state allowed such entry,
interstate bank organizations could not be formed.
Banks could not in fact cross state borders until 1982,
when Alaska, Massachusetts, and New York permitted
out-of-state BHCs to enter.

MOVES TOWARD DEREGULATION
Maine’s 1978 move to permit entry by out-of-state BHCs
marked the beginning of a fifteen-year period in which the
states relaxed barriers to bank expansion.3 By the end of
1992, the state-level deregulatory process was essentially
completed: all states but Arkansas, Iowa, and Minnesota
allowed statewide branching, and all states but Hawaii permitted out-of-state BHCs to enter.
Table 1 chronicles the steps taken by individual
states to eliminate geographic restrictions.4 The first column
presents the year in which each state authorized branching
by means of merger and acquisition.5 The second column
reports the year in which each state first permitted interstate
banking. In some cases, choosing a date for the authorization
of branching was difficult, because the states often deregulated only gradually. In most cases, the date selected reflects
the time at which the state finished the branching
deregulation process.6 In four cases, however, we chose dates
earlier than the literal end of the process of deregulation
because the remaining restrictions did not appear to impose
a meaningful constraint on branching.7

FORCES OF CHANGE
Several developments contributed to the removal of the geographic barriers to bank expansion. In the mid-1980s, the
Office of the Comptroller of the Currency took advantage of a
clause in the 1864 National Bank Act to allow nationally
chartered banks to branch freely in those states where thrifts

did not face branching restrictions. The Comptroller’s action
was instrumental in introducing statewide branching in

Table 1
THE STATES REMOVE RESTRICTIONS
ON GEOGRAPHIC EXPANSION
State
Alabama
Alaska
Arizona
Arkansas
California
Colorado
Connecticut
Delaware
District of Columbia
Florida
Georgia
Hawaii
Idaho
Illinois
Indiana
Iowa
Kansas
Kentucky
Louisiana
Maine
Maryland
Massachusetts
Michigan
Minnesota
Mississippi
Missouri
Montana
Nebraska
Nevada
New Hampshire
New Jersey
New Mexico
New York
North Carolina
North Dakota
Ohio
Oklahoma
Oregon
Pennsylvania
Rhode Island
South Carolina
South Dakota
Tennessee
Texas
Utah
Vermont
Virginia
Washington
West Virginia
Wisconsin
Wyoming

Intrastate Branching
Deregulated
1981
Before 1970
Before 1970
1994
Before 1970
1991
1980
Before 1970
Before 1970
1988
1983
1986
Before 1970
1988
1989
—
1987
1990
1988
1975
Before 1970
1984
1987
1993
1986
1990
1990
1985
Before 1970
1987
1977
1991
1976
Before 1970
1987
1979
1988
1985
1982
Before 1970
Before 1970
Before 1970
1985
1988
1981
1970
1978
1985
1987
1990
1988

Interstate Banking
Deregulated
1987
1982
1986
1989
1987
1988
1983
1988
1985
1985
1985
—
1985
1986
1986
1991
1992
1984
1987
1978
1985
1983
1986
1986
1988
1986
1993
1990
1985
1987
1986
1989
1982
1985
1991
1985
1987
1986
1986
1984
1986
1983
1985
1987
1984
1988
1985
1987
1988
1987
1987

Source: Chronology is based on information in Amel (1993).
Note: Before the passage of the Riegle-Neal Act, Iowa had not deregulated
intrastate branching and Hawaii had not deregulated interstate banking.

FRBNY ECONOMIC POLICY REVIEW / DECEMBER 1997

15

several southern states. Another impetus behind deregulation
may have been the rash of bank and thrift failures in the
1980s, which increased public awareness of the advantages of
large, well-diversified banks (Kane 1996).
Kroszner and Strahan (1997) suggest that the emergence of new technologies in both deposit taking and lending
encouraged the elimination of geographic barriers by changing the nature of banking markets. For instance, the introduction of the automated teller machine in the late 1970s and the

The initiative to relax restrictions on interstate
banking came primarily from larger banking
organizations that were well equipped to
pursue lower funding costs and better lending
opportunities in neighboring states.

development of money market mutual funds increased competitiveness in deposit markets. As a result, branching and
interstate banking restrictions could no longer offer the same
degree of protection from competition, making it less likely
that banks would lobby for the preservation of these rules. At
the same time, new information technologies diminished the
value of the specialized knowledge that long-established local
bankers might have had about the risks of borrowers in the
community. This change enhanced the ability of banks to lend
in more distant markets. Thus, a situation developed in which
protected banks’ incentive to defend restrictions on branching
and interstate banking diminished over time, while expansion-minded banks’ desire to see the restrictions fall increased.
The initiative to relax restrictions on interstate
banking came primarily from larger banking organizations
that were well equipped to pursue lower funding costs
and better lending opportunities in neighboring states.
Their efforts may have succeeded in the 1980s because it
became apparent that banks and nonbanks were already practicing interstate banking. As Savage (1993) argues, “the proliferation of loan production offices, nonbank subsidiaries of

16

FRBNY ECONOMIC POLICY REVIEW / DECEMBER 1997

bank holding companies, nonbank banks, and interstate
thrift institutions, the widespread use of credit cards, and
the provision of financial services by nonfinancial firms not
subject to geographic limitations all made the traditional restrictions on the geographic expansion of banks
more difficult to explain and justify. If so many financial
services could be provided across state lines by these various
means, why shouldn’t deposit-taking institutions be allowed
to expand as well?”
The breakdown of the geographic constraints on
banks over the last twenty years has had a significant
impact on the industry. Branching deregulation has
prompted banks to enter new markets (Amel and Liang
1992), persuaded BHCs to consolidate their subsidiaries
into branches (McLaughlin 1995), and forced smaller
institutions to exit banking (Calem 1994). Interstate
banking activity has increased dramatically, boosting the
percentage of deposits held by out-of-state BHCs in the
typical state from 2 percent to 28 percent between 1979
and 1994 (Berger, Kashyap, and Scalise 1995). Interstate
banking has also intensified the demands placed on bank
management: the compensation of managers is now tied
more closely to bank performance, and the turnover rate
among banks’ chief executive officers has increased
(Hubbard and Palia 1995).
In addition to prompting changes in the organization of the industry and the behavior of individual banks,
deregulation has had profound effects on the overall performance of the banking system. The next section looks at the
impact of deregulation on two components of bank performance: the costs of providing services and the prices
charged customers for those services.

DEREGULATION, COST EFFICIENCY,
AND PRICES
Did banks perform better when they were permitted to
operate statewide branch networks and to build multistate bank holding companies? We investigate this question
by examining whether bank costs—as measured by loan
losses (net loan charge-offs divided by total loans) and noninterest costs (noninterest expenses divided by total
assets)—declined after deregulation, creating a more effi-

cient system. We also examine changes in loan prices
(interest income on loans and leases divided by total loans
and leases) to determine whether bank customers are better
off following deregulation. We look at state-level data for the
1978-92 period to summarize the impact of deregulation on
the overall performance of the banking system.
To understand how we arrive at our measures
of the cost efficiency of the banking system, consider
New York in 1978. We construct the charge-offs ratio
by dividing the sum of loans charged off by all banks
operating in New York in 1978 by the sum of all loans
held by New York banks in 1978. We construct similar aggregates for the noninterest expense and loan
price variables in each state and year in the sample. 8
The data for these performance measures are derived
from the year-end Reports of Condition and Income,
filed by all banks with the federal banking agencies.

We use regression techniques to estimate the
impact of deregulation on bank costs and loan prices. (For a
detailed discussion of these calculations, see Box 1.) The
regression methods allow us to control for other factors
that might influence our measures of bank cost and loan
prices—most notably, the health of the state’s economy.
Bank costs, particularly those related to loan defaults,
generally move with the business cycle: borrowers tend to
pay off loans during boom times but are less able to do so
during recessions. If states deregulated branching and
interstate banking during hard times, average measures of
costs could improve after deregulation as states’ economies
recovered from recession. A simple before-and-after comparison of bank performance would show an improvement
in bank loan portfolios and profitability after deregulation, but these advances would largely reflect the timing
of deregulation. We address this possibility by controlling

BOX 1: AN EMPIRICAL MODEL OF BANK PERFORMANCE
Using the dates of deregulation reported in Table 1, we construct two indicator variables equal to 1 for states permitting

measures for banks in these two states do not reflect their branching laws, but rather the health and profitability of the credit card

branching and interstate banking. We then use these indicator variables to estimate the effects of the policy changes in

business.

the following regression model:

predicted values for our two cost measures and our measure of
loan prices in different regulatory environments. Consider

y t,i = α t + β i + γ 1 branch t, i + γ 2 bank t, i + ε t, i ,
where yt,i equals one of our two cost measures or our measure
of loan prices in the ith state in year t, brancht,i is an indicator
equal to 1 for states without restrictions on branching, and
bankt,i is an indicator equal to 1 for states that have entered
into an interstate banking agreement.
In this specification, βi measures the state-specific
component of banking performance, αt measures the effects of
the national business cycle at time t, and γ1 and γ2 measure the
changes in performance stemming from the two types of deregulation. In constructing the deregulation indicators, we drop the
year in which the deregulation went into effect. We also drop
Delaware and South Dakota from the analysis entirely. These two
states experienced a dramatic expansion in their banking sectors
during the 1980s when credit card operations relocated there to
take advantage of liberal usury laws. As a result, performance

We then use the regression model to construct average

charge-offs. We estimate the predicted value of this variable
for each state and year for each of three regulatory configurations: one in which both branching and interstate banking are
fully regulated (brancht,i = 0 and bankt,i = 0), one in which
branching is permitted but interstate banking is not
(brancht,i = 1 and bankt,i = 0), and one in which both branching
and interstate banking are permitted (brancht,i = 1 and bankt,i = 1).
This gives us a panel of predicted values for each state and
year in each of the three regulatory environments. We
then compute the simple average predicted charge-off ratio
(across states and years) for each regulatory configuration
and report each of those three averages in Chart 1 in the text.
The statistical significance reported in the text is derived
by testing the hypothesis that γ1 and γ2 estimated from the
above regression equal zero.

FRBNY ECONOMIC POLICY REVIEW / DECEMBER 1997

17

for the national business cycle in our regressions.9
Our analysis suggests that loan losses, noninterest
expenses, and loan rates decreased significantly once statewide
branching was allowed—even after we adjust for the influence
of the business cycle on bank performance and for persistent
cross-state differences in bank performance.10 Chart 1 reports
the average levels of the cost and price measures that would
have been observed during the 1978-92 sample period under
three alternative regulatory regimes: (1) restrictions in place
on both branching and interstate banking, (2) branching
Chart 1

Costs and Interest Rates Are Lower
in Deregulated Environments
Percent
1.5
Ratio of Loan Charge-offs to Total Loans
1.2%

1.0
0.6%
0.5

0.4%

0
5
4

Ratio of Noninterest Expenses to Assets
3.5%

3.3%

3.3%

3
2
1
0
12.0
11.5

Yield on Loans
11.5%
11.1%

11.0

10.8%

10.5
10.0
Branching and interstate
Branching
Branching and interstate
banking permitted
permitted; interstate
banking prohibited
banking prohibited
Source: Authors’ calculations, based on data from Federal Financial Institutions
Examination Council, Reports of Condition and Income.
Note: Chart shows the average level of price and performance measures that
would have been observed in the 1978-92 period had all states been subject
to the regulatory regimes identified along the x-axis.

18

FRBNY ECONOMIC POLICY REVIEW / DECEMBER 1997

permitted but interstate banking prohibited, and (3) both
branching and interstate banking permitted. The top panel
suggests that if no state had allowed either statewide branching or interstate banking between 1978 and 1992, the ratio of
charge-offs to total loans in the typical state in a typical year
would have been 1.2 percent. Had all states allowed statewide
branching but prohibited interstate banking in our sample
period, average charge-offs in the typical state would have
fallen by half, to 0.6 percent.11 The ratio of noninterest
expenses to assets would have fallen from 3.5 percent to
3.3 percent if branching had been permitted throughout the
period (middle panel). It appears that most of these reduced
costs were passed along to bank borrowers in the form of lower
loan rates, which in our estimates declined from 11.5 percent
to 11.1 percent on average (bottom panel).12 Each of these
improvements is statistically significant at the 5 percent level.13
Foes of bank deregulation and consolidation have
argued that the increasing concentration in the banking
industry could enhance market power. While measures of
concentration at both the state and national levels have
increased in recent years following deregulation, concentration at local levels has remained remarkably constant
(Rhoades 1996). If enhanced market power were a problem,
we would see both increased concentration and higher
prices at the local level following deregulation, neither of
which has occurred. It is true that our estimates indicate that
bank costs have fallen more than revenues, suggesting an
increase in industry profitability. Similarly, estimates of the
impact of deregulation on banks’ return on equity and
return on assets in another study (Jayaratne and Strahan
forthcoming) showed small increases in profitability that
were sometimes statistically significant (at the 10 percent
level) and sometimes not. Nevertheless, it appears that
most, or perhaps all, of the cost reductions from deregulation are passed along to customers. There is little evidence
that deregulation has increased market power.
Our regression analysis also shows that some modest improvements in bank performance have followed the
introduction of interstate banking. Although operating
costs do not decline at all (Chart 1, middle panel), chargeoffs fall from 0.6 to 0.4 percent of total loans when
interstate banking is allowed in addition to statewide

branching (top panel), and the average interest rate falls
from 11.1 percent to 10.8 percent (bottom panel).
The evidence of gains following interstate banking
deregulation, however, is much less robust than the evidence
of improvements following branching deregulation. When
we control for state business cycles (by including lags of
state-level personal income growth) as well as national
business cycles, we see no statistically significant improvements following interstate banking. This finding suggests
that the observed gains might stem from favorable banking
conditions at the time of deregulation rather than from
deregulation itself. Alternatively, robust evidence of performance improvements following interstate banking may be
lacking because most states entered interstate banking agreements around the same time, making it difficult to distinguish
the effects of deregulation from the effects of other changes.
Because of this statistical problem, we cannot determine
whether interstate banking had a significant impact on bank
performance. Consequently, we focus on branching
deregulation in the remainder of the article.

ROBUSTNESS OF THE PERFORMANCE IMPROVEMENTS
A possible explanation for the observed reduction in
loan losses and loan rates is that banks made fewer risky
loans following branching deregulation. If the output
mix of banks changed from riskier to safer loans following deregulation, then we might expect to observe
declines in both loan losses and loan rates. Changes in
banks’ output could also explain declines in noninterest
expenses if, for instance, banks provided fewer checking
accounts (which are relatively costly for banks to maintain) following deregulation. To investigate this possibility, we estimate the effects of deregulation on
noninterest expenses, loan losses, and loan prices while
controlling for banks’ output mix. In each case, we find
that the improvements in costs and the reductions in
loan losses and loan prices after branching deregulation
remain statistically significant even after controlling for
the output mix. We also find no decrease in two risky
loan categories—credit cards and commercial loans—
following branch deregulation, suggesting that banks
did not shift to safer loans after deregulation.14

It is possible, however, that within each loan category banks are making safer loans after deregulation than
they did before. So, even though the volume of credit card
loans and commercial loans has remained fairly constant,
after deregulation the loans themselves may be less risky.
This is unlikely for two reasons. First, evidence suggests
that, if anything, banks increased their risk taking after geographic deregulation because eliminating entry barriers
reduced banks’ franchise value (Keeley 1990). Second, as
we indicate below, banks with higher profits and fewer loan
losses grew faster than banks with lower profits and more
loan losses once branching was permitted. Declines in loan
losses seem to reflect not a change in the inherent riskiness of
the pool of borrowers but better screening and monitoring
of borrowers by the banking system.
We have established that bank performance in the
average state improved following statewide branching. But
did banks in only a few states experience improvements, or
was the phenomenon widespread? To answer this question, we
look at the changes in bank cost efficiency in individual
states (Chart 2). Specifically, we plot the change in banks’
ratio of charge-offs to total loans before and after deregulation relative to the corresponding change for the group
of states that did not deregulate their branching laws during
the period. This “control group” of states is used to remove the
Chart 2

Loan Charge-offs Fall after Branching Deregulation
in All but Two States
Change in loan charge-offs
0.010
NH

0.005
UT

0.000
NJ

VA OH CT

MA
PA

-0.005
AL

-0.010
-0.015
75

80

GA

TX
IL
LA
NE
FL
NM
OR WV
TN
MI IN KY
KS
MO
WA MS
WI
ND
HI
MT
OK
CO
WY

85

90

95

Year of branching deregulation
Source: Authors’ calculations, based on data from the Federal Financial
Institutions Examination Council, Reports of Condition and Income.

FRBNY ECONOMIC POLICY REVIEW / DECEMBER 1997

19

effects of nationwide shocks to bank performance. The control
group consists of the eleven states that are identified in Table 1
as having deregulated in or before 1970 and the three that are
identified as not having deregulated as of 1992.15

Reductions in loan losses following branching
deregulation are widespread; in all states but
New Hampshire and Utah, charge-offs decline
after deregulation relative to the change in
charge-offs experienced by states that did not
deregulate branching during the period.

The change in loan charge-offs for each of the thirtythree deregulating states appears as a single point plotted
above the year of deregulation for that state; multiple points
appear above a year when more than one state deregulated in
that year. Consider the example of Pennsylvania, represented
by the single point plotted in 1982. This state’s mean chargeoff ratio rose by about 0.3 percentage point after deregulation
in 1982, while all states that did not change policy in 1982
experienced a 0.7 percentage point increase in charge-offs after
1982. We therefore report a relative decline in charge-offs of
0.4 percentage point for Pennsylvania.
As the chart shows, reductions in loan losses following branching deregulation are widespread; in all
states but New Hampshire and Utah, charge-offs
decline after deregulation relative to the change in
charge-offs experienced by states that did not deregulate
branching during the period. Similar pictures emerge
for both loan prices and noninterest expenses. For loan
prices, we find declines following branching deregulation in twenty-five cases out of thirty-three. Again,
New Hampshire is a significant outlier. 16 We find that
noninterest expenses fall in nineteen out of the twentyfour deregulating states available for this analysis, again
relative to the control group of states.

20

FRBNY ECONOMIC POLICY REVIEW / DECEMBER 1997

WHY DEREGULATION IMPROVES BANK EFFICIENCY
Limits on bank expansion could have had adverse effects
on efficiency in banking for at least three reasons. First,
prohibitions on branching and interstate banking may
have limited the opportunity for the best run banks to
grow. In unregulated markets, more efficient firms have a
natural tendency to gain market share over their less productive competitors, an outcome that will increase average efficiency as the industry evolves over time. By
preventing better run banks from establishing branches,
and by preventing BHCs from expanding across state
lines, these regulations may have retarded this natural
evolution. After the geographical constraints were lifted,
the more efficient banks may have expanded, thereby
improving the performance of the average banking asset.
We call this the selection hypothesis.
Second, limited restrictions on geographic expansion
may have weakened the discipline that markets usually place
on managers of corporations. When interstate banking is prohibited, managers worry less about takeovers. Because their
jobs are more secure, they may also be less motivated to
increase shareholder value, maximize efficiency, and minimize
costs. According to this disciplining hypothesis, efficiency in
banking improves after deregulation because managers are
forced to increase shareholder value in order to preserve their
jobs. Note that the disciplining hypothesis predicts that all

Prohibitions on branching and interstate
banking may have limited the opportunity for
the best run banks to grow.

banks will improve their performance following deregulation,
since managers at all banks will come under greater pressure.
By contrast, the selection hypothesis predicts that the more
efficient banks will gain market share, not that the efficiency
of all individual banks will improve.
A third possible reason why efficiency might
improve following deregulation is that barriers to geographic expansion prevent banks from operating at the

most efficient size. There is some evidence, for instance,
that small banks can reduce average costs by expanding up
to about $500 million in total assets (Berger, Hunter, and
Timme 1993). According to the economies of scale hypothesis,
the efficiency of the banking system will improve after
deregulation as small banks grow and reduce costs. Of course,
according to this view, all of the benefits come from changes
occurring at the lower end of the bank size distribution. Since
small banks hold a relatively small share of total banking
assets, these benefits would likely be small.
Which of these three explanations best accounts for
the efficiency gains observed following deregulation? We can
rule out the economies of scale explanation on two grounds.
First, there is scant evidence of scale economies in banking
beyond about $500 million in total assets (Berger, Hunter,
and Timme 1993). The large improvements that we have
found in the state-level aggregates cannot plausibly be attributed to the fact that small banks are moving closer to
the optimal scale. In 1980, for instance, banks with under
$500 million in assets (in 1994 dollars) held less than
30 percent of total assets in the banking system. Second, we
have estimated the change in our performance measures
following branching deregulation for small banks (those
with assets under $100 million) and large banks separately. We find that the improvements are greater for
large banks than for small, a finding inconsistent with
the economies of scale explanation. 17
More difficult to evaluate is the hypothesis that management discipline accounts for the beneficial effects of
branching deregulation. Because we lack good measures of the
degree of managerial effort at banks, we cannot test this
hypothesis directly. Nevertheless, we cannot reject the possibility that disciplining played some role in the improved efficiency of banks. Hubbard and Palia (1995) find evidence of
greater managerial discipline following interstate banking:
the turnover rate for banks’ chief executive officers rises and
the pay-performance relation tightens once states allow interstate banking. Hubbard and Palia contend that these changes
result from a more active market for corporate control after
deregulation. Such changes may well have disciplined management to improve bank performance, although neither this
article nor the Hubbard and Palia study establishes this point.

The remaining explanation for bank efficiency
gains, the selection hypothesis, can readily be tested. To do
so, we examine whether better run banking companies
grow faster than their less efficient rivals following branching deregulation. First, we classify banks on the basis of
their profitability just before deregulation. We then
observe the change in the market share after deregulation
for the high-profit banking companies. If the selection
hypothesis is correct, we should find that profitable banks
increase their market share at the expense of unprofitable
banks following deregulation.
Specifically, for each state, we first rank banking
companies from highest to lowest according to their return
on equity at the end of the year prior to the year of deregulation. Next, we go down that ranking until we reach a bank
that, together with all previous banks, accounts for 50 percent of the state’s bank assets. The banking companies in this
group constitute our high-profit firms.18 We then calculate
the group’s share of state bank assets five years after
branching deregulation.19 As implied by the selection
hypothesis, we find that the high-profit banking companies
grow faster after branching deregulation (Table 2, row 1);
their share of banking assets increases, on average, by
8.5 percentage points (from 51.3 percent to 59.8 percent)—a
statistically significant increase.20

Table 2
BETTER BANKS INCREASE THEIR MARKET SHARE
AFTER BRANCHING DEREGULATION

Post-deregulation period

Initial Market
Share of
High-Profit
Banks
(Percent)
51.3

Market Share
of High-Profit
Banks Six
Years Later
(Percent)
59.8

Pre-deregulation period

49.9

51.7

Increase
in Share
(Percentage
Point Change)
8.5
(3.91)**
1.8
(0.99)

Source: Authors’ calculations, based on data from Federal Financial Institutions
Examination Council, Reports of Condition and Income.
Notes: The table reports the change in the share of total bank assets held by that
half of the banking companies with the highest return on equity at the beginning
of the specified six-year period. The post-deregulation period begins the year
before the year of deregulation; the pre-deregulation period begins seven years
before the year of deregulation. The t-statistic reported below the market share
change for each period tests the hypothesis that the change equals zero.
*Statistically significant at the 10 percent level.
**Statistically significant at the 5 percent level.

