Engle (2004)

American Economic Association
Risk and Volatility: Econometric Models and Financial Practice Author(s): Robert Engle Source: The American Economic Review, Vol. 94, No. 3 (Jun., 2004), pp. 405-420 Published by: American Economic Association Stable URL: http://www.jstor.org/stable/3592935 . Accessed: 02/04/2013 12:00
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp
.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact [email protected].
American Economic Association is collaborating with JSTOR to digitize, preserve and extend access to The American Economic Review.
http://www.jstor.org
This content downloaded from 200.37.165.17 on Tue, 2 Apr 2013 12:00:10 PM All use subject to JSTOR Terms and Conditions
Risk and Volatility: Econometric Models and FinancialPracticet

ENGLE* By ROBERT The advantageof knowing about risks is that we can change our behavior to avoid them. Of course, it is easily observed that to avoid all risks would be impossible; it might entail no flying, no driving,no walking, eating and drinking only healthy foods, and never being touched by sunshine. Even a bath could be dangerous.I could not receive this prize if I sought to avoid all risks. There are some risks we choose to take because the benefits from taking them exceed the possible costs. Optimalbehaviortakes risks that are worthwhile. This is the central paradigm of finance; we must take risks to achieve rewardsbut not all risks are equally rewarded. Both the risks and the rewardsare in the future, so it is the expectation of loss that is balanced against the expectation of reward.Thus we optimize our behavior, and in particularour portfolio, to maximize rewardsand minimize risks. This simple concept has a long history in economics and in Nobel citations. Harry M. Markowitz (1952) and James Tobin (1958) associated risk with the variancein the value of a portfolio. From the avoidance of risk they derived optimizing portfolio and banking behavior. William Sharpe (1964) developed the implicationswhen all investors follow the same objectives with the same information.This theory is called the CapitalAsset Pricing Model or CAPM, and shows that there is a naturalrelation between expected returns and variance. These contributionswere recognized by Nobel prizes in 1981 and 1990. Fisher Black and Myron Scholes (1972) and RobertC. Merton (1973) developed a model to evaluatethe pricingof options. While the theory is based on option replication arguments through dynamic trading strategies, it is also consistent with the CAPM. Put options give the owner the right to sell an asset at a particular price at a time in the future.Thus these options can be thought of as insurance.By purchasing such put options, the risk of the portfolio can be completely eliminated. But what does this insurance cost? The price of protection depends upon the risks and these risks are measuredby the variance of the asset returns.This contribution was recognized by a 1997 Nobel prize. When practitionersimplementedthese financial strategies, they required estimates of the variances.Typically the squareroot of the variance, called the volatility, was reported.They immediately recognized that the volatilities were changing over time. They found different answers for different time periods. A simple approach,sometimes called historical volatility, was, and remains, widely used. In this method, the volatility is estimated by the sample standard deviation of returns over a short period. But, what is the right period to use? If it is too long, then it will not be so relevant for today and if it is too short, it will be very noisy. Furthermore,it is really the volatility over a futureperiod that should be consideredthe risk, hence a forecast of volatility is needed as well as a measurefor today. This raises the possibility that the forecast of the average volatility over the next week might be different from the forecast over a year or a decade. Historical volatility had no solution for these problems. On a more fundamentallevel, it is logically inconsistent to assume, for example, that the variance is constant for a period such as one year ending today and also that it is constantfor the year ending on the previous day but with a differentvalue. A theory of dynamic volatilities
405
t This article is a revised version of the lecture Robert Engle delivered in Stockholm, Sweden, on December 8, 2003, when he received the Bank of Sweden Prize in Economic Sciences in Memory of Alfred Nobel. The article is copyright ? The Nobel Foundation2003 and is published here with the permission of the Nobel Foundation. * Stern School of Business, New York University, 44 West Fourth Street, New York, NY 10012 (e-mail: This paperis the resultof more than [email protected]). two decades of researchand collaborationwith many, many people. I would particularlylike to thank the audiences in B.I.S., Stockholm, Uppsala, Cornell University, and the Universite de Savoie for listening as this talk developed. David Hendry,Tim Bollerslev, Andrew Patton,and Robert Ferstenberg provided detailed suggestions. Nevertheless, any lacunas remain my responsibility.
