The impact of federal funds target changes on interest rate volatility

International Review of Economics and Finance 15 (2006) 241–259

www.elsevier.com/locate/iref

The impact of federal funds target changes on interest rate volatility

Jim Lee*

Texas A&M University-Corpus Christi, 6300 Ocean Drive, Corpus Christi, TX 78412, United States

Received 3 June 2003; received in revised form 5 July 2004; accepted 9 November 2004

Available online 6 January 2005

Abstract

This paper empirically investigates the 1-day response of interest rate volatility to a federal funds target rate

change over the period 1989–2003. Federal funds futures data are used to distinguish between anticipated and

unanticipated changes in the funds rate target. Interest rate volatility is modeled as an EGARCH process. The

volatility response to an unanticipated Fed policy action is relatively large in size and highly significant for short-

term interest rates. However, interest rates at long maturities are found to be responsive to target rate changes, even

if they are anticipated, when estimations take into account of structural change in association with the Fed’s policy

disclosure beginning in 1994, as well as asymmetrical effects between monetary easing and tightening.

D 2004 Elsevier Inc. All rights reserved.

JEL classification: E4; E5; C32

Keywords: Interest rate volatility; Federal funds futures; Federal funds rate target; EGARCH

1. Introduction

A growing body of literature suggests that monetary policy has become less powerful since the early

1990s when Fed policy decisions have become more predictable. Because the impact of monetary policy

on the overall economy begins with changes in market interest rates in response to a change in the

federal funds rate, the relationship between these interest rates is an important one in assessing the

effectiveness of monetary policy transmission.

1059-0560/$ -

doi:10.1016/j.i

* Tel.: +1 36

E-mail add

see front matter D 2004 Elsevier Inc. All rights reserved.

ref.2004.11.005

1 825 5831; fax: +1 361 825 5609.

ress: [email protected].

J. Lee / International Review of Economics and Finance 15 (2006) 241–259242

Conventional wisdom dictates two financial market characteristics in the case of a Fed policy action.

First, an increase in the funds rate target evokes an immediate increase in market interest rates and vice

versa. Cook and Hahn (1989), and Roley and Sellon (1996) find such a policy impact to be statistically

meaningful in the 1970s, but not in more recent periods. Second, in line with the expectations theory of

the term structure, the response to monetary policy tends to be weaker for interest rates of longer

maturities. Roley and Sellon (1995) show historical evidence in support of this view.

Kuttner (2001) argues that these earlier studies are misleading as they do not distinguish between

anticipated and unanticipated policy actions. Using data from the federal funds futures market to isolate

anticipated from unanticipated changes in the funds target rate, he find that the impact of anticipated

actions was small, while the impact of unanticipated actions is much larger and highly significant.

While existing studies focus primarily on interest rate levels, interest rate volatility has become an

increasing concern to policymakers and financial market participants alike. Increased market volatility is

associated with higher uncertainty about market outlooks, which also affects, among other things, the

ability of market participants to discern the monetary policy stance. If interest rate volatility, which is

particularly higher among short-term rates, is further transmitted toward longer-term rates, then

investment decisions, and thus overall economic activity, will also be affected.

Given the growing interest in financial market volatility, this paper explores the impact of monetary

policy on the volatilities of interest rates in addition to their levels. More specifically, we examine the 1-

day volatility responses of the yields for U.S. Treasury securities to a change in the funds rate target. For

estimation purposes, we use a GARCH-type (generalized autoregressive conditional heteroskedasticity)

model to capture stochastic variations in interest rate volatility and to estimate its response to a Fed

action. Using a similar model, Ayuso, Haldane, and Restoy (1997) show that volatilities in European

interest rates spill over from short- to long-term maturities, while Lee (2002) shows little volatility

spillover across the term structure. The present study attempts to examine if this phenomenon occurs in

the U.S. as well.

Another distinction of this paper is our application of Kuttner’s (2001) methodology, which uses

federal funds futures data to separate anticipated and unanticipated changes in the funds rate target

change. In line with earlier findings on interest rate levels, our empirical results indicate that the

volatility response to an unanticipated Fed policy action is relatively large in size and highly significant

for short-term interest rates, and the corresponding response to an anticipated action is substantially

smaller. However, when changes in the policy regime and policy asymmetries are taken into account,

interest rates at the long end of the yield curve become responsive to anticipated policy actions.

The remainder of the paper is organized as follows. The next section describes the data and the third

section outlines the empirical model. The fourth section presents the findings of the basic model, and the

fifth section examines the effects of different policy regimes and different directions of monetary policy.

The final section contains a summary and concluding remarks.

2. Data

The empirical methodology of this paper differs from previous studies in two dimensions. First, to the

extent that financial market volatility reflects uncertainty about the monetary policy stance, we look at

market interest rates’ volatilities in addition to their levels. The conventional measure of uncertainty is

unconditional volatility, such as standard deviation. This measure is often criticized as being ad hoc in

J. Lee / International Review of Economics and Finance 15 (2006) 241–259 243

nature. Conditional volatility, on the other hand, serves as a more convincing measure of market

participants’ perceived uncertainty regarding the future economic and policy outlooks. The present paper

estimates conditional volatility using a variant of GARCH. Previous studies using this type of model

include Ayuso et al. (1997), who examine volatility transmissions across European interest rates of

different maturities.

Second, we distinguish possible differences between the market response to an anticipated and an

unanticipated change in the funds rate target. Expectations of Fed policy actions are, of course, not

directly observable. Some studies, such as Edelberg and Marshall (1996), Evans and Marshall (1998),

and Mehra (1996), use a vector autoregression (VAR) to measure market expectations, and thus

unanticipated changes, of monetary policy. There is, however, controversy over the reliability of VAR-

based measures of policy shocks, as discussed in Evans and Kuttner (1998).

