34

Gerald P. Dwyer, Jr. and IL W. Hafer

Gerald P Dwyer Jr., a professor of economics at ClemsonUniversity, is a visiting scholar and R. W Hafer is a researchofficer at the Federa/ Reserve Bank of St Louis. Kevin L.Kliesen provided reasearch assistance.

Interest Rates and EconomicAnnouncements

HE ANNOUNCEMENT of some governmentstatistic, like the latest unemployment rate orthe nation’s most recent trade balance, often isused as the rationale for observed changes infinancial markets that day. One reporter, for ex-ample, suggests that

[iln the early 1980s, investors were overly con-cerned with credit and monetary figures, focusingon Federal Reserve data. These days, professionalsare preoccupied with inflation, the dollar and thehealth of the economy.’

Another reporter points out the unsystematicnature of such interpretations with the wrycomment that

[tihe trade deficit doesn’t matter as much anymore. At least, not to the stock market. At leastnot this month.’

Do announcements of government statisticssystematically affect financial markets? There isa substantial literature on the relationship be-tween interest rates and stock prices and an-nouncements of the money stock. Overall, thisevidence supports the conclusion that announce-ments of the money stock had an important in-fluence on interest rates in the early 1980s.’

This influence arose when the Federal Reservefirst announced in October 1979 that it woulduse the money stock as a target for monetarypolicy, then largely disappeared in 1982 and1983 when the Federal Reserve moved awayfrom monetary aggregate targeting.~

Existing studies of the relationship betweeninterest rates or stock prices and announce-ments of other economic data find littleevidence that either is affected by these an-nouncements. For example, Pearce and Roley(1985) investigate the effect of unexpectedchanges in inflation and real activity on stockprices and find little response of stock prices.Hardouvelis (1987) examines this relationship forinterest rates and stock prices and finds thatthey are systematically affected only by an-nouncements of the money stock.

One common aspect of these studies is thatthey examine subperiods associated withchanges in monetary policy. Changes in policyregimes provide an obvious basis on which toexpect changes in the effect of money stock an-nouncements on financial markets. There is noobvious reason, however, for changes in the ef-

‘Wallace (1988).‘Sease (1989).‘This is less obvious for stock prices. As Thornton (1989)has demonstrated, money announcements are not suffi-

ciently important for stock prices to be reliably associatedwith changes in the money stock.

4See Gilbert (1985) and Thornton (1988) for a discussion ofthe changes in operating procedures.

MARCHIAPRIL 1989

35

fects of announcements of other economic datato occur only when the Federal Reserve changesoperating procedures.’ It is quite possible that atemporal association between interest rates orstock prices and the announcement of a par-ticular statistic is fleeting compared to estimatesbased on multi-year sample periods of aboutthree years.

The purpose of this paper is to examine thetemporal response of short- and long-term in-terest rates to announcements of certain keygovernment statistics. Unlike previous studies,we do not constrain the investigation by usingonly time periods of alternative monetary policyoperating procedures. Rather, we attempt todetermine whether the different announce-ments vary in importance over different timeperiods, even as short as one year.

MONEY, INFLATION AND REALACTIVITY

We examine the relationship between changesin interest rates and announcements using re-gressions that can be written as

(1) AR, = a + flU, + 8,

where AR, is the change in the interest rate inperiod t, U is a vector of the unexpected partsof the announcements of some governmentstatistics, ~, is the error term, and a and j3 de-note the set of parameters to be estimated. Wefocus on the unexpected parts of the announce-

ments because, when the change in the price ofan asset like Treasury securities or stocks ismeasured over a short period, the change in theasset’s price may be affected only by the unex-pected part of the announcement.’ For the mostpart, previous empirical analyses indicate thatchanges in interest rates are systematicallyassociated only with the unexpected part ofweekly announcements of the money stock.~Inaddition to the money stock announcements, we

study the effects of announcements of inflation,real economic activity and the trade balance.

We examine the effects on both short- andlong-term interest rates. Under the expectationshypothesis of the term structure, any differen-tial response of interest rates reflects differen-ces in the impact of the unexpected change ineconomic variables on current and predictedfuture short-term interest rates.’ If the expectedchange in money, inflation, industrial produc-tion, etc., is partly transitory, then the effect onthe current short-term rate will be larger thanon the long-term rate.’

Unexpected MOneY

The evidence in previous studies clearly in-dicates that the relationship between changes ininterest rates and the unexpected part of themoney announcement in the early 1980s is posi-tive. There are three possible explanations forthis association: an “expected liquidity” effect;an “expected inflation” effect; and a “realeconomic activity” effect.’°

‘Using only these periods to look for changes in the effectof announcements on financial markets becomes increas-ingly implausible as the changes in operating proceduresbecome more distant in the past.