406
ECONOMICREVIEW THEAMERICAN
JUNE 2004
is needed; this is the role that is filled by the ARCH models and their many extensions that we discuss today. In the next section, I will describethe genesis of the ARCH model, and then discuss some of its many generalizationsand widespreadempirical support.In subsequentsections, I will show how this dynamicmodel can be used to forecast volatility and risk over a long horizon and how it can be used to value options. I. The Birth of the ARCHModel The ARCH model was invented while I was on sabbaticalat the London School of Economics in 1979. Lunch in the Senior Common Room with David Hendry, Dennis Sargan,Jim Durbin,and many leading econometriciansprovided a stimulatingenvironment.I was looking for a model that could assess the validity of a conjecture of Milton Friedman(1977) that the unpredictability of inflation was a primary cause of business cycles. He hypothesized that the level of inflation was not a problem;it was the uncertainty aboutfuturecosts andprices that from investing and would prevententrepreneurs lead to a recession. This could only be plausible if the uncertaintywere changing over time so this was my goal. Econometricians call this I had recently workedextenheteroskedasticity. sively with the Kalman Filter and knew that a likelihood function could be decomposed into the sum of its predictive or conditional densities. Finally, my colleague, Clive Granger,with whom I sharethis prize, had recently developed a test for bilinear time series models based on the dependence over time of squaredresiduals. That is, squaredresidualsoften were autocorrelated even thoughthe residualsthemselves were not. This test was frequentlysignificantin economic data; I suspected that it was detecting somethingbesides bilinearitybut I did not know what. The solution was autoregressive conditional heteroskedasticityor ARCH, a name invented by David Hendry. The ARCH model described the forecastvariancein terms of currentobservables. Instead of using short or long sample standard deviations,the ARCH model proposed taking weighted averages of past squaredforecast errors,a type of weighted variance. These weights could give more influence to recent
informationand less to the distantpast. Clearly the ARCH model was a simple generalizationof the sample variance. The big advance was that the weights could be estimated from historical data even though the true volatility was never observed. Here is how this works. Forecasts can be calculated every day or every period. By examining these forecasts for different weights, the set of weights can be found that make the forecasts closest to the variance of the next return.This procedure, based on Maximum Likelihood, gives a systematicapproachto the estimationof the optimal weights. Once the weights are determined, this dynamic model of time-varying volatility can be used to measurethe volatility at any time and to forecast it into the near and distant future. Granger's test for bilinearity turnedout to be the optimal, or LagrangeMultiplier test for ARCH, and is widely used today. There are many benefits to formulating an explicit dynamic model of volatility. As mentioned above, the optimal parameterscan be estimatedby MaximumLikelihood.Tests of the adequacyand accuracyof a volatility model can be used to verify the procedure.One-step and multi-step forecasts can be constructed using these parameters.The unconditional distributions can be establishedmathematicallyand are generally realistic. Inserting the relevant variables into the model can test economic models that seek to determine the causes of volatility. Incorporatingadditional endogenous variables and equationscan similarlytest economic models aboutthe consequences of volatility. Several applicationswill be mentioned below. David Hendry's associate, FrankSrba, wrote the first ARCH program.The application that appearedin Engle (1982) was to inflationin the United Kingdom since this was Friedman's conjecture.While there was plenty of evidence that the uncertainty in inflation forecasts was time varying, it did not correspondto the U.K. business cycle. Similar tests for U.S. inflation data, reported in Engle (1983), confirmed the finding of ARCH but found no business-cycle effect. While the trade-off between risk and return is an importantpart of macroeconomic theory, the empiricalimplications are often difficult to detect as they are disguised by other dominating effects, and obscured by the reliance on relatively low-frequency data. In fi-
VOL.94 NO. 3
ENGLE:RISKAND VOLATILITY
407
nance, the risk/returneffects are of primary importanceand data on daily or even intradaily frequencies are readily available to form accurate volatility forecasts.Thus financeis the field in which the great richness and variety of ARCH models developed.
