The Journal of Finance - December 1989 - SCHWERT - Why Does Stock Market Volatility Change Over Time

THE JOURNAL OF FINANCE * VOL. XLIV, NO.
5 * DECEMBER 1989
Why Does Stock Market Volatility Change Over

Time?
G. WILLIAM SCHWERT*
ABSTRACT
Thispaperanalyzestherelation ofstockvolatility
withrealandnominalmacroeconomic
volatility,
economicactivity,financialleverage,andstocktrading usingmonthly
activity
data from1857to 1987.An important fact,previously
notedbyOfficer (1973),is that
stockreturnvariabilitywas unusuallyhighduringthe 1929-1939GreatDepression.
Whileaggregateleverageis significantly itexplainsa relatively
withvolatility,
correlated
smallpartofthemovements in stockvolatility.
The amplitudeofthe fluctuations in
aggregatestockvolatility
is difficult
to explainusingsimplemodelsofstockvaluation,
duringtheGreatDepression.
especially
ESTIMATES OF THE STANDARD deviation of monthly stock returns vary from two
to twentypercent per month duringthe 1857-1987 period. Tests for whether
differences this large could be attributableto estimationerrorstronglyrejectthe
hypothesisofconstantvariance.Large changes in the ex ante volatilityof market
returns have important negative effectson risk-averse investors. Moreover,
changes in the level of marketvolatilitycan have importanteffectson capital
investment,consumption,and other business cycle variables. This raises the
question of whystock volatilitychanges so much over time.
Many researchershave studiedmovementsin aggregatestockmarketvolatility.
Officer(1973) relatesthese changes to the volatilityof macroeconomicvariables.
Black (1976) and Christie (1982) argue that financial leverage partly explains
this phenomenon.Recently,therehave been manyattemptsto relate changes in
stockmarketvolatilityto changesin expectedreturnsto stocks,includingMerton
(1980), Pindyck (1984), Poterba and Summers (1986), French, Schwert, and
Stambaugh (1987), Bollerslev, Engle, and Wooldridge (1988), and Abel (1988).
Mascaro and Meltzer (1983) and Lauterbach (1989) find that macroeconomic
volatilityis relatedto interestrates.
Shiller (1981a,b) argues that the level of stock marketvolatilityis too high
relativeto the ex post variabilityof dividends.In presentvalue models such as
Shiller's, a change in the volatilityof eitherfuturecash flowsor discountrates
* WilliamE. SimonGraduateSchoolof BusinessAdministration, University and

of Rochester,
NationalBureauof EconomicResearch.I receivedhelpfulcomments fromDavid Backus,Fischer
Black,MarieDavidian,HarryDeAngelo,Beni Lauterbach, Ron Masulis,GrantMcQueen,Robert
Merton,Dan Nelson,CharlesPlosser,Paul Seguin,RobertStambaugh, JeroldZimmerman, seminar
at Yale University
participants and at the Universitiesof Chicago,Michigan,Rochester,and
Washington, and threeanonymous Ken Frenchand Rene Stulzdeservespecialcreditfor
referees.
theirhelp.The BradleyPolicyResearchCenterat theUniversityofRochester providedsupportfor
thisresearch.
1115
15406261, 1989, 5, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/j.1540-6261.1989.tb02647.x by Royal Danish Library, Wiley Online Library on [12/02/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
1116 The JournalofFinance
causesa changein thevolatility ofstockreturns. Therehavebeenmanycritiques

ofShiller'swork,notablyKleidon(1986).Nevertheless, on "excess
theliterature
volatility"has notaddressedthequestionofwhystockreturnvolatility is higher
at sometimesthanat others.
This papercharacterizes thechangesin stockmarketvolatility through time.
In particular,it relatesstockmarketvolatility volatility
to thetime-varying ofa
varietyof economicvariables.Relativeto the 1857-1987period,volatility was
unusuallyhighfrom1929to 1939formanyeconomicseries,including inflation,
moneygrowth, industrial production,and othermeasuresof economicactivity.
Stockmarketvolatility increaseswithfinancialleverage,as predictedby Black
and Christie,althoughthisfactorexplainsonlya smallpartofthevariationin
stockvolatility. In addition,interestrate and corporatebond returnvolatility
are correlatedwithstockreturnvolatility. Finally,stockmarketvolatilityin-
creasesduringrecessions.None ofthesefactors, however,playsa dominantrole
in explainingthebehaviorofstockvolatility overtime.
It is usefulto thinkof the stockprice,Pt as the discountedpresentvalue of
expectedfuture cash flowsto stockholders:
Et-, Pt = Et-, E0 Dtk

kDtk
k=1[l + Rt+k]
whereDt+k is thecapitalgainplusdividendspaid to stockholders in periodt + k
and 1/[1 + Rt+kIis the discountrate forperiodt + k based on information
availableat timet- 1. (Et-,denotestheconditional expectation.) The conditional
varianceofthe stockpriceat timet - 1, vart-1(Pt),dependson the conditional
variancesofexpectedfuturecash flowsand offuturediscountrates,and on the
conditional covariancesbetweentheseseries.'
At the aggregatelevel,the value of corporateequityclearlydependson the
healthof the economy.If discountrates are constantover time in (1), the
conditional varianceofsecurity pricesis proportional to theconditional variance
oftheexpectedfuturecash flows.Thus,it is plausiblethata changein thelevel
ofuncertainty aboutfuture macroeconomic conditions wouldcausea proportional
changein stockreturnvolatility.2 If macroeconomic data provideinformation
aboutthevolatility ofeitherfutureexpected cash flows or futurediscountrates,
theycan helpexplainwhy stock return changes
volatility overtime."Fads" or
"bubbles" in stockprices would introduce additional sources ofvolatility.
I
Section describesthetimeseriesproperties ofthe data and the strategyfor
modelingtime-varying Section
volatility. II analyzes the relations of stockand
bondreturnvolatility withthevolatility ofinflation, moneygrowth, and indus-
trialproduction.SectionIII studies the relationbetween stock market volatility
'The variance of the sum of a sequence of ratios of randomvariables is not a simple functionof
the variances and covariances of the variables in the ratios,but standardasymptoticapproximations
depend on these parameters.
2 For a positivelyautocorrelatedvariable, such as the volatilityseries in Table II, an unexpected
increase in the variable implies an increase in expected futurevalues of the series formany steps
ahead. Given the discountingin (1), the volatilityseries will move almostproportionally.See Poterba
and Summers (1986) fora simple model that posits a particularARIMA process forthe behaviorof
the time-varying parametersin a relatedcontext.
WhyDoes StockMarket VolatilityChange Over Time? 1117
and macroeconomicactivity.Section IV analyzes the relationbetweenfinancial

leverage and stock returnvolatility.Section V analyzes the relation between
stock markettradingactivityand volatility.Finally, Section VI synthesizesthe
resultsfromthe precedingsections and presentsconcludingremarks.
I. The Time Series Behavior of Stock and Bond Return Volatility

A. VolatilityofStock Returns
Following French, Schwert,and Stambaugh (1987), I estimate the monthly
standarddeviation of stock returnsusing the daily returnsto the Standard and
Poor's (S&P) composite portfoliofromJanuary 1928 throughDecember 1987.
The estimatesfromFebruary1885 throughDecember 1927 use daily returnson
the Dow Jones composite portfolio.(See Schwert (1989d) for a more detailed
descriptionof these data.) The estimatorof the variance of the monthlyreturn
is the sum ofthe squared daily returns(aftersubtractingthe average daily return
in the month):
Nt
at rE1izt, (2)
wherethere are Nt daily returnsritin month t.3Using nonoverlappingsamples

of daily data to estimate the monthlyvariance creates estimationerrorthat is
uncorrelatedthroughtime.4
Daily stock returndata are not readilyavailable before1885. Also, macroeco-
nomic data are rarelymeasuredmoreoftenthan monthly.To estimatevolatility
frommonthlydata, I use the followingprocedure:
(i) Estimate a 12th-orderautoregressionforthe returns,includingdummy
variables Djt to allow fordifferentmonthlymean returns,using all data
available forthe series,
12 12
Rt = aj Djt + Rt( i + et. (3a)

j=1 i=1
(ii) Estimate a 12th-orderautoregressionforthe absolute values ofthe errors

from (3a), including dummyvariables to allow for differentmonthly
standarddeviations,
12 12
I It| = yjDjt + pi I Et-I + Ut. (3b)

j=1 i=1
'French, Schwert,and Stambaugh (1987) use one lagged cross-covariancein (2), and theymake
no adjustmentforthe mean return.Their estimatoris not guaranteedto be positive.Indeed, forone
monthin the 1885-1927period,the French,Schwert,and Stambaughestimateofvolatilityis negative.
The estimates from(2) are verysimilar to the French, Schwert,and Stambaugh estimates,except
that theyare always positive.
If the data are normallydistributed,the variance of the estimatorat is a2/2Nt,where a 2 is the
truevariance (Kendall and Stuart (1969, p. 243)). Thus, forNt = 22 and ot= 0.04, the standarderror
of at is 0.006, which is small relative to the level of ot. Since this is a classic errors-in-variables
problem,the autocorrelationsof the estimatesat will be smallerthan,but will decay at the same rate
as, the autocorrelationsof the truevalues ot.
(iii) The regressandleI is an estimateof the standarddeviationof the stock

marketreturnformontht similarto ut(althoughit uses one ratherthan
22 observations).The fittedvalues from(3b) I itIestimatethe conditional
standarddeviationof Rt,giveninformationavailable beforemontht.5
This method is a generalization of the 12-month rolling standard deviation
estimatorused by Officer(1973), Fama (1976), and Merton (1980) because it
allows the conditionalmean returnto varyover time in (3a) and allows different
weights for lagged absolute unexpected returns in (3b). It is similar to the
autoregressiveconditional heteroskedasticity(ARCH) model of Engle (1982).
Davidian and Carroll (1987) arguethat standarddeviationspecificationssuch as
(3b) are more robustthan variance specificationsbased on Et.They also argue
that iteratedweightedleast squares (WLS) estimates,iteratingbetween(3a) and
(3b), providemoreefficientestimates.Followingtheirsuggestion,I iteratethree
timesbetween (3a) and (3b) to computeWLS estimates.
Figure 1 plots the predictedstandarddeviationsfrommonthlyreturnsIestIfor
1859-1987, along with the predictedstandard deviations fromdaily returnsat
(froma 12th-orderautoregressionfor ?t as in (3b)) for 1885-1987. Volatility
predictionsfromthe daily data are much higherfollowingthe 1929 and 1987
stockmarketcrashes because therewereverylargedaily returnsin October1929
and October 1987. Otherwise,Figure 1 shows that the predictedvolatilityseries
are similar.Stock returnvolatilityis persistentover time.
B. VolatilityofBond Returns
If the underlyingbusiness riskof the firmrises,the risk of both the stock and
the bonds of the firmshould increase. Also, ifleverageincreases,both the stocks
and the bonds of the firmbecome more risky.Thus, in many instances the risk
of corporate stock and long-termcorporate debt should change over time in
similarways.
Figure 2 plots the predictedstandard deviations of long-termcorporatebond
returnsIerht for 1859-1987. It also shows the predictedstandard deviations of
stock returns 1 stj for comparison. Note that the scale of the right-handbond
returnaxis is about three times smaller than the scale of the left-handstock
returnaxis, showing that the standard deviation of monthlystock returnsis
about threetimes largerthan forbond returnsover this period. There are many
similaritiesbetweenpredictedvolatilitiesofstockand bond returns.In particular,
volatilitywas veryhigh from1929 to 1939 comparedwith the rest of the 1859-
1987 period.Moreover,bond returnswereunusuallyvolatilein the periodsduring
and immediatelyfollowingthe Civil War (1861-1865). In recenttimes,the "OPEC
oil shock" (1973-1974) caused an increase in the volatilityof stock and bond
returns.
Figure 3 plots the predictedstandard deviations of short-terminterestrates
I YrstI
for1859-1987. The volatilityof Inttmeasurestime variationin the ex ante
5 Since the expectedvalue of the absolute erroris less than the standard deviation froma normal
distribution,El stI = at(2/1r)',all absolute errorsare multipliedby the constant (2/lr)-Y?= 1.2533.
Dan Nelson suggestedthis correction.
0.25 - 0.25
0.2 --0.2
c c
* 0
0.1 5 0.15 C4
c c
o 0
o 0
0.1 - M t - R 0.1
0 ~~~~I0
'"I
0 0
1859 1869 1879 1589 1899 1909 1919 1929 1939 1949 1959 1969 1979 1989
89- eebr18
- - Monthly Returns -aur Daily Returns
Figure 1. Predictions of the monthly standard deviation of stock returns based on

monthly data (---) for 1859-1987 and on daily data (-) for 1886-1987. For monthly
returns,a 12th-orderautoregressionwith differentmonthlyinterceptsis used to model returns,and
then the absolute values of the residuals are used to estimatevolatilityin montht. For daily returns,
the returnsin the monthare used to estimatea sample deviationforeach month.To modelconditional
volatility,a 12th-orderautoregressivemodel with differentmonthlyinterceptsis used to predictthe
standarddeviationin montht based on laggedstandarddeviationestimates.This plot contains fitted
values fromthe volatilityregressionmodels.
nominal interestrate,not risk,since these securitiesare essentiallydefaultfree.6