‘The lack of generality of the proposition that asset pricesare affected only by the unexpected part of announce-ments is made by, among others, LeRoy (1982), especiallypp. 205-08. It can, however, be justified as an approxima-tion (Sims, 1984). The extension to interest rates, an in-verse function of the price, can be justified as anapproximation.7Some have found that the expected component of thechange in money also exerts a statistically significant ef-fect on changes in interest rates. See, for example, Hem(1985) and Belongia and Sheehan (1987). Several studieshave shown, however, that such results may stem fromcertain anomalies in the data. For example, Belongia,Hafer and Sheehan (1986) find that the significance of ex-pected changes in money stems from one observation inwhich a benchmark revision in the data coincided with aso-called Social Security week. The removal of this obser-vation reduces expected money’s coefficient to statisticalinsignificance. Other researchers, for example, Clark, et al.(1988), also have argued that the inclusion of Social

Security weeks leads to the spurious result that expectedchanges in money influence interest rate changes. For adiscussion of the effects of Social Security weeks on theobserved changes in money, see Hafer (1984).

‘The evidence in Flavin (1984) and Campbell and Shiller(1987), for example, indicates that the expectationshypothesis accounts for much of the variation of long-terminterest rates relative to short-term rates.‘Under the expectations hypothesis of the term structure,the change in the long-term interest rate is the sum of thediscounted change in expected future interest rates, aterm due to the return from holding the bond and termsdue to the expected short-term rates appearing in one butnot the other bond. See Flavin (1984), p. 231. If the coeffi-cient relating the changes in expected interest rates to theunexpected part of the announcement decreases with termto maturity, then the usual algebra indicates that theresponse will be less for long-term interest rates.

“Cornell (1983) and Sheehan (1985) discuss these explana-tions and provide useful surveys of the evidence.

FEDERAL RESERVE SANK OF St LOUIS

36

The expected liquidity effect is based on thesupposition that a larger forecast error is asso-ciated with an expectation that the FederalReserve will engage in more contractionaryopen market operations in the near futurerelative to what they would have done other-wise. As a result of the expected contractionaryopen market operations, near-term interestrates increase. The expectation of higher in-terest rates in the near future, though, raisescurrent rates to maturity on securities that ma-ture after the expected contractionary openmarket operations.” An unexpected increase inthe money stock is thus associated with an in-crease in interest rates.

An alternative explanation can be cast interms of expected inflation. Under this explana-tion, an unexpected increase in money leadseconomic agents to revise their expectations offuture inflation upward. Because nominal inter-est rates are the sum of the real interest rateand the expected inflation rate, an unexpectedincrease in expected inflation, ceteris paribus,leads to an increase in nominal interest rates.

The real economic activity effect predicts thatinterest rates will respond positively to an unex-pected money increase. According to this ex-planation, the money announcement reveals in-formation about money demand in the econ-omy. If the announced stock of money dependson the demand for money, an announcedmoney stock greater than expected indicatesthat money demand is greater than expected. Ifthe demand for money depends, among otherthings, on expectations of future real economicactivity, an unexpected increase in the moneystock reflects an increase in expected real activi-ty.12 Because economic activity and real interestrates are positively correlated, an unexpectedincrease in the money stock is associated withan increase in real and nominal interest rates.

Unexpected Iratian

Whether announcements of inflation arerelated to changes in interest rates due to an ef-fect on expected monetary policy or expectedinflation, an announcement that inflation isgreater than expected can result in an increasein interest rates. If an announcement of infla-

tion greater than expected for the recent periodincreases expected future inflation, there is adirect effect on the nominal interest rate. Onthe other hand, with a goal of lower inflation,the Federal Reserve may be expected to offsetthe higher inflation (or the perception of higherfuture inflation) by a more restrictive monetarypolicy. In either case, an unexpected increase ininflation increases nominal interest rates. As forthe money stock, the relative effect on short-term and long-term rates reflects how perma-nent the change in inflation is expected to be.The more transitory it is, the smaller therelative effect on long-term rates.

Real Activity

An unexpected increase in real activity raisesnominal interest rates through two channels.One is from agents’ revised expectations thatfuture real activity will be higher, thus causingexpected real interest rates and, hence, nominalrates to increase. The other is from the ex-pected reaction of the Federal Reserve. Ifeconomic agents expect the Fed to tightenmonetary policy on news of stronger futureeconomic growth, then interest rates can in-crease because of the expected liquidity effect.

Trade Balance

The trade balance is exports minus imports.When exports are less than imports, the tradebalance is negative, a situation that character-izes most of the 1980s. An announcement of alarger-than-expected trade balance can increaseor decrease nominal interest rates. A largertrade balance today is associated with largertrade balances in the future.13 Even with thisqualification, however, the effect of announce-ments of trade balances on interest rates is am-biguous. Because the trade balance is the nega-tive of the capital account, a larger trade bal-ance is associated with a smaller balance oncapital account. A larger trade balance and asmaller balance on the capital account can beassociated with either a decrease in the supplyof foreign funds to the United States or adecrease in the demand for funds in the UnitedStates. A decrease in the supply of funds wouldbe associated with an increase in interest rates

current one-day rate also can increase because of in-

tertemporal substitution.