II. Generalizing the ARCH Model
Generalizations to different weighting schemes can be estimated and tested. The very importantdevelopment by my outstandingstudent Tim Bollerslev (1986), called Generalized Autoregressive Conditional Heteroskedasticity, or GARCH, is today the most widely used model. This essentially generalizes the purely autoregressiveARCH model to an autoregressive moving average model. The weights on past squared residuals are assumed to decline geometricallyat a rate to be estimatedfrom the of data. An intuitively appealing interpretation the GARCH(1,1) model is easy to understand. The GARCH forecast variance is a weighted average of three different variance forecasts. One is a constant variance that correspondsto the long-runaverage.The second is the forecast that was made in the previous period. The third is the new informationthat was not available when the previous forecast was made. This could be viewed as a varianceforecastbased on one periodof information.The weights on these three forecasts determinehow fast the variance changes with new informationand how fast it reverts to its long-run mean. A second enormously importantgeneralization was the ExponentialGARCH or EGARCH model of Daniel B. Nelson (1991), who prematurely passed away in 1995 to the great loss of our profession as eulogized by Bollerslev and Peter E. Rossi (1995). In his short academic career, his contributionswere extremely influential. He recognized that volatility could respond asymmetricallyto past forecast errors.In a financial context, negative returnsseemed to be more importantpredictorsof volatility than positive returns. Large price declines forecast greater volatility than similarly large price increases. This is an economically interestingeffect that has wide-ranging implications to be discussed below. Furthergeneralizations have been proposed by many researchers.There is now an alphabet
soup of ARCH models that include: AARCH, APARCH, FIGARCH,FIEGARCH,STARCH, SWARCH, GJR-GARCH,TARCH, MARCH, NARCH, SNPARCH, SPARCH, SQGARCH, CESGARCH,ComponentARCH, Asymmetric ComponentARCH, Taylor-Schwert,Student-tARCH, GED-ARCH, and many others that I have regrettably overlooked. Many of these models were surveyed in Bollerslev et al. (1992), Bollerslev et al. (1994), Engle (2002b), and Engle and Isao Ishida (2002). These models recognize that there may be importantnonlinearity, asymmetry,and long memory properties of volatility and that returnscan be nonnormal with a variety of parametricand nonparametric distributions. A closely relatedbut econometricallydistinct class of volatility models called StochasticVolatility, or SV models, have also seen dramatic development. See, for example, Peter K. Clark (1973), StephenTaylor (1986), AndrewC. Harvey et al. (1994), and Taylor (1994). These models have a differentdata-generating process which makes them more convenient for some purposes but more difficult to estimate. In a linear framework, these models would simply of the same process; be differentrepresentations but in this nonlinear setting, the alternative specificationsare not equivalent, althoughthey are close approximations. III. ModelingFinancialReturns The success of the ARCH family of models is in large measureto the applications attributable in finance. While the models have applicability for many statistical problems with time series value for financialtime data,they find particular series. This is partly because of the importance of the previously discussed trade-off between risk and returnfor financialmarkets,and partly because of three ubiquitous characteristicsof financial returns from holding a risky asset. Returnsare almost unpredictable, they have surprisingly large numbersof extreme values, and both the extremes and quiet periods are clustered in time. These featuresare often described as unpredictability, fat tails, and volatility clustering. These are precisely the characteristics for which an ARCH model is designed. When volatility is high, it is likely to remainhigh, and whenit is low, it is likely to remainlow. However,
408
THEAMERICAN ECONOMICREVIEW
JUNE 2004
these periods are time limited so that the forecast is sure to eventually revert to less extreme volatilities. An ARCH process produces dynamic, mean-reverting patternsin volatility that can be forecast. It also produces a greaternumber of extremes than would be expected from a standardnormal distribution,since the extreme values during the high volatility period are greater than could be anticipated from a constant volatility process. The GARCH(1,1) specificationis the workhorse of financialapplications.It is remarkable that one model can be used to describethe volatility dynamicsof almost any financialreturnseries. This appliesnot only to U.S. stocksbut also to stocks tradedin most developed markets,to most stocks tradedin emergingmarkets,and to most indices of equity returns.It applies to exchange rates, bond returns,and commodity returns.In many cases, a slightly bettermodel can be foundin the list of models above,but GARCH is generallya very good starting point. The widespread success of GARCH(1,1) begs to be understood.What theory can explain why volatility dynamics are similar in so many different financial markets?In developing such a