Note that the right-handinterestrate volatilityscale is over 12 times smaller
than the left-handstock volatilityscale. There are periods in the 19th century
when short-terminterestrate volatilityrose for briefperiods, many of which
were associated with banking panics. (See Schwert (1989b).) It is clear from
Figures2 and 3 that long-termbond returnand short-term interestrate volatility
increased dramaticallyaround 1979. There is not a similar increase in stock
returnvolatility.As noted by Huizinga and Mishkin (1986), the Federal Reserve
Board changed its operatingproceduresto focus on monetaryaggregatetargets
at that time.
The plots in Figures 2 and 3 are consistentwith the followingsimple story.
Short-terminterestrate and long-termbond returnvolatilityhave similarities
due to inflationand monetarypolicy.Stock and long-termbond returnvolatility
have similaritiesdue to real financialand business risk.
Table I contains means, standard deviations, skewness,and kurtosiscoeffi-
6 See Fama (1976) foran analysis of the variabilityof short-termnominal interestrates.
1120 The Journal of Finance
0.25 - 0.08
-0.07
0.2
5 ~~~~~~~~~~~~~~~
0 0
-00
o 0.1
*0
C0
0.
> ~~~~~~~~~~~~~~~
>
0* 0.02 c
0.05 "t
0 0
1859 1869 1879 1889 1899 1909 1919 1929 1939 1949 1959 1969 1979 1989
January1859 - December 1987
- - Stock Returns - Corp Band Returns
Figure 2. Predictionsof the monthlystandarddeviationsof stock returns(---) and of

high-gradelong-termcorporatebond returns(-) for1859-1987. A 12th-order autoregres-
sionwithdifferent monthly interceptsis usedto modelreturns,and thentheabsolutevaluesofthe
residualsare used to estimatevolatilityin montht. To modelconditional a 12th-order
volatility,
autoregressivemodelwithdifferent monthly is usedto predictthestandarddeviationin
intercepts
montht based on laggedstandarddeviationestimates.This plot containsfittedvaluesfromthe
volatility
regressionmodels.
cients and autocorrelationsof the estimates of stock returnvolatilitybased on

monthlyand daily data, It I and ?t. It also contains summarystatistics for
estimatesof the volatilityof short-and long-termbond returns,IrstI and l,
inflation,It 1,moneygrowth,I mtI,and industrialproduction, i I*7
Table II summarizesthe autoregressionsused to predictvolatility.The sum of
the autoregressivecoefficientsmeasures the persistenceof the volatilityseries,
wherea value of unityimpliesnonstationaritv.(See Engle and Bollerslev (1986)
fora discussion of integratedconditional heteroskedasticity.)The F-test meas-
ures whetherthere is significantdeterministicseasonal variationin the average
volatilityestimates. The coefficientof determinationR2 and the Box-Pierce
(1970) statisticQ(24) measurethe adequacy of the fitof the model.
As suggestedby the analysis in footnote1, the estimates of volatilityfrom
daily data have much less errorthan the estimates frommonthlydata. The
sample standard deviation of I' I is about sixtypercent largerthan that of Ct
from1885 to 1987,thoughthe average values are similar.Moreover,the autocor-
ISeeTable Al in theAppendixfora briefdescription

ofthevariablesusedinthis-paper.
0.25 - 0.02
0.018
0.2 - -0.016
3 | | ., ,, p0.014 D
.c 0.15 -0.012 L
'X 0.01
.0>
>~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-0.006 L
I
~~~~~~~~~~
01 85 6 7 189189199 2 199199199 6 7 89
~~~~~~~~~~~~
1859 1869 1879 1889 1899 1909 1919 1929 1939 1949 1959 1969 1979 1989
January 1859 - December 1987
- - Stock Returns - Short Interest Rate
Figure 3. Predictions of the monthly standard deviations of stock returns (---) and of
short-term interest rates ( ) for 1859-1987. A 12th-orderautoregressionwith different
monthlyinterceptsis used to model returnsor interestrates, and then the absolute values of the
residuals are used to estimate volatilityin month t. To model conditional volatility,a 12th-order
autoregressivemodel with differentmonthlyinterceptsis used to predictthe standard deviation in
month t based on lagged standard deviation estimates. This plot contains fittedvalues fromthe
volatilityregressionmodels.
relationsof ?t are much largerthan those of IC'st thoughthey decay slowlyfor

both series. This slow decay shows that stock volatilityis highlypersistent,
perhaps nonstationary.(See Poterba and Summers (1986) and Schwert (1987)
forfurtherdiscussion.) The correlationbetween I'st and ?t is 0.56 from1885 to
1987, and the correlationbetween the volatilitypredictions lestj and at is 0.78
from1886 to 1987. The two methods of predictingvolatilityhave similar time
series properties.
The autocorrelationsin Table I and the summarystatisticsforthe estimated
models in Table II are similar forall the volatilityseries. The autocorrelations
are small (between 0.2 and 0.4), but they decay veryslowly.This is consistent
withconditionalvolatilitybeing an integratedmovingaverageprocess,so shocks
to volatilityhave both permanentand transitoryparts. The unit root tests in
Table II show that mostofthe sums ofthe autoregressivecoefficientsare reliably
differentfromunityusing the tables in Fuller (1976). However,Schwert (1987,
1989a) shows that the Fuller criticalvalues are misleadingin situationssuch as
this.The estimationerrorin the monthlyvolatilityestimatesbiases the unit root
on theThe
PPI order
Daily errors
est daily
rates
Monetary
Industrial
Monthly
Monthly
Monthly Monthly
Monthly
frommonthly
Bond
growth long-term
long-term stock
inflation rates stock summarySummary
base stock
stock
Volatility this
rates bond bond
rates returns
short-term autoregression
production
growth high-quality
returns
returns modelstandardReturns,
Series returns
statistics
medium-quality
returnsinter-
returns with
are
within andStatistics
estimate the
deviation
the
thedifferent
for
Period
Sample
1890-19871858-1987
1920-1987
1879-1987 1858-1987 1858-1987 month. means,
1885-1987
1858-1987
1885-1987 Growth
estimates
Formonthly
monthly
1175 13021560
816 1560 1235
1560 1560Size
1235
and Monthly
Sample the Rates
standard
of
standard Industrial
further
intercepts
0.01630.0084 0.0010
0.0127
0.01840.0080 0.0415
0.0455 Mean
0.0444 is the
details, deviations,
Box-Pierce Estimates
used
Dev.Std. seedeviations. of Table
0.02020.01020.0223
0.0161 0.0272
0.0116 0.0141 0.0450
0.0435 to
(1970) I
The skewness, the
Production,
model Producer
2.20 3.363.33
5.25 3.25 5.42
3.16
3.16
3.06 equations
Skewness thestatistic
(3a)
exception
for Price
kurtosis,
is
and growth
24 and Standard
6.67 16.01 47.9413.76 53.21
16.46 18.43
14.29 17.59 the
Kurtosis(3b) lags 1858-1987
in rates,
of Index,
r1
0.41 0.43 0.40 0.42
0.48 0.69
0.43 0.21
0.20 theestimate
andthe the
r2 of
0.31 0.34 0.25 0.32
0.37 0.34 0.19
0.20
0.58 text Deviations
then autocorrelations
r3 andstock
the at of
0.30 0.24 0.33 0.34
0.28 0.51
0.19 0.25
0.24
the lags
ril market 1, Monetary
0.20 0.300.25
0.26 0.25 0.44
0.14 0.19
0.18 autocorrelations
Autocorrelations
data absolute Stock
2,
0.19 0.29 0.24 0.22

0.24 0.44
0.16 0.16r12
0.17 11,
Q(24). Base,
values
volatility
A and
appendix.
of 12
1486 1549 1256 2589
2586 5711
1053 807913 Q(24) and
Returns,
the12th-
based of
Table II
SummaryStatistics forAutoregressivePredictive Models forthe
Volatilityof Stock Returns,Bond Returns,and the GrowthRates of
the Producer Price Index, the MonetaryBase, and Industrial
Production, 1859-1987
A 12th-orderautoregressionwith differentmonthlyinterceptsis used to model the growthrates or
returns, and then the absolute values of the errors from this model I t1 estimate the
monthlystandarddeviations.The exceptionis the estimateof stock marketvolatilitybased on daily
stock returnswithinthe month.The 12th-orderautoregressionforthe volatilityestimatesis
12 12
| =t| yj Djt + E Pi|r It1 + ut. (3b)

j=1 i=1
This table shows the sum ofthe autoregressivecoefficients(p1 + * + P12), indicatingthe persistence
of volatility.A t-testforwhetherthe sum equals unity,indicatingnonstationarity,is in parentheses
below the sum. It also shows an F-test forthe equalityof the 12 monthlyintercepts(y = * = y12) Y
and itsp-value. Finally,it shows the coefficientof determinationR2 and the Box-Pierce (1970) Q(24)
statisticforthe residual autocorrelations-(whichshould be distributedas x2 (12) in this case).
Sum of AR
Coefficients F-Test forEqual
(t-testvs. MonthlyIntercepts
VolatilitySeries one) (p-value) R2 Q(24)
Monthlystock returns 0.8471 0.97 0.132 45.8
(-3.72) (0.475)
Daily stock returns 0.9634 0.59 0.524 60.2
(-1.07) (0.838)
Monthlyshort-terminterestrates 0.7925 1.96 0.371 19.5
(-4.40) (0.028)
Monthlyhigh-qualitylong-termbond returns 0.8376 0.59 0.260 59.4
(-4.20) (0.835)
Monthlymedium-qualitylong-termbond returns 0.7769 6.78 0.280 16.6
(-3.47) (0.000)
PPI inflationrates 0.8438 0.39 0.271 53.1
(-4.29) (0.961)
Monetarybase growthrates 0.7918 0.65 0.220 37.0
(-4.74) (0.787)
Industrialproductiongrowthrates 0.8336 0.42 0.219 46.9
(-3.82) (0.948)
estimatestowardstationarity.'The resultsforthe estimateof stockvolatility

fromdailydata &tsupportthisconclusionsincethe sum of the autoregressive
is closerto unityand theteststatisticis small.
coefficients
C. MeasurementProblems-The EffectsofDiversification
Eventhoughthesetofstockscontainedinthe"market" changesover
portfolio
Therearefewstocksin thesample
is notaffected.
time,thebehaviorofvolatility
problemassociated
8Also see Pagan and Ullah (1988) for a discussion of the errors-in-variables
withmodels like (3b).
in 1857,and theyare all railroadstocks.Nevertheless, theyrepresent mostof

the activelytradedequitysecuritiesat thattime.Also,railroadsowneda wide
varietyofassetsat thattime.I havecalculatedtestsforchangesin stockvolatility
aroundthetimeswhenmajorchangesinthecomposition oftheportfolio occurred,
thereis no evidenceof significant
and, surprisingly, changes.Schwert(1989d)
analyzesseveralalternativeindicesofUnitedStatesstockreturnsforthe 19th
century and findsthatthedifferent portfolios after1834.
have similarvolatility
Thoughthe numberof securitiesand industries includedhas grownovertime,
theplotofstockreturnvolatility in Figure1 doesnotshowa downward trend.
This conclusioncontrastswiththeanalysisofunemployment, industrialpro-
duction,and grossnationalproductdata by Romer(1986a,b,1989).Also,when
theBureauofLaborStatisticshas expandedthemonthly sampleusedtocalculate
the CPI inflationseries,therehave been noticeablereductions in the volatility
of measuredinflationrates.Shapiro(1988) arguesthat the stabilityof stock
returnvolatilitybetweenthe 19thand 20thcenturiessupportsRomer'sconclu-
sions that the higherlevel of volatilityin pre-1930macroeconomic data is
primarily due to measurement problems.Nonetheless,it is perhapssurprising
thatstockreturnvolatilityis nothigherin the 19thcentury dueto measurement
problems.
II. Relations between Stock Market Volatilityand Macroeconomic