Kuttner (2001) has suggested a market-based approach as an alternative to model-based measures of

market expectations. The futures data have three major advantages over other gauges for policy

expectations. First, the data are not dependent on model selection; second, there is no problem with the

vintage of the data used to perform forecasts; third, possible generated-regressor problems in model-

based approach are absent.1

The federal funds futures rate is defined as 100 minus the price of a futures contact. Following

Kuttner (2001), we derive the anticipated and unanticipated components of an FOMC decision on the

fed funds target from the change in the futures contact’s price relative to the day prior to the policy

action. More specifically, the unexpected or bsurpriseQ policy component is calculated as the change in

the bspot-monthQ (for the month in which the target is changed) fed funds futures rate on the day of the

rate change, multiplied by a scale factor to reflect the number of the days affected by the change:

Dr̃rut ¼m

m� tf 0n;t � f 0

n;t�1

�;

�ð1Þ

where Dr̃u is the unanticipated target rate change, f 0n,t is the spot-month futures rate on day t of month n,

and m is the number of days in the month. In the case where the rate change occurs on the first day of the

month, we replace f 0n,t�1 with f 0n�1,m, the latter of which denotes the 1-month futures rate from the last

day of the previous month. In order to avoid exaggerating noise at the end of the month, if the rate

change falls within the last 3 days of the month, then we use the unscaled 1-month futures rate change

instead of the change in the spot month rate.

After deriving the unanticipated target rate change, the anticipated component of the rate change (Dr̃a)

is then computed as the actual (Dr̃) minus the unanticipated change:

Dr̃rat ¼ Dr̃rt � Dr̃rut : ð2Þ

Since the fed funds futures market was not established until 1989, our analysis will be limited to the

post-1989 period. Our daily data set spans over the period between June 6, 1989 and December 31,

2003. Over the 1989–2003 observation period, there were 57 changes in the funds rate target.

The fed funds rate is considered to be the Fed’s primary monetary policy instrument, at least over our

observation period. In reality, the Fed cannot perfectly control the fed funds rate, which is determined by

supply of and demand for borrowed reserves. The Fed simply sets a target rate and then buys and sells

1Of course, this approach using market data also has limitations. See Kuttner (2001) for a detailed discussion.

1989 1991 1993 1995 1997 1999 2001 20030.0

2.5

5.0

7.5

10.0

Fig. 1. Federal funds rate target (%).


government securities (open market operations) so that the actual funds rate is equal to the target on

average. Fig. 1 plots the target rate over time. The observation period contains two major monetary

easing periods (one from June 1989 to September 1992 and another one from January 2001 through the

end of the sample) and two monetary tightening periods (one from February 1994 to January 1995 and

another from June 1999 to May 2000).

Fig. 2 displays unanticipated changes or bsurprisesQ (actual less expected changes) in target rate

decisions based on futures market data. Negative errors occurred typically when the Fed lowered the

funds rate target and positive errors occurred when the Fed raised the target. The patterns of the forecast

errors indicate that the sizes of interest rate changes have generally been under-predicted. Not

surprisingly, the errors were large when the funds rate target changed its direction. Their sizes, however,

1989 1991 1993 1995 1997 1999 2001 2003-0.50

-0.25

0.00

0.25

0.50

Fig. 2. Unanticipated funds rate target change (%).


tended to diminish only slowly over subsequent changes even in the same directions, reflecting the

extent of persistence in the fed funds futures market.

The observation period is characterized by higher transparency about the Fed’s actions to market

participants relative to earlier periods. In particular, beginning with the February 1994 meeting, Fed

policy decisions have been announced immediately following each FOMC meeting. This procedural

change has eliminated virtually most of the timing ambiguity associated with existing Fed actions.

However, it is widely conceived that the funds rate target had already become more apparent to market

participants since the late 1980s. This view is confirmed by Kuttner (2001), who reports a substantial

reduction in the 1-day response of market interest rates to the bobservedQ target rate change throughoutthe 1990s. In addition, Kuttner find little response of interest rates to lagged target changes.

Given these perspectives, it is reasonable to assume that the decision on the target rate, either at

FOMC meetings or within the intermeeting periods, was widely known by market participants within a

day during the observation period. In other words, no surprises existed the day after the target rate

change. This aspect of policy transparency, however, does not necessarily mean that there would be no

Table 1

Unit-root test results

Lags (k) ADF Phillips-Perron

(A) Levels

FTAR 12 �1.35* �1.98*3-month 12 �1.11* �1.48*6-month 11 �1.09* �1.28*1-year 12 �1.25* �1.77*2-year 11 �1.51* �2.31*5-year 11 �2.22* �3.53*7-year 11 �2.42* �3.81*10-year 11 �2.77* �4.06*30-year 11 �2.80* �3.71*

(B) First-differences

FTAR 11 �17.75 �379.913-month 11 �17.82 �317.056-month 10 �16.83 �340.851-year 11 �17.58 �342.082-year 10 �18.22 �341.675-year 10 �18.67 �336.287-year 10 �18.86 �331.2510-year 10 �18.72 �335.9530-year 10 �17.54 �300.57FTAR denotes the federal funds rate target. The unit-root tests are based on regression of the form: Dxt=

l0+l1t+qxt�1+P

ki=1l2iDxt�i where xt is an interest rate series; t is a linear time trend; and k is a lag order determined by

the Schwartz information criterion and indicated in the second column. ADF statistics are augmented Dickey-Fuller t-statistics

for q. The Phillips and Perron (1998) tests are based on a nonparametric estimate of the spectral density of Dxt at frequency

zero, measured relative to the variance of Dxt. Critical values for ADF-t statistics are obtained from Davidson and MacKinnon

(1993); for Phillips-Perron’s t statistics, Phillips and Perron (1998).

* Denotes that the null hypothesis of a unit root cannot be rejected at the 0.01 level.


uncertainty about future changes in the target rate. Uncertainty about the future economic and policy

outlooks, which exacerbates market volatility, might persist for a prolonged period.

For market interest rates, we investigate daily Treasury yields between the 3-month and 30-year

maturity. The data are obtained from Bloomberg. To avoid spurious inferences, the data series for

regression need to be stationary. As both the augmented Dickey-Fuller (1981) and Phillips and Perron

(1998) test statistics in Table 1 indicate, the null hypothesis of a unit root cannot be rejected for all data

in levels. The corresponding statistics, however, show strong evidence of stationarity for the variables in

first-differences.