FRBNY ECONOMIC POLICY REVIEW / DECEMBER 1997

21

Of course, we would expect banks enjoying
high profits and good loan portfolios to grow relatively
faster at all times, even when branching restrictions are
in place. In other words, the fact that banks with good
balance sheets grow faster than less profitable banks
need not indicate that deregulation caused the weaker
banks to lose ground. To isolate the effects of deregulation on selection, we compare the differential growth
rates of high- and low-profit banks in a deregulated
environment with the same differential growth rates in
a regulated environment.21
A striking contrast is evident in the growth rates
achieved in regulated and deregulated environments (Table 2).
High-profit banks increase their market share by only
1.8 percentage points (from 49.9 to 51.7 percent) in the
average state over the pre-deregulation period (Table 2, row 2).
This change is so small that we cannot reject the possibility
that high-profit banks do not increase their market share at all
over the six-year period before deregulation (that is, 1.8 percent

High-profit banking companies grow faster

improving bank lending. How did these changes affect the
rest of the economy? Earlier research has shown that countries
with better developed banking systems grow faster because
savings are channeled into the highest-return investments
(King and Levine 1993). Banks can help to route savings to
the most productive uses in two ways. First, they provide
information about the profit potential of different businesses, channeling savings toward good projects and away
from bad. Second, banks monitor those firms with which
they have lending relationships to ensure that bank funds
are put to proper use (Diamond 1984).23
Branching deregulation should enhance the
ability of banks to direct savings to the best projects and
to oversee the successful execution of those projects. As
we have seen, banks function better after branching
deregulation, and their loan losses decrease sharply. The
selection hypothesis suggests that these improvements
occur because banks that are better able to screen and
monitor loans are able to expand their operations at the
expense of less effectively managed banks after deregulation. As a result, the economy can grow faster because
savings flow more consistently into profitable investment opportunities.

after branching deregulation; their share of
banking assets increases, on average, by
8.5 percentage points—a statistically
significant increase.

is not a statistically significant change). In the postderegulation period, by contrast, the market share of the
high-profit banks rises sharply. In sum, the evidence in
Table 2 strongly supports the hypothesis that branching
deregulation forced a process of selection whereby weaker
banks lost ground to better run banks.22

DEREGULATION AND ECONOMIC GROWTH
Thus far we have argued that relaxation of geographic
restrictions improved the performance of the banking system,
enhancing the efficiency of the average bank asset and

22

FRBNY ECONOMIC POLICY REVIEW / DECEMBER 1997

THE EFFECT ON STATE ECONOMIES
To investigate whether state-level rates of economic growth
did in fact increase following branching deregulation,24 we
estimate the change in the average growth rate of two measures of economic activity: real per capita personal income
and real per capita gross state product.25 These two measures
differ somewhat in concept: Personal income reflects the
income of a state’s residents, providing a measure of residents’
welfare. Gross state product, by contrast, measures the total
incomes of factors of production located within the state,
allowing us to assess the economic activity that actually occurs
there.26 As in our estimates of the effects of branching deregulation on bank performance, we control for both business cycle
effects and the effects of differences in the long-run growth
rate across states.27 Our tests of the effects of branching deregulation on the state economies show a significant acceleration
in growth: annual personal income grows about 0.51 percent-

age point faster after branching deregulation, and gross state
product, about 0.69 percentage point faster (Table 3, row 1).
This acceleration is not only statistically significant at the
5 percent level but is also economically “large” relative to the
1.6 percent annual average growth rate of real per capita personal income over the sample period.
Of course, there is uncertainty associated with this
estimate—with a 5 percent probability of error, we can
only be confident that personal income growth increased
somewhere between 0.06 and 0.97 percentage point. Moreover, these figures are estimated under the assumption that
the growth pickup persists indefinitely. One possibility is
that the economy benefits for a few years as the banking
system becomes more efficient, then growth returns to the
level that prevailed before the policy change.
We disentangle the short- and long-run effects
of deregulation on growth by assessing the average
growth rate following deregulation during three distinct time periods (Table 3, rows 2-4). We measure the
change in the growth rate during the first five years
after branching deregulation, the change in growth relative to the years before deregulation during years five to
ten, and the change from years eleven and beyond. We
find that the beneficial effects of the policy change are
greatest during the first ten years. Personal income
growth accelerates by 0.35 percentage point in the first
five years and by 0.37 percentage point in the next five

Table 3
STATES’ ECONOMIC GROWTH ACCELERATES
AFTER BRANCHING DEREGULATION

(1) Overall increase in growth
(2) Increase in growth, years 1-5
(3) Increase in growth, years 5-10
(4) Increase in growth, years 10+

Change in Personal
Income Growth
(Percentage Point)
0.51
(2.22)**
0.35
(1.75)*
0.37
(1.85)*
0.17
(0.89)

Change in Gross State
Product Growth
(Percentage Point)
0.69
(2.09)**
0.60
(2.07)**
0.65
(2.41)**
0.67
(2.48)**

Source: Jayaratne and Strahan (1996), Tables 2 and 5, rows 3 and 7.
Note: The t-statistics are given in parentheses.
*Statistically significant at the 10 percent level.
**Statistically significant at the 5 percent level.

years. But after ten years, our estimate of the growth
effect falls to 0.17 percentage point and is no longer statistically significant. In the gross state product series,
however, the increases in growth appear to last beyond
ten years. (See Box 2 for a detailed discussion of the
growth regressions used to generate these results.)

Annual personal income grows about
0.51 percentage point faster after branching
deregulation, and gross state product, about
0.69 percentage point faster.

Overall, we lack conclusive evidence on whether the
growth effects persist beyond ten years. This limitation is not
surprising, however, since we observe only about ten years of
growth experience after deregulation for most states. Nevertheless, even if the observed increases in growth do not continue indefinitely, the short-run effects appear to be large.28

ROBUSTNESS OF THE GROWTH ACCELERATION
Did many states experience a growth pickup in the wake of
branching deregulation or was the change concentrated
among a few? To evaluate whether the effects were widespread, we offer a state-by-state assessment of the growth
in personal income. Chart 3 plots the average change in
growth for each of the thirty-five states that deregulated
their branching restrictions relative to the average change
in growth for the nonderegulating states. (The latter group
of states, as in Chart 2, is used to control for nationwide
changes in growth.) Like Chart 2, Chart 3 plots these
growth changes by the year of deregulation.
The growth acceleration following deregulation is
clearly a general phenomenon. Twenty-nine of the thirtyfive states that deregulated performed better than the nonderegulators. (The exceptions are New Hampshire, Florida,
Michigan, Kansas, Colorado, and New Mexico.) Even when
the deregulating states experienced growth declines following

FRBNY ECONOMIC POLICY REVIEW / DECEMBER 1997

23

branching, the nonderegulators generally fared even worse.
This pattern suggests that when a downturn was occurring in the national business cycle at the time of branching deregulation, the downturn was at least partly offset
by the positive effects of statewide branching.
We have shown that rates of economic growth
increased following branching deregulation. The increase is
both statistically large, which suggests that we can be confident that it is not the result of chance, and economically
large, which suggests that over time economic welfare
would be raised dramatically as a consequence of the
accelerated growth. The growth acceleration is also widespread, benefiting twenty-nine of the thirty-five deregulating
states. The remaining question, however, is whether
deregulation actually caused the growth pickup. Establishing causal relationships is always difficult in empirical
economics because researchers cannot run controlled

Chart 3

Personal Income Growth Rates Accelerate
after Branching Deregulation in All but Six States
Change in growth rates
0.04
NY

0.03

NJ
HI

0.02
CT

ME

0.01
VA

0.00

OH

AL
UT

PA

TN
GA OR

-0.01

WY
WV

LA
OK
IN

MT
KY

MS
WI
ND
MO
NE
MA
TX
KS
NM
WA
IL
MI
CO
FL

-0.02
NH

-0.03
75

80
85
90
Year of branching deregulation

Source: Authors’ calculations.

BOX 2: AN EMPIRICAL MODEL OF GROWTH
To estimate the effects of branching deregulation on growth,
we use the following model:

We estimate the model with a variety of different
specifications. The simplest uses ordinary least squares

⁄ Yt – 1 , i = αt + γ 5 D t5, i + γ 10 D t10, i + γ 10 Dt10
,i
+ µ 1 [ Yt – 1 , i ⁄ Yt – 2 , i ] + µ 2 [ Yt – 2 , i ⁄ Yt – 3 , i]
+ µ 3 [ Yt – 3 , i ⁄ Yt – 4 , i ] + δYt – 1 , i + ε t, i ,

(OLS). The model is also estimated by weighted least squares

where Yt,i is a measure of real per capita income (output),
D t5, i is a branching indicator equal to 1 for states that allowed
statewide branching at most five years ago, D t10, i is a branch-

data relating to interstate commerce—is likely to be greater

+

Y
t, i

+

ing indicator equal to 1 for states that allowed statewide
+

branching six to ten years ago, and D t10, i is a branching indicator
equal to 1 for states that allowed statewide branching more
than ten years ago.

(WLS), with weights proportional to the size of the state
economy at the beginning of the period. We use WLS because
measurement error in state economic data—particularly in
for smaller states. Smaller states are also more likely to depend
on a limited number of industries, leading to greater susceptibility to industry-specific shocks. In all cases we
report heteroskedasticity-consistent standard errors (White 1980).
While there is no a priori reason to suspect that
regional business cycles will introduce a bias, we also present

In this specification, the γ coefficients measure the

estimates from an augmented version of the above model

increase in per capita economic growth stemming from
branching deregulation at different time periods. The αt

allowing the time effects (that is, the business cycle effects)

terms measure the common, economy-wide shocks to growth
such as the national business cycle. The µ terms capture the
effects of the state-specific business cycle, and δ reflects the

to vary across four broad regions of the United States. This
specification is included mainly as a robustness check. Table 1 in
the text shows that many states in the South and Midwest

extent to which poorer states grow faster (the “convergence

deregulated around the same time, leading to the possibility
that regional business cycle effects drive the estimate of the

effect” observed in Barro and Sala-I-Martin [1992]).

growth effect coefficients. To control for the regional business

24

FRBNY ECONOMIC POLICY REVIEW / DECEMBER 1997

95

BOX 2: AN EMPIRICAL MODEL OF GROWTH (Continued)
cycle, we modified the above model slightly by interacting

these models. Almost all specifications show that the increase in

the year-fixed effect with four regional indicator variables (for

growth after branching deregulation lasts up to ten years, but

the Northeast, South, West, and Midwest).

only half the models show a growth increase beyond ten years.

The table below presents the results of estimating

STATE ECONOMIES GROW MORE RAPIDLY AFTER BRANCHING DEREGULATION
Growth Effect:
Years 6-10
(2)

Growth
Effect:
Years 10+
(3)

Growtht-1
(4)

GROWTH BASED ON PERSONAL INCOME
Basic model, OLS
0.59**
(0.23)
Basic model, WLS
0.61**
(0.21)
Regional effects, OLS
0.35
(0.20)
Regional effects,WLS
0.31**
(0.16)

0.86**
(0.23)
0.86**
(0.22)
0.37*
(0.20)
0.38**
(0.19)

0.34
(0.22)
0.34**
(0.16)
0.17
(0.19)
0.21
(0.13)

0.14*
(0.08)
0.20**
(0.05)
0.08
(0.08)
0.16**
(0.05)

GROWTH BASED ON GROSS STATE PRODUCT
Basic model, OLS
0.77**
(0.30)
Basic model, WLS
0.64**
(0.26)
Regional effects, OLS
0.60**
(0.29)
Regional effects, WLS
0.43**
(0.21)

0.94**
(0.30)
0.83**
(0.33)
0.65**
(0.27)
0.57**
(0.24)

0.63**
(0.27)
0.48*
(0.26)
0.67**
(0.27)
0.59**
(0.24)

0.21**
(0.06)
0.21**
(0.05)
0.15**
(0.06)
0.23**
(0.04)

Growth Effect:
Years 1-5
(1)

Growtht-3
(6)

Lag of Per
Capita Income
(7)

Adjusted R2
(8)

-0.03
(0.06)
0.06
(0.04)
-0.03
(0.07)
0.04
(0.04)

-0.04
(0.08)
0.04
(0.04)
0.02
(0.08)
0.07
(0.05)

-0.38**
(0.13)
-0.29**
(0.08)
-0.29**
(0.11)
-0.28**
(0.09)

0.52%
(1,015)
0.73%
(1,015)
0.64%
(974)
0.79%
(974)

0.09*
(0.05)
0.13**
(0.06)
0.06
(0.05)
0.11**
(0.04)

0.03
(0.07)
0.06
(0.07)
0.07
(0.07)
0.08
(0.07)

-0.07**
(0.03)
-0.09**
(0.03)
-0.04*
(0.02)
-0.08**
(0.03)

0.41%
(521)
0.62%
(521)
0.50%
(500)
0.69%
(500)

Growtht-2
(5)

Source: Jayaratne and Strahan (1996), Table 5.
Notes: The table presents estimates of the increase in state economic growth following relaxation of intrastate branching restrictions. Delaware is dropped from all
regressions used to produce these estimates while Alaska and Hawaii are dropped from the regressions with regional effects. In addition, the year in which each state
deregulated was dropped. Growth data for personal income are from 1972-92 and for state product from 1981-91 (three years are lost with the addition of the
lagged dependent variables). In column 8, the number of observations appears in parentheses below the R2. In columns 1-7, standard errors appear in parentheses
below the coefficients. Reported standard errors are heteroskedasticity-consistent (White 1980). The coefficients on the branching indicators and the lag of income
are multiplied by 100.
*Statistically significant at the 10 percent level.
**Statistically significant at the 5 percent level.

experiments. Nevertheless, we must consider other factors that could explain our finding. One possibility is that
state governments instituted a variety of new policies at
the same time that they deregulated their banking systems. If so, these policy changes could be responsible for
the improved growth performance.
We find no evidence of such coincident policy
changes. The political control of state governments did not

change significantly around the time of branching deregulation. In only two cases out of thirty-five did control of
both houses of the state legislature and the governorship
pass from one political party to the other during the fouryear election cycle leading up to branching deregulation.
The political affiliation of both houses of the state legislature
changed only six times out of thirty-five during the fouryear window before branching deregulation.

FRBNY ECONOMIC POLICY REVIEW / DECEMBER 1997

25

Moreover, even after controlling for two measures of
state fiscal policy—the ratio of public investment by the state
government to total income and the ratio of tax receipts by the
state government to total income—we continue to find a significant growth acceleration after branching deregulation. Our
tests suggest that there were no changes in states’ tax and
other fiscal policies that coincided with branching deregulation and that could explain the observed increase in state economic growth following statewide branching.
Another possible explanation for our finding is
that state legislatures relaxed branching restrictions in
anticipation of faster growth and the need to finance attractive
projects. Why might this be the case? Perhaps when a state
has strong growth prospects, potential bank borrowers pressure
state governments to deregulate their banking systems. But if
states deregulated branching rules because they anticipated
the need to finance a future economic boom, then we should
see a sharp rise in bank lending following deregulation.
Jayaratne and Strahan (1996) demonstrate, however, that
no increase in lending occurred. Moreover, the growth
effects of branching deregulation remain largely
unchanged even after we control for loan growth.
Finally, we consider the possibility that some
unobserved set of technological changes led to branching
deregulation, improved bank performance, and increased
economic growth. For example, increased competition
from nonbank financial institutions clearly helped to spur
the removal of barriers to branching. Perhaps such financial
innovations also forced banks to improve their performance
and boosted states’ economic growth. Two considerations,
however, lead us to discount this possibility. First, if this
explanation were true, we would see an improvement in
bank performance and increased economic growth
immediately before, as well as after, deregulation. Our
data show no such pattern.29 Second, any technological
changes that occurred around the time of deregulation
should have affected all states. In that case, we should
not see any improvement in bank performance nor any

26

FRBNY ECONOMIC POLICY REVIEW / DECEMBER 1997

increase in economic growth in deregulating states relative
to nonderegulating states. Our data, of course, provided clear
evidence of such differences in the experiences of the states.
To summarize, the large increase in bank loan
quality in conjunction with little or no change in loan
growth suggests that the increase in states’ economic
growth was at least partly due to statewide branching. The
improvements in banking stemming from selection (and
possibly disciplining) appear to have had important
beneficial effects on the economy.

CONCLUSION
Restrictions on bank branching have proved to be very
costly. By preventing the more efficient banks from
expanding at the expense of their less efficient rivals, these
restrictions retarded the “natural” evolution of the industry.
As our analysis has shown, once state branching restrictions
were lifted, the efficiency of the banking system improved
as the better banks expanded into new markets. Bank
borrowers benefited from lower loan rates, while the overall economy grew faster as banks did a better job separating
the good projects from the bad and monitoring firms after
lending relationships had been established. State restrictions
on interstate banking may have created similar constraints,
although our statistical procedure has a harder time
identifying such effects.
The Riegle-Neal Act removes the remaining geographic barriers to bank expansion and permits the creation
of multistate banking franchises. This federal legislation
may produce benefits similar to those achieved through
state deregulation—reduced bank costs, lower loan rates,
and accelerated economic growth. Nevertheless, it is possible
that the latitude given banks to create branches and buy
out-of-state banks over the last two decades may have
already weeded out weaker institutions and exhausted the
benefits of geographic deregulation. Whether there is
additional room for improved efficiency through the
process of selection remains to be seen.

ENDNOTES

1. Although the act gives each state the right to prevent out-of-state banks
from owning branches there, only Texas and Montana have chosen to do so.
2. Several types of geographic restrictions have been imposed over the
years on banks, but this article focuses on limits on banks’ ability to
establish branches within their home states and on limits on BHCs’
ability to acquire banks outside their home states. We do not consider
other restrictions, such as those prohibiting the formation of multibank
BHCs, primarily because we lack the necessary data.
3. Although some states removed barriers to branching before 1978 (see
Table l), most of the state deregulatory activity was concentrated in the
1978-92 period. The focus on this period also enables us to take
advantage of the greater availability of bank data after 1978.
4. We include Delaware and South Dakota in Table 1, but we exclude
them from our analysis (see Box 1).
5. Many states also permitted de novo branching after permitting banks
to branch through mergers and acquisitions. We do not emphasize de novo
branching powers because bank expansion into new markets generally
occurs through the purchase of whole banks or branches of banks located
in those new markets, not through the opening of new branches.
6. Information on the timing of states’ deregulatory initiatives is taken
from Amel (1993).

10. The long-run average level of bank loan losses may differ across states
because banks operating in states dominated by particularly high-risk
industries will exhibit higher loan losses. Oil states such as Texas, Alaska, and
Louisiana, for instance, exhibited loan losses that exceeded the national
average during our sample period. Improvements in loan quality after
deregulation could therefore reflect a tendency for states dominated by highrisk industries to deregulate their branching and interstate banking
restrictions later than the typical state. We accounted for this possibility by
controlling for persistent cross-state differences in bank performance.
11. We find declines in loan loss provisions and nonperforming loans of
similar magnitude following branching deregulation. See Jayaratne and
Strahan (forthcoming).
12. We find no change in deposit interest rates following deregulation,
however. All of the cost declines seem to be passed along to bank
borrowers rather than depositors.
13. The estimates of the effects of deregulation on our performance
measures are based on a regression model that assumes that the changes
occur immediately following deregulation and are permanent. Because
we have only five to ten years of experience after deregulation for most
states, we cannot be sure that these effects will continue indefinitely.
Nevertheless, we find that the observed improvements in bank
performance persist more than five years after branching deregulation.
14. These results are reported in Jayaratne and Strahan (forthcoming).

7. For instance, in 1982 Pennsylvania passed a law permitting banks to
branch in the home office county, in a contiguous county, in a
bicontiguous county, or in the counties of Allegheny, Delaware,
Montgomery, and Philadelphia. In 1990, Pennsylvania permitted
unrestricted branching statewide. In the results presented below, we
assume that by 1982 Pennsylvania permitted intrastate branching
(despite the fact that the process was not finished until eight years later)
because the effect of the 1982 law brought Pennsylvania so close to
complete intrastate branch freedom. We follow a similar practice for
Ohio, Virginia, and Washington. Our results are not sensitive to the
alternative dating of deregulation in these four states.
8. The noninterest expense variable equals total noninterest expenses
incurred by all banks in a state divided by total banking assets held by
banks in that state. The loan price variable equals interest earned on all
loans and leases in a state divided by total loans plus leases held on bank
balance sheets in that state.
9. When we control for the state business cycle, the estimated effects of
statewide branching decrease but are still both statistically significant
and economically important.

NOTES

15. New York and Maine are dropped from this analysis because they
deregulated before loan charge-off data became available. As noted
earlier, Delaware and South Dakota are dropped throughout the analysis.
16. New Hampshire eliminated its branching restrictions in 1987, just
before the beginning of the New England banking crisis. This sequence of
events might explain why bank performance is observed to deteriorate after
deregulation.
17. These results are available on request.
18. When we substitute loan charge-offs for return on equity as a measure of
bank quality, we obtain similar results. To conserve space, however, we do
not include these results in this article. In addition, we do not include
noninterest expenses in this analysis, because the data are available beginning
only in 1984. The lack of earlier data means that we can conduct the exercise
in Table 2 for only three deregulating states using noninterest expense data.
19. We chose this window length because most of the observed changes
in bank structure occurred within five years after branching deregulation.