Volatility
A. VolatilityofInflationand MonetaryGrowth
The stockreturns analyzedaboveall measurenominal(dollar)payoffs. When
inflationof goods'pricesis uncertain,the volatilityof nominalasset returns
shouldreflectinflation I use thealgorithm
volatility. in equations(3a) and (3b)
to estimatemonthly inflationvolatilityfrom1858to 1987forthePPI inflation
rate.Figure4 plotsthepredicted PPI inflation volatilityIeptl
from1859to 1987.
Note thatthe right-hand PPI inflationvolatility axis is about2/3 smallerthan
theleft-hand axis.The volatility
stockvolatility ofinflation wasveryhigharound
the CivilWar (1860-1869),reflecting changesin the value of currency relative
to goldaftertheU.S. wentoffthegoldstandardin 1862.SincetheU.K. remained
onthegoldstandard, thisalso represents in theexchangeratesbetween
volatility
U.S. and U.K. currencies.The Spanish-American War (1898),WorldWar I and
its aftermath (1914-1921),and WorldWar II (1941-1946)are also periodsof
highinflationuncertainty.Another increaseininflation occurred
volatility during
the 1973-1974OPEC oil crisis.Whileinflationvolatilityincreasedduringthe
1929-1939period,thischangeis minorcomparedwiththevolatility thatoccurred
duringwars.
Figure5 plotsthepredictedvolatility ofthemonetary base growth ratesI
from1880 to 1987.The volatilityof moneybase growthratesrose duringthe
bankpanicand recessionof 1893and remainedhighuntilabout1900.The next
sharpincreasein volatilityoccurredduringthebankpanic of 1907.The period
following theformationoftheFederalReserveSystem(1914-1923)was another
of
period highvolatility. Finally,theperiodoftheGreatDepression(1929-1939)
ChangeOverTime?
WhyDoes StockMarketVolatility 1125
0.25 -- 0.16
-0.14
0.2 -
-0.12
L ~~~~~~~~~~~
11 ''1', 00.1
a:0
0.15 - ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~~.8
a
0
0
0.1~
4-nLlary ~ ~ ~ ~ ~ ~ ~~~~~~~~~~~~~~00
1859 - December1987a
0t
> C)~~~~~~~~~~~~~~~~~~
r 0.061
ft 0.05 4 4 K H ~~~~~~~~~~~~~~~~-.0
~~0.02
)fr15-98.A1t-re
~~~
uoersinwt
prdue prc ine>Ifainrts(
0
1859 1869 1879 1889 1899 1909 1919 1929 19,39 19,49 195'9 196'9 197'9 198'9
o autoregrestok
- - RetReturns
Stock ~~~~~~-
PPI Inflation Rate
gure 4. Predictionsof the monthlystandard

Fin ontn returns
devimations of stock and of
producerprice index inflation rates (-) for 1859-1987. A 12th-orderautoregressionwith
differentmonthlyinterceptsis usedto modelreturns orinflationrates,andthentheabsolutevalues
in montht. To modelconditional
oftheresidualsare used to estimatevolatility a 12th-
volatility,
modelwithdifferent
orderautoregressive monthly intercepts thestandarddeviation
is usedtopredict
in montht based on laggedstandarddeviationestimates. This plotc'ontainsfittedvaluesfromthe
models.
regression
volatility
was a periodofveryhighvolatility. Since the early1950's,thevolatility ofthe

monetary base growth rate has beenrelatively low and stable.9
BoththePPI inflation rateand the monetary base growthrateexhibitmuch
lower levelsof volatility after World War II. In each case, the sampleused to
measurethese variables has expanded over time, and therehave been major
institutionalchanges that have been intended to dampen macroeconomic fluc-
tuations.Without detailed analysissimilar to Romer's work on industrialpro-
duction,unemployment, and grossnational product, it is impossibleto tell how
important the changes in measurement techniques have been in reducing vola-
tility.
Table III containstestsofthe incremental predictive powerof12 lags ofPPI
inflationvolatilityI' tI in a 12th-order vectorautoregressive (VAR) systemfor
thatthepatternofvolatility
9It is surprising forthemoneybase growth
is so different rateand
thePPI inflationrate.Nevertheless, ofmoneysupply(M2) growth
I havealso analyzedthevolatility
andtheConsumer ratessince1915,andtheyleadto similarconclusions.
PriceIndex(CPI) inflation
The lack of relationbetweenmonetary and pricevolatility
volatility questionfor
is an interesting
futureresearch.
0.25 - 0.06
-0.05
0.2-
C
4-
* ~~~~~~~~~~~~~~~
TC 0.15-
0
-0.03~
o
c ~~~~~~~~~~~~~~~~~~~~~~~~~*0
t rd a0.02 >
autoregressive model with0different monthly intercepts is used to predict the standard deviation in
l
0~
r
~~~~~~
n
~aur 180-Dcebr18
4-
~
-. Stc euns-MnyBaeGot
141
0 0
1859 1869 1879 1889 1899 1909 1919 1929 1939 1949 1959 1969 1979 1989
- - Stock Returns - Money Base Growth
Figure 5. Predictions of the monthly standard deviations of stock returns (- )and of

money base growth rates ( )for 1880-1987. A 12th-orderautoregressionwith different
monthlyinterceptsis used to model returnsor moneygrowthrates,and then the absolute values of
the residualsare used to estimatevolatilityin montht. To model conditionalvolatility,a 12th-order
autoregressivemodel with differentmonthlyinterceptsis used to predictthe standard deviation in
month t based on lagged standard deviation estimates. This plot contains fittedvalues fromthe
volatilityregressionmodels.
stockvolatility, high-grade bondreturnvolatilityI ht and short-term interest

volatilityI'4tI thatallowsfordifferent
monthly The VAR modeluses
intercepts.
boththemonthly measureofstockreturnvolatilitylIt~jand the dailymeasure
These VAR modelsare generalizations
at10 oftheautoregressive modelin (3b),
buttheyincludelaggedvaluesof othervariablesto helppredictvolatility. The
F-testsin Table III measurethesignificance ofthelaggedvaluesofthecolumn
variablein predicting therowvariable,giventheothervariablesin themodel.F-
statisticsthat are largerthan the 0.01 criticalvalue 2.28 are indicatedwith
asterisks.
The largestF-statisticsare on the maindiagonalof thesematrices,and the
sizeofthestatisticsdecreasesawayfromthediagonal.Forexample,laggedstock
10Models usingthe volatilityof medium-grade(Baa-rated) bond returnvolatilitY,IErm,l, instead of

high-gradebond returnvolatility,yielded similar results forthe post-1920 periods. Medium-grade
bond volatilityis more stronglyrelatedto the stock volatilityand more weaklyrelatedto the short-
terminterestrate volatility,but the relationswiththe macroeconomicvolatilityseries are generally
similar.Because these data are onlyavailable from1920 to 1987 and the resultsare similar,theyare
not reported.
volatilityis the most importantvariable in predictingcurrentstock volatility.
Lagged bond returnvolatilityalso helps in most sample periods, and lagged
short-terminterestvolatilitycontributesless. Likewise, stock volatilityhelps
predictbond returnvolatilityin mostperiods,but it rarelyimprovespredictions
of interestrate volatility.In most sample periods,short-terminterestvolatility
helps predict bond returnvolatilityand vice versa. Except for monthlystock
volatilityfrom1953 to 1987, thereis littleevidencethat inflationvolatilityhelps
to predictfutureasset returnvolatility.
The present value relation in (1) is forward-looking. In an efficientmarket,
speculative prices will react in anticipation of futureevents. Thus, it is also
interestingto see whetherasset returnvolatilityhelps to forecastlater volatility
of macroeconomicvariables. Except for long-termbond returnsfrom1859 to
1987, there is no evidence that either stock or bond returnvolatilityhelps to
predictinflationvolatility.Perhaps this is because the major changes in inflation
volatilityoccur duringwars, and there seems to be littleeffectof wars on stock
or bond returnvolatility.
Table IV contains tests of the incrementalpredictivepower of 12 lags of
monetarybase growthvolatility lemtl in a 12th-orderVAR system similar to
Table III. The relations among the measures of financial returnvolatilityare
similar to Table III. There is evidence that money growthvolatilityhelps to
predictthe volatilityof long-termbond returnsfrom1885 to 1919. Also, from
1885 to 1987, 1885 to 1919,and 1920 to 1952,thereis evidencethat moneygrowth
volatilityhelps to predict the volatilityof stock returnsmeasured using daily
data. On the otherhand, from1920 to 1952 (and the sample periodsthat include
this subperiod), both measures of stock returnvolatilityhelp to predict the
volatilityof the base growthrate.
The relationsbetween inflationor money growthvolatilityand the volatility
of asset returns are not strong. It is surprisingthat these macroeconomic
measures of nominal volatilityare not more closely linked with the volatilityof
short-and long-termbond returns.
B. Real MacroeconomicVolatility
Since common stocks reflectclaims on futureprofitsof corporations,it is
plausible that the volatilityof real economic activityis a major determinantof
stock returnvolatility.In the present value model (1), the volatilityof future
expected cash flows,as well as discount rates, changes if the volatilityof real
activitychanges.
Figure 6 contains a plot of the predicted volatilityof the growthrates of
industrial production sit-l. Note that the right-handindustrial production
volatilityscale is about 2/3 smaller than the left-handstock volatility scale.
Summarystatisticsforthese estimatesare in Tables I and II. Industrialproduc-
tion volatilitywas highduringthe mid-1930's,duringWorldWar I, and especially
" I also examinedthe volatilityof bank clearingsdata

fromMacaulay (1938) and the volatilityof
the liabilitiesof business failuresdata fromDun and Bradstreet(Citibase (1978)). Neitherof these
"real activity"variables was stronglyrelatedto stock volatility.
to A
four
four
monthly
volatility,
predict
PPI PPI PPI respectively. F-Tests
Bond
Stock Bond
Stock Bond
Stock data
columns.
the
Interest Interest Interest Variable variable,
Dependent are
The includingfrom
used
0.05
F-statistics
in respective
1.33 1.41
1.16 1.67
2.16 1.93 0.80
0.86 2.07 12th-order
dummy
2.41* 9.33*
5.83* Stock andtherow Vector
greater
first vector
F-Tests0.01
than variables
four
1.51
1.04 1.24 1.04 1.46 with variables,
for
8.05* 15.74*
3.87* 2.89* 6.42* 10.82*
2.65* Bond 2.28 critical
are
Including
given
columns,
Monthlyvaluesthemonthly
autoregressive
PPIAutoregressive
1.65 0.41
1.25 0.60 0.89
1.98 0.68 0.94
1.30 forand
3.34* 21.99* 12.68* Stock
Interest indicated
other(VAR)
the
with intercepts.
measures Models
an Inflation
model Table
Volatilityof variables
The
0.51
0.48
3.67* 0.72 31.91*
1885-1919
1.14
0.65 0.95 0.94
5.70* 1.36
0.50 PPI is forIII
1885-1987 1859-1987 F-statistic
in
stock
asterisk.theF-tests
with
Stock,
12 return estimated
1.14 0.71
0.94 0.84
1.04
Volatility,
8.86* 67.92*
2.83* Stock reflect
model. for
and
the
Bond,
F-Tests200volatility
stock,
Measures and
0.70
1.20
6.03*
4.31* 1.13
2.29* 6.77*
14.06* Bondwith based
of bond,
degrees 1859-1987
on incremental
Daily of stock
daily interest
ability Interest
0.59 1.20 0.83 1.65 Stock
4.57* 2.99* 21.39*
3.75* data
freedom of
return
Interest rate,
areare the Rate
and
Volatilityused
1.80 volatility
PPI
1.95
0.86
1.26 0.61
1.27
2.10 in column
3.29* 28.83* PPI
and
thebased
2.28,
laston variable Volatility,
inflation
WhyDoes StockMarketVolatility
ChangeOverTime? 1129
PPI PPI
Bond
Stock Bond
Stock
Interest Interest
0.63 2.00
1.67 1.26 1.04
0.48
9.27*
3.35*
0.35 1.05
3.17*
5.23* 0.51
0.85 3.07*
4.49*
0.76 1.63
3.20*
5.25* 0.54 0.26
11.92*0.36
1.99
19.16*1.36
3.65* 0.21
13.05* 0.51
1.95
1953-1987 1920-1952
0.99
0.96 3.27*
6.50* 0.61
0.88 22.03*
5.92*
0.41
5.04* 1.51
3.09* 0.31
1.03 4.09*
3.52*
0.87 0.72
3.97*
4.39* 0.56 0.62
0.28
11.81*
1.52
1.90 1.55 0.21
12.55* 0.35
1.82
16.72*
A
laston
four
growth
four variable
Base Bond
to
monthly
respectively. F-Tests
Stock Base Bond Stock Base Bond Stock Base Bond Stock
Interest Interest Interest Interest Variable variable,
data volatility,
Dependent
columns.
predict from
are
The the
F-statistics
used
0.83
1.25 1.55
2.00 0.40 1.63
1.19 1.41 including
12th-order
4.05* 2.82* 1.98
7.80* 2.26 4.80* 5.36*
7.85* Stock 0.05
in Vector
greater
andtherespective
F-Tests vector
dummy
than row
first
0.01 Including
0.68 0.96 1.64
0.38 1.23
0.87 0.92 1.26 with
4.82*
3.38* 4.03*
3.16* 3.52* 3.50* 2.60* Bond 2.28 four
16.92*
arecritical variables
variables,
Monthly forautoregressive
Autoregre
columns, Monetary
valuesgiven
1.43 1.93 0.41 0.26 1.53
3.20*
6.31* 11.99*0.40 1.46
3.43* 0.44 0.55 1.79
22.30*0.78
Stock
Interestindicated
forandthe (VAR)
monthly
the
with
Base
other model
Models
an
Volatilitymeasures is Table
1.63
0.63 0.95
1.08 5.23* 1.11
0.39 0.88 0.82
2.96* 0.79
3.71* 0.93 Base
2.25
1.07
21.72*
forIV
1953-1987 1920-1952 1885-1919 1885-1987 of intercepts.
F-statistic Growth
asterisk.variables
in The
stock
with Stock,
estimated
12 the
1.11
0.94 9.22* 3.11*
3.11* 0.56