“See Fama (1982).

13There is evidence of positive autocorrelation in the data.Over the sample period used in this paper, the first sixvalues of the autocorrelation function are 0.85, 0.85, 0.83,0.80, 0.76 and 0.75.

MARCH/APRiL 1989

37

in the United States.’4 A decrease in demandwould be associated with a decrease in interestrates in the United States. Given these twopossibilities and no ancillary evidence to supporteither, the hypothesized effect on nominal inter-est rates of an unexpected increase in the tradedeficit is uncertain.

THE DATA

Daily interest rates on three-month Treasurybills and 30-year Treasury bonds are used inour empirical analysis. These rates are closingquotes supplied by the New York FederalReserve, calculated as averages of rates re-ported by primary government security dealersbetween 3:15 p.m. and 4:00 p.m. Eastern Stan-dard Time. The changes in rates are measuredas the difference between the interest rate fromone day’s close to the next.”

To estimate the unexpected part of the an-nounced values of the economic series, we usethe initial announced values of the series minusthe median response from the survey conductedby MMS International.b6 This widely used sur-vey polls approximately 50 to 60 governmentsecurities dealers weekly, asking them to indi-cate their expectation of the change in the nar-row money stock (Ml). At most a week beforean announcement of several other economicseries, the survey participants also are asked toindicate their forecasts of the change in otherseries, such as the Consumer Price Index.

In this study, we use the survey forecasts forMl, the Consumer Price Index (CPI), the Pro-ducer Price Index (PPI), the industrial produc-tion index, the unemployment rate and thetrade balance, Because the survey forecasts ofthe price indexes and industrial production are

all measured in terms of monthly percentagechanges, the unexpected part of the announcedvalues also are measured as a monthly percen-tage change. The actual and the forecasted un-employment rates are both measured as percen-tages of the number of unemployed relative tothe labor force. The forecasts of Ml and thetrade balance are stated in terms of their dollarvalues. We measure the unexpected part ofthese variables as the percentage difference be-tween actual and forecasted values relative tothe actual values.

Although other economic variables obviouslymight be included in this analysis, the series us-ed in this study represent a broad range ofeconomic activity, reflecting changes in infla-tion, real activity and foreign trade. Moreover,the variances of changes in the Treasury billrate and the Treasury bond rate are greater onthe days with these announcements than onother days.”

To abstract from the effects of intervening an-nouncements, we include in our regressionschanges in interest rates only for those days onwhich these announcements occur. Since pastintervening announcements are prior informa-tion and, under rational expectations, are uncor-related with the current unexpected change,this restriction does not bias our analysis. Afuture unexpected change in a variable will becurrently unknown and, under rational expecta-tions, also would be uncorrelated with the cur-rent unexpected change.”

Means and standard deviations of the unex-pected changes in Ml, the price indexes, in-dustrial production, the unemployment rate andthe trade balance are presented in table 1.”

‘4We assume that the United States is not small relative tothe rest of the world.

“The three-month Treasury bill rate is measured using thestandard discount interest rate formula. The bond rate isthe yield to maturity.

1OMM$ International and Douglas K. Pearce providedseveral of the series examined here. Actual changes in theseries are taken from relevant government and FederalReserve publications.

“The F-statistic for testing the hypothesis that the varianceof the change in the Treasury bill rate is the same on dayswith these announcements and days without these an-nouncements is 1.86 with 679 and 1292 degrees offreedom, which has a marginal significance level of lessthan 0.001. The F-statistic for testing this hypothesis usingthe variance of the change in the Treasury bond rate is1.43, also with 679 and 1292 degrees of freedom and amarginal significance level less than 0.001 -

“This and the prior statement assume that the forecasts areessentially the same as rational expectations. It is, ofcourse, true that our estimated coefficients can be affectedby other events on the same day that are correlated withexcluded variables.

19There are 95 months used in table 1. There are only 94observations for the unexpected change in the CPI,because the survey value is missing for the announcementin January 1986. There are 94 observations on the unex-pected change in unemployment, because the Treasurybill rate is not available for April 5, 1985, when theunemployment rate was announced. Therefore, we do notuse this observation. Finally, 93 observations are used forthe trade balance, because only 11 values were announc-ed in 1987: and two values were announced on the sameday in April 1987. We use just the announcement for themore recent month, March.