FRBNY ECONOMIC POLICY REVIEW / DECEMBER 1997

27

ENDNOTES (Continued)

Note 19 continued
For example, nearly two-thirds of the 30 percent increase in the statelevel bank asset concentration occurred within five years after branching
deregulation. Similar results are reported in Berger, Kashyap, and Scalise
(1995), who find that most changes to bank structure occur within five
years after geographic deregulation. (Some states entered interstate
banking agreements during the five-year window. For these states, we
use the year just prior to the year in which the state entered the interstate
banking agreement as the end of the window. We dropped four states—
West Virginia, Tennessee, Oregon, and New Hampshire—that entered
interstate banking agreements in the same year or one year after
branching was deregulated.)
20. Although high-profit banks are defined to have 50 percent of a
state’s bank assets at the beginning of the deregulation period, we can
only approximate this target because no group of banks in a state will
contain exactly one half of that state’s total bank assets. Thus, in Table 2,
high-profit banks are shown to have 51.3 percent of the average state’s
bank assets, not 50 percent.
21. We define high-profit banking companies before deregulation in
much the same way we defined high-profit banking companies after
deregulation. Banking companies are identified as high-profit on the
basis of their return on equity at the end of the year seven years before the
year of deregulation. We then measure their change in market share over
the next six years.
22. Recall that we found only weak evidence that overall bank profits
increased after branching deregulation. This earlier finding does not
conflict with the fact that high-profit banks grew faster than low-profit
banks. Two forces are operating. Because the high-profit banks tend to
grow at the expense of their less efficient competitors after deregulation,
aggregate profits should increase, all else being equal. At the same time,
however, because the high-profit banks are likely to have achieved their
superior growth rates in part by charging customers less, aggregate
profits should drop. These two forces are approximately offsetting: thus,
overall profits changed little following deregulation.

writing and exercising of such covenants allow banks to monitor their
borrowers effectively (Morgan 1995).
24. We focus here on branching deregulation, rather than interstate
banking, because once we controlled for the business cycle, we found
sharp improvements in bank performance associated with statewide
branching but not with interstate banking. Although we looked for
evidence of changes in economic growth associated with interstate
banking, we found none.
25. Statistics on personal income and gross state product are published
annually by the U.S. Department of Commerce. Annual state population
figures are from the U.S. Bureau of the Census. We convert nominal
personal income to constant dollars using a national price deflator, the
consumer price index.
26. The difference between personal income and gross state product is
apparent in how the two measures treat capital income. Capital income
is allocated to personal income according to the state of residence of the
owner of capital, while for gross state product, capital income is allocated
according to the physical location of the capital itself. Real per capita
personal income grew 1.6 percent per year during our analysis period
(1972-92), while gross state product grew 1.4 percent per year between
1978 and 1992. (Because the Commerce Department changed the base
year for the industry price deflators in 1977, we could not construct a
consistent growth series prior to 1978 using gross state product.)
27. To control for regional business cycle effects, we include a set of time
dummy variables that vary across four broad regions. For details, see
Jayaratne and Strahan (1996), Table 2.
28. Note that there are theoretical reasons to believe that reductions in
financial market frictions can increase the steady-state growth rate of the
economy. For a survey of the relevant models, see Galetovic (1994) and
Pagano (1993).
29. These results are available from the authors upon request.

23. For instance, banks write loan covenants that restrict firms’ ability
to engage in certain activities during periods of financial distress. The

The views expressed in this article are those of the authors and do not necessarily reflect the position of the Federal Reserve
Bank of New York or the Federal Reserve System. The Federal Reserve Bank of New York provides no warranty, express or
implied, as to the accuracy, timeliness, completeness, merchantability, or fitness for any particular purpose of any information
contained in documents produced and provided by the Federal Reserve Bank of New York in any form or manner whatsoever.

28

FRBNY ECONOMIC POLICY REVIEW / DECEMBER 1997

NOTES

REFERENCES

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

Jayaratne, Jith, and Philip E. Strahan. 1996. “The Finance-Growth Nexus:
Evidence from Bank Branch Deregulation.” QUARTERLY JOURNAL OF
ECONOMICS 111: 639-70.

Amel, Dean, and Nellie Liang. 1992.“The Relationship between Entry into
Banking Markets and Changes in Legal Restrictions on Entry.”
ANTITRUST BULLETIN 37: 631-49.

———. Forthcoming. “Entry Restrictions, Industry Evolution and
Dynamic Efficiency: Evidence from Commercial Banking.” JOURNAL
OF LAW AND ECONOMICS.

Barro, Robert, and Xavier Sala-I-Martin. 1992. “Convergence.” JOURNAL
OF POLITICAL ECONOMY 100: 223-51.

Kane, Edward. 1996. “De Jure Interstate Banking: Why Only Now?”
JOURNAL OF MONEY, CREDIT, AND BANKING 28 (May): 141-61.

Berger, Allen N., William C. Hunter, and Stephen G. Timme. 1993. “The
Efficiency of Financial Institutions: A Review and Preview of Research
Past, Present and Future.” JOURNAL OF BANKING AND FINANCE 17:
221-49.

Keeley, Michael. 1990. “Deposit Insurance, Risk and Market Power in
Banking.” AMERICAN ECONOMIC REVIEW 80: 1183-200.

Berger, Allen N., Anil K Kashyap, and Joseph M. Scalise. 1995. “The
Transformation of the U.S. Banking Industry: What a Long Strange Trip
It’s Been.” BROOKINGS PAPERS ON ECONOMIC ACTIVITY, no. 2: 55-218.
Calem, Paul. 1994. “The Impact of Geographic Deregulation on Small
Banks.” Federal Reserve Bank of Philadelphia BUSINESS REVIEW,
December.
Calomiris, Charles. 1993. “Regulation, Industrial Structure, and
Instability in U.S. Banking: An Historical Perspective.” In Michael
Klausner and Lawrence White, eds., STRUCTURAL CHANGE IN
BANKING, 19-116. New York: New York University.
Chapman, John, and Ray Westerfield. 1942. BRANCH BANKING. New
York: Harper and Brothers.
Diamond, Douglas. 1984. “Financial Intermediation and Delegated
Monitoring.” REVIEW OF ECONOMIC STUDIES 51: 393-414.
Economides, Nicholas, R. Glenn Hubbard, and Darius Palia. 1995. “The
Political Economy of Branching Restrictions and Deposit Insurance:
A Model of Monopolistic Competition among Small and Large
Banks.” NBER Working Paper no. 5210.
Galetovic, Alexander. 1994. “Finance and Growth: A Synthesis and
Interpretation of the Evidence.” Unpublished paper, Princeton University.
Hubbard, R. Glenn, and Darius Palia. 1995. “Executive Pay and
Performance: Evidence from the U.S. Banking Industry.” JOURNAL OF
FINANCIAL ECONOMICS 39: 105-30.

NOTES

King, Robert, and Ross Levine. 1993. “Finance and Growth: Schumpeter
Might Be Right.” QUARTERLY JOURNAL OF ECONOMICS 108: 717-38.
Kroszner, Randall S., and Philip E. Strahan. 1997. “The Political Economy
of Deregulation: Evidence from the Relaxation of Bank Branching
Restrictions in the United States.” Federal Reserve Bank of New York
Research Paper no. 9720, June.
McLaughlin, Susan. 1995. “The Impact of Interstate Banking and
Branching Reform: Evidence from the States.” Federal Reserve Bank
of New York CURRENT ISSUES IN ECONOMICS AND FINANCE 1, no. 2.
Morgan, Donald. 1995. “Banks as Monitors: New Evidence Using the
Financial Covenants in Bank Loan Commitments.” Columbia
University Business School Working Paper no. PW-95-12.
Pagano, Marco. 1993. “Financial Markets and Growth: An Overview.”
EUROPEAN ECONOMIC REVIEW 37: 613-22.
Rhoades, Stephen. 1996. “Bank Mergers and Industrywide Structure,
1980-94.” Board of Governors of the Federal Reserve System Staff
Study no. 169.
Savage, Donald. 1993. “Interstate Banking: A Status Report.” FEDERAL
RESERVE BULLETIN, December: 1075-89.
Sylla, Richard, John Legler, and John Wallis. 1987. “Banks and State
Public Finance in the New Republic: The United States, 1790-1860.”
JOURNAL OF ECONOMIC HISTORY 47: 391-403.
White, Halbert. 1980. “A Heteroskedasticity-consistent Covariance Matrix
Estimator and a Direct Test for Heteroskedasticity.” E CONOMETRICA 48:
817-30.

FRBNY ECONOMIC POLICY REVIEW / DECEMBER 1997

29

What Moves the Bond Market?
Michael J. Fleming and Eli M. Remolona*

T

o what extent can movements in the financial
markets be attributed to the arrival of new
information? In a landmark 1989 study of
the stock market, David Cutler, James
Poterba, and Lawrence Summers found that it was surprisingly difficult to identify information that could account
for the largest price movements. No similar effort has been
made, however, to explain the largest price movements in
the bond market, although both theory and a large literature on announcement effects suggest that the results for
this market should be more promising.
In this article, we take a close look at a single year
in the U.S. Treasury securities market (which we refer to as
the bond market) and attempt to identify information that
may account for the sharpest price changes and the most
active trading episodes. Sharp price moves may be attributed to changes in expectations shared by investors, and

*Michael J. Fleming is an economist and Eli M. Remolona a research
officer at the Federal Reserve Bank of New York.

surges in trading activity to a lack of consensus on prices. 1
To explain the price changes and trading surges, we examine how closely these events correlate with the release times
of macroeconomic announcements.
We also investigate whether the bond market’s
behavior is related to factors affecting the informational
value of the announcements—specifically, the type of
announcement and the magnitude of the surprise in the data
released. While other studies have examined announcement
effects in the bond market, our use of high-frequency market
data and precise announcement release times allows us to
identify such effects more precisely than most earlier studies.
In addition, our analysis of the role of uncertainty in assessing the impact of macroeconomic announcements goes
beyond the scope of earlier bond market studies. To represent
the bond market in our analysis, we focus on the five-year
U.S. Treasury note, one of the most actively traded U.S. Treasury securities.
For the period examined—August 23, 1993, to
August 19, 1994—we find that each of the twenty-five

FRBNY ECONOMIC POLICY REVIEW / DECEMBER 1997

31

sharpest price changes and each of the twenty-five greatest
trading surges can be associated with a just-released
announcement. We also show that the market differentiates
among announcements containing different information,
with the employment, producer price index (PPI), federal
(fed) funds target rate, and consumer price index (CPI)
announcements eliciting the most pronounced responses in
terms of both price movements and trading activity. In addition, our precise data allow us to document for the first time
a significant market impact from U.S. Treasury security auction results. Finally, we demonstrate that the market’s reactions depend on the surprise component of a given
announcement and on conditions of market uncertainty.

The apparently weak informational effects found in
the stock market are not entirely surprising. Much of the
observable information likely to be relevant to the stock
market as a whole takes the form of macroeconomic
announcements. The theoretical effects of such announcements are often ambiguous for stocks, but not for bonds. The

Theory says that movements in financial asset
prices should reflect new information about
fundamental asset values. In the case of the
stock market, however, such theory has been

PREVIOUS STUDIES
The literature on announcement effects in the stock and
bond markets is quite extensive. Our review of this literature
serves two purposes: it pulls many of the different strands of
the literature together for the first time and it suggests the
extent to which our empirical results—based on a one-year
sample—can be generalized to other periods.

STOCK MARKET STUDIES
Theory says that movements in financial asset prices
should reflect new information about fundamental asset
values. In the case of the stock market, however, such
theory has been difficult to confirm. Most notably, in an
analysis of the fifty largest one-day price moves in the
Standard and Poor’s Composite Stock Index since 1946,
Cutler, Poterba, and Summers (1989) find that in most
cases the information cited by the press as causing the
market move “is not particularly important.” In earlier
studies, Schwert (1981), Pearce and Roley (1985), and
Hardouvelis (1987) find little evidence that the stock
market responds to macroeconomic news other than
monetary information (such as money supply and discount rate announcements). More recently, McQueen
and Roley (1993) find a stronger relationship between
stock prices and news after controlling for different
stages of the business cycle. Even with their best effort,
however, McQueen and Roley are able to explain only
3.9 percent of the daily variation in the S&P 500 Index.

32

FRBNY ECONOMIC POLICY REVIEW / DECEMBER 1997

difficult to confirm.

reason is that stock prices depend on both cash flows and the
discount rate, while bond prices—for which cash flows are
fixed in nominal terms—depend only on the discount rate.
An upward revision of expected real activity, for example,
raises the discount rate for both stocks and bonds, which
would reduce prices. At the same time, however, the revision
raises expected cash flows for stocks, an outcome that
increases stock prices. The net effect on bond prices of such
an announcement is clearly negative, but the net effect on
stock prices will depend on whether the cash flow effect or
the discount rate effect dominates.

BOND MARKET STUDIES
Earlier findings on announcement effects in the bond
market suggest that it will be easier to relate this market’s
movements to information arrival.2 Indeed, studies over
the years have documented a significant bond market
impact from numerous macroeconomic announcements,
including money supply, industrial production, PPI, CPI,
unemployment rate, and nonfarm payroll employment
numbers (Table 1). Market movements in these studies
are typically based on daily interest rates, and announcements are measured by the extent of the surprise each
entails—that is, the difference between the forecast and

the actual number released. Forecasts are either derived
by the studies’ authors from the time series of the variables
or generated by the market analysis firm MMS International Inc. from surveys conducted a few days before the
announcements.

Table 1
STUDIES FINDING THAT MACROECONOMIC
ANNOUNCEMENTS SIGNIFICANTLY AFFECT
INTEREST RATES
Announcement
Money supply

Study Author
Berkman (1978)
Grossman (1981)
Urich and Wachtel (1981)

Thornton (1989)
Strongin and Tarhan (1990)
McQueen and Roley (1993)

Sample Period
Jul. 1975 - Jun. 1977
Sep. 1977 - Sep. 1979
Jan. 1974 - Dec. 1977
Jan. 1979 - Sep. 1979
Oct. 1979 - Dec. 1981
Sep. 1977 - Nov. 1981
Sep. 1977 - Oct. 1982
Sep. 1977 - Oct. 1982
Nov. 1977 - Jul. 1982
Oct. 1979 - Oct. 1982
Oct. 1979 - Aug. 1984
Feb. 1980 - Dec. 1981
Jan. 1983 - Dec. 1983
Jan. 1978 - Jan. 1984
May 1980 - Jan. 1984
Sep. 1977 - May 1988

Roley and Troll (1983)
Harvey and Huang (1993)
McQueen and Roley (1993)
Edison (1996)

Sep. 1977 - Oct. 1979
Dec. 1981 - Apr. 1988
Sep. 1977 - May 1988
Feb. 1980 - Feb. 1995

Urich and Wachtel (1984)
Smirlock (1986)
Hardouvelis (1988)
Dwyer and Hafer (1989)
McQueen and Roley (1993)
Edison (1996)

Oct. 1979 - Jul. 1982
Oct. 1979 - Dec. 1983
Oct. 1979 - Aug. 1984
Feb. 1980 - Dec. 1980
Sep. 1977 - May 1988
Feb. 1980 - Feb. 1995

Smirlock (1986)
Hardouvelis (1988)
McQueen and Roley (1993)
Edison (1996)

Oct. 1979 - Dec. 1983
Oct. 1982 - Aug. 1984
Sep. 1977 - May 1988
Feb. 1980 - Feb. 1995

Cornell (1982, 1983)
Roley (1982)
Roley (1983)
Roley and Troll (1983)
Urich and Wachtel (1984)
Roley and Walsh (1985)
Hardouvelis (1988)
Dwyer and Hafer (1989)

Industrial production

Producer price index

Consumer price index

Durable goods orders

Hardouvelis (1988)

Oct. 1982 - Aug. 1984

Retail sales

Hardouvelis (1988)
Edison (1996)

Oct. 1982 - Aug. 1984
Feb. 1980 - Feb. 1995

Unemployment rate

Hardouvelis (1988)
Cook and Korn (1991)
McQueen and Roley (1993)
Prag (1994)
Edison (1996)

Oct. 1982 - Aug. 1984
Feb. 1985 - Apr. 1991
Sep. 1977 - May 1988
Jan. 1980 - Jun. 1991
Feb. 1980 - Feb. 1995

Nonfarm payroll
employment

Cook and Korn (1991)
McQueen and Roley (1993)
Edison (1996)
Krueger (1996)

Feb. 1985 - Apr. 1991
Sep. 1977 - May 1988
Feb. 1980 - Feb. 1995
Feb. 1979 - Apr. 1996

Notes: The table lists those studies that have found a statistically significant
relationship between the surprise component of an announcement and U.S.
interest rates. For studies that examine the impact on several interest rates,
we consider only the results for the longest maturity rate. Studies are not
listed in which the impact of an announcement is found to have a sign opposite to
that
predicted.

The literature provides evidence of a “flavor-ofthe-month” aspect to the bond market’s behavior, in
which different announcements are regarded as important
in different periods. Starting with Berkman (1978), studies
from the late 1970s to the mid-1980s document a significant impact of money supply announcements. However,
Dwyer and Hafer (1989) show a diminishing significance
for such announcements in the mid-1980s. Studies in the
1980s, such as Urich and Wachtel (1984) and Smirlock
(1986), begin to demonstrate the importance of the PPI,
CPI, and unemployment rate announcements. More
recent studies, particularly Cook and Korn (1991) and
Krueger (1996), establish the ascendant importance of
the nonfarm payrolls number in the Bureau of Labor
Statistics’ (BLS) employment report.
It is noteworthy that the bond market studies
that consider several announcements tend to find that
relatively few of them have significant effects on the

Earlier findings on announcement effects in
the bond market suggest that it will be easier
to relate this market’s movements to
information arrival.

market. 3 One possible reason for this finding is that
the daily interest rate data on which these studies rely
are not of sufficiently high frequency to capture the
market’s reaction cleanly. As Hardouvelis (1988)
points out, researchers ought to measure the market
change from just before to just after the announcement. Another possible reason for the lack of significance is that the effect of a given announcement
surprise may vary even over short periods of time,
depending on what else is going on in the economy.
Prag (1994), for example, shows that the effect of
unemployment rate announcements on interest rates
depends on the existing level of unemployment.

FRBNY ECONOMIC POLICY REVIEW / DECEMBER 1997

33

BOND MARKET STUDIES USING INTRADAY DATA
The recent availability of high-frequency intraday price
data has increased the power of researchers’ efforts to
estimate announcement effects. Ederington and Lee
(1993), for instance, use such data on Treasury bond
futures to examine the impact of monthly economic
announcements. They find that nine out of sixteen
announcements have significant price effects, with the
greatest impact coming from the employment, PPI,
CPI, and durable goods orders releases. More recently,
Fleming and Remolona (1997) analyze intraday cash
market Treasury securities data and find that eight out
of nineteen announcements have a significant impact on
price and eleven out of nineteen have a significant
impact on trading volume. Instead of measuring surprise components, both studies rely on dummy variables
for announcement days to isolate the announcements’
effects. They therefore measure the average impact of
the announcements without regard for the particular
numbers released in any given report.
If an announcement’s impact depends only on
the unexpected part of the released information, then
accounting for the sign and magnitude of the unexpected
component should improve the estimates of announcement
effects. Nonetheless, intraday studies relying on such
surprises do not identify more significant announcements
than do studies relying only on announcement dummy
variables. For example, Becker, Finnerty, and Kopecky
(1996) find that nonfarm payroll employment and CPI
surprises affect the fifteen-minute returns on bond futures
significantly, while housing starts and merchandise trade
surprises do not. In addition, Balduzzi, Elton, and Green
(1996) conclude that surprises from only six of twentythree monthly announcements have a significant price
impact on the ten-year U.S. Treasury note.

in weeks of earnings announcements reflect “changes in
the expectations of the market as a whole” while surges in
trading activity reflect “a lack of consensus regarding the
price.” Morse (1981) provides evidence that earnings
announcements affect daily trading volume, but Jain
(1988) finds that macroeconomic news has no effect on
hourly trading volume. Moreover, Woodruff and Senchack
(1988) find that the effects of earnings announcements on
prices and trading volume depend on the magnitude of
the surprises.
As hypothesized by Beaver (1968), an increase in
trading activity after announcements may largely
reflect differences of opinion among market participants.4
Other literature on trading activity has focused on the idea
that both price changes and trading activity reflect the
arrival of private information.5 The conveyance of private
information through trading is probably not that important in the bond market, however, since much of the
information relevant to the market is released to the public
through scheduled announcements. An explanation for
changes in trading activity that is more pertinent to the
bond market is that investors with duration targets or
dynamic hedging strategies rebalance their portfolios after
price changes.6
In summary, macroeconomic announcements cannot account for the largest price moves in the stock market
and, in fact, are typically found to have an insignificant
impact on stock prices. In contrast, numerous studies find
a significant impact on bond prices, although no study
prior to this one has explicitly tried to account for the
largest price movements. As for the effects of announcements on trading activity, differences of opinion among
traders or portfolio rebalancing might lead to a surge in
trading activity after a release, but studies have been
limited largely to the stock market and the results so far
have been mixed.

STUDIES OF TRADING ACTIVITY
Much of the research on trading activity has been limited
to the stock market, with the early literature focusing on
the difference between the effects of earnings announcements on prices and the effects on trading activity. Beaver
(1968) argues, for example, that stock price movements

34

FRBNY ECONOMIC POLICY REVIEW / DECEMBER 1997

METHODOLOGY AND DATA
ANALYTICAL APPROACHES
Our analysis of the U.S. Treasury securities market combines
the different approaches offered by the literature on announce-

ment effects. First, we follow Cutler, Poterba, and Summers
(1989) in examining the largest price changes and determining the extent to which these changes coincide with the
release times of announcements. Second, like Ederington and
Lee (1993), we run dummy-variable regressions to measure

We employ high-frequency price and trading
data from the U.S. Treasury securities market, as
well as data on the dates and exact release times
of various macroeconomic announcements. These
data allow us to correlate market movements
closely with information releases and to identify
the market impact of announcements precisely.

the extent to which the market systematically differentiates
among the different types of announcements to reflect the
inherent differences in the information released. Third, we
follow Becker, Finnerty, and Kopecky (1996) and other studies in investigating whether measured surprises in the
announcements help explain the market’s responses. Finally,
following McQueen and Roley (1993), we analyze the possible effects of market conditions on the impact of a given
announcement surprise.
In applying each of these approaches, we employ
high-frequency price and trading data from the U.S. Treasury securities market, as well as data on the dates and exact
release times of various macroeconomic announcements.
These data allow us to correlate market movements closely
with information releases and to identify the market impact
of announcements precisely. In addition, we utilize data on
the market’s expectations for each announcement in our
analyses of the effects of announcement surprises. Finally, we
depend on quantitative measures of uncertainty for our analysis of the impact of market conditions. The specific data we
use are described in detail in the rest of this section.

U.S. TREASURY SECURITIES DATA
Our U.S. Treasury securities data cover one year of tick-bytick trading activity in the interdealer broker market.
Our data source is GovPX, Inc., a joint venture set up
by the primary dealers and interdealer brokers in 1991
to improve the public’s access to U.S. Treasury securities
prices (Wall Street Journal 1991). GovPX consolidates
and posts real-time quote and transaction data from five
of the six major interdealer brokers, which together
account for roughly two-thirds of the interdealer broker
market. Posted data include the best bids and offers,
trade prices and sizes, and the aggregate volume of trading for all Treasury bills, notes, and bonds. GovPX data
are distributed electronically to the public through several on-line vendors.
Our sample period runs from August 23, 1993, to
August 19, 1994, giving us a year with 250 trading days
after excluding ten holidays. The period is somewhat
unusual in that it covers a time when the Federal Reserve
was particularly active in monetary tightening, raising its
fed funds target rate five times (Chart 1). We choose the
on-the-run five-year U.S. Treasury note to represent the
U.S. Treasury securities market in our analysis. On-the-run

Chart 1

Federal Funds Target Rate and Five-Year U.S. Treasury
Note Yield
August 23, 1993, to August 19, 1994
Target rate (percent)

Yield (percent)

5.5

Five-year U.S. Treasury note yield

7.5

Scale
5.0

7.0

4.5

6.5

4.0

6.0
5.5

3.5
Federal funds target rate

Scale
3.0

5.0

2.5
Aug Sep

4.5
Oct Nov Dec
1993

Jan

Feb

Mar

Apr

May

Jun
1994

Jul Aug

Sources: Federal Reserve Bank of New York; GovPX, Inc.
Note: Federal Open Market Committee meeting dates are indicated by
the blue vertical lines.