4.78* 0.92 1.04
0.96 1.40
0.88 Stock for
23.21* 8.80* 2.76*
62.39* returnF-tests
and
model.stock,Bond,
Volatility,
F-Tests200 reflect
thebond,and
volatility
0.85 1.02 0.22
1.72
3.53*
4.77* 3.60*
3.64* 3.21*1.54 3.83* 1.79
3.09* 2.04 5.83* Bondwith
15.88*
Measures
degrees
based
Daily of of
on interest
stock 1885-1987
Interest
1.26 0.68 0.52 0.27
0.86 1.03 1.17 0.85 1.00 Stock incremental
5.36*
3.45* 12.08* 4.61* 2.95* 22.61*
3.39* daily rate,
Interest freedom
return
aredata and Rate
ability
Volatilityare of
1.80
1.74
0.90 1.15
1.08 0.42
1.16 0.85
1.28 0.60
1.36 the
6.22* 2.44* 2.47* 2.28* 18.73* 4.83* Base used
and volatility
monetary
in
2.28,
thebasedbase
column Volatility
Why13oes$tockMarket VolatilityChange Over Time? 1131
0.25 Q.10
' C~~~~~~~~~.14
0.2 -
Ii ~~~~~~~~~~~~~~~
.0
0l 0,1
c 0.15 -' A-
0 n
-- - Stock Re Ind'l Productlon0.04

o 0
A A' ~~~~~~~~~~~~~~~
.2
'A ~~~~~~~~~~~~~~~~
Fiur 6. Prdcin ftemnhysadr evain fsokrtrs(-)ado

volatility, a 12th-order autoregressi" Ioe wt d e m
01utra prdcto gotrae( )f,r19-87A12hrdrautrgesinwt
thobouevle ftersdasaeue oetmt oaiiyi ot t. To moe codtoal

0 Q
1859 1869 1879 1889 1899 1909 1919 1929 1939 1949 1959 1969 1979 1989
Januory1891 - December 1987_
- - Stock Retu.rns -IndI Production
Figure 6. Predictionhof the monthlystandarddeviationsof stocia returns(-otcn and of.

industrialproductiongrowthrates (.-) fpr 1891-1987. A 12th-order autoregression with
monthly
diffe,rent is usedto modelreturns
intercepts orindustrialproductiongrowth rates,andthen
theabsolutevaluesoftheresidualsareusedto estimnate in montht,To modelconditional
volatility
a 12th-order
volatility, withdifferent
mnodel
autoregressivp monthly is usedto predictthe
intercepts
standarddeviationin montht basedon laggedstandarddeviation Thisplotcontainsfiltted
estimates.
valuesfromthevolatility models.
regression
duringthe post-WorldWar II period.There is a small increasein volatility

duringthe 1973-1974recession.Romer (1986b) arguesthat data collection
procedures causepartofthehighervolatility ofthisseriesbefore1929.
Table V containstests of the incremental predictivepowerof 12 lags of
industrialproduction volatilityI i,l in a 12th-orderVAR systemsimilarto those
in Tables III and IV. The resultsforthefinancialvariablesare similarto those
reportedin Table III, The F-statistics measuring theabilityofindustrialproduc-
tionvolatilitytopredictfinancial volatility aresmall.Thereis somewhat stronger
evidencethatstockreturnvolatility predictsindustrial production for
volatility
the1891-1987and 1920-1952periods.
Thus,thereis weakevidencethatmacroeconomic providesincremen-
volatility
tal information aboutfuturestockreturnvolatility. Thereis somewhatstronger
evidencethatfinancial volatilityhelpstopredictmacroeconomic While
volatility.
manyof the macroeconomic volatilityseriesare highduring1929-1939,none
increasesbya factorofthreeas stockreturn did.This "volatility
volatility puzzle"
willremainafterall thesubsequentanalysis.
A
are
1.80 the
four
andused
volatility
IP IP IP IP in productionF-Tests
column
Bond
Stock Bond
Stock Bond
Stock Bond
Stock 2.28,
Interest Interest Interest Interest Variable thebased
Dependent variable,
on growth from
last variable
to
four
respectively.
monthly
1.42 1.85
1.22 1.72 0.52 2.25
0.95 1.26
0.98 1.43 12th-order
2.65* 8.46*
3.82* 3.18* 5.12*
9.47* Stock Vector
volatility,
predict
data
columns.
Including
F-Tests the vector
are
F-statistics
The including
0.79 0.92 1.31
0.38 0.85
1.46 1.59 0.76 with
4.70*
3.54* 4.20*
3.32* 6.30* 2.66* Bond
15.66*
3.63* used
0.05
in
greater respective
andthe dummy Industrial
Monthly row autoregressive
than Autoregress
0.22 0.01
first
6.65* 1.64
2.88* 0.57 0.27
12.25*0.39 0.66
2.88* 0.60
1.24 0.75 1.85
21.87*0.85
Stock
Interest2.28
four variables
(VAR)
arecritical
variables,
for
Models
Production
Volatility givenmodel Table
1.14
3.48* 1.21
0.49 0.79
0.48
5.60* 1.29 1.24
0.37
3.27* 0.98 0.51
24.07* 1.03 IP
0.49 columns,
values is
forV
1953-1987 1920-1952 1891-1919 1891-1987 indicated monthly
forandthe
with
the
other
Growth
Stock,
an
0.71
0.77 0.72
1.08 1.72
0.72
0.96 0.81
1.33 estimated
9.77*
3.08* 5.18*
22.36* 7.38* 64.93* Stock
2.68* intercepts.
measures
for
of
The
F-statistic
asterisk.variables Bond,
F-Tests
in
stock stock,
with Volatility,
and
1.02
4.42* 1.08
3.66* 1.23
0.20 3.74* 1.70
3.50* 1.22 3.57* 0.89
4.29* 2.11 6.25* Bondwith 12 theF-tests
14.41*
return bond,
Daily and
reflect
model.
200 the Interest
0.29 0.72 0.58 0.23
0.65 0.67 1.20 1.25 Stock interest
5.67*
3.31* 12.21* 3.97* 2.70* 0.55 21.92*
3.43* volatility 1891-1987
Interest
rate, Rate
Measures
degrees
of based
of
Volatilityon and
incremental
0.61
1.20 0.70 0.56
0.66
0.72 0.58 0.35
0.70 0.62
0.99
2.24 IP stock
3.37* 4.80* 2.73* 19.79*
daily
freedom
ability
aredata
return Volatility,
of industrial
III. Volatilityand the Level of Economic Activity
A. VolatilityDuring Recessions
The previoustestsanalyzedtherelationsamongvariousmeasuresofvolatility.
Thereis also reasonto believethatstockreturnvolatility is relatedto thelevel
ofeconomicactivity. For example,iffirmshavelargefixedcosts,netprofits will
fallfasterthan revenuesif demandfalls.This is oftencalled "operating lever-
age."12 Table VI containsa testofthe relationbetweenstockvolatility and the
level of macroeconomic activity.It containsestimatesof the coefficient of a
to
dummyvariableadded equation (3b) equal to unity during recessions as
definedby the NationalBureau of Economic Research (NBER) and equal to
zerootherwise. is
Ifthiscoefficient above
reliably zero,the of
volatility the series
is largerduringrecessionsthanduringexpansions."3
Table VI showsthatvolatilityis higherduringrecessionssincemostof the
estimatesare positiveand none is morethan 1.2 standarderrorsbelowzero.
Exceptfor1859-1919,all the estimatesforstockvolatilityare morethan 1.8
standarderrorsabovezero.Moreover, the estimatesofthe percentage increase
in volatilityin recessionscompared with expansions, in braces below the t-
are large(up to
statistics, 277 percent in 1920-1952 using the dailyestimates of
volatility).Along with the measures of stock marketvolatilityle tj and St, the
volatility productionICjtjshowsthe mostreliableincreasesduring
of industrial
Thereis weakerevidencethatbondreturns,
recessions. short-term interestrates,
and moneygrowth rateshavehighervolatility duringrecessions.
Figure7 showsthe plot of predictedmonthlystockvolatilitylike Figure1,
exceptthat the periodsof NBER recessionsare drawnas solid lines and
expansionsare drawnas dottedlines.It is clearfromthisplotthatvolatility is
higherduringrecessions.This phenomenon
generally is notlimitedto theGreat
Depression.
Thus,stockmarketvolatility is relatedto thegeneralhealthoftheeconomy.
One interpretation of this evidenceis that it is caused by financialleverage.
Stockpricesare a leadingindicator, so stockpricesfall(relativeto bondprices)
beforeand duringrecessions.Thus,leverageincreasesduringrecessions, causing
an increasein the volatilityof leveragedstocks.The analysisbelowaddresses
thisquestiondirectly.
B. Volatilityand CorporateProfitability
and severalmeasures
I havealso analyzedtherelationbetweenstockvolatility
Recently,
ofcorporateprofitability. Fama and French(1988b)and othershave
shownthatvariablessuch as dividend(DIP) or earningsyields(E/P) predict
stock returnsfor horizonsas far as five years into the future.Keim and
Stambaugh(1986) and Fama and French(1989) showthatspreadsbetweenthe
12I am gratefulto Fischer Black forsuggestingthis interpretation.

13Since the NBER announces the timingof recessions and expansions six to nine monthsafter
theyhave begun,this evidence does not implythat the recessionvariable can be used to help predict
futurevolatility.
Table VI
Estimates of the Relations Between Business Cycles and Financial and
MacroeconomicVolatility,1859-1987
Estimates of dummyvariable coefficientsare added to the autoregressivemodel forvolatility.The t-
statisticsin parenthesesuse White's (1980) heteroskedasticityconsistentstandarderrors.A dummy
variable equal to one duringmonthsdesignatedas recessionsby the National Bureau of Economic
Research is added to a regressioncontaining12 monthlydummyvariables and 12 lags of volatility.
The estimatesrepresentthe increasein averagevolatilityduringperiodsof recession.The percentage
increase in volatilityduringrecessionsrelativeto expansions is in braces below the standarderrors.
The estimatesin the firsttwo columns use as much data as are available forthe respectiveseries.
Coefficientestimatesmorethan two standarderrorsfromzero are indicatedwithan asterisk.
Dependent Variable 1859-1987 1859-1919 1920-1952 1953-1987
Monthlystock returns 0.0063* -0.0014 0.0195* 0.0139*
(2.93) (-0.55) (3.09) (3.12)
$61%} 1-6%) 1234%1 168%J
Daily stock returns 0.0038* 0.0014 0.0077* 0.0037
(3.05) (0.92) (2.55) (1.81)
199%} {8%} 1277%1 145%I
High-gradelong-termbond re- 0.00065 0.00019 0.00234 0.00160
turns (1.21) (0.39) (1.68) (0.99)
142%} 114%} 1161%}j 70%)
Short-terminterestrates 0.00008 0.00007 0.00004 0.00031
(1.22) (0.88) (0.33) (1.41)
129%o} 115%o1 116%} 1134%)
PPI inflationrates 0.00024 -0.00070 -0.00067 -0.00052
(0.31) (-0.58) (-0.64) (-1.16)
110$7o} 1-13%} 1-15%1 1-57%1
Monetarybase growthrates 0.0015* 0.0017 0.0010 -0.0002
(2.47) (1.77) (0.81) (-0.51)
1125%} 154%1 142%$ {-11%}
Industrialproductiongrowthrates 0.0032* 0.0011 0.0022 0.0026*
(2.58) (0.48) (0.96) (2.35)
183%I} 8%) 130%1 152%}
yields on low versus high-gradelong-termcorporate debt also predict stock

returns.Where such variables track time-varyingexpected returns,they may
also predicttime-varying volatility.
The relationsbetweenstock volatilitywith eitherdividendor earningsyields
are sometimes positive and sometimes negative. These opposite associations
suggestthat thereis no stable relationbetweenearningsor dividendpolicy and
stockvolatility.To limitthe numberof reportedresults,I only summarizethese
tests here.
Table VII contains estimates of the coefficientsof the spread between the
yieldson Baa- versusAa-ratedcorporatebonds when added to the autoregressive
modelssummarizedin Table II. All of the estimatesare positive,and several are
more than two standard errorsabove zero. Thus, the differencebetween the
0.25 - 0.25
0.2 ~~~~~~~~~~~~~~~
L L
.c 0.15 o.1s2
o 0
o 0
.2 i ~~~~~~~~~~~
0. -0.1 4-
r
0.1 - - E n sR T
18 1-86'9 16 8918'9 991''129 1939 1949 1959' 19"69 1979 1989

-Expansions - Reces~ions
Figure 7. Predictiois of the monthlystandard deviationof stock returnsduringNB3ER
recessions ( a) and duringexpansions (--) for 1859-1987. A 12th-order autoregression
withdifferent imonthly intercepts
is used to modelreturns,and thenthe absolutevalues of the
residualsare used to estimatevolatility
in montht. To modelconditional a 12th-order
volatility,
autoregressivemodelwithdifferqnt monthly intercepts
is usedto predictthestandarddeviationin
montht based on laggedstandarddeviationestimates.This plot containsfittedvaluesfromthe
volatility
regres'iopmodel,shownseparately forrecessionsandexpansions.
yields on bonds of differenltquality is directlyrelated to subsequentlyobserved

stockvolatility.This is not surprising,since the differencein promisedyieldson
bonds of differentquality should be largerin periQdswhen defaultriskis high.
TV, Effects of Leverage Qn Stock Market Volatility

One explanation of time-varyingstock volatilityis that leverage changes as
relativestock and bond prices change. In particular,the variance of the return
to2 the assets
th of a firma.t is a functionofthe variances ofthe retprnsto the stock
0St and the bonds ab, apd the covariance of the returns Cov(RStIRbt):
vt = ($/V)tl oSt + (B/V)It bt + 2 (S/V)t-1 (B/V)t_l cov(RSt,Rbt), (4)
where (S/V)tXl and (B/V)t-lrepresentthe fractionof the marketvalue of the

firmdue to stocks and bonds at time t - 1. Consider a firmwith riskless debt
(Fbt = cov(Rt,Rbt) 0), wherethe variance ofthe assets ofthe flrm . is constant
Table VII
Estimates of the Relation Between the
Standard Deviation of Stock Returns and the
Corporate Bond Quality Yield Spreads, 1920-
1987
The previousmonth'sspreadbetweentheMoody'sBaa long-term
corporatebond yield and the Aa yield,(YBaa - YAa)t- , is includedin
an autoregressive
modelforvolatility,
12 12
U,t E ti + E 3j Ust-j + -Y(YBaa YAa)t-1 + Ut.

i=1 j=1
Onlythe coefficientof the yieldspread-yis shown.Asymptotic

standarderrorsare in parenthesesunderthecoefficient
estimates.