FEDERAL RESERVE BANK OF St LOUIS

38

Table 1Descriptive Statistics onMeasures of the Unexpected Partof Announcements (February 1980 toDecember 1987)

Standard Number ofVariable Mean deviation observations

Narrow moneystock 0 08%- 46% 376

Consumer

price index —001 20 94

Producerpnce index -0.10 34 95

Unemploymentrate 001 21 94

Industrialproduction —0 03 .39 95

Trade balance 1318 9823 93

Uken the 111111111 it ii p1 etbiOO UI lilt’ ilota ,uid

11w - i/C of the i. son tied - tandzir d er I ols all

hut oiti’ cil the Iue.u~toni-usE i’rio~ loin

iflI’,i5LIiP~ i~l lw Uile\I)et ted COm[)UIWIIt. ot the

cillIUltiltI (iticilEsi itt not dil terent It-urn tel u.

l-c~r c’~aiiii~lt’liii tnielii1llc.,\ tttt’i~t I ale i’, in—

iiotii~ii’d I’, —I ~n-i-ttriligt-Sil\ .‘ -l ~itr ((‘Ill if_fl

I)o111t~Jitti I. lorc—i inli’tI to dii -‘iitie IN (‘I tilriiiiiiiiit al pt ii kiun. \ neati oF Cl I) I I ‘er-u

\\ itliiti tlit 1)1 l(I5ll)ii ol the data. C )nI’. (till’ iii

tlit “I’ - l’Iie~, hit’ ~il titliji ri Ut’ ii~cli’x ha-, alilt—_Ill ~altu’th it i— —,u

4rtilit-111hl\ chIli-tent. licitli

litilliCflC all_k and tatt’~ttcahI~ Ii tJlli let 0 at the

j)el cent ITIaI gititI t~,riilIt .111CC ht~el this i in —,oi

illtU’\ sis ot thic’ ddl,I alum. u itli nthei not k iii—chic-alt’’, th,it the c’ ~ut~e_~data ‘iii usiltill)lli\lliiltlOlI’ ijl I ilititi.il l’\

1)lttilLOli’, ititl 1I1 Ill

used to estimate the unexpected parts of theseries being announced.’°

An issue that generally is not dealt with whenusing these data is the fact that the measure-ment of the expected and actual changes insome of the variables is only in increments of0.1. That is, forecasts and actual values for theCPI, the PPI and the industt’ial production indexare all collected as monthly percentage changeswith only one digit after the decimal point. Be-cause there is a relatively small range of fore-cast errors at this level of precision, there are alimited number of values that the forecast er-rors actually take. Even so, the information intable I indicates that there is sufficient variationto estimate a meaningful relationship betweenthese data and changes in interest rates.’1

EMPIRICAL RESULTS

The period used in our analysis runs fromFebruary 1980 through December 1987. Thebeginning of the period is dictated by the lackof survey forecasts of the trade balance prior tothat time. The end of the period is dictated bydata availability. The vector of observations foreach right-hand-side variable includes zeros forthose days when a series is not announced.

Regressions by Year

Previous analyses generally estimate equationI over all of the available data and for periodscorresponding to changes in the Federal Re-serve’s operating procedures.” Because we areconcerned with the pattern of the coefficientestimates on money and other variables overtime, we ignore these particular periods andestimate equation 1 for the full period and for

each year. Because we have no a priori informa-tion that dictates the correct periods, this ap-proach allows us to gauge the effects of the

“For other analyses of this data, see, among others, Pearceand Roley (1985).

“For example, the forecast errors for the CPI measure ofthe inflation rate range in increments of 0.1 from -0.6 to0.5. The modal error is zero, and the forecast errors aredispersed around this value, not (as would be possible) vir-tually always .1 or .2 in absolute value. Similar commentsapply to the PPI and the industrial production index. Theforecast errors of the unemployment rate, also measuredto a precision of 0.1 percentage points, range from -0.5 to0.7 by increments of 0.1.

22As noted above, previous researchers generally delineatesample periods by changes in monetary policy operatingprocedures. These include the October 6, 1979, shift away

from emphasizing the behavior of the federal funds rateand placing more importance on the behavior of themonetary aggregates and the October 1982 shift back tointerest rates. While statistical tests generally do not rejectthe use of these breakpoints, it has been questionedwhether the procedures used are adequate to reject the apriori break point being tested. That is, if October 6, 1979,is not the true breakpoint in the relationship but anotherrelatively close date is, the test procedures used will notreject October 6 as the break. Indeed, evidence presentedin Hafer and Sheehan (1989), based on time-varyingparameter estimates, indicates that the often-used October1979 and October 1982 sample breaks are not consistentwith the data.

MARCH/APRlL 1989

39

~1A~ ~ ~ ~ ~ ~ ~/?~ ~/ ~ ~“~“ ~Le\/ \c~:/~~\ \~ ~ ~ ~

‘ \ , ~

i’~ ‘\~ ~~O\ ~ ~ ~ ~

~b~SY ~~ %~ ~ ~</ \~at r ~A~r ~ ~\

\ ,~ ~r2~2a~r ~ :~/ ~1~ ~~ ~-~s~nUc:~~~/rC~ ~ . <\c~ ~ ~ k~\ ~ ~ ~ ~~

~ t/ - :~~ ~ .>~ ~ s-~i \Z~ ~

*/ ~ct/~ /// N. Q~r

~ /~C’ , \:s~

:~s~$~~s / ~ ~-:isn~:S ~z4~~t~ ~7/~

~ ~ , ,0 ~

~ ~ / /

~

V~ , ~ .