3. Empirical model

Given the unit-root test results in Section 2, the conditional mean equation of the GARCH system for

each market interest rate it (expressed in basis points) is represented by a first-differenced process, or

I(1):

Dit ¼ c1Dfet þ c2Df

ut þ et; etf 0; htð Þ ð3Þ

where D is a first-difference operator, et is an innovation series, and Dfte and Dft

u represent the anti-

cipated and unanticipated changes in the target rate, respectively.

To estimate the impact of Fed policy on interest rates, Kuttner (2001) estimates Eq. (3)

without allowing for ARCH effects, as captured by ht.2 Table 2 shows the results for Engle’s (1983)

Lagrange multiplier tests for ARCH in residuals from estimating Eq. (3) by ordinary-least squares. The

strong evidence of ARCH in high orders offers support for a generalized ARCH process instead of

ARCH.

Since Bollerslev (1986), GARCH models have been commonly used to take into account high

persistence in the conditional variances of financial time series. In characterizing the dynamics of

financial market volatility, the standard GARCH models do suffer from a major drawback: they are

symmetric in the sense that the signs of innovations are irrelevant. In other words, an unanticipated drop

in an interest rate is assumed to generate the same impact on its future volatility as an unanticipated rise

of the same magnitude. A symmetry constraint on the effect of past innovations in the conditional

variance equation appears to be inappropriate, given the strong evidence of asymmetries found in

financial instruments. For example, Jones, Lamont, and Lumsdaine (1998) illustrate the different

responses of the volatility in Treasury bond returns to bgoodQ versus bbadQ economic news

announcements.

From the above perspectives, we apply an exponential GARCH process to estimate the conditional

variances of the Treasury yields.3 This type of model is introduced by Nelson (1991) to characterize the

2The first-difference form of the conditional mean equation is estimated such that the results can be directly compared with those in Kuttner

(2001). An error-correction representation, which includes the levels of the variables, can be used to estimate possible cointegration between

interest rates. Adding the level terms, nonetheless, did not alter the qualitative results for the conditional variance equation reported here.3In a preliminary analysis across a host of GARCH-type models, the EGARCH(1,1) model outperformed other models, including an

IGARCH as described in Section 4, an asymmetric GARCH model suggested by Glosten, Jagannathan, and Runkle (1993), and the inclusion of

additional autoregressive terms. The results are not reported here to conserve space.

Table 2

ARCH test statistics

Maturity v2(5) v2(10)

3-month 87.95* 154.55*

6-month 45.36* 74.72*

1-year 15.10* 31.22*

2-year 42.96* 62.47*

5-year 28.38* 39.99*

7-year 30.71* 45.31*

10-year 21.99* 30.72*

30-year 16.04* 28.48*

Figures are Engle’s (1983) Lagrange multiplier tests for ARCH in residuals of an order in parentheses from estimating Eq. (3)

by OLS.

* Denotes statistical significance at the 0.01 level.


asymmetric behavior of asset prices and is now widely used in financial economics. The conditional

variance equation of an EGARCH(1,1) process can be expressed as:

ln htð Þ ¼ xþ ant�1 þ bln ht�1ð Þ þ h jnt�1j �ffiffiffiffiffiffiffiffi2=p

p� �þ u1Df

et þ u2Df

ut ð4Þ

where ntuet=ffiffiffihp

t is the normalized innovation series. The log-linear form of Eq. (4) has an advantage.

For the conditional variance to be strictly positive, the conditional variance matrix of standard GARCH

models must be non-negative. This restriction is not required in the EGARCH specification due to its

flexible log-linear functional form.

The key feature of the EGARCH model is the basymmetryQ term (a), which reflects different impacts

of positive and negative innovations on conditional variances. A positive a estimate implies that a

positive shock increases volatility more than a negative shock of an equal magnitude. This may reflect a

stylized fact of financial markets in which investors are more likely to be affected by interest rate hikes

due to finance constraints than interest rate cuts.

To explore the possible impacts on both the levels and volatilities of interest rates, the policy variables

enter both the conditional mean and conditional variance equations. Their contemporary as opposed to

lagged values capture the responses within the same day of a monetary policy change. As indicated

above, we assume that the Fed’s decision on the funds rate target is known by market participants within

the same day of its official announcement.

The EGARCH system of (3) and (4) is estimated using the maximum-likelihood method in which the

log-likelihood function is expressed as:

XTt¼1

Kt Hð Þ ¼ � 1

2T ln 2pð Þ þ

XTt¼1

lnht þ e2t =ht� �" #

t ¼ 1; 2; N ;T ð5Þ

where Hu[c1, c2, x, a, b, h, u1, u2] is the vector of all parameters in the system. Estimations are

conditional on the assumption of normality. As shown in Table 3, the Jarque-Bera statistics provide

strong evidence against normality in the residual term of Eq. (3), largely due to high levels of skewness

Table 3

Normality test results

Maturity Skewness Kurtosis Jarque-Bera v2(2)

3-month 2.48* 6.14* 9882.76*

6-month 2.46* 5.43* 8346.95*

1-year 2.30* 4.41* 6424.51*

2-year 2.23* 3.75* 5380.02*

5-year 1.98* 2.27* 3297.45*

7-year 1.91* 1.88* 2877.50*

10-year 1.86* 1.62* 2617.75*

30-year 1.53* 1.76* 1598.27*

The data are residuals from estimating Eq. (3) with OLS.

* Denotes statistical significance at the 0.01 level.


and kurtosis (thick-tailed distributions). Given these findings, we estimate the model using Bollerslev

and Wooldridge’s (1992) quasi-maximum likelihood procedure, which produces consistent parameter

estimates that are robust to non-normality.

4. Estimation results

For each Treasury rate series, we estimate the EGARCH(1,1) systems (3) and (4) by the quasi-

maximum likelihood method as described above. The estimation results for the period between June 6,

1989 and December 31, 2003 are reported in Table 4.4 The left panel of the table shows the results for

the conditional mean equation. The estimates for the responses of interest rate levels to target rate

changes are similar to those reported in Kuttner (2001, Table 5).5 Essentially, the estimates for the

anticipated component of target rate changes are all qualitatively trivial. By comparison, the estimates

for the unanticipated component are significant at the 0.01 level for yields of 7-year maturity and

shorter.