estimatesmorethantwostandarderrorsfromzeroare
Coefficient
withan asterisk.
indicated
StandardDeviation StandardDeviation
Sample fromMonthly fromDailyStock
Period StockReturns Returns
1920-1987 14.83* 3.937*
(5.82) (1.85)
1920-1952 18.07* 4.256
(8.00) (2.15)
1953-1987 5.649 3.950
(8.29) (3.14)
over time. The standard deviation of the stock returnis Tst= UV; (V/S)t-1. This
shows how a change in the leverageof the firmcauses a change in the volatility
of stock returns.
Figure8 plots the predictionsof stock marketvolatilityat fromFigure 1 along
withthe estimatesimpliedby changingleverage (V/S)t-l scaled to have a mean
equal to the average-of at for 1900-1987. Changing leverage explains a small
portion of the increase in stock marketvolatilityin the early 1930's and mid-
1970's. It cannot explain most of the variationin at.14
Christie (1982) proposes regressiontests forthe effectsof changingleverage
on the volatilityof stock returns.He notes that, if the volatilityof the value of
the firma, is constant,(4) impliesthe regressionmodel:
ast = ao + a,1 (B/S),j + ut, (5)
where ao0= a1 = av, in the riskless debt case. With risky consol bonds containing
protective covenants, as modeled by Black and Cox (1976), Christie shows that
o = av > a,.
Table VIII contains generalized least squares (GLS) estimates of equation (5)
for 1901-1987, 1901-1952, and 1953-1987. There is strong residual autocorrela-
tion using ordinary least squares; hence, the GLS estimates use an ARMA(1,3)
14 A plotusingthemonthly
measureofvolatility
I'sl yieldssimilarconclusions.
0.25 - 0.25
0.2 - -0.2
0 0
0 ~~~~~~~~~~~
o 0~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~I
an 1 - D I
0 II I~~~~~~~~~~~~~~~
1900 1910 1920 1930 1940 1950 1960 1970 1980

- Effect of Leverage - - Stock Volatility
Figure 8. Predictionsofthe monthlystandarddeviationof stockreturnsbased on daily

data (-- -) and the level of stock returnvolatilityimpliedby changingfinancialleverage
( ~) for 1900-1987. The dailyreturnsin the monthare used to estimatea samplestandard
deviationforeach month.To modelconditional volatility,
a 12th-order autoregressive
modelwith
differentmonthly interceptsis usedto predictthe standarddeviationin montht based on lagged
standarddeviationestimates. This plotcontainsfittedvaluesfromthevolatility regression
model.
The effectofleverageis estimatedbyassumingthatthevolatilityoftheassetsofthefirmis constant
andthatdebtis riskless. Then,thestandarddeviationofstockreturns changesin proportion
to the
valueofthefirmdividedbythevalueofthestock(V/S)t-1.Whenstockpricesfallrelativeto bond
prices,stockvolatility
increases.Thus,the"effectofleverage"plotis a timeseriesofaggregate
firm
value(stockplusbondvalue)dividedbystockvalue,scaledto havethesamemeanas thepredictions
ofvolatility
fromtheregression model.
model for the errors.This is similar to the French, Schwert,and Stambaugh

(1987) model forSt. The resultsdepend on the volatilitymeasure and the sample
period.For the daily volatilitymeasure ,t,the interceptaO is always greaterthan
the slope a,, as predictedbythe riskydebtmodel.The estimatesofao are between
0.03 and 0.04 per month,and theyare over threestandarderrorsfromzero. This
would be the estimateof firmvolatilitya, in Christie'smodel. All the estimates
of a, are positive, showingthat stock volatilityrises when leverage rises. The
standarderrorsare large,however,so the t-statisticstestinga, = 0, or testinga,
= ao, are small. The t-testin the last column of Table VIII tests the hypothesis
that the slope equals the intercept(aO = a). The p-value in parenthesesis for
the two-sidedalternativehypothesis(aO aj).
The estimatesforthe monthlyvolatilitymeasure IestI for1953-1987 are similar
1138 TheJournalofFinance
Table VIII
Estimates of the Relation Between the Standard Deviation of Stock
Returns and Leverage, 1901-1987
Regressionsof stock volatilityon debt/equityratios,
ast -= ao + a1 (B/S)t1 + Ut, (5)

where(B/S)t-l is an estimateof the debt/equityratio forthe aggregatestock marketportfolioat the
end of month t - 1. Generalized least squares estimates include an ARMA (1, 3) process for the
errors.Asymptoticstandard errorsare in parentheses under the coefficientestirnatXs.S (u) is the
standard deviation of the errors,R2 is the coefficientof determinationincluding the effectsof
estimatingthe ARMA (1, 3) process forthe errors,and Q(24) is the Box-Pierce (1970) statisticfor
24 lags of the residual autocorrelations,whichshould be distributedAs x2 (20). The t-testfora0oa- a,
tests whetherthe risklessdebt model is an adequate approximationto the effectof leverageon stock
returnvolatility,where a0 > a, is impliedby the riskydebt model. Coefficientestimatesmore than
two standarderrorsfromzero are indicatedwithan asterisk.The p-values forthe Box-Pierce statistic
and forthe two-sidedalternativeao0# a, are in parenthesesunderthe test statistics.
Sample t-test
Period ao a, S(u) R2 Q(24) ao a,
Standard Deviation fromMonthlyReturns
1901-1987 0.0269* 0.0512* 0.0424 0.165 56.2 -0.87
(0.0101) (0.0193) (0.0000) (0.383)
1901-1952 0.0232 0.0700* 0.0475 0.194 50.3 -1.08
(0.0157) (0.0300) (0.0002) (0.279)
195341987 0.0315* 0.0221 0.0336 0.055 16.9 0.45
(0.0066) (0.0146) (0.657) (0.651)
Standard Deviation ftomDaily Returns
1901-1987 0.0376* 0.0154 0.0187 0.571 24.6 1.01
(0.0093) (0.0147) (0.216) (0.311)
1901-1952 0.0402* 0.0168 0.0205 0.606 35.0 0.71
(0.0135) (0.0225) (0.020) (0.479)
1953-1987 0.0317* 0.0101 0.0157 0.296 12.7 1.03
(0.0073) (0.0147) (0.890) (0.301)
to thedailyestimates.For 1901-1952and 1901-1987,however, theestimatesof

ao are less than the estimatesof a1, a resultthat is inconsistent
withall the
leveragemodels.Again,the standarderrorsare large,so the t-testfora0oa1 is
notlarge.Theseestimatesofa, are reliablyabovezero,showingthatan increase
in the debt/equity ratio(B/S),-1leads to an increasein stockreturnvolatility.
Theseregressions also havestrongresidualautocorrelation.An obviousinterpre-
tationis thatthevolatility ofthevalueofthefirmo-,tis notconstantoverthese
samples.Rather,it rose at the same timethatleveragerose duringthe Great
Depression,so the largeestimatesof a, are caused by omittinga correlated
regressor.Again,this evidenceshowsthat leveragealone cannotexplainthe
historicalmovements in stockvolatility.
V. Stock Market Trading and Volatility

Thereareat leastthreetheoriesthatpredicta positiverelationbetweenvolatility
and volume.First,ifinvestors
haveheterogeneous beliefs,newinformation will
cause both price changesand trading.Second, if some investorsuse price
movementsas information on whichto make tradingdecisions,large price
movements willcause largetradingvolume.Finally,ifthereis short-term
"price
in secondarytradingmarkets,largetradingvolume
pressure"due to illiquidity
thatis predominantly eitherbuyorsellorderswillcausepricemovements. There
has beenmuchpreviousresearchon therelationbetweenvolatility and trading
volume,butmostofit has focusedon thebehaviorofindividualsecurities. The
timeseriesbehaviorof volatilityand tradingvolumeforthe aggregatestock
marketprovidesa different on thesequestions.
perspective
A. TradingDays and Volatility
Frenchand Roll (1986) showthat stockvolatilityis higherwhenstockex-

changesare openfortrading.In particular,theyfindthatthevarianceofstock
returns and
overweekends holidays is much lessthana typicalone-dayvariance
the
times number of calendardays since tradinglast occurred.Mostpeculiarly,
during1968,when the NYSE closed on Wednesdaysdue to the "paper-work
crunch,"thevariance ofTuesday Thursdayreturns
to was notmuchlargerthan
a one-day This
variance. occurred
even though the stockexchangesweretheonly
economic institutions
takingholidays. Table IX containsregressions,
(st = ao + a,lDayst + Ut, (6)
whereDaystis thenumberoftradingdaystheNYSE was openduringmontht.

If varianceis proportionalto tradingtime,a, representsthestandarddeviation
per tradingday and ao shouldequal zero. If volatilityis unrelatedto trading
theintercept
activity, a0 estimatestheaveragemonthly standarddeviationand
a, shouldequal zero.Table IX containsGLS estimatesofequation(6) for1885-
1987,1885-1919,1920-1952,and 1953-1987.These estimatesdo not provide
strongsupportfor eitherhypothesis, but the French-Rollscenariois more
consistentwiththe data. All but one of the estimatesof the tradingtime
a1 arepositive,and severalarealmosttwostandarderrorsabovezero.
coefficient
On theotherhand,manyoftheestimatedintercepts are negative,and onlyone
is morethan two standarderrorsabove zero. Thus, NYSE tradingactivity
explainspartofthevariationin stockvolatility. thisrelationdoes
Nevertheless,
notexplainmuchofthevariationin volatility throughtime.
B. Trading Volumeand Volatility
Anothermeasureof stocktradingactivityis sharetradingvolume.Karpoff

on the relationbetweenvolatilityand
(1987) surveysthe extensiveliterature
volume.Table X containsestimatesoftheregression
ast=aa
Ast = + Volt+ ut, (7)
-4(1-6L)
Table IX
Estimates of the Relation Between the Standard Deviation
of Stock Returns and the Square Root of the Number of
Trading Days, 1885-1987
ofstockvolatility
Regressions on thesquarerootofthenumber
oftradingdaysper
month,
ast = a0 + a1 IDayst + Ut, (6)
wherevDas is thesquarerootoftheNYSE trading daysinthemonth.Generalized

leastsquaresestimatesincludean ARMA (1,3) processfortheerrors.Asymptotic
standarderrorsare in parentheses estimates.S(u) is the
underthe coefficient
standarddeviationoftheerrors, R2is thecoefficient
ofdeterminationincludingthe
effectsofestimating theARMA (1,3)processfortheerrors, and Q(24) is theBox-
whichshouldbe
Pierce(1970) statisticfor24 lagsoftheresidualautocorrelations,
distributedas x2(20), withthep-valuein parentheses underthe test.Coefficient
estimatesmorethantwostandarderrorsfromzeroare indicated withan asterisk.
Sample
Period ao a, S(u) R2 Q(24)
StandardDeviationfromMonthly Returns
1885-1987 -0.0276 0.0152* 0.0418 0.142 56.1
(0.0357) (0.0073) (0.0003)
1885-1919 -0.0703 0.0224 0.0347 0.028 16.1
(0.0612) (0.0122) (0.708)
1920-1952 0.0224 0.0065 0.0545 0.194 45.7
(0.0764) (0.0152) (0.0009)
1953-1987 -0.0514 0.0202 0.0336 0.055 16.3
(0.0567) (0.0124) (0.697)
StandardDeviationfromDailyReturns
1885-1987 0.0341* 0.0018 0.0186 0.538 21.9
(0.0150) (0.0029) (0.347)
1885-1919 0.0251 0.0027 0.0157 0.226 10.9
(0.0231) (0.0046) (0.950)
1920-1952 0.0632 -0.0025 0.0231 0.622 38.7
(0.0318) (0.0056) (0.007)
1953-1987 -0.0038 0.0087 0.0157 0.300 13.1
(0.0225) (0.0048) (0.872)
whereVol,is the growthrateofvolumefrommontht - 1 to montht,and the

errorsutfollowan ARMA(1,3)process.This modelrelatesstockvolatility to a
distributedlag of past sharevolumegrowth, wherethe coefficient of volume
growth decreasesgeometrically.'5
The estimatesin Table X also showa positive
relationbetweenstockvolatility All theestimatesoff are
and tradingactivity.
morethantwostandarderrorsabove zero.The estimatesof a are all positive.
For the estimatesofvolatility
based on dailydata ,t,theyare severalstandard
15
bythepatternofregression
This modelwas suggested in an unrestricted
coefficients regression
on current
ofvolatility L is thelagoperator,
and fourlagsofvolumegrowth. LkXt = Xt-k.
ChangeOverTime?
Table X
Estimates of the Relation Between the Standard Deviation
of Stock Returns and Stock Market Trading Volume,
1885-1987
Distributedlag regressions on the growthrateof NYSE share
of stockvolatility
tradingvolume(Vol,),
Ost -= a + Volt+ u'. (7)

(1 - 3L)
Generalizedleastsquaresestimatesincludean ARMA (1,3) processfortheerrors.

Asymptotic standarderrorsare in parentheses underthecoefficient estimates.The
lag modelfortheeffect
distributed of current
and laggedsharevolumegrowth on
themonthly standard deviation ofstockreturnsimpliesgeometric decay.The implied
forlag k is j3k. L is thelag operator,
coefficient LkXt = Xt_k. S(u) is thestandard
oftheerrors,
deviation R2is thecoefficientofdeterminationincluding theeffects
of
estimatingthe ARMA (1,3) processforthe errors,and Q(24) is the Box-Pierce
(1970)statisticfor)24lagsoftheresidualautocorrelations, whichshouldbe distrib-
utedas X2(20),withthep-valueinparentheses underthetest.Coefficient estimates
morethantwostandarderrorsfromzeroare indicated withan asterisk.
Sample
Period ao , S(u) R2 Q(24)
StandardDeviationfromMonthly
Returns
1885-1987 0.0454* 0.0473* 0.1561 0.0394 0.237 55.4
(0.0049) (0.0038) (0.0800) (0.0000)
1885-1919 0.0410* 0.0331* 0.3484* 0.0328 0.127 19.2
(0.0023) (0.0047) (0.1320) (0.509)