FRBNY ECONOMIC POLICY REVIEW / DECEMBER 1997

35

securities are the most recently issued securities of a given
maturity and account for the majority of interdealer
trading volume.7 Fleming (1997) reports that among the
on-the-run issues, the five-year note is the most actively
traded security among the brokers reporting to GovPX.
During our sample period, GovPX posted a daily average
of 2,167 bid-ask quotations and 659 trades for this note.8

ANNOUNCEMENT DATES AND RELEASE TIMES
We also collected data on the dates and release times
of twenty-one different macroeconomic announcements
(Table 2). These include the nineteen monthly announcements that regularly appear in “The Week Ahead” section of
Business Week, as well as fed funds target rate announcements
and announcements of U.S. Treasury security auction
results.9 Nineteen of the announcements come from government agencies and two come from the private sector.

Table 2
MACROECONOMIC

Eighteen of the nineteen monthly announcements are
released at regularly scheduled times of the day, with ten
released at 8:30 a.m. eastern time, one at 9:15 a.m., six at
10 a.m., and one at 2 p.m.10 Announcement times vary for
one monthly announcement (consumer credit), for the fed
funds target rate announcements, and for the Treasury security auction results announcements. We rely on Bloomberg
for the precise release times of these announcements.
As for release dates, consumer confidence is the
first report to be released with information about a given
month and is actually released at the end of the same
month it is covering (Chart 2). The NAPM survey, the
other private-sector report in our sample, is typically the
next report released—on the first business day of the
month following the month covered. The employment
report, usually released on the first Friday of the month, is
the first government report to be announced with informa-

ANNOUNCEMENTS

Time
8:30 a.m.
8:30 a.m.
8:30 a.m.
8:30 a.m.
8:30 a.m.
8:30 a.m.
8:30 a.m.a
8:30 a.m.
8:30 a.m.
8:30 a.m.

Short Title
Consumer price index (CPI)
Durable goods orders
Employment
Gross domestic product (GDP)
Housing starts
Leading indicators
Personal income
Producer price index (PPI)
Retail sales
Trade balanceb

Full Title
Consumer Price Index
Advance Report on Durable Goods Manufacturers’ Shipments and Orders
The Employment Situation
Gross Domestic Product
Housing Starts and Building Permits
Composite Indexes of Leading, Coincident, and Lagging Indicators
Personal Income and Outlays
Producer Price Indexes
Advance Retail Sales
U.S. International Trade in Goods and Services

9:15 a.m.

Industrial Production and Capacity Utilization

10 a.m.
10 a.m.
10 a.m.
10 a.m.
10 a.m.

Industrial production
and capacity utilization
Business inventories
Construction spending
Consumer confidence
Factory inventories
NAPM survey

10 a.m.
2 p.m.
Variesc
Variesd
Variese

New single-family home sales
Federal budget
Consumer credit
Federal funds target rate
Treasury security auction results

New One-Family Houses Sold and For Sale
Treasury Statement (the Monthly “Budget”)
Consumer Installment Credit
N.A.
Treasury Security Auction Results

Manufacturing and Trade: Inventories and Sales
Value of New Construction Put in Place
Consumer Confidence Index
Manufacturers’ Shipments, Inventories, and Orders
National Association of Purchasing Management Report on Business

Reporting Entity
Bureau of Labor Statistics
Bureau of the Census
Bureau of Labor Statistics
Bureau of Economic Analysis
Bureau of the Census
Bureau of Economic Analysis
Bureau of Economic Analysis
Bureau of Labor Statistics
Bureau of the Census
Bureau of the Census, Bureau
of Economic Analysis
Federal Reserve Board
Bureau of the Census
Bureau of the Census
Conference Board
Bureau of the Census
National Association
of Purchasing Management
Bureau of the Census
Department of the Treasury
Federal Reserve Board
Federal Reserve Board
Department of the Treasury

Notes: The table reports the announcement time, title, and reporting entity for eighteen regularly scheduled announcements and three announcements with varying release times. All times are eastern.
a

Personal income was reported at 10 a.m. for the first three announcements in the period of analysis and at 8:30 a.m. thereafter.

b

This report replaced the Census Bureau’s Report of U.S. Merchandise Trade in March 1994.

c

Eight of the twelve announcements in our sample were made at 4 p.m. The others were made at 2:12 p.m., 2:45 p.m., 3:14 p.m., and 3:55 p.m.

d

The six announcements in our sample were made at 10:06 a.m., 11:05 a.m., 1:17 p.m., 2:18 p.m., 2:20 p.m., and 2:26 p.m.

e

All of the auction results in our sample were announced between 12:30 p.m. and 2:15 p.m., with most reported between 1:30 p.m. and 2 p.m.

36

FRBNY ECONOMIC POLICY REVIEW / DECEMBER 1997

Chart 2

EXPECTATIONS AND ANNOUNCEMENTS

Macroeconomic Announcement Release Dates

Market expectations for the nineteen monthly announcements are obtained from the Wall Street Journal. Every
Monday, the Journal publishes consensus forecasts provided
by Technical Data, a market analysis firm, for the coming
week’s announcements. Technical Data produces the forecasts from a survey of twenty-five economists conducted
the Friday before.13 We refer to Barron’s (which also relies
on Technical Data) for forecasts unavailable in the Wall
Street Journal and to Business Week (which relies on MMS
International) for forecasts that we could not get from the
first two sources. We obtained a complete set of forecasts
for eighteen of our nineteen monthly announcements and a
partial set (eight out of twelve) for the remaining one (factory inventories). Actual announcement data are retrieved
from these same three sources and are supplemented by
data from Bloomberg when necessary.
Expectations for the fed funds target rate are cal,
culated using
the rates on fed funds futures contracts.
,
Since the settlement
price of a fed funds futures contract
is based on the average effective overnight fed funds rate
over an entire month, the rate at any point during a
f
month i is a weighted average of the actual fed funds
a
rate to date i and the rate expected to prevail for the rest
N – T- × i m ,
T- × i a + ------------of the month, i m. Specifically, i f = --N
N
where T is the number of days passed to date and N is the
number of days in the month. The fed funds target rate
expected to prevail after an FOMC meeting is then calculated by solving for i m using the daily rate data up to each
FOMC announcement.14
We can measure expectations for the Treasury
security auction results much more precisely than other
expectations. Our measure is the yield in the when-issued
market (extracted from the GovPX data set) at the time of
the auction. Actual results are then measured by the auction yield as reported in the next day’s Wall Street Journal.15

August 23, 1993, to August 19, 1994

Consumer confidence
NAPM survey
Employment
Producer price index
Consumer price index
Retail sales
Industrial production and capacity utilization
Housing starts
Federal budget
Gross domestic producta
Durable goods orders
Personal income
New single-family home sales
Leading indicators
Factory inventories
Construction spending
Consumer credit
Business inventories
Trade balance
22 25 28 31 3
Concurrent
month

6

9 12 15 18 21 24 27 30 2 5 8 11 14 17 20 23
One month previous
Two months previous

Release dates
Sources: Business Week; Office of Management and Budget, Schedule of
Release Dates for Principal Federal Economic Indicators (1993, 1994); Bloomberg L.P.
Note: The chart shows the range of release dates for scheduled monthly
economic announcements and indicates the month of economic data
included in the report.
aAlthough

gross domestic product is a quarterly statistic, the advance,
preliminary, and final estimates are released in successive months. The advance
statistic is released roughly one month after the end of the quarter.

tion about a given month.11 It is followed by releases of the
PPI, the CPI, retail sales, and industrial production and
capacity utilization. The remaining twelve monthly reports
are released in the second half of the month following the
month covered, or in the month after that.
Our year of data contains twelve releases for each of
the nineteen monthly announcements. In 1994, the Federal
Reserve began making fed funds target rate announcements, the first one at its February 1994 Federal Open Market Committee (FOMC) meeting. This study provides the
first intraday analysis of the fed funds target rate announcements, of which there are six in our sample.12 The impact of
the Treasury security auction results announcements, which
are scheduled at regular intervals, are considered separately
for each coupon security of a given maturity. Our year of
data contains results of two thirty-year-bond auctions, four
ten-year-note auctions, twelve five-year-note auctions, four
three-year-note auctions, and twelve two-year-note auctions. In total, our sample contains 268 announcement
releases on 173 separate days, leaving 77 days with no
announcement.

MARKET UNCERTAINTY
Our analysis of market conditions relies on two measures of market uncertainty (Chart 3). One is the
implied volatility derived from options on U.S. Treasury
futures traded on the Chicago Board of Trade. Specifi-

FRBNY ECONOMIC POLICY REVIEW / DECEMBER 1997

37

Chart 3

Table 3
SHARPEST PRICE CHANGES FOR THE FIVE-YEAR
U.S. TREASURY NOTE

Measures of Market Uncertainty
August 23, 1993, to August 19, 1994

Price
Change
(Percent)
-0.590

Date
August 5, 1994

Time
8:30-8:35 a.m.

-0.536

May 6, 1994

8:30-8:35 a.m.

-0.440

July 8, 1994

8:30-8:35 a.m.

-0.412

April 1, 1994

8:30-8:35 a.m.

0.407

July 29, 1994

8:30-8:35 a.m.

-0.406

September 3, 1993

8:30-8:35 a.m.

-0.384

May 12, 1994

8:30-8:35 a.m.

Source: Authors’ calculations, based on data from the Chicago Board of Trade
and the Federal Reserve Bank of New York.

-0.343

May 27, 1994

8:35-8:40 a.m.

Notes: The expected federal funds rate change is defined as the difference
between the federal funds futures rate (drawn from the contract expiring at
the end of the month two months ahead) and the federal funds target rate.
Implied volatility is an annualized measure derived from futures options on
ten-year U.S. Treasury notes.

-0.332

November 9, 1993

8:30-8:35 a.m.

-0.315

February 4, 1994

8:30-8:35 a.m.

-0.313

September 10, 1993

8:30-8:35 a.m.

-0.282

January 7, 1994

8:30-8:35 a.m.

-0.266

August 16, 1994

1:45-1:50 p.m.

-0.265

June 3, 1994

8:40-8:45 a.m.

-0.259

February 4, 1994

11:05-11:10 a.m.

-0.255

April 1, 1994

8:40-8:45 a.m.

-0.253

July 14, 1994

8:30-8:35 a.m.

-0.249

September 14, 1993

8:30-8:35 a.m.

-0.224

April 13, 1994

8:30-8:35 a.m.

-0.223

May 11, 1994

1:40-1:45 p.m.

-0.223

April 1, 1994

8:35-8:40 a.m.

-0.223

February 11, 1994

8:30-8:35 a.m.

-0.222

July 12, 1994

8:30-8:35 a.m.

-0.221

May 17, 1994

2:35-2:40 p.m.

-0.218

December 9, 1993

8:30-8:35 a.m.

Volatility (percent)

Expected rate change (percent)
1.2

10

Implied volatility

Scale
1.0

9

0.8

8

0.6

7

0.4

6

5

0.2
Expected federal funds rate change

Scale
0
Aug Sep

4
Oct Nov
1993

Dec

Jan

Feb

Mar

Apr

May Jun
1994

Jul Aug

cally, the volatility measure equals the average of six
individual implied volatilities calculated using the
nearest-to-the-money calls and puts on futures contracts
on ten-year U.S. Treasury notes. The second measure is
the expected change in the fed funds rate—defined as
the difference between the fed funds futures rate (drawn
from the contract expiring at the end of the month two
months ahead) and the fed funds target rate. The
expected fed funds rate change is positive for our entire
sample year because the question during this period was
largely whether the Federal Reserve was going to raise
rates, and if so, by how much.

THE LARGEST MARKET MOVES
To account for the sharpest price changes and the greatest
surges in trading activity in the bond market, we selected
the twenty-five largest price changes and the twenty-five
most active trading episodes from every five-minute interval across the global trading day from August 23, 1993, to
August 19, 1994 (Tables 3 and 4).16

Announcement
(Time)
Employment
(8:30 a.m.)
Employment
(8:30 a.m.)
Employment
(8:30 a.m.)
Employment,
personal income
(8:30 a.m.)
Gross domestic
product
(8:30 a.m.)
Employment,
leading indicators
(8:30 a.m.)
Producer price
index, retail sales
(8:30 a.m.)
Gross domestic
product
(8:30 a.m.)
Producer price index
(8:30 a.m.)
Employment
(8:30 a.m.)
Producer price index
(8:30 a.m.)
Employment
(8:30 a.m.)
Federal funds
target rate
(1:17 p.m.)
Employment
(8:30 a.m.)
Federal funds
target rate
(11:05 a.m.)
Employment,
personal income
(8:30 a.m.)
Retail sales
(8:30 a.m.)
Consumer price
index, retail sales
(8:30 a.m.)
Consumer price
index, retail sales
(8:30 a.m.)
Ten-year-note
auction results
(1:42 p.m.)
Employment,
personal income
(8:35 a.m.)
Producer price
index, retail sales
(8:30 a.m.)
Producer price index
(8:30 a.m.)
Federal funds target
rate (2:26 p.m.)
Producer price index
(8:30 a.m.)

Source: Authors’ calculations, based on data from GovPX, Inc.

PRICE SHOCKS
It is striking that the twenty-five sharpest price changes in
the bond market all occurred on announcement days.17
38

FRBNY ECONOMIC POLICY REVIEW / DECEMBER 1997

Notes: The table reports the largest percentage price changes by five-minute
interval for the five-year U.S. Treasury note along with associated announcements (and announcement times). The largest price changes are chosen from
all
five-minute intervals across the global trading day for the period August 23,
1993, to August 19, 1994. All times are eastern.

Moreover, all but one came within fifteen minutes of an

Table 4
MOST ACTIVE TRADING INTERVALS FOR THE FIVE-YEAR
U.S. TREASURY NOTE

announcement’s release. The largest shock was a price
decline of 0.59 percent (a yield increase of 14 basis points)

Number
of Trades
35

Date
July 29, 1994

Time
8:50-8:55 a.m.

30

September 14, 1993

8:40-8:45 a.m.

sales report, three a personal income report, two a CPI

29

July 20, 1994

8:35-8:40 a.m.

report, and two a GDP report. In eight instances, the

28

January 7, 1994

8:45-8:50 a.m.

shocks came after the concurrent release of two reports.

28

February 11, 1994

8:35-8:40 a.m.

28

February 11, 1994

9:00-9:05 a.m.

It is striking that the twenty-five sharpest

27

May 27, 1994

8:45-8:50 a.m.

price changes in the bond market all occurred

27

July 14, 1994

8:35-8:40 a.m.

on announcement days. Moreover, all but

26

May 6, 1994

9:20-9:25 a.m.

26

May 13, 1994

8:50-8:55 a.m.

25

November 5, 1993

8:35-8:40 a.m.

25

January 7, 1994

8:35-8:40 a.m.

25

January 28, 1994

8:40-8:45 a.m.

25

March 1, 1994

10:50-10:55 a.m.

announcement and one trailed a release of auction results

25

March 15, 1994

8:35-8:40 a.m.

for the ten-year U.S. Treasury note.

25

April 20, 1994

8:45-8:50 a.m.

25

June 3, 1994

8:35-8:40 a.m.

25

June 10, 1994

9:00-9:05 a.m.

immediately upon the release of the August 5, 1994,
employment report. Nine other shocks were found to follow an employment report, six a PPI report, five a retail

one came within fifteen minutes of an
announcement’s release.

Three other shocks followed a fed funds target rate

The fact that price shocks in the bond market are
so explainable stands in contrast to the difficulty of
explaining them in the stock market. It is true that we

25

July 8, 1994

8:40-8:45 a.m.

a

March 4, 1994

8:45-8:50 a.m.

24a

April 20, 1994

9:40-9:45 a.m.

24

June 29, 1994

9:15-9:20 a.m.

of hindsight in the analysis. Cutler, Poterba, and Summers
rely on explanations offered by the New York Times after the

24a

July 8, 1994

8:45-8:50 a.m.

a

24

July 12, 1994

8:35-8:40 a.m.

events.18 Because these are ex post explanations, the authors

24a

July 12, 1994

8:40-8:45 a.m.

attempt to explain only a year in the bond market,

24

while Cutler, Poterba, and Summers (1989) seek to
explain more than forty years in the stock market. However, it is important to note that our explanations are based
on an ex ante list of announcements, thus reducing the bias

a

focus on whether the explanations are convincing. Although
our analysis is limited to a single year, it is a year for which
we are able to verify precise release times for announcements that we have reason to believe a priori contain information relevant to the market.

Announcement
(Time)
Gross domestic
product (8:30 a.m.)
Consumer price
index, retail sales
(8:30 a.m.)
Housing starts
(8:30 a.m.)
Employment
(8:30 a.m.)
Producer price index,
retail sales
(8:30 a.m.)
Producer price index,
retail sales
(8:30 a.m.)
Gross domestic
product
(8:30 a.m.)
Retail sales
(8:30 a.m.)
Employment
(8:30 a.m.)
Consumer
price index
(8:30 a.m.)
Employment
(8:30 a.m.)
Employment
(8:30 a.m.)
Gross domestic
product (8:30 a.m.)
NAPM survey,
construction
spending
(10:00 a.m.)
Producer price index
(8:30 a.m.)
Housing starts
(8:30 a.m.)
Employment
(8:30 a.m.)
Producer price index
(8:30 a.m.)
Employment
(8:30 a.m.)
Employment,
leading indicators
(8:30 a.m.)
Housing starts
(8:30 a.m.)
Gross domestic
product
(8:30 a.m.)
Employment
(8:30 a.m.)
Producer price index
(8:30 a.m.)
Producer price index
(8:30 a.m.)

Source: Authors’ calculations, based on data from GovPX, Inc.
Notes: The table reports the highest number of trades by five-minute interval
for the five-year U.S. Treasury note along with associated announcements
(and announcement times). The most active intervals are chosen from all fiveminute intervals across the global trading day for the period August 23,
1993, to
August 19, 1994. All times are eastern.
a

Eight intervals with twenty-four trades are in the sample; we report the six
with the largest number of bid-ask quotations.

FRBNY ECONOMIC POLICY REVIEW / DECEMBER 1997

39

TRADING SURGES

movements tend to be concentrated in the second half of

It is similarly striking that the twenty-five greatest surges in
trading activity all occurred on announcement days. The evidence linking each surge to an announcement may seem less
compelling than the corresponding evidence for price shocks
because a longer lag separates these surges from the time of
announcement. Nonetheless, all of the surges in activity

the period: sixteen of the twenty-five price shocks and

It is similarly striking that the twenty-five
greatest surges in trading activity all occurred
on announcement days. . . . All of the surges
in activity came within seventy minutes of
an announcement’s release, nineteen of
them within half an hour.

came within seventy minutes of an announcement’s release,
nineteen of them within half an hour.19 The greatest surge
consisted of thirty-five transactions worth a total of
$240 million (in face value) in a five-minute interval
twenty minutes after the July 29, 1994, GDP report.20
Eight of the other surges followed an employment report,
six a PPI report, four a GDP report, four a retail sales
report, three a housing starts report, and two a CPI report.
In five instances, the surges followed the concurrent release
of two reports.

eighteen of the twenty-five trading surges. Federal
Reserve target rate changes and market uncertainty over
those changes may explain this pattern, a hypothesis we
explore later.
The association between announcement release
times and the largest price shocks and trading surges
reflects a more general intraday pattern seen on most
announcement days. In general, pronounced market movements follow announcement releases. On an average
announcement day, we find that price volatility spikes just
after the release times and that these spikes are absent on
nonannouncement days (Chart 4).21 This pattern has also
been documented by Ederington and Lee (1993) and
Fleming and Remolona (1997). In addition, we find that
the average number of trades following release times on
announcement days exceeds the average on nonannouncement days (Chart 5). Trading volume, which accounts for
the size as well as the number of trades, follows a similar
pattern, as documented in Fleming and Remolona (1997).

Chart 4

Intraday Price Volatility on Announcement
and Nonannouncement Days
Price volatility
1.4
1.2
Announcement days
1.0
0.8
0.6

INTRADAY ANNOUNCEMENT PATTERNS
The largest movements in prices and surges in trading

0.4

activity exhibit certain regularities. First, we account

0.2

for all these movements with only twelve announce-

0

ments. Among these, the employment, PPI, and retail
sales announcements appear to be consistently important for both price shocks and trading surges, fed funds
target rate actions for price shocks, and housing starts
announcements for trading surges. Second, the large

40

FRBNY ECONOMIC POLICY REVIEW / DECEMBER 1997

Nonannouncement days
8 a.m. 9 a.m. 10 a.m. 11 a.m. Noon 1 p.m. 2 p.m. 3 p.m. 4 p.m. 5 p.m.

Time of day
Source: Authors’ calculations, based on data from GovPX, Inc.
Notes: The chart shows the standard deviation of log price changes by
five-minute interval for the five-year U.S. Treasury note for days with at
least one of the twenty-one announcements listed in Table 2 and days with
none of these announcements. The standard deviation equals the actual
3
standard deviation times 10 . The period of analysis is August 23, 1993, to
August 19, 1994. Times shown are interval starting times (eastern).

Chart 5

minute interval following an announcement, with prices

Intraday Trading Activity on Announcement
and Nonannouncement Days

defined as the midpoints between bid and ask quotes.22

Number of trades

We measure trading activity as the number of transactions
during the one-hour interval following the announcement.

14

The longer interval for trading activity is consistent with

12
Announcement days

Fleming and Remolona’s (1997) results suggesting that

10

prices adjust rapidly while high trading activity persists for

8

an extended period after an announcement.

6

For our explanatory variables, we define announcement dummy variables D knt , where D knt = 1 if announcement k is made on day n just before interval t and D knt = 0

4
Nonannouncement days
2
0
8 a.m.

9 a.m. 10 a.m. 11 a.m. Noon

1 p.m. 2 p.m. 3 p.m. 4 p.m. 5 p.m.

Time of day
Source: Authors’ calculations, based on data from GovPX, Inc.
Notes: The chart shows the mean number of interdealer trades by five-minute
interval for the five-year U.S. Treasury note for days with at least one of the
twenty-one announcements listed in Table 2 and days with none of these
announcements. The period of analysis is August 23, 1993, to August 19, 1994.
Times shown are interval starting times (eastern).

WHICH ANNOUNCEMENTS HAVE
THE MOST RELEVANCE?
If the market’s movements represent a reaction to new information, some types of announcements should induce a
stronger reaction than others because of inherent differences in
the information contained about the economy. We now test
whether the market’s price movements and trading activity
serve to differentiate among the various announcements, and
to the degree they do, which announcements matter the most.
While differences from expectations in a given announcement
may be an important determinant of the market’s response—
an issue we explore in the next section of the article—our first
step is simply to determine which announcements consistently affect the market and to what extent.