~~S~&~ t / ~\\

~ ~- ~ ~g~a7 ~ 7 / // ~‘ ~ ~ &~~ ~-~~TIcJ~.~~t!’~!ttunexpected parts in the announcements overtime. While the choice of a year is admittedlyarbitrary, it is long enough that some precisionin the regression coefficients is possible butshort enough that it is unlikely to miss an esti-mable transitory change in the coefficients.”

The regression results are reported in table 2for the change in the Treasury bill rate.’4 Basedon a 5 percent marginal significance level, onlyunexpected money (UM1) has a statistically sig-nificant coefficient in the full-period regressions.This result does not mean, however, that othereconomic variables do not influence Treasurybill rate changes during the period. On the con-trary, the annual regression results indicate that

unexpected unemployment (UU) is marginallystatistically significant at the 5 percent level for1983. It also appears that unexpected changes inthe industrial production index (UIP) are associ-ated with increases in the short-term interestrate in 1985 and 1986. In none of the annualregressions, however, do unexpected changes inthe Consumer Price Index (UCPI), the ProducerPrice Index (UPPI) or the trade balance (UTB)have statistically significant coefficients.

The regressions using the change in the 30-year Treasury bond rate are presented in table3. The regression for the full period again has astatistically significant estimated coefficient forthe unexpected part of Ml. The magnitude of

23It is possible, of course, that the estimated coefficientschange with each announcement. Without the impositionof constraints on the way that the coefficients change,however, such a specification is not estimable. Our regres-sion coefficients can be interpreted as estimates of theaverage coefficient in a given year.

‘4Again note that the estimation uses only unexpectedchanges in the variables. Since correlations between theexpected and unexpected values reveal that the two seriesare uncorrelated, omitting the expected values does notbias the estimated regression coefficients.

FEDERAL RESERVE SANK OF ST. LOUIS

40

< K N~ ~cN ~

~mp~cN~ \~*%r : Y~ P111k ‘~S;2N ~s

- ~KK K KKeat ~ K.t7Z~t r

K.K VIII~

It lit ~~ i~~n ~SMan~ —~K. K K K.K~K~. N

~- T -~ 11$

N

~c~’

the estimated coefficient is less than one-halfthat of the short-term Treasury bill rate, a re-sult that is consistent with previous work” Inaddition to money announcements, the full-period regression suggests that unexpectedchanges in the PPI have a positive and statistic-ally significant effect on the change in theTreasury bond rate.

Except for 1980, however, the separate annualresults provide little evidence that the unex-pected changes in these economic variableshave much effect on changes in the long-terminterest rate. In the results for 1980, unexpec-ted money and inflation measured by the PPIare statistically significant at the 5 percent level.The coefficient on industrial production is signi-ficant at the 5 percent level in 1987. Note, how-ever, that the sign of this estimated coefficient(negative) is incorrect. The estimated coefficientfor the unexpected part of the trade balance issignificant at the 6 percent marginal significancelevel in 1987. Interestingly, while the coefficient

often is negative, it is positive for 1987. For theother years, the estimation results are consis-tent with the proposition that unexpected partsof announcements of variables besides moneyhave little effect on the change in the 30-yearrate.

Stability Tests

An important aspect of the regression resultsin tables 2 and 3 is the variability in the estim-ated coefficients over time. For example, con-sider the magnitude of the estimated coeffici-ents on unexpected money from 1980 to 1987in table 2. Based on the annual regression re-sults, the estimated coefficient peaks at 0.35 in1981 and declines to essentially zero in 1987.This result is consistent with the hypothesis thatunexpected changes in the money stock areassociated with changes in interest rates earlyin the period but not recently.

‘5For example, see Cornell (1983).

MARCH/APRtL 1989

41

To investigate whether the estimated coeffi-cients from the various years are statisticallydifferent, two tests are conducted. One testdetermines whether the coefficients for eachvariable change over time. We test whethereach variable’s coefficients are equal from 1980to 1987. The results of these tests, regardless ofthe interest rate used, are consistent with thehypothesis that only the coefficients on unex-pected money vary across years. The F-statisticfor unexpected money when the change in theTreasury bill rate is used is 4.48. The result us-ing the Treasury bond rate is an F-statistic of2.68. Both are statistically significant at less thanthe 1 percent marginal significance level.b6 TheF-statistics for the remaining variables are in-significant: they almost never even exceedunity.’~

While this test has reasonable power againstthe alternative hypothesis that the coefficientsare nonzero and change over several of theyears, it generally has low power against thealternative that a variable has a nonzero coeffi-cient for a relatively short period such as oneyear. Consequently, testing the coefficients oversingle years is a useful additional test.