theory, we must first understandwhy asset prices change. Financial assets are purchased and owned because of the futurepayments that can be expected. Because these payments are uncertainand depend upon unknowable future developments, the fair price of the asset will require forecasts of the distribution of these payments based on our best informationtoday. As time goes by, we get more informationon these future events and revalue the asset. So at a basic level, financialprice volatility is due to the arrivalof new information.Volatility clustering is simply clusteringof informationarrivals. The fact that this is common to so many assets is simply a statementthat news is typically clustered in time. To see why it is naturalfor news to be clustered in time, we must be more specific about the informationflow. Consideran event such as an invention that will increase the value of a firmbecause it will improvefutureearningsand dividends. The effect on stock prices of this event will depend on economic conditions in the economy and in the firm. If the firm is near the effect can be very large and if it bankruptcy, is already operating at full capacity, it may be
small. If the economy has low interestrates and surplus labor, it may be easy to develop this new product. With everything else equal, the response will be greaterin a recession than in a boom period.Hence we are not surprisedto find higher volatility in economic downturnseven if the arrival rate of new inventions is constant. This is a slow moving type of volatility clustering that can give cycles of several years or longer. The same invention will also give rise to a high-frequency volatility clustering. When the invention is announced, the market will not immediatelybe able to estimate its value on the stock price. Agents may disagree but be sufficiently unsure of their valuations that they pay attention to how others value the firm. If an investorbuys until the price reacheshis estimate of the new value, he may revise his estimate after he sees others continue to buy at successively higher prices. He may suspect they have better informationor models and consequently raise his valuation. Of course, if the others are selling, then he may revise his price downward. This process is generally called price discovery and has been modeled theoreticallyand empirIt leads to volically in marketmicrostructure. at a much atility clustering higher frequency than we have seen before. This process could last a few days or a few minutes. But to understand volatilitywe must think of morethanone invention.While the arrival rateof inventions not have clear other may patterns, types of news surelydo. The news intensityis generally high duringwars and economic distress.During or reguimportant global summits,congressional or central board bank elections, latory hearings, meetings,therearelikely to be manynews events. These episodes are likely to be of mediumduration, lastingweeks or months. The empirical volatility patternswe observe are composed of all three of these types of events. Thus we expect to see ratherelaborate volatility dynamics and often rely on long time series to give accurate models of the different time constants. IV. Modelingthe Causesand Consequences of FinancialVolatility Once a model has been developed to measure volatility, it is naturalto attemptto explain the
VOL.94 NO. 3
409
causes of volatility and the effects of volatility on the economy. There is now a large literature examining aspects of these questions. I will only give a discussion of some of the more limited findings for financial markets. In financialmarkets,the consequencesof volatility are easy to describe although perhaps difficult to measure. In an economy with one risky asset, a rise in volatility should lead investors to sell some of the asset. If there is a fixed supply, the price must fall sufficiently so that buyers take the other side. At this new lower price, the expected returnis higherby just enough to compensate investors for the increased risk. In equilibrium, high volatility should correspond to high expected returns. Merton (1980) formulated this theoretical model in continuous time, and Engle et al. (1987) proposed a discrete time model. If the price of risk were constantover time, then rising conditional variances would translate linearly into rising expected returns.Thus the mean of the return equation would no longer be estimated as zero, it would depend upon the past squaredreturnsexactly in the same way thatthe conditional variance depends on past squared returns.This very strong coefficient restriction can be tested and used to estimate the price of risk. It can also be used to measure the coefficient of relative risk aversion of the representative agent under the same assumptions. Empiricalevidence on this measurementhas been mixed. While Engle et al. (1987) find a positive and significant effect, Ray-Yeutien Chou et al. (1992) and Lawrence R. Glosten et al. (1993) find a relationship that varies over time and may be negative because of omitted variables. Kenneth R. French et al. (1987) showed that a positive volatility surpriseshould and does have a negative effect on asset prices. There is not simply one risky asset in the economy and the price of risk is not likely to be constant;hence the instability is not surprising and does not disprove the existence of the risk returntrade-off, but it is a challenge to better modeling of this trade-off. The causes of volatility are more directly modeled. Since the basic ARCH model and its many variantsdescribe the conditionalvariance as a function of lagged squared returns,these are perhapsthe proximatecauses of volatility. It is best to interpretthese as observablesthathelp