ESTIMATION OF ANNOUNCEMENT IMPORTANCE
To establish the importance of the various announcements,
we run regressions of price volatility and trading activity
on dummy variables representing each of the announcements listed in Table 2. We measure price volatility by the
absolute value of the change in log prices in the five-

otherwise.23 We rely on an additional set of dummy variables
Dt to control for intraday patterns of price volatility and tradj

ing activity. We denote the dependent variables by Y nt ,
where the superscript j indicates whether the variable is price
volatility or trading activity. Our regression equation
j
j
j
j
is then Ynt = a0 + Σ Tt =– 11 a t Dt + Σ kK = 1 bk Dknt + e ntj ,
where T=22 (the number of different intervals corresponding
to the release times of the different announcements) and K=25
(the number of announcements we analyze). The coefficient of
interest is bkj , which measures the impact of announcement k.

ANNOUNCEMENTS AFFECTING PRICE
Our results suggest that the bond market differentiates
among the various types of announcements through the
magnitude of its price movements. Nine of the twenty-five
announcements examined are found to have a significant
impact on price, six showing significant effects at the
1 percent level and three at the 5 percent level (Table 5). In
order of importance, the significant announcements with
the greatest effects on price are: (1) employment, (2) PPI,
(3) fed funds target rate, (4) retail sales, (5) CPI, (6) NAPM
survey, (7) five-year-note auction results, (8) industrial
production and capacity utilization, and (9) consumer
confidence. This list of significant announcements is
longer than any such list in previous studies.
Our regression results are noteworthy for several
other reasons. First, we document for the first time a significant market impact from U.S. Treasury security auction
results. Second, bond prices react so consistently to four
announcements—the NAPM survey, five-year-note auction

FRBNY ECONOMIC POLICY REVIEW / DECEMBER 1997

41

results, industrial production and capacity utilization, and
consumer confidence—that these announcements are significant even when absent from the twenty-five largest
price shocks. Third, although GDP releases account for
two of our twenty-five largest price shocks, such releases
fail to induce a price reaction consistently and hence are
not found to be significant in our regressions.24
Our results, in conjunction with those of earlier
researchers, also provide evidence of stability in the

Table 5
IMPACT OF
Rank
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25

ANNOUNCEMENTS ON PRICE

Announcement
Employment
Producer price index
Federal funds target rate
Gross domestic product
Retail sales
Consumer price index
Thirty-year-bond auction results
Ten-year-note auction results
NAPM survey
Five-year-note auction results
Industrial production and capacity utilization
Consumer confidence
New single-family home sales
Durable goods orders
Construction spending
Three-year-note auction results
Trade balance
Housing starts
Personal income
Business inventories
Consumer credit
Factory inventories
Two-year-note auction results
Federal budget
Leading indicators

Memo:
Adjusted R2
χ2 statistica
Number of observations

Coefficient
26.10**
13.71**
11.00**
7.19
7.04*
6.75**
6.48
5.84
4.12*
3.62**
3.42**
3.09*
2.58
1.78
1.78
1.76
1.68
1.34
1.15
1.14
0.86
0.70
0.26
0.03
-3.32

0.40**
362**
5,323**

Source: Authors’ calculations, based on data from GovPX, Inc.
Notes: The table presents the regression coefficients indicating the average
difference in price volatility for the five-year U.S. Treasury note for the fiveminute period after an announcement as compared with the same period on
nonannouncement days. Volatility is defined as the absolute value of the log
price change times 104. Coefficient significance is based on two-sided t-tests using
heteroskedasticity-consistent (White) standard errors. The period of analysis is
August 23, 1993, to August 19, 1994.
The χ2 statistic tests whether all model coefficients equal zero and is computed using the heteroskedasticity-consistent covariance matrix.

a

*Significant at the .05 level.
**Significant at the .01 level.

42

FRBNY ECONOMIC POLICY REVIEW / DECEMBER 1997

announcements that have relevance to bond prices. In their
analysis of bond futures prices from November 1988 to
November 1991, Ederington and Lee (1993) find the
employment, PPI, CPI, and durable goods orders reports to
be the most important regularly scheduled announcements. The continued significance of the employment
report may be explained by the fact that it still offers the
market the first comprehensive look at the economy’s
strength, with data on nonfarm payroll employment, the
unemployment rate, and average hourly earnings.25 The
PPI and CPI reports also continue to be significant. Of
Ederington and Lee’s most important announcements, only
the durable goods orders report has lost its significance.26

ANNOUNCEMENTS AFFECTING TRADING ACTIVITY
Our results, in conjunction with those of earlier researchers, also suggest that the bond market differentiates
among announcements through the extent of trading
activity elicited. Fourteen of the announcements have a
significant positive impact on trading activity, twelve at
the 1 percent level and two at the 5 percent level (Table 6).
In order of importance, the announcements that generate significant trading activity are: (1) employment,
(2) fed funds target rate, (3) thirty-year-bond auction
results, (4) PPI, (5) ten-year-note auction results, (6) CPI,
(7) NAPM survey, (8) GDP, (9) retail sales, (10) threeyear-note auction results, (11) new single-family home
sales, (12) factory inventories, (13) business inventories,
and (14) industrial production and capacity utilization.
We note that, first, the announcements that matter
for price also tend to matter for trading activity. The employment report, for example, has the greatest impact on both
price and trading activity. Second, housing starts releases
account for three of the twenty-five greatest trading surges
but do not consistently produce a rise in trading activity.
Third, eight announcements consistently lead to additional
trading activity even when they do not account for any of the
twenty-five greatest trading surges: fed funds target rate,
thirty-year-bond auction results, ten-year-note auction
results, three-year-note auction results, new single-family
home sales, factory inventories, business inventories, and
industrial production and capacity utilization.

Table 6
IMPACT OF
Rank
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25

ANNOUNCEMENTS ON TRADING ACTIVITY

Announcement
Employment
Federal funds target rate
Thirty-year-bond auction results
Producer price index
Ten-year-note auction results
Consumer price index
NAPM survey
Gross domestic product
Retail sales
Three-year-note auction results
New single-family home sales
Factory inventories
Business inventories
Industrial production and capacity utilization
Housing starts
Trade balance
Leading indicators
Consumer confidence
Personal income
Two-year-note auction results
Durable goods orders
Consumer credit
Construction spending
Federal budget
Five-year-note auction results

Memo:
Adjusted R2
χ2 statistica
Number of observations

Coefficient
87.93**
72.14**
63.55**
58.29**
46.50**
45.92**
39.72**
39.47**
38.21**
36.24**
30.05**
26.14**
23.53*
23.02*
15.37
13.54
6.46
5.35
3.72
0.72
-0.32
-0.35
-1.21
-7.03
-10.42*

0.38
6,721**
5,386

Source: Authors’ calculations, based on data from GovPX, Inc.
Notes: The table presents the regression coefficients indicating the average
difference in trading activity for the five-year U.S. Treasury note for the onehour period after an announcement as compared with the same period on nonannouncement days. Trading activity is defined as the number of interdealer
broker transactions reported by GovPX. Coefficient significance is based on
two-sided t-tests using heteroskedasticity-consistent (White) standard errors.
The period of analysis is August 23, 1993, to August 19, 1994.
The χ2 statistic tests whether all model coefficients equal zero and is computed using the heteroskedasticity-consistent covariance matrix.

a

*Significant at the .05 level.
**Significant at the .01 level.

reports—consumer confidence and the NAPM survey—are even more timely than the employment
report. Although both reports significantly affect the
market, the bond market evidently regards their information about the economy as somewhat less important
than the information in the government’s employment,
PPI, CPI, and retail sales reports. As we will demonstrate, the degree of surprise in a given announcement
and conditions of market uncertainty also influence an
announcement’s importance.

ANNOUNCEMENT SURPRISES
AND MARKET CONDITIONS
DOES THE MAGNITUDE OF SURPRISE MATTER?
A bond market that truly responds to the arrival of information should not only differentiate among the various
types of announcements but also react more sharply to
larger surprises in a given announcement.27 Many bond
market announcement studies focus on the surprise component because information is believed to have value only to
the extent that it is unexpected. For example, an unexpectedly strong nonfarm payrolls number should cause a fall in
bond prices, with a greater surprise causing a greater fall.
The effect on trading activity is less clear, however, because
a larger surprise would not necessarily lead to wider disagreement among traders about the appropriate price
adjustment, although we might expect it to lead to greater
portfolio rebalancing if the larger surprise is accompanied
by a greater price change.
To measure the impact of unexpected information,
we regress five-year U.S. Treasury note price changes and

TIMELINESS
The timeliness of an announcement—that is, how soon
data are released after the period covered ends—helps
to explain its impact on prices and trading activity. Of
the government reports, the most timely are employment, PPI, CPI, and retail sales, in that order (Chart 2).
This order of timeliness is nearly matched by the
reports’ order of importance for both price shocks and
trading activity. Timeliness, however, is not the sole
determinant of market impact. The two private sector

trading activity on the surprise components of announcements. We define surprises Sknt ≡ Aknt – Fknt , where Aknt is
the actual number released in announcement k on day n in
interval t and Fknt is the corresponding forecast number
(Sknt=0 on days and in intervals without a release of
announcement k). Although each announcement typically
reveals several pieces of information, we limit our analysis to
surprises in the headline number. For the employment
report, we therefore focus on nonfarm payroll employment
surprises; for industrial production and capacity utilization,

FRBNY ECONOMIC POLICY REVIEW / DECEMBER 1997

43

we focus on industrial production surprises. To facilitate a

whole, these results suggest that larger announcement
surprises do not systematically widen the divergence in
traders’ views or lead to greater portfolio rebalancing.

comparison of announcement effects and to ensure that our
estimated coefficients are representative of a typical
announcement, we scale the surprises by the mean absolute
1
surprise S k = ------Σ n S knt , where Nk is the number of
Nk
releases of announcement k in our sample.

Table 7
IMPACT OF

Hence, our regression equation for bond prices is
Sknt
P = aP + ΣK
P --------- + u P , where Z P is
given by Z nt
k = 1 ck S
nt
0
nt
k
the signed price change. In the case of trading activity, our
Sknt
Q,
equation is ZQ = a Q + Σ tT=–11 aQ Dt + Σ K=1 c Q ----------- + u nt
nt
k
t
0
k
Sk
where Z Q is trading activity, Dt are dummy variables to connt
trol for intraday patterns of trading activity, and S knt are the
Q
absolute surprises.28 The coefficients c kP and ck , which mea-

Announcement
Employment
(nonfarm payrolls)
Producer price index
Ten-year-note
auction results
Thirty-year-bond
auction results
Retail sales
Consumer price index
New single-family
home sales
Federal funds
target rate
Consumer confidence
Five-year-note
auction results
NAPM survey
Industrial production
Housing starts
Gross domestic product
Trade balance
Construction spending
Consumer credit
Durable goods orders
Two-year-note
auction results
Leading indicators
Federal budget
Business inventories
Personal income
Three-year-note
auction results
Factory inventories

sure the effects of announcement surprises on prices and
trading activity, respectively, are reported in Table 7 along
with the mean absolute surprise for each announcement.
In general, the surprise components provide more
precise estimates of announcement effects on bond prices,
indicating a market that is indeed reacting to the arrival of
information. Taking account of the magnitude and sign of
the surprise lends significance to six announcements not
found to be significant in the regressions with announcement dummy variables, adding to an already long list of
significant announcements. The six additional announcements are the auction results for the ten-year U.S Treasury
note and thirty-year U.S. Treasury bond, new single-family
home sales, housing starts, the trade balance, and consumer
credit. The fed funds target rate and retail sales announcements, however, lose their significance because their price
effects do not bear a consistent sign. Increases in the fed
funds target rate, in particular, often had a strong effect on
bond prices during the period, but the effects were at times
positive and at times negative.29
In the case of trading activity, it is much less
clear that taking account of the magnitude of the surprise
helps explain the bond market’s response to announcements. A comparison of Tables 6 and 7 shows that the
absolute surprises add significance to the effects of the
business inventories releases but reduce significance for
the new single-family home sales releases. Unlike the
effects on prices, the significance of fed funds target rate
actions for trading activity remains the same. On the

ANNOUNCEMENT SURPRISES
Mean
Absolute Surprise

Price
Coefficient

Trading Activity
Coefficient

92,000 jobs
0.23%

-23.10**
-8.59**

60.52**
27.87**

0.02%

-8.05**

34.21**

0.03%
0.46%
0.10%

-7.71**
-6.51
-6.48**

40.41**
39.03**
24.56**

63,000 homesa

-5.08**

23.97*

0.13%
3.92

-4.61
-4.42**

60.80**
9.62

0.01%
0.93%
0.18%
62,000 homesa
0.36%
$1.04 billion
0.94%
$2.10 billion
1.03%

-4.20**
-4.17**
-3.87**
-3.42**
-3.20
-2.50**
-1.79
-1.70**
-1.41

-7.86*
35.83**
17.81*
12.05
29.04**
4.94
-5.35
2.24
-5.10

-1.25
-0.46
-0.29
0.05
0.19

6.37
2.28
-1.86
24.88**
-1.66

0.01%
0.09%
$1.33 billion
0.22%
0.19%

Memo:
Adjusted R2
χ2 statisticb
Number of observations

FRBNY ECONOMIC POLICY REVIEW / DECEMBER 1997

1.06
1.61*

27.40**
27.55**

0.27**
996**
5,319

0.37
5,655**
5,382

Source: Authors’ calculations, based on data from GovPX, Inc.
Notes: The table presents the regression coefficients indicating the impact of
announcement surprises on price and trading activity for the five-year U.S.
Treasury note. Announcement surprises are the actual numbers announced
minus the forecast numbers divided by the mean absolute surprise for each
announcement type. The impact on price is examined with signed surprises
while surprise magnitudes are used for trading activity. Price is defined as the
log price change times 104 for the five-minute period immediately after
announcement; trading activity is defined as the number of transactions in the
one-hour period after an announcement. Coefficient significance is based on
two-sided t-tests using heteroskedasticity-consistent (White) standard errors.
The period of
analysis is August 23, 1993, to August 19, 1994.
a

Figure reported is at an annual rate.

The χ2 statistic tests whether all model coefficients equal zero and is computed using the heteroskedasticity-consistent covariance matrix.
b

*Significant at the .05 level.
**Significant at the .01 level.

44

0.02%
0.14%

Pi

The largest price shock in our sample followed an employment report that contained relatively little surprise. Specifically, on August 5, 1994, the price of the five-year U.S.
Treasury note fell 0.59 percent within five minutes of the
release of a nonfarm payrolls number that exceeded the
forecast by only 54,000 jobs.30 The period seems to have
been a time of great uncertainty, with previous announcements giving mixed signals about the strength of the economy and bond market participants trying to guess whether
the Federal Reserve was about to raise rates for the fifth
time in six months. Hence, the issue we examine is
whether market participants attach more significance to
the same information during times of greater uncertainty.
To analyze the impact of market uncertainty, we
run regressions that allow the surprise variables to interact
with our uncertainty variables. As described earlier, our
measures of uncertainty are the implied volatility from
Treasury futures options and the expected change in the fed
funds rate. We specify the announcement surprise coeffii
cients to depend on uncertainty, c Pk = g Pk + h Pi
k Vn
i
and c Qk = g kQ + h Qk i Vni , where V n is one of our two mea-

Table 8
IMPACT OF

Qi

sures of uncertainty and the coefficients h k and h k measure the influence of uncertainty on announcement effects.
The regression equation for bond prices then becomes

DO MARKET CONDITIONS MATTER?

Sknt
Sknt P
P = P + K
- + Σ Kk = 1 hP i Vni -------- + unt
a 0 Σ k = 1 g kP -------Znt
k
Sk
Sk
and the equation for trading activity becomes
S knt
Q
K
ZntQ = a Q + Σ Tt =–11 a t D t + Σ k = 1 g Q -------k S
0
k
S knt Q
+ Σ Kk = 1 h Qk i Vni -----------+ unt .
Sk
Table 8 presents the results of these regressions for the
8:30 a.m. announcements, identifying the announcement
Pi
Qi
surprises for which h k and h k are significant. Because the
two measures of uncertainty are highly correlated, we analyze them in separate regressions.31
Our results show that the price response to a
given announcement surprise is frequently greater under
conditions of increased uncertainty. Uncertainty in the
form of implied volatility from Treasury futures options
helps explain the bond market’s price reaction to durable
goods orders, GDP, and housing starts surprises, while
uncertainty in the form of an expected fed funds rate

MARKET CONDITIONS ON ANNOUNCEMENT RESPONSES
a

Model
1

Dependent Variable
Price

Interaction Terms
None

Interaction χ 2
N.A.

2

Price

Implied volatility

45**

3

Price

22*

4

Trading activity

Expected federal
funds rate change
None

N.A.

5

Trading activity

Implied volatility

82**

6

Trading activity

Expected federal
funds rate change

60**

Significant Interaction Coefficientsb
N.A.
Durable goods orders**,
gross domestic product*, housing starts**
Durable goods orders**,
employment (nonfarm payrolls)*
N.A.
Consumer price index**,
producer price index**, trade balance**
Consumer price index**, durable goods orders*,
employment (nonfarm payrolls)*,
personal income**, producer price index**

Model R2
0.42

Number of
Observations
250

1,050**

0.44

250

248**

0.44

250

158**

0.29

250

671**

0.34

250

514**

0.32

250

Model χ
219**

2

c

Source: Authors’ calculations, based on data from GovPX, Inc.
Notes: The table presents the results from regressions of price and trading activity on announcement surprises and two variables interacted with announcement surprises for the five-year U.S. Treasury note. All results are derived from analyses of the 8:30 a.m. monthly announcements. The price regressions are run with signed
announcement surprises and with signed price changes for the 8:30 a.m.-8:35 a.m. interval. The trading activity regressions are run with absolute announcement
surprises and with trading activity measured as the number of trades in the 8:30 a.m.-9:30 a.m. interval. Coefficient significance is based on two-sided t-tests using
heteroskedasticity-consistent (White) standard errors. The period of analysis is August 23, 1993, to August 19, 1994.
a
This χ2 statistic tests whether all interaction terms equal zero and is computed using the heteroskedasticity-consistent covariance matrix. The statistic is calculated excluding any significant interaction terms that have a sign opposite to that predicted.
b
c

The list of coefficients excludes significant interaction terms that have a sign opposite to that predicted.

This χ2 statistic tests whether all model coefficients equal zero and is computed using the heteroskedasticity-consistent covariance matrix.

*Significant at the .05 level.
**Significant at the .01 level.

FRBNY ECONOMIC POLICY REVIEW / DECEMBER 1997

45

change helps explain the reaction to durable goods orders
and employment surprises.
For trading activity, market uncertainty often
heightens the trading surge that follows announcement
surprises. Uncertainty as measured by implied volatility
helps explain the rise in trading activity in the wake of
CPI, PPI, and trade balance surprises, while uncertainty
as measured by the expected fed funds rate change helps
explain the increase in activity after CPI, durable goods
orders, employment, personal income, and PPI surprises. These results suggest that uncertain market conditions contribute to the divergence in traders’
interpretations of announcement surprises.

CONCLUSION
Our finding that the largest price shocks and the greatest
surges in trading activity in the bond market stem from
the arrival of public information is reassuring. Over the
August 23, 1993, to August 19, 1994, sample period,
each of the twenty-five sharpest price changes and each of
the twenty-five greatest surges in trading activity can be
associated with a just-released announcement. These
results suggest that U.S. Treasury securities prices react
largely to the arrival of public information about the
economy. The surge in trading activity following the
price shocks suggests a lack of consensus among market

46

FRBNY ECONOMIC POLICY REVIEW / DECEMBER 1997

participants over whether the initial price change is precisely the appropriate adjustment to the new information,
although portfolio rebalancing may also be important.
It is also reassuring to find that various measures
of the information content of the different announcements generally help explain such market responses. In
particular, the market distinguishes among announcements with inherently different information, reacting
most dramatically—through both price movements and
trading activity—to the employment, PPI, fed funds target rate, and CPI announcements. U.S. Treasury security
auction results are also found to have significant effects
on both price and trading activity.
Moreover, we find that the bond market’s reactions
depend on the unexpected component of a given announcement and on conditions of uncertainty. Taking account of
the surprise component in a report’s announced numbers
extends our list of announcements that significantly affect
bond prices from nine to thirteen, longer than any such list
in previous studies. Greater market uncertainty also leads
to a stronger market response, particularly in the form of
increased trading activity. These results suggest that the
bond market’s price and trading reactions reflect differences of informational content in and among the varying
announcements under changing market conditions.

ENDNOTES

The authors benefited from the comments of Peter Antunovich, Pierluigi Balduzzi,
Richard Cantor, Richard Clarida, Jennifer Conrad, Clifton Green, Spence
Hilton, Frank Keane, Sandra Krieger, Grant McQueen, Richard Peach,
Deborah Perelmuter, James Poterba, Tony Rodrigues, Christopher Sims, Charles
Steindel, and participants at the Federal Reserve Bank of New York’s Friday
seminar series. They thank Christina Hubbard for research assistance and
GovPX, Inc., for providing the market data.
1. See Beaver (1968).
2. However, in an analysis similar to the Cutler, Poterba, and Summers
(1989) study of the U.S. stock market, Elmendorf, Hirshfeld, and Weil
(1996) find it difficult to relate the largest movements in U.K. bond prices
from 1900 to 1920 to news arrival.
3. Roley and Troll (1983), for example, find no significant announcement
effects from the CPI, the unemployment rate, and the PPI; Hardouvelis
(1988) finds none from consumer credit, housing starts, industrial
production, leading indicators, merchandise trade, or personal
income; and Dwyer and Hafer (1989) find none from the CPI, industrial
production, the unemployment rate, or merchandise trade.
4. Kim and Verrechia (1991) and He and Wang (1995) show theoretically
how heterogeneity of views among investors can generate speculative
trading activity.
5. French and Roll (1986), for example, attribute the fact that stock return
volatilities are higher when the exchanges are open than when they are
closed to the effect of private information conveyed through trading.
6. This is the argument used by Fleming and Remolona (1997) to explain
the persistence of trading volume beyond price volatility in the Treasury
market after an announcement. That study, as well as earlier stock market
studies by Jain and Joh (1988) and Gallant, Rossi, and Tauchen (1992),
suggests that price volatility causes trading activity.
7. Fleming (1997) finds that 64 percent of interdealer trading is in
on-the-run issues, 24 percent is in off-the-run issues, and 12 percent is
in when-issued securities. Off-the-run securities are issued securities that
are no longer active; when-issued securities are securities that have been
announced for auction but not yet issued.

10. Included in the 8:30 a.m. count is the personal income
announcement, which was released at 10 a.m. for the first three
announcements in our sample but at 8:30 a.m. thereafter.
11. The employment report was released on the second Friday in October
1993 and in July 1994.
12. Five announcements occurred after the regularly scheduled February,
March, May, July, and August 1994 FOMC meetings. The other
announcement occurred in April 1994, when the fed funds target rate was
increased without an FOMC meeting. Cook and Hahn (1989), Pakko and
Wheelock (1996), and Roley and Sellon (1996) use daily data to examine
the impact of fed funds target rate changes.
13. Ideally, we would like to use forecasts that are based on expectations
right before each announcement since expectations can change over the
course of a week. Our use of weekly forecasts may bias the coefficients of our
estimates toward zero in those regressions that depend on announcement
surprises.
14. Krueger and Kuttner (1996) show that the fed funds futures rate is
effective at identifying changes in the fed funds rate. Our methodology
follows that of Pakko and Wheelock (1996), using effective fed funds rate
data from the Federal Reserve Bank of New York and fed funds futures data
from the Chicago Board of Trade.
15. The three-, ten-, and thirty-year securities are issued at pricediscriminating auctions, so for these securities the yield corresponding to
the lowest accepted price is used. The two- and five-year securities are
issued at uniform-price auctions.
16. Andersen and Bollerslev (forthcoming) perform a similar exercise with
Deutsche mark–dollar exchange rates and find that fifteen of the twenty-five
largest five-minute absolute returns from October 1992 to September 1993
are directly associated with the release of economic news.
17. Note that there are seventy-seven nonannouncement days on which
purely random shocks could have taken place. With a sample of 250 days,
the probability that all 25 of the shocks occur on an announcement day
purely by chance is 0.01 percent.