Testing the hypothesis that a coefficient inany single year is the same as in the remainderof the years provides at most marginal evidenceof coefficient instability across the period.’8 Us-ing a S percent marginal significance level, testsusing the Treasury bill regressions indicate thatthe coefficient on the unexpected part of thetrade balance in 1980 is statistically differentfrom the coefficients in the rest of the period:the estimated t-statistic is —2.33.’°With the ex-ception of unexpected money, each of the other

coefficients for the Treasury bill rate is equalover time. Besides unexpected money, only theunexpected part of the industrial production in-dex in 1980 has a coefficient for the Treasurybond rate that is statistically different from theremaining years (t = 2.30).

These test results are largely consistent withthe hypothesis that the response of interestrates to unexpected changes in the variablesother than money are constant and equal tozero.

Rolling Regression Estimates

Breaking the eight years into annual segmentsto estimate the changes in the coefficients overtime may obscure changes that occur duringthe years. To investigate the evolution of the es-timated coefficients, it is worthwhile to examinethe coefficients in a relatively unrestricted man-ner. This can be done by estimating regressionsthat roll through the sample. Equation 1 is esti-mated for successive 12-month periods, adding amonth and dropping a month as the estimationof the regression coefficients rolls through thefull period.” The first 12-month period beginsin February 1980 and ends in January 1981; thelast sample ends with December 1987. While us-ing a 12-month period for the rolling regres-sions still has an arbitrary element, theestimated coefficients for any 12-month periodare readily available and can be examined.”

To show how the estimated coefficients haveevolved over the period, the estimated coeffi-cients for both the Treasury bill and Treasurybond rates are plotted in figure 1. In interpret-ing these plots, it is important to note that, be-

2 The results from a standard F-test are consistent with thenull hypothesis of overall coefficient stability. In this test,each variable including the constant is allowed to take dif-ferent values for each year. This unrestricted equation iscompared with the equation where all estimates are fixedfor the full period. The calculated F-statistic for changes inthe Treasury bill rate is 1.24. When the change in theTreasury bond rate is used, the F-statistic is 1.06. Both ofthese values are less than the 5 percent critical value.Such a test, however, may mask changes in one or twovariables’ coefficients. Given the number of variables andtime periods, changes in the estimated coefficient forsome variable can be swamped by the stability of theothers.

‘7Using the change in the Treasury bill rate, the variablesand corresponding F-statistic are: CPI (0.08); PPI (0.95);unemployment (0.64); industrial production (0.61); andtrade balance (0.85). Using the change in the Treasurybond rate, the F-statistics are: CPI (0.25); PPI (1.08);unemployment (0.36); industrial production (1.24); andtrade balance (0.98).

‘~Thistest is first run with the coefficients of all othervariables allowed to be different, then with the coefficientsof all the other variables besides money set equal for all ofthe years. In the text, we report the results with the coeffi-cients of other variables besides money set equal to eachother for all of the years. The results with other coeffi-cients allowed to vary are little different than thosediscussed.

~ the multiple tests across variables and years, thereare good reasons to use a smaller significance level. Ifone desires an overall 5 percent significance level for allthe tests combined, the correct significance level fortesting the stability of the coefficients for each year andeach variable is about one-tenth of 1 percent.

‘°Loeys(1985) examines the effects of unexpected moneyon interest rates in a similar manner.

‘1We also estimated the rolling regressions using successive18-month periods. There are only minor changes in theresults.

FEDERAL RESERVE BANK OF St LOUIS

Figure 1Panel AThe Coefficients of the Unexpected Component ofMl for the 3-Month Treasury Bill

42

The Coefficients of the Unexpected Component ofMl for the 30-Year Treasury Bond

NOTE: A dashed line indicates significance at the 5 percent level.

The Coefficients of the Unexpected Component ofCPI for the 30-Year Treasury Bond

1991 92 99 94 95 ‘9 ~997

Panel BThe Coefficients of the Unexpected Component ofCPI for the 3-Month Treasury Bill

-4

-3

-0

Panel CThe Coefficients of the Unexpected Component ofPPI for the 3-Month Treasury Bill

-2

-3

The Coefficients of the Unexpected Component ofPPI for the 30-Year Treasury Bond

19S~ 92 63 66 65 66 997 1861 62 63 64 65 66 997

MARCH/APRIL 1989

43

-Panel DThe Coefficients of the Unexpected Component ofThe Unemployment Rate for the Treasury Bill

Panel EThe Coefficients of the Unexpected Component ofIndustrial Production for the 3-Month Treasury Bill

The Coefficients of the Unexpected Component ofThe Unemployment Rate for the Treasury Bond

The Coefficients of the Unexpected Component ofIndustrial Production for the 30-Year Treasury Bond

Panel FThe Coefficients of the Unexpected Component ofThe Trade Balance for the 3-Month Treasury Bill

5691 92 53 94 65 65 5997

1995 55 5997 9661 62 63 94 65 65 5987

The Coefficients of the Unexpected Component ofthe Trade Balance for the 30-Year Treasury Bond

5985 62 63 54 65 ‘5 ~967 1565 62 63 94 65 56 5967

FEDERAL RESERVE BANK OF St LOUIS

44

cause common observations in regressions areseparated by less than 12 months, the estimatedregression coefficients within a 12-month periodare not independent. This implies that somesmoothness in the plotted variation of the coef-

ficient estimates is to be expected even if allestimated coefficients are zero and any varia-tion is random.