1920-1952 0.0545* 0.0629* 0.0597 0.0502 0.316 40.9
(0.0150) (0.0074) (0.1188) (0.004)
1953-1987 0.0395* 0.0539* 0.3061 0.0324 0.124 19.9
(0.0025) (0.0092) (0.1684) (0.462)
StandardDeviationfromDailyReturns
1885-1987 -0.0246 0.0168* 0.9984* 0.0179 0.568 21.9
(0.0560) (0.0019) (0.0012) (0.346)
1885-1919 0.0372* 0.0123* 0.9536* 0.0151 0.281 13.7
(0.0020) (0.0023) (0.0299) (0.845)
1920-1952 0.0484* 0.0203* 0.9007* 0.0223 0.650 36.0
(0.0165) (0.0037) (0.1002) (0.016)
1953-1987 0.0351* 0.0182* 0.5952* 0.0154 0.324 14.0
(0.0041) (0.0044) (0.2431) (0.832)
errorsabovezero.Fortheestimatesofvolatility basedon monthly data I stjS the

estimatesof 6 are closerto zero,thoughfor1885-1919it is overtwostandard
errorsabovezero.Thus,the evidencein Table X supportstheproposition that
stockmarketvolatility is higher.
is higherwhentradingactivity
Table XI containstestsoftheincremental powerof12 lagsofNYSE
predictive
sharevolumegrowthVol,in a 12th-order VAR systemforstockvolatility, high-
gradebond returnvolatilityI ht,j and short-terminterestvolatilityI4stI that
allowsfordifferent monthlyintercepts. This modelis similarto thoseused in

Tables III, IV, and V. The F-statisticsmeasuringthe abilityof sharevolume
growth are small,exceptfor1885-1919and 1885-
to predictfinancialvolatility
1987 using the daily measure of stock volatilitySt. There is little evidence that
financial helpstopredictfuture
volatility tradingvolumegrowth, exceptforstock
from1920to 1952.
volatility
The maindifference betweenthe distributedlag modelsin Table X and the
VARmodelsinTable XI is thatthedistributed lagmodelsincludethecorrelation
of contemporaneous volumeand volatilityand the VAR modelsdo not. The
strongrelationsin Table X and the weak ones in Table XI pointto a strong
Table XI
F-Tests fromVector AutoregressiveModels for Stock, Bond, and
InterestRate Volatility,Including Growthin NYSE Share Trading
Volume, 1885-1987
A fourvariable,12th-order vectorautoregressive (VAR) modelis estimatedforstock,bond,and
interestratevolatility,
andNYSE sharetrading volumegrowth (Vol),including
dummy variablesfor
monthly The F-testsreflect
intercepts. theincremental abilityofthecolumnvariableto predictthe
respectiverowvariables,giventheothervariablesin themodel.Measuresofstockreturnvolatility
basedon monthly data are used in the firstfourcolumns,and measuresofstockreturnvolatility
basedon dailydata are used in thelast fourcolumns.The 0.05 and 0.01 criticalvaluesfortheF-
with12 and 200degreesoffreedom
statistic are 1.80and 2.28,respectively.
F-statistics than
greater
2.28areindicated withan asterisk.
F-testswithMonthly
Stock F-testswithDailyStockVolatility
Dependent Volatility
Variable Stock Bond Interest Vol Stock Bond Interest Vol
1885-1987
Stock 10.29* 2.75* 0.84 1.55 77.94* 6.97* 1.38 7.65*
Bond 5.26* 17.70* 2.01 0.87 2.59* 17.37* 3.47* 0.86
Interest 1.62 3.68* 22.33* 0.70 0.87 2.22 22.36* 0.45
Vol 1.97 1.11 0.63 11.25* 1.30 1.60 0.66 10.94*
1885-1919
Stock 1.42 1.37 0.38 1.11 9.40* 4.35* 2.29* 3.97*
Bond 1.37 8.78* 1.22 0.83 1.09 6.14* 1.17 0.90
Interest 2.05 1.48 3.24* 0.96 0.87 1.18 4.45* 0.98
Vol 1.42 1.36 0.97 4.12* 1.86 1.10 0.76 3.55*
1920-1952
Stock 4.07* 3.80* 0.54 2.13 22.39* 3.35* 0.63 1.01
Bond 8.88* 4.06* 0.33 1.14 6.08* 3.40* 0.30 1.48
Interest 0.51 0.52 11.87* 0.52 0.62 0.35 11.87* 0.51
Vol 2.28* 2.13 0.62 4.20* 2.71* 0.99 0.59 3.53*
1953-1987
Stock 2.18 1.06 2.23 1.58 10.10* 1.18 0.75 0.54
Bond 1.87 3.57* 3.20* 0.39 2.85* 3.58* 3.58* 0.36
Interest 1.59 5.15* 6.02* 1.17 0.84 5.17* 5.66* 0.99
Vol 0.50 0.46 0.64 7.25* 1.58 0.49 0.82 7.50*
correlationbetweenthe "shocks" to volume and volatility.Unexpected changes
in volatilityand volume are highlypositivelycorrelated.Given the historyof
volatility,there is not much correlationbetweenvolatilityand lagged values of
tradingvolume.
In general,high tradingactivityand high volatilityoccur together.Of course,
these regressionscannot show whetherthis relationis due to "tradingnoise" or
to the flowof informationto the stock market.
VI. Summaryand Conclusions

This paper analyzes many factorsrelatedto stock volatility,but it does not test
forcauses of stock price volatility.Rather, the hypothesesinvolve associations
between stock volatilityand other variables. For example, the analysis of the
volatilityof bond returns,inflationrates, moneygrowth,and industrialproduc-
tion growth,along withstock volatility,seeks to determinewhetherthese aggre-
gate volatilitymeasures change togetherthroughtime. In most general equilib-
rium models, fundamentalfactorssuch as consumptionand productionoppor-
tunitiesand preferenceswould determineall these parameters(e.g.,Abel (1988)).
Nevertheless,the process of characterizingstylizedfactsabout economicvolatil-
ityhelps to definethe set of interestingquestions,leadingto tractabletheoretical
models.
A. JointEffectsofLeverageand MacroeconomicVolatility
Most of the tests above analyze stock volatilityalong with one other factor.
To summarizeall these relations,Table XII contains estimatesof the multiple
regression:
ln Jst = ae + arDrt + 31ln I eptI + 32 n I emtl
+ 33 ln I sitI (8)
+ zyln (V/S)t-l + ut.
In (8), ae representsthe constant term during expansions, and (ae + ar)
representsthe constanttermduringrecessions.The slope coefficientsj1 through
33 representthe elasticities of stock returnvolatilitywith predicted inflation
volatility,predictedmoneygrowthvolatility,and predictedindustrialproduction
volatility,respectively.The coefficientzymeasuresthe effectof financialleverage
on volatility.Table XII shows estimates of equation (8) for both measures of
stock returnvolatility.There is no correctionforautocorrelationin the errors
from(8), although the standard errorsuse Hansen's (1982) heteroskedasticity
and autocorrelationconsistentcovariance matrix.16
Equation (8) measures the contributionsof macroeconomicconditionalvola-
16 Since many of the regressorsin (8) are fittedvalues from firststage regressions(3b), the
"generatedregressors"problem discussed by Pagan (1984) is relevant here. In brief,to the extent

that there are omittedvariables that could be used to help to predictthe volatilityof some of these
series, the coefficientsof all of these second stage regressorswill be biased. Experimentationwith
instrumentalvariables estimation,the technique recommendedby Pagan, yieldedsimilarresults.
Table XII
Estimates of the Relation of the Standard Deviation of Stock Returns
to the Predicted Volatilityof MacroeconomicVariables, and the Effect
of Leverage, 1900-1987
The regressionmodel,
In a,t = a + a, Drt+ /3 in lIpti + /2 In IlmtI+ %3 In iliti + y ln(V/S)t-l + ut, (8)
includes a constant a, (not shown in this table), a dummyvariable Drtequal to unityduringNBER

recessions,the logarithmsof the predictedstandard deviationsof PPI inflationIIptl,of moneybase
growthIsme l, and of industrialproductionIitI,and the logarithmof leverage(V/S)tX1.The predicted
standard deviations are fittedvalues fromthe autoregressivemodels in Table II. The logarithmof
the stockreturnvolatilitymeasuresare the regressands.Asymptoticstandarderrorsare in parentheses
underthe coefficientestimates.All tests use Hansen's (1982) heteroskedasticity and autocorrelation
consistentcovariancematrix,using 12 lags and leads and a dampingfactorof0.7. R2 is the coefficient
of determinationand Q(24) is the Box-Pierce (1970) statisticfor24 lags of the residual autocorrela-
tions,which should be distributedas x2(24) in this case, with the p-value in parenthesesunder the
test.The columnlabeled Sum containsthe sum ofthe coefficientsofpredictedvolatilities.Coefficient
estimatesmorethan two standarderrorsfromzero are indicatedwithan asterisk.
PredictedMacroeconomicVolatility
Sample Recessions PPI Base IP Leverage
Period a?r /l3 /2 /33 Sum ly R2 Q(24)
Standard Deviation fromMonthlyReturns
1900-1987 0.256* -0.035 0.103 0.079 0.147 0.275 0.022 70.5
(0.120) (0.076) (0.072) (0.059) (0.088) (0.440) (0.000)
1900-1952 0.193 -0.035 0.261* 0.145 0.371* -0.370 0.027 44.7
(0.148) (0.096) (0.088) (0.079) (0.131) (0.634) (0.006)
1953-1987 0.479* 0.112 -0.183 0.180* 0.109 0.256 0.050 38.1
(0.082) (0.128) (0.133) (0.083) (0.179) (0.644) (0.034)
Standard Deviation fromDaily Returns

1900-1987 0.182 0.087 0.210* 0.031 0.328* 0.091 0.208 2905
(0.096) (0.045) (0.062) (0.043) (0.068) (0.316) (0.000)
1900-1952 0.177 0.077 0.273* 0.099 0.450* -0.273 0.168 1795
(0.125) (0.057) (0.080) (0.052) (0.116) (0.465) (0.000)
1953-1987 0.248* 0.151* 0.119 -0.009 0.262 0.047 0.120 540
(0.109) (0.053) (0.111) (0.052) (0.137) (0.316) (0.000)
alongwithleverage,
tilityfactors, in explainingthetimeseriesvariationin stock
From(4), st ( V/S)t-
returnvolatility. sincethevarianceofbondreturns
andthecovarianceofbondreturns withstockreturns willbe muchsmallerthan
a2t.Thus,equation(8) is an approximation of(4), wherethepredicted
volatilities
ofthemacroeconomic factorsaffectfirmvolatility{2t.The sumoftheelasticities
(i31 + /32 + /33) measures the response of firmvolatilityto a one percentincrease
in the volatilityof all the macroeconomic
factors.The elasticitywithleverage
shouldbe y = 1.
The averagelevel of volatilityis muchhigherduringrecessions(consistent
withTable VI). The columnlabeled"Recessions"inTable XII containsestimates
ofar, thedifferential intercept duringrecessions.Theyarebetween0.17and 0.50
acrossthedifferent measuresofstockvolatility and different
periods,and many
are reliablyabovezero.If therecessiondummyvariableproxiesforvariationin
operating leverage,it is interestingthatit remainsimportant forstockvolatility
evenwhenotherfactorsare included.
The effect offinancialleverageis small.The estimatesusingdailyreturnsare
reliablybelowunity.Perhapsthisreflects theimperfectproxiesforthisandother
regressors and thecollinearity amongthem.
Most ofthe estimatesof the predictedmacroeconomic volatilitycoefficients
are positive,and some are reliablyabove zero. For example,usingthe stock
volatilitymeasurefromdailydata ln ?' for1900-1952,all thesecoefficients are
at least 1.35 standarderrorsabove zero.The sumof thesecoefficients is 0.45,
witha standarderrorof 0.12. Thus, if the volatilityof inflationrates,money
growth, and industrialproduction all increaseone percent,stockvolatilityin-
creases by 0.45 percent.Across both monthlyand daily measuresof stock
volatilityand acrossall subperiods, thecoefficientestimatesofpredictedmoney
base growth volatilityare reliablypositivemostoften.
B. Summary
Manyeconomicseriesweremorevolatilein the 1929-1939GreatDepression.
Nevertheless, stockvolatilityincreasedby a factoroftwo or threeduringthis
periodcomparedwiththe usual levelof the series.(See Figure1.) Thereis no
otherseriesin thispaperthatexperienced similarbehavior.
Second,thereis evidencethatmanyaggregate economicseriesaremorevolatile
duringrecessions(Table VI). This is particularlytrueforfinancialassetreturns
and formeasuresof real economicactivity.One interpretation ofthisevidence
is that"operating leverage"increasesduringrecessions.
Third,thereis weakevidencethatmacroeconomic volatilitycan helptopredict
stockandbondreturn volatility
(TablesIII, IV,andV). The evidenceis somewhat
stronger that financialasset volatilityhelps to predictfuturemacroeconomic
This is notsurprising
volatility. sincethepricesofspeculative assetsshouldreact
quicklyto newinformation abouteconomicevents.
Fourth,financialleverageaffectsstock volatility.When stock prices fall
relativeto bondprices,or whenfirmsissuenewdebtsecuritiesin largerpropor-
tionto new equitythan theirpriorcapitalstructure, stockvolatilityincreases
(Table VIII). However, thiseffectexplainsonlya smallproportion ofthechanges
in stockvolatilityovertime(Figure8).
Fifth,thereseemsto be a relationbetweentradingactivity and stockvolatility.
The numberoftradingdaysin themonthis positively relatedto stockvolatility,
especiallyin 1953-1987(Table IX). This reinforces the evidencein Frenchand
Roll (1986). Also, sharetradingvolumegrowthis positivelyrelatedto stock
volatility
(Tables X and XI).
C. The Volatility
Puzzle
Major episodesin UnitedStateseconomichistoryare associatedwithlarger
suchas theCivilWar,WorldWarI, theGreatDepression,WorldWar
volatility,
II, theOPEC oil shock,and thepost-1979period.The puzzlehighlighted bythe
resultsin thispaperis thatstockvolatility is notmorecloselyrelatedto other
measuresof economicvolatility.For example,the volatilityof inflationand