The sizes of the estimated responses to policy bsurprisesQ are also markedly larger than the

corresponding estimates for anticipated policy actions. For example, the response of the 3-month T-bill

yield to a 1% bsurpriseQ in the target rate change is 79 basis points. This estimate is substantially larger

than the estimated 5 basis points for the anticipated rate change, or the 27-basis-point estimate reported

by Cook and Hahn (1989) for the brawQ change in the funds rate target. The estimated responses decline

across maturities, supporting the expectations theory that longer-term rates are simply averages of

expected future shorter-term rates.

The right panel of Table 4 shows the coefficient estimates for the conditional variance equation.

Volatility persistence is measured by b. If it equals one, then the conditional variance will follow an

integrated process of first order, i.e., I(1). Most estimates appear to be fairly close to one, indicating high

levels of volatility persistence. We have attempted to impose an I(1) restriction but found little effect on

4To avoid possible misspecification, an intercept is included in estimating the conditional mean equation. Estimates for all intercepts are

nonetheless statistically insignificant.5Discrepancies between our estimates and Kuttner’s (2001) are attributable to: (1) the consideration of ARCH residuals, (2) the inclusion of

days with no official decisions on the funds rate target, and (3) the slightly longer estimation period in our paper.

Table 4

EGARCH(1,1) estimation results, 1989–2003

Maturity Conditional mean Conditional variance

Target change x a b h Target change

Expected Surprise Expected Surprise

c1 c2 u1 u2

3-month 0.05 (1.30) 0.79 (10.27)* 0.02 (15.77)* 0.28 (13.36)* 0.97 (50.24)* 0.31 (2.03)** 0.25 (0.56) 1.43 (36.28)*

6-month 0.07 (1.27) 0.70 (11.40)* 0.05 (2.87)* 0.21 (9.27)* 0.98 (21.95)* 0.20 (1.99)** 1.07 (1.38) 1.41 (1.94)***

1-year 0.001 (0.02) 0.59 (5.33)* 0.68 (22.56)* 0.15 (5.50)* 0.97 (11.06)* 0.15 (1.73)*** 0.41 (0.68) 1.18 (1.79)***

2-year 0.02 (0.79) 0.42 (4.96)* 2.34 (8.54)* 0.12 (5.23)* 0.97 (10.95)* 0.08 (0.64) 0.14 (0.25) �0.25 (0.41)

5-year 0.07 (1.12) 0.49 (6.61)* 1.50 (5.60)* 0.20 (16.95)* 0.94 (26.42)* 0.01 (0.08) 0.46 (0.83) 0.04 (0.25)

7-year 0.06 (1.47) 0.32 (6.05)* 0.46 (8.66)* 0.10 (7.78)* 0.98 (57.62)* �0.06 (0.41) 0.33 (1.03) �0.33 (0.64)

10-year 0.03 (1.63) 0.14 (1.28) 5.21 (7.76)* 0.05 (9.90)* 0.98 (15.81)* �0.12 (0.75) 0.02 (0.09) 0.21 (0.44)

30-year �0.02 (0.68) 0.13 (1.58) 0.06 (1.21) 0.08 (4.25)* 0.96 (59.58)* �0.15 (1.47) 0.41 (0.85) 0.95 (1.39)

Estimates are based on Bollerslev and Wooldridge’s (1992) quasi-maximum likelihood estimations of the system of equations Dit=c0+c1Dfte+c2Dft

u+et andln htð Þ ¼ xþ ant�1 þ bln ht�1ð Þ þ h jnt�1j �

ffiffiffiffiffiffiffiffi2=p

p� �þ u1Df

et þ u2Df

ut where ntuet=

ffiffiffiffihtp

. Absolute values of t-statistics are indicated in parentheses.

*, **, *** denote statistical significance at the 0.01, 0.05, and 0.1 levels, respectively.

J.Lee

/Intern

atio

nalReview

ofEconomics

andFinance

15(2006)241–259

249

Table 5

Structural break test results

Maturity Sup-WT p-value Exp-WT p-value Date

3-month

Constant 6.62 0.12 1.18 0.15 1992:10:01

c1 4.87 0.26 1.81 0.07 1994:02:04

c2 2.95 0.56 0.84 0.25 1994:04:18

All coefficients 10.44 0.18 3.17 0.14 1992:10:01

6-month

Constant 5.03 0.24 0.78 0.27 1992:10:01

c1 5.21 0.23 1.36 0.12 1994:02:04

c2 4.67 0.28 1.17 0.15 1994:04:18


1-year

Constant 4.13 0.35 0.55 0.40 1992:10:01

c1 3.99 0.38 1.07 0.18 1994:02:04

c2 5.31 0.22 1.50 0.10 1994:04:18


2-year

Constant 2.54 0.66 0.28 0.67 1992:10:01

c1 5.91 0.17 1.51 0.10 1994:02:04

c2 6.05 0.16 1.63 0.08 1994:04:18


5-year

Constant 1.27 0.96 0.13 0.96 1998:10:05

c1 3.69 0.42 0.72 0.30 1994:02:04

c2 4.59 0.29 1.08 0.17 1994:04:18


7-year

Constant 1.43 0.92 0.11 0.99 1998:10:05

c1 5.03 0.24 1.13 0.16 1994:02:04

c2 3.93 0.38 0.83 0.26 1994:04:18


10-year

Constant 1.79 0.84 0.12 0.99 1998:10:05

c1 4.66 0.29 1.05 0.18 1994:02:04

c2 4.08 0.36 0.92 0.22 1994:04:18


30-year

Constant 2.01 0.78 0.12 0.99 1998:10:05

c1 5.48 0.20 1.45 0.11 1994:02:04

c2 5.34 0.21 1.38 0.12 1994:04:18




other coefficient estimates in that corresponding IGARCH model, particularly the estimates for the

policy variables. Estimates for the bleverageQ term (a) capture different impacts of positive and negative

innovation on conditional variances. The positive estimates imply that a positive interest rate innovations

evokes a larger effect on market volatility than a negative innovations.