8. Appendix B of Fleming and Remolona (1997) details the data
cleaning and processing.

18. The explanation for the 20 percent decline on October 19, 1987, for
example, is “worry over dollar decline and trade deficit, fear of U.S. not
supporting dollar.”

9. We count the announcement of gross domestic product (GDP) as a
monthly release. Although GDP is a quarterly measure, the Bureau of
Economic Analysis issues advance, preliminary, and final estimates in
successive months.

19. Fleming and Remolona (1997) analyze the adjustment patterns of
trading volume after major announcements. They find an appreciable lag in
the surge in trading volume after the initial price shock and a persistence of
high volume for a few hours afterward.

NOTES

FRBNY ECONOMIC POLICY REVIEW / DECEMBER 1997

47

ENDNOTES (Continued)

20. We use the number of transactions as our measure of trading activity
instead of the face value of securities traded. We base this decision on
Jones, Kaul, and Lipson’s (1994) finding that transaction size has no
information content beyond that contained in the frequency of trades.
21. On the days with 8:30 a.m. announcements, the price change in the
first five minutes after the announcement explains 31 percent of the whole
day’s (7:30 a.m. to 5 p.m.) price change.
22. We could also use transaction prices, but using the bid-ask midpoints
allows us to avoid complications associated with the “bid-ask bounce,” in
addition to providing us with more observations.
23. For announcements released in the final minute of an interval, we
begin the analysis at the start of the next interval. For all other
announcements, the analysis begins in the same interval. For example,
a 1:34 p.m. release time implies an analysis based on the 1:35 p.m.1:40 p.m. interval for price and the 1:35 p.m.-2:35 p.m. interval for
trading activity, while a 1:33 p.m. release time implies an analysis based
on the 1:30 p.m.-1:35 p.m. interval for price and the 1:30 p.m.2:30 p.m. interval for trading activity.
24. As noted earlier, the releases consist of advance, preliminary, and final
estimates of quarterly GDP announced in successive months. An advance
estimate accounted for one of the two largest price shocks associated with
GDP releases; a preliminary estimate accounted for the other.
25. As Krueger (1996) notes, the BLS now collects the nonfarm payroll
employment data from a sample of more than 200,000 establishments that
offers wide geographic and industry coverage. We document the
employment report’s importance for the bond market; Harris and Zabka
(1995) and Andersen and Bollerslev (forthcoming) show its importance for
the foreign exchange market.

however, the reduction of that lag time has made such projections difficult.
Compounding the problem, orders have increased for goods whose prices
are changing rapidly, particularly computers; this price volatility has made
it harder for the durable goods report to assess the quantity of goods
ordered, since the report measures orders only in dollar terms.
27. We do not address issues of rationality or market efficiency in this
article—that is, we do not test whether market prices properly reflect all
available information, nor whether they adjust to such information in an
appropriately rapid fashion.
28. Absolute surprises are used for the trading activity regression (and not
the price regression) because we are testing whether the magnitudes of
announcement surprises are correlated with changes in trading activity.
For example, we suspect that nonfarm payroll surprises of 100,000 jobs
and -100,000 jobs would have contrary effects on price, but that both
would be associated with an increase in trading activity relative to
smaller magnitude surprises.
29. Pakko and Wheelock (1996) discuss why the effects change in sign.
30. The average absolute nonfarm payroll employment surprise in the
sample was 92,000 jobs (Table 7) and was as large as 206,000 on
April 1, 1994. Other components of the employment report do not seem to
explain the market’s sharp August 5 response—the announced
unemployment rate of 6.1 percent was expected, manufacturing overtime
hours were unchanged at 4.6, average manufacturing hours actually
declined to 41.9 from 42.0 the previous month, and the previous month’s
nonfarm payroll employment was revised down from 379,000 to 356,000.
Nonfarm payroll employment was not the only sign of strength, however;
average hourly earnings increased by 4¢ to $11.12.
31. The correlation between our implied volatility measure and the
expected fed funds rate change is 0.73.

26. The decreased significance of durable goods orders may reflect their
declining reliability as an indicator of future manufacturing activity.
Because an increasing share of durable goods are now shipped almost
immediately, much of the lag time that existed between order receipt and
shipment has been eliminated. In the past, that lag time enabled analysts
to use durable goods orders to predict future manufacturing activity. Now,

The views expressed in this article are those of the authors and do not necessarily reflect the position of the Federal Reserve
Bank of New York or the Federal Reserve System. The Federal Reserve Bank of New York provides no warranty, express or
implied, as to the accuracy, timeliness, completeness, merchantability, or fitness for any particular purpose of any information
contained in documents produced and provided by the Federal Reserve Bank of New York in any form or manner whatsoever.

48

FRBNY ECONOMIC POLICY REVIEW / DECEMBER 1997

NOTES

REFERENCES

Andersen, Torben G., and Tim Bollerslev. Forthcoming. “DM-Dollar
Volatility: Intraday Activity Patterns, Macroeconomic Announcements,
and Longer Run Dependencies.” JOURNAL OF FINANCE.

Edison, Hali J. 1996. “The Reaction of Exchange Rates and Interest
Rates to News Releases.” Board of Governors of the Federal Reserve
System International Finance Discussion Papers, no. 570, October.

Balduzzi, Pierluigi, Edwin J. Elton, and T. Clifton Green. 1996. “Economic
News and the Yield Curve: Evidence from the U.S. Treasury Market.”
Unpublished paper, New York University, November.

Elmendorf, Douglas W., Mary L. Hirschfeld, and David N. Weil. 1996. “The
Effect of News on Bond Prices: Evidence from the United Kingdom,
1900-1920.” REVIEW OF ECONOMICS AND STATISTICS 78: 341-4.

Beaver, William H. 1968. “The Information Content of Annual Earnings
Announcements.” EMPIRICAL RESEARCH IN ACCOUNTING: SELECTED
STUDIES. Supplement to JOURNAL OF ACCOUNTING RESEARCH 6:
67-92.

Fleming, Michael J. 1997. “The Round-the-Clock Market for U.S. Treasury
Securities.” Federal Reserve Bank of New York ECONOMIC POLICY
REVIEW 3, no. 2: 9-32.

Becker, Kent G., Joseph E. Finnerty, and Kenneth J. Kopecky. 1996.
“Macroeconomic News and the Efficiency of International Bond
Futures Markets.” JOURNAL OF FUTURES MARKETS 16, no. 2: 131-45.
Berkman, Neil G. 1978. “On the Significance of Weekly Changes in M1.”
NEW ENGLAND ECONOMIC REVIEW, May-June: 5-22.
Cook, Timothy, and Thomas Hahn. 1989. “The Effect of Changes in the
Federal Funds Rate Target on Market Interest Rates in the 1970s.”
JOURNAL OF MONETARY ECONOMICS 24: 331-51.
Cook, Timothy, and Steven Korn. 1991. “The Reaction of Interest Rates to
the Employment Report: The Role of Policy Anticipations.” Federal
Reserve Bank of Richmond ECONOMIC REVIEW, SeptemberOctober: 3-12.
Cornell, Bradford. 1982. “Money Supply Announcements, Interest Rates,
and Foreign Exchange.” JOURNAL OF INTERNATIONAL MONEY AND
FINANCE 1: 201-8.

Fleming, Michael J., and Eli M. Remolona. 1997. “Price Formation and
Liquidity in the U.S. Treasury Market: Evidence from Intraday
Patterns around Announcements.” Federal Reserve Bank of New York
Staff Reports, no. 27, July.
French, Kenneth R., and Richard Roll. 1986. “Stock Return Variances: The
Arrival of Information and the Reaction of Traders.” JOURNAL OF
FINANCIAL ECONOMICS 17: 5-26.
Gallant, A. Ronald, Peter E. Rossi, and George Tauchen. 1992. “Stock Prices
and Volume.” REVIEW OF FINANCIAL STUDIES 5, no. 2: 199-242.
Grossman, Jacob. 1981. “The ‘Rationality’ of Money Supply Expectations
and the Short-Run Response of Interest Rates to Monetary Surprises.”
JOURNAL OF MONEY, CREDIT, AND BANKING 13, no. 4: 409-24.
Hardouvelis, Gikas A. 1987. “Macroeconomic Information and Stock
Prices.” JOURNAL OF ECONOMICS AND BUSINESS 39: 131-40.

———. 1988. “Economic News, Exchange Rates, and Interest Rates.”
JOURNAL OF INTERNATIONAL MONEY AND FINANCE 7: 23-35.

———. 1983. “The Money Supply Announcements Puzzle: Review
and Interpretation.” AMERICAN ECONOMIC REVIEW 73, no. 4: 644-57.
Cutler, David M., James M. Poterba, and Lawrence H. Summers. 1989.
“What Moves Stock Prices?” JOURNAL OF PORTFOLIO MANAGEMENT
15: 4-12.
Dwyer, Gerald P., and R.W. Hafer. 1989. “Interest Rates and Economic
Announcements.” Federal Reserve Bank of St. Louis REVIEW,
March-April: 34-46.
Ederington, Louis H., and Jae Ha Lee. 1993. “How Markets Process
Information: News Releases and Volatility.” JOURNAL OF FINANCE 48,
no. 4: 1161-91.

NOTES

Harris, Ethan S., and Natasha M. Zabka. 1995. “The Employment Report
and the Dollar.” Federal Reserve Bank of New York CURRENT
ISSUES IN ECONOMICS AND FINANCE 1, no. 8.
Harvey, Campbell R., and Roger D. Huang. 1993. “Public Information and
Fixed Income Volatility.” Unpublished paper, Duke University, July.
He, Hua, and Jiang Wang. 1995. “Differential Information and Dynamic
Behavior of Stock Trading Volume.” REVIEW OF FINANCIAL STUDIES 8,
no. 4: 919-72.
Jain, Prem C. 1988. “Response of Hourly Stock Prices and Trading Volume
to Economic News.” JOURNAL OF BUSINESS 61, no. 2: 219-31.

FRBNY ECONOMIC POLICY REVIEW / DECEMBER 1997

49

REFERENCES (Continued)

Jain, Prem C., and Gun-Ho Joh. 1988. “The Dependence between
Hourly Prices and Trading Volume.” J OURNAL OF F INANCIAL
AND Q UANTITATIVE A NALYSIS 23, no. 3: 269-83.
Jones, Charles M., Gautam Kaul, and Marc L. Lipson. 1994. “Transactions, Volume,
and Volatility.” REVIEW OF FINANCIAL STUDIES 7, no. 4: 631-51.
Kim, Oliver, and Robert E. Verrecchia. 1991. “Market Reaction to
Anticipated Announcements.” JOURNAL OF FINANCIAL ECONOMICS
30, no. 2: 273-309.
Krueger, Alan B. 1996. “Do Markets Respond More to More Reliable
Labor Market Data? A Test of Market Rationality.” Unpublished
paper, Princeton University, September.
Krueger, Joel T., and Kenneth N. Kuttner. 1996. “The Fed Funds Futures
Rate as a Predictor of Federal Reserve Policy.” JOURNAL OF FUTURES
MARKETS 16, no. 8: 865-79.
McQueen, Grant, and V. Vance Roley. 1993. “Stock Prices, News, and Business
Conditions.” REVIEW OF FINANCIAL STUDIES 6, no. 3: 683-707.
Morse, Dale. 1981. “Price and Trading Volume Reaction Surrounding
Earnings Announcements: A Closer Examination.” JOURNAL OF
ACCOUNTING RESEARCH 19, no. 2: 374-83.
Pakko, Michael R., and David C. Wheelock. 1996. “Monetary Policy and
Financial Market Expectations: What Did They Know and When Did
They Know It?” Federal Reserve Bank of St. Louis REVIEW, JulyAugust: 19-32.
Pearce, Douglas K., and V. Vance Roley. 1985. “Stock Prices and Economic
News.” JOURNAL OF BUSINESS 58, no. 1: 49-67.
Prag, Jay. 1994. “The Response of Interest Rates to Unemployment
Rate Announcements: Is There a Natural Rate of Unemployment?”
JOURNAL OF MACROECONOMICS 16, no. 1: 171-84.
Roley, V. Vance. 1982. “Weekly Money Supply Announcements and the
Volatility of Short-Term Interest Rates.” Federal Reserve Bank of
Kansas City ECONOMIC REVIEW, April: 3-15.

Roley, V. Vance, and Gordon H. Sellon. 1996. “The Response of the Term
Structure of Interest Rates to Federal Funds Rate Target Changes.”
Federal Reserve Bank of Kansas City Research Working Paper
no. 96-08, December.
Roley, V. Vance, and Rick Troll. 1983. “The Impact of New Economic
Information on the Volatility of Short-Term Interest Rates.” Federal
Reserve Bank of Kansas City ECONOMIC REVIEW, February: 3-15.
Roley, V. Vance, and Carl E. Walsh. 1985. “Monetary Policy Regimes, Expected
Inflation, and the Response of Interest Rates to Money Announcements.”
QUARTERLY JOURNAL OF ECONOMICS 100, Supplement: 1011-39.
Schwert, G. William. 1981. “The Adjustment of Stock Prices to Information about Inflation.” JOURNAL OF FINANCE 36, no. 1: 15-29.
Smirlock, Michael. 1986. “Inflation Announcements and Financial Market
Reaction: Evidence from the Long-Term Bond Market.” REVIEW OF
ECONOMICS AND STATISTICS 68: 329-33.
Strongin, Steven, and Vefa Tarhan. 1990. “Money Supply Announcements
and the Market’s Perception of Federal Reserve Policy.” JOURNAL OF
MONEY, CREDIT, AND BANKING 22, no. 2: 135-53.
Thornton, Daniel L. 1989. “The Effect of Unanticipated Money on the
Money and Foreign Exchange Markets.” JOURNAL OF INTERNATIONAL
MONEY AND FINANCE 8: 573-87.
Urich, Thomas, and Paul Wachtel. 1981. “Market Response to the Weekly
Money Supply Announcements in the 1970s.” JOURNAL OF FINANCE 36,
no. 5: 1063-72.

———. 1984. “The Effects of Inflation and Money Supply
Announcements on Interest Rates.” JOURNAL OF FINANCE 39, no. 4:
1177-88.
Wall Street Journal. 1991. “Several Firms Plan to Start Service on Bond
Prices.” June 12.
Woodruff, Catherine S., and A. J. Senchack. 1988. “Intraday Price-Volume
Adjustments of NYSE Stocks to Unexpected Earnings.” JOURNAL OF
FINANCE 43, no. 2: 467-91.

———. 1983. “The Response of Short-Term Interest Rates to Weekly
Money Announcements.” JOURNAL OF MONEY, CREDIT, AND
BANKING 15, no. 3: 344-54.

50

FRBNY ECONOMIC POLICY REVIEW / DECEMBER 1997

NOTES

Is There an Inflation Puzzle?
Cara S. Lown and Robert W. Rich*

H

istorically, inflation has followed a fairly
predictable course in relation to the business cycle. Inflation typically rises during
an economic expansion, peaks slightly after
the onset of recession, and then continues to decline
through the first year or two of recovery. During the
present U.S. expansion, however, inflation has taken a
markedly different path. Although more than six years
have passed since the 1990-91 recession, inflation in the
core CPI (the consumer price index excluding its volatile
food and energy components) has yet to accelerate (Chart 1).
Moreover, during the last three years, inflation has
remained stable despite projections of higher expected
inflation from the Blue Chip Consensus forecast and
contrary to traditional signals such as the run-up in commodity prices experienced from late 1993 to early 1995.

* Cara S. Lown is a research officer and Robert W. Rich an economist at
the Federal Reserve Bank of New York.

Economists and policymakers have referred to the
restrained behavior of prices during this long expansion as
an “inflation puzzle.” In a recent interview, Robert T.
Parry, president of the Federal Reserve Bank of San Francisco, commented, “I have a question mark, and it leads me
to recommend vigilance with regard to inflation, but I do
have to note that things have turned out well. . . . [We’ve]
either been lucky, in which case the old relationships will
reassert themselves, or [we’ve] got a new regime under way.
And I don’t think we know enough at this point to know
which of those two things is operative.”1 As Parry suggests, two different types of explanations could account for
the recent behavior of inflation. The failure of inflation to
accelerate may reflect the effects of temporary factors
unique to this expansion. Alternatively, the unexpectedly
low level of inflation may indicate a permanent change in
the way inflation reacts to economic growth and other
related variables.
Each of these explanations holds important implications for the conduct of monetary policy. The Phillips

FRBNY ECONOMIC POLICY REVIEW / DECEMBER 1997

51

curve, the principal tool used by economists to explain
inflation, has been subject to systematic overprediction
errors during the past few years. If these errors reflect the
influence of temporary factors, then the Phillips curve
relationship should ultimately regain its stability. However, if these errors reflect a permanent change in the
dynamics of the inflation process, then economists
could no longer view the Phillips curve as a reliable guide
in forecasting inflation.
Because labor costs are an important factor in
determining prices, the recent slowdown in compensation
growth has been cited in both types of explanations for the
inflation puzzle. Some commentators argue that this slowdown in compensation growth, attributable largely to declining benefit costs, has acted as a supply shock and has
temporarily lowered inflation relative to its historical proximate determinants. Others contend that a permanent
change in compensation growth, resulting from heightened
job insecurity and its constrictive effect on wage growth,
has led to a fundamental shift in the inflation process.
This article explores the inflation puzzle and
investigates whether compensation has acted as either a
temporary restraint on inflation or as the underlying source
of a new inflation regime.2 After reviewing the recent

Chart 1

Core CPI
Percentage Change from a Year Ago
Percent
15

12

9

6

behavior of inflation, we specify and estimate a traditional
price-inflation Phillips curve model over the 1965-96
period. Our results show that in late 1993 the model
begins to systematically overpredict inflation and appears
to break down.
We then modify our traditional Phillips curve
specification by incorporating compensation growth as an
additional determinant of inflation. With this variable, the
model’s explanatory power improves significantly, and it
tracks inflation much more accurately over the current

Our findings indicate that compensation growth
has been weak during this expansion, especially
from late 1992 through early 1995, a period
that corresponds to the observed breakdown in
our traditional Phillips curve specification.

expansion. The restored stability of the model appears to
rule out the view that inflation’s recent behavior reflects a
fundamental shift in the inflation process.
Finally, we specify and estimate a wage-inflation
Phillips curve model quantifying the restraint in compensation growth over the post-1991 period. Our findings
indicate that compensation growth has been weak during
this expansion, especially from late 1992 through early
1995, a period that corresponds to the observed breakdown
in our traditional Phillips curve specification. This coincidence further supports our conclusion that compensation’s
slow growth has temporarily restrained inflation during
this expansion.

3

0
1961

65

70

75

80

85

90

95

Source: U.S. Department of Labor, Bureau of Labor Statistics.
Note: Shaded areas indicate periods designated recessions by the National
Bureau of Economic Research.

52

FRBNY ECONOMIC POLICY REVIEW / DECEMBER 1997

THE EMERGENCE OF THE INFLATION
PUZZLE
Contrary to expectations, inflation has not accelerated since
the end of the 1990-91 recession. Yet variables commonly
regarded as inflation indicators have remained at levels that

usually coincide with an inflation pickup. The level of the
actual unemployment rate relative to the nonaccelerating
inflation rate of unemployment (NAIRU) is one such variable. The NAIRU represents the rate of unemployment
that is consistent with stable inflation. Unemployment
rates below (above) the NAIRU are thought to signal
higher (lower) inflation in wages and prices. As the upper
panel of Chart 2 shows, the unemployment rate has been
below 6 percent—the consensus estimate of the NAIRU at
the beginning of this expansion—since late 1994. Even if
the NAIRU has declined below 6 percent during the
1990s, as some analysts argue, there is little direct evidence
suggesting that it has tracked the unemployment rate
or fallen low enough to be consistent with the level of
inflation observed since 1995.3
Like the NAIRU, the capacity utilization rate has
stayed at levels that typically signal higher future inflation
(bottom panel of Chart 2). In the past, capacity utilization

Chart 2

Unemployment and Capacity Utilization Rates
Percent
8.0
Unemployment Rate
7.5
7.0

rates in excess of 82 to 84 percent were associated with rising inflation because of the onset of supply shortages and
bottlenecks in production (Boldin 1996). Capacity utiliza-

The Phillips curve’s recent failure in
forecasting price changes contrasts sharply with
its long-standing reliability in predicting
short-run movements in inflation.

tion has moved down from its peak of almost 85 percent;
still, it has stayed above or close to 83 percent since 1994.
Consistent with these two indicators, the Blue
Chip Consensus forecast overpredicted inflation from 1992
to 1995 by progressively larger margins of error each year
(Chart 3). Estimated price-inflation Phillips curves have
also systematically overpredicted inflation in the past
couple of years. The Phillips curve’s recent failure in
forecasting price changes contrasts sharply with its longstanding reliability in predicting short-run movements in
inflation. We now turn to a discussion of the Phillips curve
and its recent record in forecasting inflation.

6.5
6.0

Chart 3

5.5

CPI Inflation, Actual and Forecast

5.0
86

Percent
4.5

Capacity Utilization Rate

84
4.0
82

Forecast
3.5

80
78
1986

87

88

89

90

91

92

93

94

95

96

Sources: Board of Governors of the Federal Reserve System; U.S. Department
of Labor, Bureau of Labor Statistics.
Notes: The dashed line marks the level at which unemployment or capacity
utilization will likely begin to exert upward pressure on inflation. The period
from the third quarter of 1990 to the first quarter of 1991, shaded in the chart,
is designated a recession by the National Bureau of Economic Research.

3.0
Actual
2.5
1991

92

93

94

95

96

97

Sources: Blue Chip Economic Indicators, various December issues; U.S.
Department of Labor, Bureau of Labor Statistics.