moneygrowthrates is veryhigh duringwar periods,as is the volatilityof
industrialproduction.Yet thevolatility ofstockreturnsis notparticularly high
duringwars.17Thereweremany"financialcrises"or "bankpanics"duringthe
19thcentury in theUnitedStatesthatcausedveryhighand volatileshort-term
interestrates.Schwert(1989b) showsthat stockvolatilityincreasesforbrief
periodsduringand immediately following the worstpanics,but therewereno
long-term effectson volatility.
On the otherhand,the evidencein thispaperreinforces the argument made
by Officer(1973) that the volatilityof stockreturnsfrom1929 to 1939 was
unusuallyhighcomparedwitheitherprioror subsequentexperience. For many
yearsmacroeconomists havepuzzledabouttheinability oftheirmodelsto explain
the data fromthe GreatDepression.The resultsin thispaperpose a similar
challengeto financialeconomists.Moreover,based on evidencein Fama and
French(1988a) and Poterbaand Summers(1988),the 1929-1939periodplaysa
crucialrolein the evidencefor"meanreversion" in stockprices.I suspectthat
an analysisof Shiller's(1981a,b)varianceboundstestswouldrevealthat the
1929-1939periodis responsibleforthe inference of "excessvolatility"
of stock
prices.Indeed,thespiritofthepreceding discussionsuggests thatstockvolatility
wasinexplicably highduringthisperiod.I amhesitantto cedeall thisunexplained
behaviorto socialpsychologists as evidenceoffadsorbubbles.
RobertMertonhas suggested thatthe Depressionwas an exampleofthe so-
called"Peso problem," in the sensethattherewas legitimate uncertaintyabout
whetherthe economicsystemwouldsurvive.The RussianRevolutionoccurred
only12 yearsbeforethe 1929stockmarketcrash,and thereweremajorpolitical
and economicupheavalsoccurring throughout Europein the interim. Withthe
benefitofhindsight, we knowthattheU.S. andworldeconomiescameoutofthe
Depressionquitewell.At thetime,however, investorscouldnothave had such
confident expectations. Uncertainty about whetherthe "regime"had changed
adds to the fundamental uncertainty reflected in past and futurevolatilityof
macroeconomic data. Hamilton's(1988) regime-switching modelformalizes this
notion.Schwert(1989b)and Turner,Startz,and Nelson(1989) use Hamilton's
modeltorepresent stockreturn It is notpossible,however,
volatility. todetermine
whether volatilitywas "toohigh"duringtheDepressionwithoutsomemodelof
the possibleoutcomesthat did not occur.Thus, thereremainsa challengeto
boththeorists and empiricists to explainwhythisepisodewas so unusual.
Appendix: Data Series Used in This Paper

A. CommonStockReturns,1857-1987
I use themonthly stockreturnindexfromSchwert(1989d).For 1926-1987,I
use the returnsincludingdividendsto the value-weighted of all New
portfolio
17If investorsknewthatthe warswould have onlyshort-term effects,it is likelythat stockvolatility
would be affectedless than the volatilityof inflationor othermacroeconomicvariables.
Table Al
Variables Used in This Paper
Sample Period,
Series Description (Source) Size
Stock Monthlyreturnto a value-weightedportfolioof New 1/1857-12/1987
York Stock Exchange stocks (CRSP/Cowles/Macau- T = 1572
lay/Smithand Cole)
art Volatilityof monthlystock returnsfromdaily returnsin 2/1885-12/1987
the month(Dow Jones/Standard& Poor's) T = 1235
Interest Short-terminterestrate on low riskdebt instrument 1/1857-12/1987
(CRSP/Macaulay) T= 1572
YAa,Bond Yield or returnon high-gradelong-termcorporatedebt 1/1857-12/1987
(Moody's Aa/Macaulay) T = 1572
YBaa Yield on medium-gradelong-termcorporatedebt 1/1919-12/1987
(Moody's Baa) T = 828
PPI Inflationof producerprice index forall commodities 1/1857-12/1987
(BLS/Warren and Pearson) T = 1572
Base Growthrate of monetarybase (high-poweredmoney) 7/1878-12/1987
(Friedman and Schwartz/NBER/FederalReserve) T= 1314
IP Growthrate of the index of industrialproduction(sea- 2/1889-12/1987
sonally adjusted - Federal Reserve) T = 1187
V/S Market value of firmdividedby the value of stock for 1/1900-12/1987
S&P composite(Holland and Myers) T = 1056
Vol NYSE share tradingvolume (S&P/NYSE) 4/1881-12/1987
T= 1280
Days Numberof NYSE tradingdays per month (Dow Jones/ 2/1885-12/1987
S&P) T= 1235
York Stock Exchange (NYSE) stocks constructedby the CenterforResearch in

SecurityPrices (CRSP) at the Universityof Chicago. For 1885-1925, I use the
capital gain returnsto the Dow Jones composite index (1972) and add the
dividendyield fromthe value-weightedportfolioof NYSE stocks constructedby
the Cowles Commission (1939, pp. 168-169), as correctedby Wilson and Jones
(1987, p. 253, with erratum).For 1871-1885, I use the Cowles returns,corrected
for the effectsof time-averagingby Schwert (1989d). For 1857-1870, I use
Macaulay's (1938, pp. A142-A161) index of railroad stock prices to calculate
capital gain returnsand then add an estimateofthe dividendyieldfromSchwert
(1989d). This is equivalent to adding a dividendyield of 0.56 percentper month
(6.7 percentper year) to the percentchanges in railroadstock prices.
B. Short-Term InterestRates, 1857-1987

For 1926-1987,I use the monthlyyieldson the shortesttermU.S. Government
security(with no special tax provisions) which matures after the end of the
monthfromthe GovernmentBond File constructedby CRSP. For 1857-1925, I
use the fourto six monthcommercialpaper rates in New York fromMacaulay
(1938, Table 10, pp. A141-A161). The commercialpaper yields are adjusted so
that the level of the series is comparable to the Treasury yields, using the
ofCRSP yieldson Macaulayyieldsfrom1926to 1937:
regression
CRSPt = -0.000761 + 0.9737368 Macaulayt + ut,
(.000085) (0.0309330)
wherestandarderrorsare in parenthesesunderthecoefficient
estimates.This is
equivalentto subtractingan averageriskpremiumof 0.076percentper month
(0.91 percentper year) fromthe Macaulay yieldsto reflecta small default
premiumin commercialpaper. The correlationbetweenthe CRSP and the
Macaulayyieldsis 0.94 for1926-1937.
InterestRates,1857-1987
C. Long-Term
The high-gradecorporatebondyieldfor1919-1987is the Moody'sAa bond
yield(FederalReserve(1976a,Table 128,pp. 468-471) for1919-1940,Federal
Reserve(1976b,Table 12.12,pp. 720-721)for1941-1947,and Citibase(1978)for
1948-1987).For 1857-1918,I use Macaulay's(1938,Table 10,pp. A141-A161)
railroadbondyieldindex,adjustedto splicewiththe Moody'sseriesusingthe
averageratiooftheyieldsduring1919,(RR/Aa)= 0.964372.
D. ReturnstoLong-Term Bonds,1857-1987
Corporate
The capitalgainor loss fromholdingthebondduringthemonthis estimated
fromyieldsassumingthat,at thebeginning ofthemonth,thebondhas a 20-year
maturity, a price equal to par, and a coupon equal to the yield,using the
conventional bondpricingformula and endingprices.The
to calculatebeginning
monthly incomereturnis assumedto be one twelfth ofthe coupon.Since the
Moody'syieldsareaveragesoftheyieldswithinthemonth, thesereturnsarenot
comparable to returns basedon end-of-month data.To correctforthisproblem,
movingaverageprocessforthereturns:
I estimatea first-order
R*bt = a + et - 0 et_,
and thenthe "corrected" returnsare definedas Rbt = a + Et. This correction
eliminatesthepositiveautocorrelation at lag one inducedby the within-month
aggregationofyields.(See Working(1960).)Note,however, thatcorrected returns
arenotgoodestimatesofactualreturns basedon end-of-month pricessincetheir
withothervariablesare stillaffected
cross-correlations by timeaggregation of
theyields.18
E. Inflation
Rates,1857-1987
For 1890-1987,I use the Bureau of Labor Statistics'ProducerPrice Index
rate,notseasonallyadjusted.For 1857-1889,I use the inflation
(PPI) inflation
rateoftheWarrenand Pearson(1933)indexofproducer prices.I am gratefulto
GrantMcQueenformakingthesedata availableto me.
18 Schwert(1989d) develops a correctionsimilarto this one forreturnscalculated fromindexes of
time-averagedstockprices.
F. StockMarketShare TradingVolume,1881-1987
NYSE sharetradingvolume
Standard& Poor's(1986,p. 214) reportsmonthly
Citibase(1978)containssimilardata for1986-1987.The NYSE
for1883-1985.19
provideddata fromApril1881through1882.I measurethe numberoftrading
dayspermonthfor1885-1987fromthedailydata on theDow Jonesindexesin
Dow Jones(1972)-andon theStandard& Poor'scompositeindexin Standard&
Poor's(1986,pp. 134-187).
G. FinancialLeverage,1900-1987
Taggart(1986)discussesmanyestimatesoftheequityto totalcapitalratio(S/
V) forpubliccorporations in the United States for 1900-1979.Holland and
Myers(1979) estimatethe capital structureof corporations using National
IncomeAccountsdata on dividendand netinterest payments fromnonfinancial
corporations.Theycapitalizetheseflowsusingthe S&P dividendyieldand the
Moody'sBaa bondyield,respectively. Thesedataareavailableannuallyfor1929-
1945and quarterly for1946-1987.For 1926,I use theestimatefromCiccoloand
Baum(1986),basedon themarketvalueofdebtandpreferred and commonstock
fora sample of about 50 manufacturing firms.For 1900, 1912, and 1922, I
multiplyestimatesof the book value of S/V fromGoldsmith,Lipsey,and
Mendelson(1963,Tables III-4,and III-4b,pp. 140-141,146-147)bytheaverage
ratioof theseestimatesdividedby the Holland-Myers estimatesforthe years
1929,1933,1939,and 1945-1958,(HM/Goldsmith) = 1.226.Thus,I haveannual
estimatesof S/V for 1900, 1912, 1922, 1926, and 1929-1945and quarterly
estimatesfor1946-1987.
I createa monthlyseries(S/V), usingtheratesofreturn to thestockportfolio,
R,,,describedabove and the returnsto corporatebonds from Ibbotson(1986),
Rbt.Before1926,I estimatecorporate bondreturns usingtheyieldson high-grade
long-term bondsdescribedabove.I interpolate forward,
+ Rst)+ Bt_1(1+ Rbt)I1,
(S/V)t+= {St_-(1+ Rsd)/[St_1(1
and backward,
(S/V)t- = tS,+1/(1 + Rst+1)+ B,+1/(1+ Rbt+l)]}
+ Rst+1)/[St+l/(1
and thenuse theaverageoftheseestimatesforthemonthlyleverageestimate,
(S/V)t = I(S/V)t+ + (S/V) t-/2.
H. StockReturnVolatility,
1885-1987
FollowingFrench,Schwert,and Stambaugh(1987),I use the dailyreturnsto
theStandard& Poor'scomposite for1928-1987toestimatethestandard
portfolio
deviationof monthlystock returns.The estimateof the monthlystandard
19The New York Stock Exchange was closed fromAugustthroughmid-December,1914 due to the
outbreakof World War I. For purposes of this paper, I interpolateshare volume growthduringthis
period.
deviationis
rNt ' /2
= E rt
whereri,is the returnto the S&P portfolioon day i in montht (aftersubtracting

the sample mean forthe month) and there are Nt tradingdays in month t. For
1885-1927 I use a comparable estimatorbased on the daily values of the Dow
Jonescompositeportfolio.See Schwert(1989c,d) formoreinformationabout the
daily stock returnsand volatilityestimates.
L IndustrialProduction,1889-1987
For 1919-1987, I use the index of industrial production fromthe Federal
Reserve Board (1986) and Citibase (1978). For 1889-1918, I use Babson's Index
of the physicalvolume of business activityfromMoore (1961, p. 130), adjusted
to splice with the industrialproductiondata using the average ratio of Babson
to adjustedindustrialproductionfor1919-1939 (7.372662). I am gratefulto Grant
McQueen forprovidingthese data.
J. Money Supply,1867-1987
I use the monetarybase (called high-poweredmoneyin Friedmanand Schwartz
(1963)). For 1867-1960, I use data fromFriedmanand Schwartz (1963, Table B-
3, column (1), pp. 799-808) for the base. For 1961-1987, I use the seasonally
adjusted monetarybase reportedby the Federal Reserve Board fromCitibase
(1978). These series are spliced using the average ratio of the respectiveseries
during 1960. Thus, the base data since 1960 are multipliedby 1.127538. The
Friedmanand Schwartzdata are reportedon a monthlybasis beginningin May
1907. From June 1878 throughApril 1907, I use a monthlymonetarybase series
fromthe National Bureau of Economic Research (NBER), multipliedby the
average ratio of the Friedmanand Schwartzseries to the NBER series for1878-
1914, 1.006948.These data wereprovidedby ProfessorRobertBarro. Thus, there
are monthlydata on growthrates of the base for1878-1987.
REFERENCES
Abel, Andrew,1988, Stock prices undertime-varying dividendrisk:An exact solutionin an infinite-

horizongeneralequilibriummodel,JournalofMonetaryEconomics22, 375-393.
Black, Fischer,1976, Studies of stockprice volatilitychanges,in: Proceedingsofthe 1976 Meetingsof
theBusiness and EconomicsStatisticsSection,AmericanStatistical Association,177-181.
and JohnC. Cox, 1976,Valuingcorporatesecurities:Some effectsofbond indentureprovisions,
JournalofFinance 31, 351-367.
Bollerslev,Tim, Robert F. Engle, and JeffreyM. Wooldridge,1988, A capital asset pricing model