Similar to the conditional mean equation, estimates for the anticipated change in the target rate are all

statistically insignificant. For the unanticipated target rate change, on the other hand, the estimate is

significant at the 0.1 level or better for short-term interest rates. The size of the volatility response to

policy bsurprisesQ also tends to decline with maturity. For each Treasury yield, a Wald test of equal

responses between the anticipated and unanticipated target rate change is also conducted for the

conditional mean (i.e., c1=c2) and the conditional variance (i.e., u1=u2) equations. In all cases, the null

hypothesis of equal responses is rejected at the 0.05 level or better.

Taken together, the results in Table 4 highlight the importance of distinguishing anticipated from

unanticipated Fed policy actions. Anticipated changes in the target rate deliver virtually no meaningful

effect on either interest rate levels or volatilities. An unanticipated change in the target rate, on the other

hand, not only leads to an immediate change in market interest rates in the same direction, but it also

raises market volatility.

There is, however, a major distinction between the policy impacts on levels and impacts on

volatilities. Estimates from the conditional mean equation show statistically significant level responses

for Treasury yields up to the 7-year maturity. By contrast, estimates from the conditional variance

equation suggest that volatility responses of Treasury yields can be found in maturities only up to the 1-

year duration. In other words, long-term yields respond to a funds rate target surprise largely with a one-

time level shift. The scant evidence of volatility spillover to interest rates at longer maturities stands in

contrast to Ayuso’s et al. (1997) findings on European interest rates.

The above findings, however, might be misleading if structural change occurs in the

relationship between monetary policy and Treasury returns. In particular, a structural break in

any of the ci’s in the conditional mean equation can possibly induce ARCH-type behavior in the

residuals. To test for possible structural breaks in Eq. (3), we apply a procedure developed by

Andrews and Ploberger (1994). More precisely, the procedure involves testing for a single structural

break at an unknown point within the middle 85% of the data set using a series of LM statistics for all

possible break points.

Table 5 reports the Sup-WT and Exp-WT statistics, which denote the maximum and exponentially

weighted average Wald statistics, respectively. Most of the Sup-WT statistics cannot reject the null

hypothesis of a structural break in any of the coefficients in the conditional mean equation. The Exp-WT

statistics, on the contrary, offer some evidence of structural change. Interestingly, the majority of the

maximum LM statistics are located in early 1994. This timing coincides with a change in the Fed’s

operation procedure. Before February 1994, the FOMC released information about its policy decisions

only later a 45-day delay. Since the meeting on February 4, 1994, the FOMC has announced its policy

decision immediately after each meeting. Thornton (1988) shows that such a procedural change has

Notes to Table 5:

The Sup-WT and Exp-WT are, respectively, the maximum and exponentially weighted average of the Wald statistics for testing

structural change within 0.15VTV0.85 of the data sample. The tests are described in Andrews and Ploberger (1994) with p-

values reported in Hansen (1997). The dates on the last column indicate the locations of the maximum LM statistics.


made monetary policy more predictable in not only the timing but also the size of potential movements

in the funds rate target.

5. Policy disclosure and asymmetrical effects

5.1. Policy disclosure

In light of the stability test results in Section 4, we investigate plausible changes in estimation results

associated with the FOMC’s immediate announcements of target rate changes beginning February 4,

1994. Table 6 illustrates how the monetary policy environments differ in periods before and after

February 4, 1994. Over the entire observation period 1989–2003, the mean absolute size of funds rate

target changes was 50 basis points, and rate increases largely offset rate decreases. For the post-February

1994 subperiod, the absolute size of rate changes averaged at 46 basis points, which was 9 basis points

less than the average for the earlier subperiod.

Along with the Fed’s 1994 procedural change, monetary policy actions have appeared to be more

predictable. Over the entire observation period, the futures market over- or under-predict about 30% of

the actual target rate changes. By comparison, the amount of policy surprises over the post-1994

subperiod was less than half of that over the pre-1994 subperiod.

The directions of Fed policy actions were also quite different between the two subperiods. During the

pre-1994 subperiod, most of the actions were to lower the funds rate target. The post-1994 subperiod,

however, was witness to a rather even split between easing and tightening policy actions. This latter

subperiod, therefore, allows us to investigate possible differences in policy effects between monetary

easing and monetary tightening.

To examine if our empirical results are sensitive to the Fed’s 1994 procedural change, we repeat the

EGARCH estimation exercises for the two subperiods divided by February 4, 1994. Table 7 compares

the coefficient estimates for the policy variables across the two subperiods.6 For the conditional mean

equation, the estimated impacts of a policy change, particularly the surprise component, is found to be

weaker for the post-1994 subperiod than those for the pre-1994 subperiod. Surprisingly, some coefficient

estimates for the expected policy variable (columns 2 and 6) are statistically significant. This finding

stands in contrast to the results for the full observation period (Table 4).

For the conditional variance equation, the estimation results for both expected and unexpected policy

changes vary considerably across the two subperiods. For the pre-1994 subperiod, even the expected

policy variable (column 4) appears to have an impact on volatilities, particularly of longer-term rates. As

explained in more detail below, the finding of an impact from an anticipated policy action on longer-term

rates can be explained by the expectations theory of interest rates.

The results for the post-1994 subperiod are more in line with those in Table 4 (full sample)—the

coefficient estimates for expected policy actions (column 8) are overall smaller in size and statistically

weaker than those for unexpected actions (column 9). Another interesting finding for the post-1994

subperiod is the estimated impact of policy surprises on the yields of 10-year T-notes and 30-year T-

bonds: a surprise target change not only affects these long-term interest rates’ levels but makes them

more volatile as well.

6Full estimation results are available upon request. Likelihood ratio tests reject parameter constancy across the two subperiods.

Table 6

Federal funds rate target changes

1989:6–2003:12 1989:6–1994:2 1994:2–2003:12

Absolute FTAR change 0.50% 0.55% 0.46%

Surprise/FTAR change 0.30 0.46 0.20

Policy change frequency

Total no. of FTAR changes 57 24 33

No. of FTAR increase (tightening) 14 0 14

No. of FTAR decrease (easing) 43 24 19

FTAR denotes the federal funds rate target.