FRBNY ECONOMIC POLICY REVIEW / DECEMBER 1997

53

A TRADITIONAL PRICE-INFLATION
PHILLIPS CURVE
The origin of the Phillips curve can be traced back to
the 1950s, when A.W. Phillips documented an inverse
relationship between the rate of change of nominal
wages and the level of unemployment in the United
Kingdom. His findings were interpreted as establishing
a wage adjustment process in which low levels of unemployment represent tight labor markets that signal, or
coincide with, accelerating wage growth. Although the
term “Phillips curve” still refers to the posited relationship between nominal wage or price changes and various
indicators of real economic activity, the econometric
modeling of this relationship has changed considerably
over the years.4
Modern versions of the Phillips curve incorporate
several features that differentiate them from earlier descriptions of the behavior of nominal wages and prices.5 For
example, in current models the output gap (the log ratio of

[Our specification for the traditional Phillips
curve] embodies the “triangle” model of
inflation: the set of explanatory variables is
meant to capture the effects of demand, inertia,
and supply considerations on inflation.

actual to potential real GDP) and the unemployment gap
(the difference between the actual rate of unemployment
and the NAIRU) figure importantly as measures of excess
aggregate demand pressure in the economy. In addition,
current models recognize the role that expected inflation
plays in wage bargaining and price setting and typically
include past rates of inflation as a proxy for this expectation.6 Finally, modern Phillips curve models include
variables to control for supply shocks such as the oil price
increases of the 1970s. As Fuhrer (1995) notes, many of
these developments were anticipated by Phillips in his
original discussion.

54

FRBNY ECONOMIC POLICY REVIEW / DECEMBER 1997

We begin our empirical analysis by specifying a
traditional price-inflation Phillips curve model. The model
allows for a more formal investigation of the stability of the
Phillips curve relationship during the current expansion.
In addition, the model will serve as a benchmark to
evaluate compensation growth’s role in explaining recent
movements in inflation.
Our traditional Phillips curve model is given by:
(1) INFt = α0 + α1GDPGAPt – 1 + α2( ∆GDPGAPt – 1)
+

3

2

i=1

i=1

∑ α2 + iINFt – i + ∑ α5 + iOILGt+– i + εt ,

where
INF = inflation measured by the growth rate of the
core CPI,
GDPGAP = the output gap measured by the log ratio of
actual to potential real GDP,
∆ GDPGAP = the first difference or change in the output
gap,
+
OILG = the net positive change in the real price of
oil, and
ε = a mean zero, serially uncorrelated random
disturbance term.
Equation 1 provides a general specification for the rate of
change in prices and is similar to other models currently
used in the Phillips curve literature.7 In the terminology of
Gordon (1996), the specification embodies the “triangle”
model of inflation: the set of explanatory variables is meant
to capture the effects of demand, inertia, and supply considerations on inflation.
The model uses the output gap (the percentage
deviation of real GDP from potential GDP), shown in
Chart 4, as a measure of excess aggregate demand pressure.8
A positive (negative) output gap indicates that the economy is operating above (below) potential GDP and would
thus generate upward (downward) inflationary pressure on
prices. Following the methodology in Gordon (1977,
1996) and Fuhrer (1995), we also include the quarterly
change in the output gap variable to allow for a rate-ofchange effect so that the pressure on prices depends on how
quickly the output gap narrows or widens.

The remaining basic determinants of inflation
include its own lagged values and oil prices. To incorporate
price inertia effects, we include lagged inflation terms in
the model. In the past, researchers used lagged inflation
rates as a proxy for expected inflation. In modern versions
of the Phillips curve, however, this interpretation has been
deemed overly restrictive (Gordon 1996). Instead, past
inflation rates are viewed as capturing the dynamics of
price adjustment related to expectations formation as well
as the importance of institutional factors such as wage and
price contracts and delivery lags in the economy.
Our benchmark model also includes a measure of
the net positive change in real oil prices to account for the
influence of supply shocks.9 This oil price variable is the
only notable departure from other conventional Phillips
curve specifications and allows for an asymmetric effect of
oil price changes on inflation (Chart 5). In other words,
while oil price increases appear to affect inflation, oil price
decreases do not seem to be important.10 The construction
of the supply shock variable follows the approach in
Hamilton (1996) and is designed not only to model the
asymmetric effects of oil price changes, but also to account
for the observed increase in the volatility of oil prices over
the post-1986 period. Because the core CPI has no energy

Chart 4

The Output Gap
Percentage Difference between Actual and Potential GDP

Chart 5

Net Positive Change in Real Oil Prices
Percent
60
50
40
30
20
10
0
1965

70

75

80

85

90

95

Source: Authors’ calculations, based on Department of Energy, Monthly
Energy Review.
Note: Shaded areas indicate periods designated recessions by the National
Bureau of Economic Research.

price component, our supply shock variable attempts to
capture any indirect effect of oil price increases on inflation.
Although our traditional price-inflation Phillips
curve takes real oil prices as exogenous, we include only
lagged values of the output gap as regressors in order to
avoid simultaneity bias arising from the endogeneity of
this variable. The lag lengths in equation 1 are selected by
maximizing adjusted R2 (a measure of the model’s ability
to explain inflation), by searching over one to four lags of
inflation and the output gap, and by searching over zero to
four lags for the net positive change in the real price of oil.11

Percent
8

MODEL ESTIMATION

6
4
2
0
-2
-4
-6
-8
-10
1965

70

75

80

85

90

95

Source: Federal Reserve Bank of New York staff estimate.
Note: Shaded areas indicate periods designated recessions by the National
Bureau of Economic Research.

We estimate equation 1 using the method of ordinary least
squares (OLS) for quarterly data from the first quarter of
1965 to the third quarter of 1996. Parameter estimates are
presented in Table 1. For the full sample period, the value
of the adjusted R2 indicates that the model can explain a
high proportion of the variation in inflation. In addition,
the Ljung-Box (1978) Q-test statistic—a general test for
serial correlation in the regression residuals—does not
reveal any evidence of model misspecification.
The estimation results also indicate that both the
level of the output gap variable and the rate-of-change effect

FRBNY ECONOMIC POLICY REVIEW / DECEMBER 1997

55

Table 1
TRADITIONAL AND
CURVE MODELS

MODIFIED PRICE-INFLATION PHILLIPS
Traditional Model

Modified Model

Variable
CONSTANT

Estimate
(0.0786
(0.0782)

p-Value
0.3146

Estimate
(0.0532*
(0.0720)*

p-Value
0.4601

GDPGAPt-1

(0.0339**
(0.0107)**

0.0016

(0.0190
(0.0108)*

0.0783

∆ GDPGAPt-1

(0.1452**
(0.0511)**

0.0045

(0.2620**
(0.0537)*

0.0000

INFt-1

(0.4080**
(0.1209)**

0.0007

(0.2610*
(0.1064)*

0.0142

INFt-2

(0.1296
(0.1168)**

0.2672

(0.1252
(0.1046)*

0.2312

INFt-3

(0.3487**
(0.1227)**

0.0045

(0.2913**
(0.1011)*

0.0040

OILGt-1

(0.0186**
(0.0056)**

0.0009

(0.0167**
(0.0046)*

0.0003

+
OILGt-2

(0.0242**
(0.0071)**

0.0007

(0.0228**
(0.0058)*

0.0001

The dynamic simulation provides strong evidence
of instability in the traditional price-inflation

UNITG+
t-1

—

—

(0.1901**
(0.0380)*

0.0000

UNITGt-2

—

—

(0.0732
(0.0390)*

0.0609

Memo:
Adjusted R2
Q-test statistic

0.776
22.731
(0.859)

parameters against the alternative hypothesis of a onetime
shift in the parameters at some specified date. One test
compares the estimates obtained using the data from one
subperiod (1965-91) with the estimates using the full
sample.12 Another test employs dummy variables for the

0.815
27.572
(0.643)

Notes: Asymptotic standard errors for the parameter estimates are reported
in parentheses and are computed using the procedure of White (1980). The
Ljung-Box Q-test statistic for serial correlation of the regression residuals
is distributed asymptotically as χ 2 with thirty-one degrees of freedom.
Probability values for the test statistics are reported in parentheses.
*Significant at the 5 percent level.
**Significant at the 1 percent level.

are highly significant and have the expected positive signs.
The two lagged values of the net positive change in the real
price of oil are also highly significant with the anticipated
positive signs. The three lags of the inflation rate are
generally significant, and we are unable to reject the
hypothesis that the sum of the coefficients equals unity
( α3 + α4 + α5 = 1 ) at conventional significance levels. The
latter restriction follows from the natural rate hypothesis and
has been previously imposed in the estimation of Phillips
curves to make the level of potential output (or the unemployment rate) independent of inflation in the long run.

Phillips curve during the current expansion.

entire parameter vector for one subperiod (1992-96) and
then tests the joint significance of the dummy variables.13
As shown by the reported value of the two test statistics in
Table 2, we fail to reject the null hypothesis of parameter
stability for the post-1991 period at conventional significance levels.14
As a second exercise, we construct dynamic
out-of-sample forecasts from the traditional price-inflation
Phillips curve. This simulation provides a more stringent
test of model stability by relying on lagged predicted
values of inflation rather than the lagged actual values of
inflation to construct the subsequent one-quarter-ahead
forecasts of inflation. In addition, the Chow tests may
suffer from low power because they are conducted over a
relatively small part of the sample period (1992-96). For
this part of the analysis, we estimate equation 1 using data
from the first quarter of 1965 through the fourth quarter of
1991. We then use the estimated equation to forecast
inflation over the 1992-96 period.

Table 2
TRADITIONAL AND MODIFIED
Chow Test Results for 1992-96

F-Statistic

Likelihood Ratio
Statistic

Traditional Phillips curve

0.192)
(0.999)

4.539)
(0.999)

Modified Phillips curve

0.244)
(0.999)

5.860)
(0.998)

Model

MODEL STABILITY OVER THE 1992-96 PERIOD
We conduct two exercises to examine the stability of the
model from 1992 to 1996. First, we apply Chow (1960)
split-sample tests to test the null hypothesis of constant

56

FRBNY ECONOMIC POLICY REVIEW / DECEMBER 1997

PHILLIPS CURVE MODELS

Note: Probability values for the test statistics are reported in parentheses.

The dynamic simulation provides strong evidence
of instability in the traditional price-inflation Phillips
curve during the current expansion (Chart 6). Specifically,
the out-of-sample forecasts systematically overpredict
inflation beginning in the third quarter of 1993. In
addition, the forecasted inflation series is characterized by
a rising trend and generates prediction errors that
increase over time. This excerise is robust to the choice of
starting dates.15
The results of our dynamic simulation appear to
show a shift in the Phillips curve relationship and are
consistent with commentators’ claims that inflation has
remained unexpectedly low during this expansion. We now
examine the role of compensation growth in the recent
behavior of inflation.

EXAMINING THE ROLE OF COMPENSATION
GROWTH
Because labor costs represent about two-thirds of the total
cost of production, some economists have suggested that
inflation’s recent behavior may be linked to movements in
compensation growth and its two components, benefits
and wages (Chart 7). Since the end of the 1990-91 reces-

sion, the growth rates for total compensation, benefits, and
wages have not only failed to display any significant
acceleration, but have generally displayed a downward
trend. This downward trend is particularly apparent for
benefit costs, where the four-quarter change has fallen from
6 percent to about 2 percent during the 1990s. These
observed patterns support the view that labor costs may be a
key factor in understanding recent movements in inflation.
Meyer (1997), for example, poses two explanations
relating compensation growth to inflation’s puzzling
behavior. First, he suggests that declining benefit costs
have caused a temporary slowdown in compensation
growth, which has acted as a supply shock. By lowering
the increase in overall labor costs, this shock has reduced
the pressure on firms to raise prices. Because most priceinflation Phillips curves exclude the effects of compensation
growth altogether, their forecasting ability appears to
break down and the models overpredict inflation.
Alternatively, Meyer suggests, the slowdown in
compensation growth may reflect a long-term change in
the behavior of the labor market. In particular, Meyer questions whether heightened job insecurity has permanently
diminished workers’ ability to obtain wage increases and has
consequently altered the link between changes in com-

Chart 6

Out-of-Sample Forecast of Core CPI Inflation

Chart 7

Traditional Phillips Curve Model

Employment Cost Index for Private Industry
Percentage Change from a Year Ago

Percent
6

Percent
15

Forecast
5

12
4

9
Actual

Compensation
6

3

Benefits

3

2
1986

87

88

89

90

91

92

93

94

95

Wages and salaries

96
0
85

Sources: Authors’ calculations; U.S. Department of Labor, Bureau of Labor
Statistics.

1981

Note: The period from the third quarter of 1990 to the first quarter of 1991,
shaded in the chart, is designated a recession by the National Bureau of
Economic Research.

Source: U.S. Department of Labor, Bureau of Labor Statistics.

90

95

Note: Shaded areas indicate periods designated recessions by the National
Bureau of Economic Research.

FRBNY ECONOMIC POLICY REVIEW / DECEMBER 1997

57

pensation (and other macroeconomic variables) and price
changes. According to this view, the recent breakdown in
price-inflation Phillips curves reflects a fundamental shift in
the inflation process emanating from the labor market.16
Although we do not look at the decline in benefit
costs or the behavior of wages individually, we investigate
the role of total compensation growth in restraining infla-

Since the end of the 1990-91 recession, the
growth rates for total compensation, benefits,
and wages have not only failed to display any
significant acceleration, but have generally
displayed a downward trend. This downward
trend is particularly apparent for benefit costs.

tion.17 Our methodology is designed to evaluate whether
this role has been temporary or permanent in nature.
If compensation growth has acted as a temporary
supply shock, we would expect the forecasting performance
and the stability of the Phillips curve over the current
expansion to be restored by incorporating the effects of
compensation growth. Moreover, because a “shock” implies
an unexpected event, we would also likely observe some
evidence of unusual restraint in the recent behavior of compensation growth. However, if a change in the behavior of
compensation growth has permanently altered the Phillips
curve relationship, we should find evidence of a breakdown, rather than stability, in the relationship between
the inflation process and compensation growth during the
current expansion. We now turn to our modified Phillips
curve equation.

MODIFYING THE TRADITIONAL MODEL
Within our Phillips curve framework, we include the
growth rate of unit labor costs—compensation (benefits and
wages) divided by productivity—as an additional determi-

58

FRBNY ECONOMIC POLICY REVIEW / DECEMBER 1997

nant of inflation. Unit labor costs provide a measure of
compensation that controls for the effects of productivity.18
During this expansion, growth in unit labor costs
has been weak and a persistent gap has been evident
between unit labor cost growth and core CPI inflation
(Chart 8). The decline in unit labor cost growth could
suggest either falling compensation growth or rising
productivity growth. As Chart 9 shows, however, productivity growth has not been unusually strong in the current
expansion. Although from late 1991 to early 1992 the
series rose at roughly a 3 percent rate, contributing to
weaker growth in unit labor costs, since then productivity
has typically grown at rates below 1 percent.
By contrast, compensation growth fell to
around 2 percent fairly early in the expansion and hovered
around that rate for more than two years before showing
signs of a modest pickup. This 2 percent growth rate is
below any rate recorded in the past thirty-five years. Thus,
we can conclude that the growth rate of unit labor costs
over the post-1991 period has been primarily driven by slow
compensation growth rather than high productivity growth.
This finding ensures that our approach will pick up the
effect of slow compensation growth, not the effect of high
productivity growth, on inflation during this expansion.

Chart 8

Core CPI and Unit Labor Costs
Percentage Change from a Year Ago
Percent
15
Core CPI
10

5

0
Unit labor costs
-5
1961

65

70

75

80

85

90

95

Source: U.S. Department of Labor, Bureau of Labor Statistics.
Note: Shaded areas indicate periods designated recessions by the National
Bureau of Economic Research.

Our modified price-inflation Phillips curve model
is given by:
INFt = α0 + α1GDPGAPt – 1 + α2( ∆GDPGAPt – 1)

(2)

3

+

2

∑ α2 + iINFt – i + ∑ α5 + iOILGt+– i
i=1

i=1

2

+

∑ α7 + iUNITGt – i + εt ,
i=1

where UNITG is the growth rate of unit labor costs in
the nonfarm business sector. In our modified model, unit
labor costs provide an explicit channel by which slow
compensation growth may have acted to offset other
sources of inflationary pressures over the current expansion,
resulting in lower inflation rates than those predicted
using the traditional model.19

MODEL ESTIMATION
We estimate equation 2 by the method of OLS using
quarterly data from the first quarter of 1965 to the third
quarter of 1996. Parameter estimates are presented in
Table 1. The two lagged values of unit labor cost growth
enter with the anticipated positive sign. The inclusion of
the unit labor cost terms improves the fit of the model over

the full sample period by almost 5 percent relative to the
traditional model, and the Q-test statistic does not suggest evidence of model misspecification.
The results for all other explanatory variables are
broadly similar across the traditional and modified models,
although the modified Phillips curve suggests that the output gap has a smaller level effect and a larger rate-of-change
effect on core CPI inflation. Like the traditional model, the
estimated version of the modified model does not constrain
the sum of the coefficients on lagged inflation to equal
unity ( α3 + α4 + α5 = 1 ). As shown in the Equation Appendix, however, we can eliminate compensation growth from
the system consisting of equation 2 and our estimated
wage-inflation Phillips curve to yield a reduced form of a
price-inflation Phillips curve. The resulting model is characterized by coefficients on lagged inflation whose sum is
not statistically different from unity, and it associates an
acceleration in inflation with a positive output gap and a
negative unemployment gap.

MODEL STABILITY OVER THE 1992-96 PERIOD
Does the inclusion of unit labor costs and the effects of
compensation growth correct the instability of our benchmark model over the post-1991 period? An examination
of the dynamic simulation for the modified price-inflation Phillips curve suggests that it does (Chart 10). 20

Chart 9

Once we incorporate the effects of unit labor costs in the

Productivity and Hourly Compensation

model, the simulated values track inflation closely over

Percentage Change from a Year Ago

the post-1991 period and display no significant sign of
Percent
15

model instability. Despite a notable error in the fourth
quarter of 1995, the equation regains its predictive accuracy over the next two quarters. 21 Because the dynamic

10
Hourly compensation

simulation uses forecasted values of inflation, however,
the error in the fourth quarter of 1995 continues to affect

5

the subsequent quarters’ forecasts and contributes to the
error in the third quarter of 1996.

0
Productivity
-5
1961

65

70

75

80

85

90

Source: U.S. Department of Labor, Bureau of Labor Statistics.
Note: Shaded areas indicate periods designated recessions by the National
Bureau of Economic Research.

95

Overall, the evidence from the modified priceinflation Phillips curve is compelling. Indeed, slow
compensation growth appears to be a key force in
restraining inflation over the current expansion. By
including unit labor costs as an additional explanatory

FRBNY ECONOMIC POLICY REVIEW / DECEMBER 1997

59

wage-inflation Phillips curve proposed by Englander and
Los (1983):
2

Chart 10

Out-of-Sample Forecast of Core CPI Inflation
Modified Phillips Curve Model

(3)

LXNGt = β0 +

Percent
6

3

+
Actual

4

3

2
88

89

90

91

92

93

94

95

96

Sources: Authors’ calculations; U.S. Department of Labor, Bureau of Labor
Statistics.
Note: The period from the third quarter of 1990 to the first quarter of
1991, shaded in the chart, is designated a recession by the National Bureau
of Economic Research.

variable, the multiperiod forecast performance of the
model improves dramatically, and we seem to eliminate
the sharp divergence between actual and predicted inflation. Thus, the restored stability of the model resulting
from the inclusion of unit labor costs appears to rule
out the view that inflation’s recent behavior reflects a
fundamental shift in the Phillips curve relationship. The
analysis, however, has yet to provide any specific insights
into compensation growth and its recent behavior. We
explore these issues in the next section.

THE BEHAVIOR OF COMPENSATION
GROWTH
The results from our modified price-inflation Phillips
curve reveal compensation growth’s role in lowering inflation since 1991. In this section, we analyze compensation’s
level of restraint compared with expected levels during
the present expansion. The comparison allows us to determine if the recent slowdown in compensation growth has
been particularly severe. We show that while restraint in
compensation growth appears to be easing, compensation
growth was unexpectedly low from late 1992 to early 1995.
To analyze the behavior of compensation growth,
we specify a model that represents a modified version of the

60

∑ β3 + iINFt – i + β7SOCt + β8UIRt – 1
+ β9DUMt + ηt ,

Forecast

87

i=1

i=1

5

1986

∑ βiLXNGt – i + β3Ut – 1

FRBNY ECONOMIC POLICY REVIEW / DECEMBER 1997

where
LXNG= the growth rate of compensation per hour in
the nonfarm business sector,
U = the unemployment rate for males aged
twenty-five to fifty-four,
INF = inflation measured by the growth rate of the
CPI (all items, urban consumers),
SOC = the change in employer Social Security
contributions,
UIR = the income replacement ratio from
unemployment insurance benefits,
DUM = dummy variable for the wage and price
controls of the 1970s, and
η = a mean zero, serially uncorrelated random
disturbance term.
Equation 3 principally links the movements in compensation growth to the unemployment rate and other labor
market variables.22 The unemployment rate of prime-age
males is used as a measure of labor market tightness. We
enter the variable in its level form and thereby abstract
from any explicit discussion of the NAIRU, except to note
that the specification can be viewed as implicitly assuming
a constant value for the NAIRU over the sample period.23
Equation 3 does not include a rate-of-change effect for the
unemployment rate; the estimated coefficient on a second
lag of the unemployment rate was found to be quantitatively and statistically insignificant and therefore was
omitted from the specification.24
The remaining determinants of compensation
growth include the change in employer Social Security tax
contributions, a component of hourly compensation. The
income replacement ratio from unemployment insurance
benefits attempts to capture changes in compensation
growth related to job search. A dummy variable accounts

for the restraining effect of wage and price controls in the
fourth quarter of 1971 and for the rebound effect after the
relaxation of the controls in the first quarter of 1972.25 We
include lagged values of compensation growth and price

While restraint in compensation growth

generate predicted values for compensation growth over
the 1992-96 period.
The evidence from the dynamic simulation
indicates that compensation growth has displayed unexpected restraint during this expansion. The out-of-sample
forecasts consistently overpredict compensation growth
beginning in the fourth quarter of 1992 (Chart 11). In addi-

appears to be easing, compensation growth
was unexpectedly low from late 1992 to
early 1995.

Table 3
WAGE-INFLATION PHILLIPS CURVE
FOR COMPENSATION GROWTH
Variable

Estimate

p-Value

CONSTANT

(0.3884**
(0.2155)**
(0.1359**
(0.0861)**

0.0715

LXNGt-1

inflation to incorporate wage and price inertia effects.
Finally, we include only lagged values of the unemployment rate and inflation rate as regressors because of
endogeneity considerations.