in forecastingvolatility ratherthan as causes. If the true causes were included in the specification, then the lags would not be needed. A small collection of papershas followed this route. Torben G. Andersen and Bollerslev (1998b) examinedthe effects of announcements on exchange rate volatility. The difficulty in finding important explanatory power is clear even if these announcementsare responsible in importantways. Another approachis to use the volatility measured in other markets. Engle et al. (1990b) find evidence that stock volatility causes bond volatility in the future.Engle et al. (1990a) model the influenceof volatility in markets with earlier closing on markets with later closing. For example, they examine the influence of currencyvolatilities in European,Asian markets, and the prior day U.S. market on today's U.S. currency volatility. Yasushi Hamao et al. (1990), Pat Bums et al. (1998), and others have applied similartechniquesto global equity markets. V. An Example To illustrate the use of ARCH models for financial applications, I will give a rather extended analysis of the Standard& Poors 500 Compositeindex. This index representsthe bulk of the value in the U.S. equity market. I will look at daily levels of this index from 1963 through late November 2003. This gives a sweep of U.S. financialhistory that provides an ideal setting to discuss how ARCH models are used for risk management and option pricing. All the statistics and graphs are computed in 4.1. EViewsTM The raw data are presentedin Figure 1 where prices are shown on the left axis. The rather smooth lower curve shows what has happened to this index over the last 40 years. It is easy to see the great growth of equity prices over the period and the subsequentdecline after the new millennium.At the beginning of 1963 the index was priced at $63 and at the end it was $1,035. That means that one dollar invested in 1963 would have become $16 by November 21, 2003 (plus the stream of dividends that would have been received, as this index does not take account of dividends on a daily basis). If this investor were clever enough to sell his position on March 24, 2000, it would have been worth
410
JUNE 2004
SP500
SPRETURNS |
SP500
- SPRETURNS I
FIGURE 1. S&P 500 DAILY PRICES AND RETURNS FROM
2. S&P 500 DAILY BEFORE1987 FIGURE
2003 1963 TO NOVEMBER JANUARY

%f n t u.ua
$24. Hopefully he was not so unluckyas to have purchasedon that day. Although we often see pictures of the level of these indices, it is obviously the relative price from the purchasepoint to the sale point that matters.Thus economists focus attentionon returnsas shown at the top of the figure.This shows the daily price change on the right axis (computedas the logarithmof the price today divided by the price yesterday).This returnseries is centeredaroundzero throughout the sample period even though prices are sometimes increasing and sometimes decreasing. Now the most dramatic event is the crash of October 1987 which dwarfs all other returnsin the size of the decline and subsequent partial recovery. Other importantfeatures of this data series can be seen best by looking at portions of the whole history. For example, Figure 2 shows the that same graphbefore 1987. It is very apparent the amplitude of the returns is changing. The magnitude of the changes is sometimes large and sometimes small. This is the effect that ARCH is designed to measureand that we have called volatility clustering. There is, however, another interesting feature in this graph. It is clear thatthe volatility is higher when prices are falling. Volatility tends to be higher in bear markets.This is the asymmetricvolatility effect that Nelson described with his EGARCH model. Looking at the next subperiodafter the 1987 crash in Figure 3, we see the record low vola-
i ,? i ,:? j 'i
Ii.il: ii
-0.04 -0.00
I;;;
- 0.04 --0.08
1,600800400200_ , I 1988
} 5 ? ,
t i
4s ,
.I
l {
*X
I,
* * . I
J l
1990 I
-
1992
1994
1996
1998
2000
SP500 .
SPRETURNS
3. S&P 500 1988 TO 2000 FIGURE
tility period of the middle 1990's. This was accompanied by a slow and steady growth of equity prices. It was frequently discussed whether we had moved permanentlyto a new era of low volatility. History shows that we did not. The volatility began to rise as stock prices got higher and higher,reachingvery high levels from 1998 on. Clearly, the stock market was risky from this perspective but investors were willing to take this risk because the returnswere so good. Looking at the last period since 1998 in Figure 4, we see the high volatility continue as the marketturneddown. Only at the end of the sample, since the official conclusion of the Iraq war, do we see substantialdeclines in volatility. This has apparentlyencouraged investors to come back into the market which has experienced substantialprice increases.