Table 7 also reveals noticeable changes in the overall sizes of the coefficient estimates for the

expected and unexpected policy variables between the two subperiods. This observation can be

explained by changes in the sizes of the expected versus unexpected policy actions. More specifically, if

the size of interest rate volatility remained the same across the two subperiods, then a reduction in the

size of policy surprises after 1994 would generate a larger coefficient estimate for the policy surprise

variable over that subperiod.

5.2. Asymmetrical effects

In Table 4 above, the estimates for the leverage term (a) in EGARCH estimations reveals the

significance of asymmetries in the conditional variances of interest rates. It is, therefore, interesting

to explore whether the direction of a Fed policy action also affects interest rate volatility. This type

of asymmetries complements a stylized fact in financial markets in which investors are more likely

to be affected by monetary policy tightening due to finance constraints than policy easing (Nelson,

1991).

To explore monetary policy asymmetries, we replace each of the expected and unexpected policy

variables with two separate positive and negative funds rate target changes. As such, a positive expected

policy variable (Dfte+) takes the value of an actual target rate increase as anticipated by market

participants and zero otherwise, whereas a negative expected policy variable (Dfte�) takes the value of an

actual target rate decrease as anticipated and zero otherwise. Correspondingly, a positive surprise policy

variable (Dftu+) takes the value of an unanticipated target rate increase and zero otherwise, whereas a

negative surprise policy variable (Dftu�) takes the value of an unanticipated target rate cut and zero

otherwise. These specifications also capture the different directions of actual target rate changes because

the majority of these policy directions over the observation period were correctly anticipated by the

futures market, even though the sizes of the moves might be under- or over-predicted.

Table 8 lists the coefficient estimates for the four alternative policy specifications over the full

estimation period 1989–2003. For the conditional mean equation (left panel), negative policy surprises

are found to affect interest rates across all maturities, while the estimated effects of positive policy

surprises are limited to 1-year and shorter maturities. The statistically weaker estimates for the impact of

target rate increases may be attributed to the estimation period in which the Fed reduced the target rate

much more often (recall Table 6). This is particularly true for the pre-1994 subperiod, in which all Fed

actions were expansionary.

Table 7

Subsample estimation results for target rate change variables

1989:6–1994:2 1994:2–2003:12

Conditional mean Conditional variance Conditional mean Conditional variance

Expected Surprise Expected Surprise Expected Surprise Expected Surprise

c1 c2 u1 u2 c1 c2 u1 u2

3-month 0.20 (2.42)** 0.75 (8.64)* 1.42 (1.55) 1.40 (2.74)* 0.02 (0.47) 0.76 (3.98)* 0.49 (0.77) 1.31 (2.18)**

6-month 0.23 (2.54)* 0.73 (7.02)* 1.28 (1.37) 0.50 (0.69) 0.06 (1.90)** 0.50 (4.09)* 0.81 (1.15) 1.46 (1.68)***

1-year 0.16 (1.89)*** 0.65 (4.44)* 0.69 (0.82) 0.04 (0.10) �0.02 (0.54) 0.39 (3.60)* 0.39 (0.80) 1.73 (0.93)

2-year 0.19 (1.75)*** 0.50 (10.44)* 0.33 (0.28) 0.24 (0.44) �0.004 (0.25) 0.15 (1.35) 0.12 (0.26) 0.61 (0.38)

5-year 0.03 (1.17) 0.49 (4.29)* 2.23 (2.11)** 0.30 (0.28) �0.01 (0.36) �0.02 (0.11) 0.04 (0.08) 0.78 (0.74)

7-year 0.05 (0.84) 0.40 (4.33)* 1.86 (2.12)** 0.66 (0.74) �0.05 (1.54) �0.08 (0.51) 0.001 (0.01) 0.62 (0.71)

10-year 0.04 (0.70) 0.33 (4.93)* 3.30 (2.02)** �1.60 (1.30) 0.03 (1.02) 0.81 (6.66)* 0.70 (1.01) 0.68 (8.74)*

30-year 0.03 (0.56) 0.25 (3.32)* 1.69 (1.59) 0.70 (0.89) �0.03 (1.05) 0.16 (2.49)** 0.23 (0.78) 1.30 (10.47)*

Estimates are based on Bollerslev and Wooldridge’s (1992) quasi-maximum likelihood estimations of the system of equations Dit=c0+c1Dfte+c2Dft

u+et andln htð Þ ¼ xþ ant�1 þ bln ht�1ð Þ þ h jnt�1j �

ffiffiffiffiffiffiffiffi2=p

p� �þ u1Df

et þ u2Df

ut where ntuet=

ffiffiffiffihtp



J.Lee

/Intern

atio

nalReview

ofEconomics

andFinance

15(2006)241–259

254

Table 8

Full sample estimation results for target change asymmetries, 1989–2003

Conditional mean Conditional variance


Positive Negative Positive Negative Positive Negative Positive Negative

c1+ c1

� c2+ c2

� u1+ u1

� u2+ u2

�

3-month �0.02 (0.61) 0.14 (1.59) 0.86 (2.87)* 0.74 (9.53)* �0.23 (0.27) 0.28 (0.48) 3.71 (1.89)*** �1.47 (2.14)**

6-month 0.00 (0.02) 0.11 (1.48) 0.59 (1.91)*** 0.70 (6.53)* 0.99 (0.83) 0.80 (1.42) 2.65 (1.66)*** �1.67 (1.94)**

1-year �0.02 (0.49) 0.04 (0.81) 0.43 (1.76)*** 0.61 (5.33)* 0.84 (0.88) �0.08 (0.13) 0.97 (0.32) �1.22 (1.40)

2-year 0.00 (0.08) 0.05 (0.70) 0.53 (1.38) 0.41 (3.66)* 0.78 (1.00) �0.49 (0.86) 0.51 (0.17) �0.05 (0.08)

5-year �0.05 (0.92) 0.03 (1.04) 0.31 (0.80) 0.31 (9.96)* 0.91 (1.22) �0.26 (1.63) 0.33 (0.13) �0.20 (0.45)