in addition, the coefficient estimates for thetwo interest rates are not statistically indepen-dent. The simple correlation of the change inthe bill rate and the bond rate is 0.658 for 1980through 1987 on days with announcements, acorrelation that is statistically significant at theS percent level. This means that, if theestimated coefficient of industrial production,for instance, increases in the regression for thebill rate, the estimated coefficient of industrial

production in the regression for the bond rate

is likely to increase as well, even if the increaseis due to random variation. Despite thesecaveats, these estimates are useful because theymake it possible to examine the inter-yearchanges in the estimates for all possible dates.

Panel a of figure 1 shows the estimated coeffi-cients on unexpected changes in Ml.” The esti-mated coefficients in the regression for the billrate and the bond rate track each other with alarger estimated coefficient for the bill rate until1984, when the estimated coefficients convergeto zero. Finding that the effect of unexpectedmoney on changes in the interest rate becomessmaller after the shift in the Federal Reserve’soperating procedure in late 1982 is consistentwith previous work.’~

Panels b and c show the estimated coefficients

of unexpected increases in the inflation rate asmeasured by the CPI and the PPI. In the regres-sions for the Treasury bill rate, not one esti-mated coefficient is statistically significant forany 12-month period using a 5 percent marginalsignificance level. In the regressions for theTreasury bond rate, only estimated coefficientsfor the PPI in nine months in 1981 and threemonths in 1984 are statistically significant using

a standard 5 percent marginal significance level.There is no evidence that the unexpected partof announcements of the CPI affect interestrates for any period as long as 12 months from1980 to 1987. One interpretation consistent withthese regression results is that there is some

evidence of a relationship between the unex-pected part of inflation as measured by the Pro-ducer Price Index in 1980, but little afterwards.Such an interpretation requires that the pointestimates of the regression coefficients be view-ed as indications that the unexpected parts ofthe announcements had stronger implicationsfor inflation over a period longer than the

three-month maturity of the Treasury bill rate.

Real activity is represented by the unemploy-ment rate (an inverse indicator) and industrialproduction. The estimated coefficients of theunemployment rate are presented in panel d. Inthe regressions for the Treasury bill rate, thecoefficient of the unemployment rate is negativeand statistically significant at the 5 percentmarginal significance level during late 1983 andearly 1984. While the estimated coefficient forthe bond rate is not statistically significant dur-ing this period, the negative and smaller (inmagnitude) coefficient is consistent with thehypothesis that the unexpected part of an-

nouncements of the unemployment rate affectinterest rates in this period. The estimated coef-ficients for industrial production (panel e) alsoprovides some evidence consistent with thehypothesis that announcements of it have af-fected interest rates. In particular, frommid-1985 through much of 1986, the Treasurybill and bond rates both have sharply increasingestimated coefficients on unexpected increasesin industrial production. For the Treasury billrate equations, these coefficients are significantat the 5 percent level. The positive sign is con-sistent with a rationalization in terms ofmonetary policy, with higher growth beingfollowed by expectations of relatively contrac-tionary monetary policy, and in terms of ex-pected higher future growth signaling higherreal interest rates.

Finally, the estimated coefficients for the tradebalance, shown in panel f, never provide muchsupport for a systematic relationship except per-haps at the start of the sample. Not one of the168 estimated coefficients in the regressions forthe bill and bond rates is statistically significantusing a 5 percent marginal significance level.

CONCLUSION

How do financial markets respond to theunexpected part of announcements of govern-

‘2Dashed lines denote statistical significance at standard 5percent marginal significance level.

“See, for example, Hardouvelis (1987) and Hafer andSheehan (1989).

MARCH/APRIL 1989

45

ment statistics? Based on the evidence presentedin this article, the answer is, very little. Using

regression analysis, statistical tests of coefficientstability and rolling regressions to detect coeffi-cient variability from 1980 to 1987, we find thatonly the unexpected changes in the moneystock have a systematic effect on interest rates.Even then, it appears that the significant effectspeter out by late 1982.

For none of the other variables examined dowe find evidence of a reliable effect on interestrates over the period. This set of variables in-cludes measures of inflation, real economic acti-vity and foreign trade. Failing to find any sys-tematic relationship between interest ratechanges and these non-monetary variables hastwo implications. One is that explanations of theresponse of interest rates to monetary an-nouncements that emphasize changes in econ-omic agents’ expectations of future inflation andreal economic activity may be off the mark for1980 through 1982. If these explanations werecorrect, such effects should be evident when in-flation and real variables themselves are used.Our results, however, reveal little effect fromthe unemployment rate or industrial production.Theories that are premised on the response ofinterest rates to expected changes in monetarypolicy are more consistent with our empiricalresults.