VOL.94 NO. 3
4
A
411
3-
210
0-1-
o 1998 1999 -2000 2001 2002 2003 - SPRETURNS
-2-3o
0o
SP500
A -4
4. S&P 500 1998 TO2003 FIGURE
-0.08
-0.04
0.00
0.04
0.08
TABLE 1-S&P 500 RETURNS Sample Mean Standarddeviation Skewness Kurtosis Full 0.0003 0.0094 -1.44 41.45 Since 1990 0.0003 0.0104 -0.10 6.78
SPRETURNS
FIGURE 5. QUANTILE PLOT OF S&P 500 RETURNS
1990 POST
We now show some statistics that illustrate the three stylized facts mentionedabove: almost unpredictable returns, fat tails, and volatility clustering. Some features of returnsare shown in Table 1. The mean is close to zero relative to the standarddeviation for both periods. It is 0.03 percent per trading day or about 7.8 percent per year. The standarddeviation is slightly higher in the 1990's. These standarddeviations correspondto annualizedvolatilities of 15 percent and 17 percent. The skewness is small throughout. The most interesting feature is the kurtosis which measuresthe magnitudeof the extremes. If returnsare normallydistributed, then the kurtosis should be three. The kurtosis of the nineties is substantial at 6.8, while for the full sample it is a dramatic41. This is strong evidence that extremes are more substantialthan would be expected from a normalrandomvariable. Similar evidence is seen graphically in Figure 5, which is a quantile plot for the post1990 data. This is designed to be a straightline if returnsare normallydistributedand will have an s-shape if there are more extremes. of returnsand the clusThe unpredictability
tering of volatility can be concisely shown by looking at autocorrelations. Autocorrelations are correlationscalculatedbetween the value of a randomvariabletoday and its value some days in the past. Predictabilitymay show up as sigin returns,and volatilnificant autocorrelations show will up as significant ity clustering in squaredor absolute returns. autocorrelations Figure 6 shows both of these plots for the post1990 data. Under conventional criteria,' autocorrelationsbigger than 0.033 in absolute value would be significant at a 5-percent level. Clearly, the return autocorrelationsare almost all insignificantwhile the squarereturnshave all autocorrelations significant. Furthermore,the are all positive, squaredreturnautocorrelations which is highly unlikely to occur by chance. This figuregives powerfulevidence for both the of returnsand the clusteringof unpredictability volatility. Now we turn to the problem of estimating volatility. The estimates called historical volatility are based on rolling standarddeviations of returns. In Figure 7 these are constructed for five-day,one-year,and five-yearwindows.While each of these approachesmay seem reasonable,
'The actual critical values will be somewhat greateras the series clearly are heteroskedastic.This makes the case in returnseven stronger. for unpredictability
412 0.25 0.2 0.15 0.1
JUNE 2004
l I
0.05
0
, EI
.I
-0.05
-0.1
mSP Returns SQ Returns

FIGURE 6. RETURN AND SQUARED RETURN AUTOCORRELATIONS
2.U
1.6
1.2
0.8
0.4
1
i
I
p.:
.
_' -. _ . . . . . .
I;
: I
"
. . .
,.,ja g - 'I
'
I t
'
LU L -w,
x.
. . K
.. I .
65I 65
. 70
70
75
75
80 80
. . I . 85
85
I , 90
90
95
95
X0
..
V5 ---
V260 --V1,3001
FIGURE 7. HISTORICAL VOLATILITIES WITH VARIOUS WINDOWS
the answers are clearly very different.The fiveday estimate is extremely variable while the other two are much smoother. The five-year estimate smooths over peaks and troughs that the other two see. It is particularly slow to recover after the 1987 crash and particularly slow to reveal the rise in volatility in 19982000. In just the same way, the annualestimate fails to show all the details revealed by the five-day volatility. However, some of these details may be just noise. Without any true measure of volatility, it is difficult to pick from these candidates. The ARCH model provides a solution to this dilemma. From estimatingthe unknownparam-
eters based on the historicaldata, we have forecasts for each day in the sample period and for any period after the sample. The natural first model to estimate is the GARCH(1,1). This model gives weights to the unconditionalvariance, the previous forecast, and the news measured as the square of yesterday's return.The weights are estimated to be (0.004, 0.941, 0.055), respectively.2 Clearly the bulk of the informationcomes from the previous day forecast. The new informationchanges this a little and the long-run average variance has a very small effect. It appears that the long-run variance effect is so tiny that it might not be important. This is incorrect. When forecasting many steps ahead, the long-run variance eventually dominatesas the importanceof news and other recent informationfades away. It is naturally small because of the use of daily data. In this example, we will use an asymmetric volatility model that is sometimes called GJRGARCH for Glosten et al. (1993) or TARCH for Threshold ARCH (Jean Michael Zakoian, 1994). The statisticalresults are given in Table 2. In this case there are two types of news. There is a squaredreturnand there is a variable that is the squaredreturnwhen returnsare negative, and zero otherwise. On average, this is