7-year 0.17 (1.46) �0.06 (0.81) 0.22 (1.47) 0.31 (8.48)* 1.15 (1.45) �0.22 (0.37) 6.43 (1.36) �0.11 (0.14)

10-year �0.07 (1.10) 0.01 (0.26) 0.24 (0.62) 0.19 (1.82)*** 1.07 (1.23) �0.33 (0.73) 0.65 (0.22) �0.49 (0.69)

30-year �0.11 (1.44) 0.03 (1.11) 0.16 (0.44) 0.11 (1.79)*** 1.27 (1.17) �0.38 (0.51) �0.10 (0.02) �0.79 (1.06)

Estimates are based on Bollerslev and Wooldridge’s (1992) quasi-maximum likelihood estimations of the system of equations Di t=c0+c1+Df t

e++c1�Df t

e�+c2+Df t

u++c2�Df t

u�+et and ln htð Þ ¼ xþ ant�1 þ bln ht�1ð Þ þ h jnt�1j �ffiffiffiffiffiffiffiffi2=p

p� �þ uþ1 Df

eþt þ u�1 Df

e�t þ uþ2 Df

uþt þ u�2 Df

u�t where

ntuet=ffiffiffihp

t. Absolute values of t-statistics are indicated in parentheses.


J.Lee

/Intern

atio

nalReview

ofEconomics

andFinance

15(2006)241–259

255

Table 9

Subsample estimation results for target change asymmetries, 1994–2003

Conditional mean Conditional variance


Positive Negative Positive Negative Positive Negative Positive Negative

c1+ c1

� c2+ c2

� u1+ u1

� u2+ u2

�

3-month 0.01 (0.16) 0.12 (1.46) 0.60 (1.11) 0.73 (9.51)* �2.00 (1.20) 0.15 (0.20) 8.60 (1.78)*** �1.71 (1.82)***

6-month 0.03 (0.62) 0.12 (1.61) �0.20 (0.96) 0.55 (5.59)* 0.52 (0.55) 0.83 (1.27) 1.80 (1.66)*** �1.70 (1.64)***

1-year �0.03 (1.03) 0.01 (0.23) 0.18 (0.71) 0.45 (2.46)** 0.39 (0.66) 0.14 (0.17) 1.41 (0.81) �2.12 (1.26)

2-year 0.001 (0.05) �0.01 (0.38) �0.05 (0.13) 0.19 (1.11) 0.44 (0.57) �0.79 (0.96) �1.08 (0.27) �0.27 (0.17)

5-year �0.04 (0.44) 0.05 (0.77) �0.28 (0.62) �0.01 (0.06) 0.51 (0.74) �0.64 (0.88) �1.42 (0.39) �0.44 (0.40)

7-year �0.06 (0.30) �0.02 (0.30) �0.32 (0.51) �0.06 (0.18) 0.59 (0.47) �0.68 (1.11) �1.30 (0.27) �0.23 (0.18)

10-year �0.05 (8.70)* �0.01 (0.26) �0.33 (5.41)* �0.13 (0.73) 0.46 (1.00) �0.85 (2.11)** �0.73 (0.29) �0.32 (0.42)

30-year �0.09 (2.04)** 0.03 (0.57) �0.26 (0.69) �0.15 (1.11) 0.50 (1.12) �0.83 (3.97)* �0.62 (0.32) �0.53 (0.85)

Estimates are based on Bollerslev and Wooldridge’s (1992) quasi-maximum likelihood estimations of the system of equations Di t=c0+c1+Df t

e++c1�Df t

e�+c2+Df t

u++c2�Df t

u�+et and ln htð Þ ¼ xþ ant�1 þ bln ht�1ð Þ þ h jnt�1j �ffiffiffiffiffiffiffiffi2=p

p� �þ uþ1 Df

eþt þ u�1 Df

e�t þ uþ2 Df

uþt þ u�2 Df

u�t where

ntuet=ffiffiffiffihtp



J.Lee

/Intern

atio

nalReview

ofEconomics

andFinance

15(2006)241–259

256


The right panel of Table 8 lists the estimation results for the policy variables in the conditional

variance equation. The overall sizes of the coefficient estimates for two surprise policy variables (last

two columns) confirm the popular view about the larger volatility impact from a positive target rate

change (monetary tightening) than a negative rate change (monetary easing). Similar to the results

without the consideration of policy asymmetries (Table 4), statistically significant estimates are confined

to the short end of the yield curve. The negative coefficient estimates for unanticipated target rate cuts

indicate that an unanticipated monetary easing action raises interest rate volatility.

It is interesting to see whether the finding of policy asymmetries also holds for the post-1994

subperiod.7 Table 9 displays the corresponding estimation results for that subperiod. By omitting the pre-

1994 sample, in which most Fed actions were to lower the target rate, two major changes emerge among

the estimates for the conditional mean equation (columns 2 to 5). First, the impact of an unanticipated

rate cut on interest rate levels (column 5) is reduced to 1-year and shorter maturities. Second, some long-

term interest rates are correlated inversely with an anticipated target rate increase (column 2). This latter

finding can be interpreted as follows. If market participants anticipate a monetary tightening action, then

immediately after the Fed’s action that confirms their expectation and eliminates the uncertainty about

such an action, they will form expectations about the effects of the observed policy action on economic

activity. As a result, even though short-term interest rates do not respond to a given Fed action that is

already anticipated, the expectation of lower future inflation attributable to the Fed tightening policy

might lead to lower long-term interest rates.

For the conditional variance equation (right panel of Table 9), the estimated effects of policy

surprises are qualitatively similar to those for the full sample period—most of the impacts are

confined to short-term rates. For expected policy actions, on the other hand, coefficient estimates are

statistically significant for interest rates at 10- and 30-year maturities—the long end of the yield

curve.