withtime varyingcovariances,JournalofPoliticalEconomy96, 116-131.
Box, GeorgeE. P. and David Pierce, 1970,Distributionof residualautocorrelationsin autoregressive-
integrated-moving averagetime series models,JournaloftheAmericanStatisticalAssociation65,
1509-1526.
Christie,AndrewA., 1982, The stochasticbehaviorof common stock variances: Value, leverageand
interestrate effects,JournalofFinancial Economics 10, 407-432.
Ciccolo, John C., Jr. and ChristopherF. Baum, 1986, Changing balance sheet relationshipsin the
U.S. manufacturingsector,1926-77, in Benjamin M. Friedman,ed: The ChangingRoles ofDebt

and Equity in Financing U.S. Capital Formation(Universityof Chicago Press, Chicago), 81-109.
Citibase Economic Database, 1946-86 [machine-readablemagneticdatafile], 1978 (Citibank, N.A.,
New York).
Cowles, Alfred III and Associates, 1939, Common Stock Indexes, 2nd ed., Cowles Commission
Monographno. 3 (Principia Press, Inc., Bloomington,IN).
Davidian, M. and R. J. Carroll,1987,Variance functionestimation,JournaloftheAmericanStatistical
Association82, 1079-1091.
Dow Jones & Co., 1972, The Dow JonesAverages,1885-1970, Maurice L. Farrell,ed. (Dow Jones &
Co., New York).
Engle, RobertF., 1982, Autoregressiveconditionalheteroskedasticity withestimatesof the variance
of United Kingdom inflation,Econometrica50, 987-1007.
- and Tim Bollerslev, 1986, Modeling the persistenceof conditional variances, Econometric
Reviews5, 1-50.
Fama, Eugene F., 1976, Inflationuncertaintyand expected returnson Treasury bills, Journal of
PoliticalEconomy84, 427-448.
and Kenneth R. French,1988a, Permanentand transitorycomponentsofstockprices,Journal
ofPoliticalEconomy96, 246-273.
and Kenneth R. French, 1988b, Dividend yields and expected stock returns,Journal of
Financial Economics22, 3-25.
and Kenneth R. French,1989, Business conditionsand expectedreturnson stocksand bonds,
JournalofFinancial Economics,Forthcoming.
Federal Reserve Board, 1976a, Banking and Monetary Statistics, 1914-1941 (U.S. Government
PrintingOffice,Washington,DC).
1976b, Banking and Monetary Statistics, 1941-1970 (U.S. GovernmentPrinting Office,
Washington,DC).
-~ 1986, Industrial Production,1986 Edition (U.S. GovernmentPrinting Office,Washington,
DC).
French,Kenneth R. and Richard Roll, 1986, Stock returnvariances: The arrivalof informationand
the reactionof traders,JournalofFinancial Economics 17, 5-26.
- , G. William Schwert,and Robert F. Stambaugh,1987, Expected stock returnsand volatility.
JournalofFinancial Economics 19, 3-29.
Friedman,Milton and Anna J. Schwartz,1963,A MonetaryHistoryof the United States, 1867-1960
(PrincetonUniversityPress, Princeton,NJ).
Fuller,Wayne A, 1976, Introductionto StatisticalTime Series (JohnWiley,New York).
Goldsmith,Raymond W., Robert E. Lipsey, and Morris Mendelson, 1963, Studies in the National
Balance Sheet of the UnitedStates, Vol. II (PrincetonUniversityPress, Princeton,NJ).
Hamilton, James D., 1988, Rational-expectationseconometricanalysis of changes in regime:An
investigationof the termstructureof interestrates,JournalofEconomicDynamics and Control
12, 385-423.
Hansen, Lars P., 1982, Large sample properties of generalized method of moments estimators,
Econometrica50, 1029-1054.
Holland, Daniel M. and Stewart C. Myers,1979, Trends in corporateprofitability and capital costs,
in R. Lindsay, ed.: The Nation's Capital Needs: Three Studies (Committee for Economic
Development,New York).
Huizinga, John and Frederic S. Mishkin, 1986, Monetary policy regime shifts and the unusual
behavior of real interestrates, Carnegie-Rochester ConferenceSeries on Public Policy 24, 231-
274.
Ibbotson Associates, 1986, Stocks,Bonds, Bills, and Inflation:1986 Yearbook(Ibbotson Associates,
Chicago).
Karpoff,Jonathan M., 1987, The relation between price changes and trading volume: A survey,
JournalofFinancial and QuantitativeAnalysis22, 109-126.
Keim, Donald B. and RobertF. Stambaugh,1986, Predictingreturnsin the stock and bond markets,
JournalofFinancial Economics 17, 357-390.
Kendall, Maurice G. and Alan Stuart,1969, The Advanced TheoryofStatistics,Vol. I, 3rd ed. (Hafner,
New York).
Kleidon, Allan W., 1986, Variance bounds tests and stockprice valuation models,JournalofPolitical
Economy94, 953-1001.
Lauterbach, Beni, 1989, Consumptionvolatility,productionvolatility,spot rate volatilityand the
returnson Treasurybills and bonds, JournalofFinancial Economics,Forthcoming.
Macaulay, FrederickR., 1938, The MovementsofInterestRates, Bond Yields and Stock Prices in the
UnitedStates Since 1856 (National Bureau of Economic Research,New York).
Mascaro, Angelo and Allan H. Meltzer, 1983, Long- and short-terminterestrates in a riskyworld,
JournalofMonetaryEconomics 12, 485-518.
Merton,Robert C., 1980, On estimatingthe expected returnon the market:An exploratoryinvesti-
gation,JournalofFinancial Economics8, 323-361.
Moore, GeoffreyH, ed., 1961, Business Cycle Indicators, Vol. II, Basic Data on CyclicalIndicators
(PrincetonUniversityPress, Princeton,NJ).
Officer,RobertR., 1973, The variabilityof the marketfactorof New York Stock Exchange, Journal
ofBusiness 46, 434-453.
Pagan, Adrian, 1984, Econometric issues in the analysis of regressionswith generatedregressors,
InternationalEconomicReview25, 221-247.
and Aman Ullah, 1988,The econometricanalysis ofmodelswithriskterms,JournalofApplied
Econometrics3, 87-105.
Pindyck,RobertS., 1984, Risk, inflation,and the stock market,AmericanEconomicReview74, 335-
351.
Poterba, James M. and Lawrence H. Summers,1986, The persistenceof volatilityand stock market
fluctuations,AmericanEconomicReview 76, 1142-1151.
and Lawrence H. Summers, 1988, Mean reversion in stock prices, Journal of Financial
Economics22, 27-59.
Romer,ChristinaD., 1986a, Spurious volatilityin historicalunemploymentdata, JournalofPolitical
Economy94, 1-37.
, 1986b,Is the stabilizationofthe postwareconomya figmentofthe data? AmericanEconomic
Review76, 314-334.
, 1989, The prewar business cycle reconsidered:New estimates of Gross National Product,
1869-1908,JournalofPoliticalEconomy97, 1-37.
Schwert,G. William, 1987, Effectsof model specificationon tests forunit roots in macroeconomic
data, JournalofMonetaryEconomics20, 73-103.
, 1989a, Tests forunit roots: A Monte Carlo investigation,JournalofBusiness and Economic
Statistics7, 147-159.
, 1989b, Business cycles, financialcrises and stock volatility,Carnegie-Rochester
Conference
Series on Public Policy 31, Forthcoming.
, 1989c, Stock volatilityand the crash of '87, ReviewofFinancial Studies,Forthcoming.
, 1989d,Indexes of United States stockprices from1802-1987,JournalofBusiness, Forthcom-
ing
Shapiro, MatthewD., 1988, The stabilizationof the U.S. economy:Evidence fromthe stock market,
AmericanEconomicReview 78, 1067-1079.
Shiller,Robert J., 1981a, Do stock prices move too much to be justifiedby subsequent changes in
dividends,AmericanEconomicReview75, 421-436.
, 1981b, The use of volatilitymeasures in assessing marketefficiency, JournalofFinance 36,
291-304.
Smith,Walter B. and ArthurH. Cole, 1935, Fluctuationsin AmericanBusiness, 1790-1860 (Harvard
UniversityPress, Cambridge).
Standard & Poor's, 1986,SecurityPrice Index Record,1986 ed. (Standard & Poor's Corp.,New York).
Taggart, Robert A., 1986, Secular patterns in the financingof U.S. corporations,in Benjamin M.
Friedman, ed.: The ChangingRoles of Debt and Equity in Financing U.S. Capital Formation
(Universityof Chicago Press, Chicago).
Turner,ChristopherM., Richard Startz, and Charles R. Nelson, 1989, A Markov model of heteros-
kedasticity,riskand learningin the stock market,JournalofFinancial Economics,Forthcoming.
ChangeOverTime?
Warren, in BulletinNo. 572,
G. F. andF. A. Pearson,1933,Wholesaleprices,1720to 1889,inclusive,
111-114(BureauofLaborStatistics, U.S. Government PrintingQffice,Washington,DC).
covariancematrixestimator
White,Halbert,1980,A heteroskedasticity-consistent and a directtest
Econometrica
forheteroskedasticity, 48,817-838.
Wilson,JackW. and CharlesP. Jones,1987,A comparison ofannualcommonstockreturns: 1871-
1925with1926-85,JournalofBusiness60,239-258.
Working,Holbrook, 1960,A noteon thecorrelation ofaveragesin randomchains,
offirstdifferences
Econometrica28, 916-918.

The Journal of Finance - December 1989 - SCHWERT - Why Does Stock Market Volatility Change Over Time

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

The Journal of Finance - December 1989 - SCHWERT - Why Does Stock Market Volatility Change Over Time

Uploaded by

Copyright:

Available Formats

THE JOURNAL OF FINANCE * VOL. XLIV, NO.

Why Does Stock Market Volatility Change Over

* WilliamE. SimonGraduateSchoolof BusinessAdministration, University and

causesa changein thevolatility ofstockreturns. Therehavebeenmanycritiques

Et-, Pt = Et-, E0 Dtk

and macroeconomicactivity.Section IV analyzes the relationbetweenfinancial

I. The Time Series Behavior of Stock and Bond Return Volatility

wherethere are Nt daily returnsritin month t.3Using nonoverlappingsamples

Rt = aj Djt + Rt( i + et. (3a)

(ii) Estimate a 12th-orderautoregressionforthe absolute values ofthe errors

I It| = yjDjt + pi I Et-I + Ut. (3b)

(iii) The regressandleI is an estimateof the standarddeviationof the stock

Figure 1. Predictions of the monthly standard deviation of stock returns based on

nominal interestrate,not risk,since these securitiesare essentiallydefaultfree.6

Figure 2. Predictionsof the monthlystandarddeviationsof stock returns(---) and of

cients and autocorrelationsof the estimates of stock returnvolatilitybased on

ISeeTable Al in theAppendixfora briefdescription

relationsof ?t are much largerthan those of IC'st thoughthey decay slowlyfor

0.19 0.29 0.24 0.22

| =t| yj Djt + E Pi|r It1 + ut. (3b)

estimatestowardstationarity.'The resultsforthe estimateof stockvolatility

in 1857,and theyare all railroadstocks.Nevertheless, theyrepresent mostof

II. Relations between Stock Market Volatilityand Macroeconomic

gure 4. Predictionsof the monthlystandard

was a periodofveryhighvolatility. Since the early1950's,thevolatility ofthe

Figure 5. Predictions of the monthly standard deviations of stock returns (- )and of

stockvolatility, high-grade bondreturnvolatilityI ht and short-term interest

10Models usingthe volatilityof medium-grade(Baa-rated) bond returnvolatilitY,IErm,l, instead of

" I also examinedthe volatilityof bank clearingsdata

-- - Stock Re Ind'l Productlon0.04

Fiur 6. Prdcin ftemnhysadr evain fsokrtrs(-)ado

01utra prdcto gotrae( )f,r19-87A12hrdrautrgesinwt

thobouevle ftersdasaeue oetmt oaiiyi ot t. To moe codtoal

Figure 6. Predictionhof the monthlystandarddeviationsof stocia returns(-otcn and of.

duringthe post-WorldWar II period.There is a small increasein volatility

12I am gratefulto Fischer Black forsuggestingthis interpretation.

yields on low versus high-gradelong-termcorporate debt also predict stock

18 1-86'9 16 8918'9 991''129 1939 1949 1959' 19"69 1979 1989

yields on bonds of differenltquality is directlyrelated to subsequentlyobserved

TV, Effects of Leverage Qn Stock Market Volatility

vt = ($/V)tl oSt + (B/V)It bt + 2 (S/V)t-1 (B/V)t_l cov(RSt,Rbt), (4)

where (S/V)tXl and (B/V)t-lrepresentthe fractionof the marketvalue of the

U,t E ti + E 3j Ust-j + -Y(YBaa YAa)t-1 + Ut.

Onlythe coefficientof the yieldspread-yis shown.Asymptotic

1900 1910 1920 1930 1940 1950 1960 1970 1980

Figure 8. Predictionsofthe monthlystandarddeviationof stockreturnsbased on daily

model for the errors.This is similar to the French, Schwert,and Stambaugh

ast -= ao + a1 (B/S)t1 + Ut, (5)

to thedailyestimates.For 1901-1952and 1901-1987,however, theestimatesof

V. Stock Market Trading and Volatility

A. TradingDays and Volatility

Frenchand Roll (1986) showthat stockvolatilityis higherwhenstockex-

(st = ao + a,lDayst + Ut, (6)

whereDaystis thenumberoftradingdaystheNYSE was openduringmontht.

B. Trading Volumeand Volatility

Anothermeasureof stocktradingactivityis sharetradingvolume.Karpoff

ast = a0 + a1 IDayst + Ut, (6)

wherevDas is thesquarerootoftheNYSE trading daysinthemonth.Generalized

whereVol,is the growthrateofvolumefrommontht - 1 to montht,and the

Ost -= a + Volt+ u'. (7)

Generalizedleastsquaresestimatesincludean ARMA (1,3) processfortheerrors.

errorsabovezero.Fortheestimatesofvolatility basedon monthly data I stjS the

allowsfordifferent monthlyintercepts. This modelis similarto thoseused in