MODEL ESTIMATION AND MODEL STABILITY OVER
THE 1992-96 PERIOD
We estimate equation 3 using the method of OLS for
quarterly data from the second quarter of 1967 to the third
quarter of 1996. The parameter estimates are presented in
Table 3. As the table indicates, the lagged values of both
compensation growth and price inflation are generally significant. The unemployment rate is highly significant and
has the expected negative sign. Further, the variables
reflecting other labor market conditions are all significant
with the expected signs. The adjusted R2, although not
quite as high as the values reported in Table 1, also
indicates that the estimated equation fits the data quite
well over the full sample period. In addition, the regression
residuals display little evidence of serial correlation over
the full sample period.
We also conduct Chow tests and a dynamic simulation. The Chow tests do not reject the null hypothesis of
parameter stability at conventional significance levels
(Table 4). For the dynamic simulation, we estimate
equation 3 from the second quarter of 1967 to the fourth
quarter of 1991; we then use the estimated equation to

MODEL

0.1144

LXNGt-2

(0.2621**
(0.0689)**

0.0001

Ut-1

-0.0672**
(0.0218)**

0.0021

INFt-1

(0.2018**
(0.0692)**

0.0036

INFt-2

(0.0175
(0.0832)**

0.8332

INFt-3

(0.1257
(0.0698)**

0.0720

SOCt

(0.0849**
(0.0186)**

0.0000

UIRt-1

(1.4288**
(0.6666)**

0.0321

DUMt

-0.7442**
(0.0790)**

0.0000

Memo:
Adjusted R2
Q-test statistic

(0.709**
28.109
(0.838)**

Notes: Asymptotic standard errors for the parameter estimates are computed
using the procedure of White (1980) and are reported in parentheses. The
Ljung-Box Q-test statistic for serial correlation of the regression residuals is
distributed asymptotically as χ 2 with twenty-nine degrees of freedom.
Probability values for the test statistics are reported in parentheses.
*Significant at the 5 percent level.
**Significant at the 1 percent level.

Table 4
COMPENSATION GROWTH
Chow Test Results for 1992-96
Model
Compensation growth
Phillips curve

MODEL

F-Statistic

Likelihood Ratio
Statistic

(0.879
(0.609)

20.287
(0.377)

Note: Probability values for the test statistics are reported in parentheses.

FRBNY ECONOMIC POLICY REVIEW / DECEMBER 1997

61

tion, the size of the errors at times is quite large. For
example, our dynamic simulation predicts that compensation growth should have been about 2 percent higher from
the end of 1992 through the end of 1994. After 1994,
however, the size of the forecast errors begins to diminish, a
pattern that supports the temporary supply shock hypothesis. If a permanent change in compensation growth had
occurred, we would expect the large disparity between the
model’s simulated values and actual growth to continue,
as it did in the traditional price-inflation Phillips curve
model.
Evidence from the dynamic simulation corroborates our earlier finding that the modified price-inflation
Phillips curve model, which incorporates the effects of
compensation growth, appears to resolve the inflation puzzle. The slowdown in compensation growth is most pronounced from the end of 1992 to early 1995, the same
period during which the traditional Phillips curve starts to
display evidence of model instability. Thus, not surprisingly, variables and relationships that ignore compensa-

Chart 11

Out-of-Sample Forecast of Compensation Growth
Percent
8

6
Forecast
4

2
Actual
0
1986

87

88

89

90

91

92

93

94

95

96

Sources: Authors’ calculations; U.S. Department of Labor, Bureau of Labor
Statistics.
Note: The period from the third quarter of 1990 to the first quarter of
1991, shaded in the chart, is designated a recession by the National Bureau
of Economic Research.

62

FRBNY ECONOMIC POLICY REVIEW / DECEMBER 1997

tion growth’s influence (such as the inflation indicators in
Charts 2 and 3 and the traditional Phillips curve) begin to
break down in late 1993 and 1994.

CONCLUSION
Contrary to its behavior in previous expansions, price inflation
has not accelerated in the six years since the 1990-91 recession. This article focuses on compensation’s role in the
inflation puzzle, investigating whether a temporary slowdown
in compensation growth has lowered the level of inflation
or if a more permanent change in compensation growth has
fundamentally altered the inflation process. We present two
pieces of evidence suggesting that slow compensation growth
has acted as a temporary restraining force on inflation.
We begin our investigation by estimating a
traditional price-inflation Phillips curve model over the
1965-96 period. Although the model tracks inflation quite
well over most of the period, it begins to break down in
late 1993. We then modify the traditional Phillips curve
model to include the effects of compensation growth. With
this addition, the model tracks inflation much more
accurately over the current expansion and displays no
significant evidence of instability. This finding provides
the first piece of evidence suggesting that no fundamental
change in the inflation process has occurred.
To arrive at the second piece of evidence supporting the notion that the low level of inflation has resulted
from a temporary slowdown in compensation growth, we
look at compensation growth itself. By estimating a wageinflation Phillips curve model, we find that compensation
growth showed unusual restraint from late 1992 to early
1995. This period of restraint appears to be temporary and
coincides with the observed breakdown in the traditional
Phillips curve model and in other inflation indicators.
Thus, taking compensation growth into account appears to
explain inflation’s behavior during the current expansion.
Still uncertain, however, is the reason for the dramatic
slowdown in compensation growth during the early 1990s.
The solution to this puzzle must await further investigation.

EQUATION APPENDIX: DERIVATION OF THE ACCELERATIONIST PHILLIPS CURVE MODEL

This appendix briefly examines the derivation of the
accelerationist model of the Phillips curve from equations
2 and 3. The key features of this model can be illustrated
by examining the relationship between the output gap
(and the unemployment gap with a constant NAIRU) and
the inflation rate. Abstracting from the influence of other
terms, we note that the system of equations 2 and 3 can be
rewritten as
3
(4)

INFt = α1GDPGAPt – 1 +

i=1

2

+

∑ α2 + iINFt – i

∑ α7 + i(LXNGt – i) ,
i=1

and
(5)

3

β3 U t – 1 +

∑ β3 + iINFt – i

i=1
LXNGt = -------------------------------------------------------,
2
( 1 – β1 L – β 2 L )

where we substitute for the definition of the growth rate of
unit labor costs (compensation growth less productivity
growth) in equation 4, and L denotes the lag operator in
k
equation 5 such that L Xt = Xt – k .
We can substitute equation 5 into equation 4 to
obtain an expression relating current inflation to the
output gap, the unemployment gap, and past rates of
inflation. If the sum of the coefficients on lagged inflation
equals unity, then there is a “natural rate” value of the output gap (and unemployment gap) of zero that is consistent
with a constant rate of inflation. Alternatively, the model
would associate a permanent positive value for the output

APPENDIX

gap with an ever-accelerating inflation rate. Within our
system of equations, the condition that the sum of the
coefficients on lagged inflation equals unity is given by
(6)

( α8 + α 9 ) ( β4 + β 5 + β 6 )
α3 + α4 + α5 + --------------------------------------------------- =1 .
( 1 – β1 – β2 )

The hypothesis that the coefficients on lagged
inflation sum to unity can be tested using the OLS
estimates of equations 2 and 3 to construct estimates for
the expression on the left-hand side of equation 6 and its
standard error. The standard error is the standard error of a
function of several estimated parameters and can be
computed using the delta method approximation (Greene
1993, p. 297):
∂g- ⋅ VAR(θ ) ⋅ ----∂g- ,
SE[ g(θ)] = -----∂θ'
∂θ
where θ denotes the parameters in equation 6, g( θ) is
the function of the parameters in 6, and VAR ( θ ) is the
variance-covariance matrix of those parameters.
Because of the slight disparity in the sample
periods for Tables 1 and 2, we estimate equation 2 and
equation 3 from the second quarter of 1967 to the third
quarter of 1996. The estimate for the expression on the
left-hand side of equation 6 is 0.87, with an estimated
standard error of 0.08. Thus, we are unable to reject the
null hypothesis that the sum of the coefficients in equation 6 is equal to unity at the 5 percent significance level.

FRBNY ECONOMIC POLICY REVIEW / DECEMBER 1997

63

DATA APPENDIX

This appendix defines the variables and the data sources
used to estimate our traditional Phillips curve model, modified Phillips curve model, and compensation growth
model. All data in our analysis include revisions through
August 12, 1997.

INFLATION EQUATION VARIABLES
INF = the growth rate of the core CPI for all urban consumers as reported by the Department of Labor, Bureau of
Labor Statistics. Data are released monthly and are seasonally adjusted.
UNITG = the growth in unit labor costs for the nonfarm
business sector as reported by the Department of Labor,
Bureau of Labor Statistics. Data are released quarterly and
are seasonally adjusted.
GDPGAP = the logarithmic ratio of GDP to POTGDP,
where GDP equals quarterly real gross domestic product
and POTGDP, quarterly potential GDP. Both variables are
in 1987 dollars until the third quarter of 1987. They are
in chain-weighted 1992 dollars from the fourth quarter of
1987 to the present. The GDP data are from the National
Income and Product Accounts. Potential GDP is a Federal
Reserve Bank of New York staff estimate.
OILG+ = the net positive change in the real price of oil,
calculated as the percentage change in the current real
price of oil from the previous year’s maximum (if that
change is positive, zero otherwise). Data for the price of oil
are an extension of Mork’s (1989) series, which reflects corrections for the effects of price controls during the 1970s.
The real price of oil is defined as the nominal oil price
index deflated by the GDP deflator.

COMPENSATION EQUATION VARIABLES
LXNG = the growth rate of compensation per hour for the
nonfarm business sector as reported by the Department of
Labor, Bureau of Labor Statistics. Compensation comprises
wages and salaries for workers plus employers’ contributions for Social Security insurance and private benefit
plans. The series also includes an estimate of wages,
salaries, and supplemental payments for self-employed
workers. Data are released quarterly and are seasonally
adjusted.
INF = the growth rate of the CPI for all urban consumers
as reported by the Department of Labor, Bureau of Labor
Statistics. Data are released monthly and are seasonally
adjusted.
U = the unemployment rate for males aged twenty-five to
fifty-four as reported by the Department of Labor, Bureau
of Labor Statistics. Data are released monthly and are
seasonally adjusted.
UIR = unemployment insurance per job loser, normalized
by the average annual earnings of a manufacturing worker.
This variable can be thought of as a replacement ratio, that
is, the fraction of earnings of manufacturing workers
replaced by unemployment insurance payments. Manufacturing workers are the most likely workers to collect
unemployment insurance. UIR is constructed as (YPTU/
LUJL)/(YPWF/LAMANU), where
YPTU = government unemployment insurance
benefits according to the National Income and
Product Accounts. Data are reported quarterly and
are seasonally adjusted.
LUJL = job losers and persons who have completed
temporary jobs as reported by the Department of
Labor, Bureau of Labor Statistics. Data are released
monthly and are seasonally adjusted.

64

FRBNY ECONOMIC POLICY REVIEW / DECEMBER 1997

APPENDIX

DATA APPENDIX (Continued)

YPWF = wage and salary disbursements in
manufacturing according to the National Income
and Product Accounts. Data are reported quarterly
and are seasonally adjusted.
LAMANU = nonfarm payroll employees in
manufacturing as reported by the Department of
Labor, Bureau of Labor Statistics. Data are
reported monthly.

APPENDIX

SOC = a measure of the direct effect of changes in payroll
tax rates for Social Security and Medicare. The quarterly
data are Federal Reserve Bank of New York staff estimates.
DUM = 1 in the fourth quarter of 1971, -0.6 in the first
quarter of 1972, and 0 elsewhere. This variable accounts
for the restraining effect of the wage and price freeze in the
fourth quarter of 1971 and the rebound effect after the
wage and price controls were relaxed in the first quarter
of 1972.

FRBNY ECONOMIC POLICY REVIEW / DECEMBER 1997

65

ENDNOTES

The authors are grateful to J.S. Butler, Gabriele Galati, Steve Kamin, Jonathan
Mc Carthy, Richard Peach, Charles Steindel, and two anonymous referees for
helpful comments. We also benefited from the suggestions of conference participants
at the Bank for International Settlements. Beethika Khan provided excellent
research assistance.
1. Dow Jones News Service, January 7, 1997.
2. Our analysis expands on results that we presented in two earlier
papers. See Lown and Rich (1997a, 1997b).
3. Gordon (1996), however, obtains an estimate of 5.3 percent for the
NAIRU starting in 1996.
4. Gordon’s work (1970, 1975, 1977, 1982, 1990) is prominent in the
literature on the estimation of the Phillips curve.
5. See King and Watson (1994), Tootell (1994), Fuhrer (1995), King,
Stock, and Watson (1995), and Gordon (1996).
6. The estimation of “expectations-augmented” Phillips curves is the
result of work by Phelps (1967) and Friedman (1968), who developed the
natural rate hypothesis and drew the distinction between the short-run
and long-run Phillips curve trade-off.
7. For detailed definitions and sources of data, see the Data Appendix.
8. The results are little affected when the unemployment rate instead of
the output gap is used to measure aggregate demand pressure. Potential
GDP measures the full-employment level of output or the output level
at which there is no tendency for inflation to accelerate or decelerate. The
level of potential GDP grows over time because of the increased
availability of resources (land, labor force, capital stock, and the level of
technology). Because potential GDP is not directly observable, several
techniques have been developed to calculate estimates of the series. A
complete review of these techniques and an evaluation of the alternative
potential GDP series are beyond the scope of this paper. As noted in the
Data Appendix, we employ a staff estimate of potential GDP to construct
the output gap variable.
9. Commodity prices and/or an exchange rate term have been used as
supply shock variables in some price-inflation Phillips curve models. We
do not include these terms in our specification, however, because we
found their effects to be small and statistically insignificant. The absence
of a strong link between commodity prices and inflation is consistent
with evidence presented by Blomberg and Harris (1995), who document
a recent decline in the predictive power of commodity prices for inflation.

66

FRBNY ECONOMIC POLICY REVIEW / DECEMBER 1997

10. We exclude the net negative real oil price change variable from
equation 1 because the variable displays quantitatively and statistically
insignificant effects.
11. The compensation growth Phillips curve described later in the text
includes dummy variables to capture the effects from the imposition and
relaxation of wage and price controls during the 1970s. We exclude these
dummy variables from the traditional price-inflation Phillips curve
because they were found to be statistically insignificant. Alternative
dating schemes for the dummy variables (Gordon 1982) also proved to be
unimportant in explaining the dynamics of inflation during the 1971-75
period.
12. This test yields an F-statistic, which is distributed asymptotically as
F with (m, n-k) degrees of freedom under the null hypothesis. The values
of n and n+m refer to the number of observations in the first subperiod
and the total sample, respectively. The value of k refers to the number of
parameters in the model.
13. This test yields a likelihood ratio statistic, which is distributed
asymptotically as chi-square with k degrees of freedom under the null
hypothesis.
14. We also looked for evidence of parameter instability using the
CUSUM and CUSUMSQ tests proposed by Brown, Durbin, and Evans
(1975). The tests are based on recursive residuals, with the CUSUM test
primarily used to detect gradual structural change and the CUSUMSQ
test used to detect sudden structural change. The tests provided no
evidence of parameter instability.
15. The dynamic simulation yielded similar results for the 1994-96
period.
16. Meyer (1997) notes that the declines in computer prices and import
prices over the current expansion may also be acting as temporary supply
shocks helping to restrain inflationary pressures in the economy.
Moreover, as an additional explanation for the inflation puzzle, he cites
firms’ inability to raise prices because of increased international
competitive pressures. We do not address these factors in this paper and
instead restrict our attention to the two explanations that concern labor
market phenomena. Further, while our analysis is not exhaustive, we
nevertheless believe that it is instructive to evaluate these explanations
before considering alternative hypotheses.
17. Our focus on compensation growth is also motivated by the idea
that the pricing decision of a firm should be based on a consideration of
its total labor costs rather than the behavior of the wage and benefit

NOTES

ENDNOTES (Continued)

Note 17 continued
components of these costs. In addition, the data preclude us from
obtaining observations on wages and benefits separately over the full
sample period. The employment cost index, which provides measures of
wages and benefits, is only available beginning in 1980 for the nonfarm
sector.
18. We modify the traditional price-inflation Phillips curve to include
unit labor costs rather than compensation per hour because it is the
behavior of compensation growth relative to productivity growth that is
relevant for describing the dynamics of the inflation process. That is,
greater productivity growth will act to offset the inflationary pressure on
prices arising from an increase in compensation growth.
19. Note that our model does not allow us to examine whether a shift in
the Federal Reserve’s inflation fighting credibility has changed the
inflation process by directly altering inflation expectations. Such an
examination is beyond the scope of this paper and would involve
estimating a separate equation for inflation expectations and including
some measure of Federal Reserve credibility as an explanatory variable.
Previous evidence, however, suggests that such a shift has not taken
place. Blanchard (1984) notes that similar types of Phillips curves
remained stable even after the 1979 change in Federal Reserve operating
procedures.

NOTES

20. As the value of the test statistics in Table 2 indicates, the Chow tests
fail to reject the null hypothesis of parameter stability at conventional
significance levels. However, this result is not particularly informative
because the Chow tests also failed to reject the null hypothesis of model
stability for the traditional Phillips curve.
21. The increase in the forecasted value for inflation primarily reflects
the influence of a change in the output gap and the oil price variable.
22. For definitions of the data and their sources, see the Data Appendix.
23. For example, we could follow the approach of Fuhrer (1995), who
assumes a value of 6 percent for the NAIRU, and use the unemployment
gap (the difference between the actual level of unemployment and the
NAIRU) instead of the unemployment rate as an explanatory variable in
equation 3. This approach, however, would not affect the regression
results other than to change the estimated value of the constant term.
24. Fuhrer (1995) also finds an absence of significant rate-of-change
effects for the unemployment rate in wage-inflation Phillips curve
models.
25. The definition of the dummy variable is from Englander and Los
(1983).

FRBNY ECONOMIC POLICY REVIEW / DECEMBER 1997

67

REFERENCES

Blanchard, Olivier J. 1984. “The Lucas Critique and the Volcker
Deflation.” AMERICAN ECONOMIC REVIEW 74, no. 2: 211-5. Papers
and Proceedings of the 96th Annual Meeting of the American
Economic Association, December 1983.
Blomberg, S. Brock, and Ethan S. Harris. 1995. “The CommodityConsumer Price Connection: Fact or Fable?” Federal Reserve Bank of
New York ECONOMIC POLICY REVIEW 2, no. 2: 21-38.
Boldin, Michael D. 1996. “Problems in Determining the CapacityInflation Link.” Unpublished manuscript, the Conference Board.
Brown, R.L., J. Durbin, and J.M. Evans. 1975. “Techniques for Testing
the Constancy of Regression Relationships over Time.” JOURNAL OF
THE ROYAL STATISTICAL SOCIETY 37: 149-92.
Chow, Gregory C. 1960. “Tests of Equality between Sets of Coefficients in
Two Linear Regressions.” ECONOMETRICA 28: 591-605.
Eggert, Robert, ed. 1991-95. BLUE CHIP ECONOMIC INDICATORS. Capitol
Publications, Inc.
Englander, Steven A., and Cornelis A. Los. 1983. “The Stability of the
Phillips Curve and Its Implications for the 1980s.” Federal Reserve
Bank of New York Research Paper no. 8303, February.
Friedman, Milton. 1968. “The Role of Monetary Policy.” AMERICAN
ECONOMIC REVIEW 58: 1-17.
Fuhrer, Jeff. 1995. “The Phillips Curve Is Alive and Well.” Federal
Reserve Bank of Boston NEW ENGLAND ECONOMIC REVIEW, MarchApril: 41-56.
Goodfriend, Marvin. 1993. “Interest Rate Policy and the Inflation Scare
Problem: 1979-1992.” Federal Reserve Bank of Richmond QUARTERLY
REVIEW 79, no. 1: 1-24.
Gordon, Robert J. 1970. “The Recent Acceleration of Inflation and Its
Lessons for the Future.” BROOKINGS PAPERS ON ECONOMIC
ACTIVITY, no. 1: 8-41.
———. 1975. “The Impact of Aggregate Demand on Prices.”
BROOKINGS PAPERS ON ECONOMIC ACTIVITY, no. 3: 613-62.
———. 1977. “Can the Inflation of the 1970s Be Explained?”
BROOKINGS PAPERS ON ECONOMIC ACTIVITY, no. 1: 253-77.
———. 1982. “Inflation, Flexible Exchange Rates, and the Natural
Rate of Unemployment.” In Baily, M.N., ed., WORKERS, JOBS AND
INFLATION, pp. 88-152. Washington, D.C.: Brookings Institution.

68

FRBNY ECONOMIC POLICY REVIEW / DECEMBER 1997

———. 1990. “U.S. Inflation, Labor’s Share, and the Natural Rate of
Unemployment.” In König, Heinz, ed., ECONOMICS OF WAGE
DETERMINATION, pp. 1-34. Berlin and New York: Springer-Verlag.
———. 1996. “The Time-Varying NAIRU and Its Implications for
Economic Policy.” National Bureau of Economic Research Working
Paper no. 5735, August.
Greene, William H. 1993. ECONOMETRIC ANALYSIS. New York:
Macmillan Publishing Company.
Hamilton, James D. 1996. “This Is What Happened to the Oil Price–
Macroeconomy Relationship.” JOURNAL OF MONETARY ECONOMICS
38, no. 2: 215-20.
King, Robert G., and Mark W. Watson. 1994. “The Post-War U.S. Phillips
Curve: A Revisionist Econometric History.” CARNEGIE-ROCHESTER
CONFERENCE SERIES ON PUBLIC POLICY 41 (December): 157-219.
King, Robert G., James H. Stock, and Mark W. Watson. 1995. “Temporal
Instability of the Unemployment-Inflation Relationship.” Federal
Reserve Bank of Chicago ECONOMIC PERSPECTIVES 19, no. 3 (MayJune): 2-12.
Ljung, G.M., and G.E.P. Box. 1978. “On a Measure of Lack of Fit in Time
Series Models.” BIOMETRIKA 67: 297-303.
Lown, Cara S., and Robert W. Rich. 1997a. “A Look at the U.S. Inflation
Puzzle.” In MONETARY POLICY AND THE INFLATION PROCESS. Bank
for International Settlements CONFERENCE PAPERS 4 (July).
———. 1997b. “Is There an Inflation Puzzle?” Federal Reserve Bank of
New York Research Paper no. 9723.
Meyer, Laurence H. 1997. “The Economic Outlook and Challenges for
Monetary Policy.” Remarks delivered to the Charlotte Economics
Club on January 16, Charlotte, N.C.
Mork, Knut A. 1989. “Oil and the Macroeconomy When Prices Go Up
and Down: An Extension of Hamilton’s Results.” JOURNAL OF
POLITICAL ECONOMY 97, no. 3: 740-4.
Phelps, Edmund S. 1967. “Phillips Curves, Expectations of Inflation, and
Optimal Unemployment over Time.” ECONOMICA 34: 254-81.
Phillips, A.W. 1958. “The Relation between Unemployment and the
Rate of Change of Money Wage Rates in the United Kingdom, 18611957.” ECONOMICA 25: 283-300.

NOTES

REFERENCES (Continued)

Tootell, Geoffrey M.B. 1994. “Restructuring, the NAIRU, and the
Phillips Curve.” Federal Reserve Bank of Boston N EW ENGLAND
ECONOMIC REVIEW, September-October: 31-44.

White, Halbert. 1980. “A Heteroskedasticity-consistent Covariance
Matrix Estimator and Direct Test for Heteroskedasticity.”
ECONOMETRICA 48, no. 4: 817-38.

The views expressed in this article are those of the authors and do not necessarily reflect the position of the Federal Reserve
Bank of New York or the Federal Reserve System. The Federal Reserve Bank of New York provides no warranty, express or
implied, as to the accuracy, timeliness, completeness, merchantability, or fitness for any particular purpose of any information
contained in documents produced and provided by the Federal Reserve Bank of New York in any form or manner whatsoever.

NOTES

FRBNY ECONOMIC POLICY REVIEW / DECEMBER 1997

69