At first glance, the finding of an impact from anticipated monetary policy appears to be at odds

with the theory of rational expectations or market efficiency. Our finding of anticipated monetary

policy effects, particularly on longer-term interest rates, can nevertheless be explained by the forward-

looking behavior of financial markets. As monetary policy becomes more predictable, interest rates

tend to be less responsive to observed Fed policy actions. However, fewer monetary policy surprises

do not necessarily mean less economic uncertainty. After an FOMC announcement of a target rate

change virtually eliminates the uncertainty that surrounds the existing monetary policy stance,

financial markets may become overshadowed by expectations and thus uncertainty about how

economic activity will react to the policy action (monetary policy transmission). As a result, even a

fully anticipated target rate change can affect market interest rates, only at this time the responses fall

largely on longer-term rates that rely to a greater extent on the outlook of future economic conditions

and inflation rather than on the existing monetary policy stance. In other words, an anticipated target

rate increase can be followed by lower long-term interest rates when market participants expect lower

inflation in the future. Given the long lag associated with monetary policy effects, higher interest rate

volatility may also emerge to reflect uncertainty about how economic activity responds to a given Fed

policy action.

7The Fed only lowered the funds rate target during the pre-February 1994 subperiod so that no policy asymmetries can be assessed for that

subsample.


6. Summary and concluding remarks

This paper has attempted to estimate the 1-day impact of Fed policy on both the levels and volatilities

of a multitude of Treasury yields. To capture plausible asymmetrical effects, interest rate volatility is

modeled as an EGARCH process. Following a recent study by Kuttner (2001), we used federal funds

futures data to separate unanticipated changes in the target rate from anticipated changes. We have found

this market-based measure to be a useful indicator for gauging monetary policy shocks.

The estimation results using data between June 1989 and December 2003 are in line with Kuttner’s

findings on interest rate levels. The responses of interest rate volatility to the unanticipated component

of Fed policy are large and highly significant, particularly among rates at the short end of the yield curve.

The corresponding responses to the anticipated component, on the contrary, are substantially smaller

both in size and in statistical importance.

Arguably, fewer monetary policy surprises since the early 1990s, particularly after the implementation

of immediate policy disclosure in February 1994, might have affected the effectiveness of monetary

policy. There is indeed evidence of a structural break in association with such a procedural change.

Moreover, there is evidence of policy asymmetries that supports the stronger effects of unanticipated

monetary policy tightening than easing on interest rate volatilities.

When these two factors are taken into account, long-term interest rates become responsive to an

observed monetary policy action, even if it is anticipated. This phenomenon can be explained by market

expectations about the dynamic effects of monetary policy on economic activity. The finding on the

anticipated policy effect on long-term interest rates highlights the distinction between monetary policy

uncertainty and economic uncertainty as well as the forward-looking behavior of financial markets under

economic uncertainty. In essence, more predictable monetary policy does not necessarily mitigate the

effect of a given policy action on the levels or volatilities of interest rates at longer maturities that rely to

a greater extent the future economic outlook.

References

Andrews, D. W. K., & Ploberger, W. (1994). Optimal tests when a nuisance parameter is present only under the alternative.

Econometrica, 62, 1383–1414.

Ayuso, J., Haldane, A. G., & Restoy, F. (1997). Volatility transmission along the money market yield curve.Weltwirtschaftliches

Archiv, 133, 56–75.

Bollerslev, T. (1986). Generalized autoregressive conditional heteroskedasticity. Journal of Econometrics, 31, 307–327.

Bollerslev, T., & Wooldridge, J. M. (1992). Quasi maximum likelihood estimation of dynamic models with time varying

covariance. Econometric Reviews, 11, 143–172.

Cook, T., & Hahn, T. (1989). The effect of changes in the federal funds rate target on market interest rates in the 1970s. Journal

of Monetary Economics, 24, 331–351.

Davidson, R., & MacKinnon, J. G. (1993). Estimation and inference in econometrics. New York7 Oxford University Press.

Dickey, D. A., & Fuller, W. A. (1981). Likelihood ratio statistics for autoregressive time series with a unit root. Econometrica,

49, 1057–1072.

Edelberg, W., & Marshall, D. (1996). Monetary policy shocks and long-term interest rates, Federal Reserve Bank of Chicago.

Economic Perspectives, 20, 2–7.

Engle, R. F. (1983). Estimates of the variance of U.S. inflation based upon the ARCH model. Journal of Money, Credit, and

Banking, 15, 266–301.

Evans, C.L., & Kuttner, K.N. (1998). Can VARs describe monetary policy? Federal Reserve Bank of Chicago, Working Paper

98-19.


Evans, C. L., & Marshall, D. (1998). Monetary policy and the term structure of nominal interest rates: Evidence and theory.

Carnegie-Rochester Conference Series on Public Policy, 49, 53–111.

Glosten, L. R., Jagannathan, R., & Runkle, D. E. (1993). On the relation between the expected value and the volatility of the

nominal excess return on stocks. Journal of Finance, 68, 1779–1801.

Hansen, B. E. (1997). Approximate asymptotic p-values for structural change tests. Journal of Business and Economic

Statistics, 15, 60–67.

Jones, C. M., Lamont, O., & Lumsdaine, R. L. (1998). Macroeconomic news and bond market volatility. Journal of Financial

Economics, 47, 315–337.

Kuttner, K. N. (2001). Monetary policy surprises and interest rates: Evidence from the fed funds futures market. Journal of

Monetary Economics, 47, 523–544.

Lee, J. (2002). Federal funds rate target changes and interest rate volatility. Journal of Economics and Business, 54, 1–33.

Mehra, Y. P. (1996). Monetary policy and long-term interest rates, Federal Reserve Bank of Richmond. Economic Quarterly,

82, 27–50.

Nelson, D. B. (1991). Conditional heteroskedasticity in asset returns: A new approach. Econometrica, 59, 347–370.

Phillips, P. C. B., & Perron, P. (1988). Testing for a unit root in times series regression. Biometrika, 75, 335–346.

Roley, V. V., & Sellon, G. H., Jr. (1995). Monetary policy actions and long term interest rates, Federal Reserve Bank of Kansas

City. Economic Quarterly, 80, 77–89.

Roley, V.V., & Sellon, G.H., Jr. (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 96-08.

Thornton, D. L. (1988). Tests of the market’s reaction to federal funds rate target changes, Federal Reserve Bank of St. Louis.

Review, 80, 25–36.

Documents

The impact of federal funds target changes on interest rate volatility