The other implication concerns the effect oninterest rates perceived by financial marketanalysts when government statistics are an-nounced. We find no consistent response of in-terest rates, either short term or long term, tounexpected changes in the different non-mone-tary variables. We do find evidence consistentwith the hypothesis that the unexpected partsof announcements of the Producer Price Index

in 1980, the unemployment rate in 1983 and in-dustrial production in 1980, 1985 and 1986 areassociated with changes in interest rates. Therelative infrequency of these significant effectscan be interpreted in one of two ways. Thefirst is that, of the 80 estimated annual coeffi-cients, it is hardly surprising that five arestatistically significant at the 5 percent marginalsignificance level. A conclusion that all of thecoefficients are zero is therefore consistent withthe results. The second is that, except for an-nouncements of the money stock in the early1980s, responses of interest rates to an-nouncements are episodic, short-lived affairs.

REFERENCESBelongia, Michael t, R. W. Hafer, and Richard G. Sheehan.“On the Temporal Stability of the Interest Rate-WeeklyMoney Relationship:’ Review of Economics and Statistics(August 1988), pp. 516-20._______ - “A Note on the Temporal Stability of the In-terest Rate-Weekly Money Relationship,” Working Paper86-002, Federal Reserve Bank of St. Louis (1986).

Belongia, Michael T., and Richard G. Sheehan. “The Infor-mational Efficiency of Weekly Money Announcements: AnEconometric Critique:’ Journal of Business and EconomicStatistics (July 1987), pp. 351-56.

Campbell, John Y., and Robert J. Shiller. “Cointegration andTests of Present Value Models,” Journal of PoliticalEconomy (October 1987), pp. 1062-88.

Clark, Truman A., Douglas H. Joines, and 0. MichaelPhillips. “Social Security Payments, Money Supply An-nouncements, and Interest Rates:’ Journal of MonetaryEconomics (September 1988), pp. 257-78.

Cornell, Bradford. “The Money Supply Announcements Puz-zle: Review and Interpretation:’ American Economic Review(September 1983), pp. 644-57.

Fama, Eugene F. “Inflation, Output, and Money,” Journal ofBusiness (April 1982), pp. 201-32.

Flavin, Marjorie. “Time Series Evidence on the ExpectationsHypothesis of the Term Structure’ in Karl Brunner andAllan H. Meltzer, eds., Monetary and Fiscal Policies andTheir Applications, Carnegie-Rochester Conference Serieson Public Policy, Volume 20 (Amsterdam: North-Holland,1984).

Gilbert, R. Alton, “Operating Procedures for ConductingMonetary Policy:’ this Review (February 1985), pp. 13-21.

Hater, R. W. “Comparing Time-Series and Survey Forecastsof Weekly Changes in Money: A Methodological Note:’Journal of Finance (September 1984), pp. 1207-la

Hafer, R.W., and Richard 0. Sheehan. ‘On the Response ofInterest Rates to Unexpected Weekly Money: Are PolicyChanges Important?” Federal Reserve Bank of St. LouisResearch Paper 87-005 (revised, February 1989).

Hardouvelis, Gikas A. “Macroeconomic Information andStock Prices’ Journal of Economics and Business (May1987), pp. 131-40.

Hem, Scott E. “The Response of Short-Term Interest Ratesto Weekly Money Announcements: A Comment:’ Journal ofMoney, Credit and Banking (May 1985), pp. 264-71.

LeRoy, Stephen F. “Expectations Models of Asset Prices: ASurvey’ Journal of Finance (March 1982), pp. 185-217.

Loeys, Jan 0. “Changing Interest Rate Responses to MoneyAnnouncements: 1977-83:’ Journal of Monetary Economics(May 1985), pp. 323-32.

Pearce, Douglas K., and V. Vance Roley. “Stock Prices andEconomic News:’ Journal of Business (January 1985), pp.49-67.

FEDERAL RESERVE BANK OF St LOUIS

46

Sease, Douglas R. “Trade Deficit’s Impact Declines Despite Sims, Christopher A. “Martingale-like Behavior of Prices andJitters,” Wall Street Journal, February 21, 1989. Interest Rates,” University of Minnesota Center for Eco-

nomic Research Discussion Paper No. 205.

Sheehan, Richard G. ‘Weekly Money Announcements: New Thornton, Daniel L. “Why Do Market Interest Rates RespondInformation and Its Effects:’ this Review (AugustlSeptember to Money Announcements?” Federal Reserve Bank of St.1985), pp. 25-34. Louis Research Paper No. 88-002, 1988.

_______ “The Borrowed-Reserves Operating Procedure:Shiller, Robert J., John Y. Campbell, and Kermit L. Theory and Evidence:’ this Review (January/February

Schoenholtz. “Forward Rates and Future Policy: Inter- 1988), pp. 30-54.preting the Term Structure of Interest Rates,” Brookings Wallace, Anise C. “The Numbers that Move the Market,”Papers on Economic Activity (1: 1983), pp. 173-217. New York Times, June 5, 1988.

MARCH/APR1L 1989