2 For a conventional GARCH model defined as ht+1 -I o + ar2 + P3ht, the weights are ((1 - a - 3), 3, a).
VOL.94 NO. 3
TABLE2-TARCH DATA OF S&P 500 RETURN ESTIMATES
413
Dependent variable:NEWRET_SP Method: ML-ARCH (Marquardt) Date: 11/24/03 Time: 09:27 Sample (adjusted):1/03/1963-11/21/2003 Included observations: 10,667 after adjustingendpoints Convergence achieved after 22 iterations Variance backcast:ON Coefficient C 0.000301 Variance equation C ARCH(1) (RESID < 0)*ARCH(1) GARCH(1) 4.55E-07 0.028575 0.076169 0.930752 5.06E-08 0.003322 0.003821 0.002246 8.980473 8.602582 19.93374 414.4693 0.0000 0.0000 0.0000 0.0000 Standarderror 6.67E-05 z-statistic 4.512504 Probability 0.0000
1.4 1.2 1.0 0.80.60.10 ,,
0.40.2 1990 0 .0 , 65 , 70 , . 75 , 80 , , 85 . 90 , 95 , 00 , |-1992 1994 1996 198 2000 2002 3GARCHSTD SPRETURNS ---3*GARCHSTD
GARCHVOL
THREE INTERVALS: FIGURE 9. GARCH CONFIDENCE DEVIATIONS STANDARD
FIGURE 8. GARCH VOLATILITIES
half as big as the variance,so it mustbe doubled implying that the weights are half as big. The weights are now computed on the long-run average, the previous forecast, the symmetric news, and the negative news. These weights are estimated to be (0.002, 0.931, 0.029, 0.038) respectively.3Clearly the asymmetryis important since the last termwould be zero otherwise. In fact, negative returnsin this model have more than threetimes the effect of positive returnson future variances. From a statistical point of view, the asymmetry term has a t-statistic of almost 20 and is very significant. The volatility series generatedby this model is given in Figure 8. The series is more jagged
3 If the model is defined as ht = wo +- h_ + ar_ + - y/2, 3, a, -y t_It_l<0, then the weights are (1 - a y/2).
than the annual or five-year historical volatilities, but is less variable than the five-day volatilities. Since it is designed to measure the volatility of returnson the next day, it is natural to form confidence intervalsfor returns.In Figure 9 returnsare plotted against plus and minus three TARCH standarddeviations. Clearly the confidence intervals are changing in a very believable fashion. A constantband would be too wide in some periods and too narrowin others. The TARCH intervalsshould have 99.7-percent probabilityof including the next observationif the data are really normally distributed. The expected numberof times that the next returnis outside the interval should then be only 29 out of the more than 10,000 days. In fact, there are 75 indicating that there are more outliers than would be expected from normality. Additional information about volatility is available from the options market.The value of
414
JUNE 2004
two widely used applications. In the presentation, some novel implications of asymmetric volatility will be illustrated. VI. FinancialPractice-Value at Risk Every morning in thousands of banks and financial services institutionsaroundthe world, the Chief Executive Officer is presentedwith a risk profileby his Risk ManagementOfficer.He is given an estimate of the risk of the entire portfolio and the risk of many of its components. He would typically learnthe risk faced by the firm's European Equity Division, its U.S. TreasuryBond Division, its CurrencyTrading Unit, its Equity Derivative Unit, and so forth. These risks may even be detailed for particular trading desks or traders. An overall figure is then reportedto a regulatorybody although it may not be the same number used for internal purposes.The risk of the company as a whole is less than the sum of its parts since different portions of the risk will not be perfectly correlated. The typical measure of each of these risks is Value at Risk, often abbreviatedas VaR. The VaR is a way of measuring the probabilityof losses that could occur to the portfolio. The 99-percent one-day VaR is a numberof dollars that the manager is 99 percent certain will be worse than whatever loss occurs on the next day. If the one-day VaR for the currencydesk is $50,000, then the risk officer asserts that only on one day out of 100 will losses on this portfolio be greater than $50,000. Of course this means that on about 2.5 days a year, the losses will exceed the VaR. The VaR is a measure of risk that is easy to understandwithout knowing any statistics. It is, however, just one quantile of the predictive distribution and therefore it has limited informationon the probabilitiesof loss. Sometimes the VaR is defined on a multi-day basis. A 99-percentten-day VaR is a numberof dollars that is greaterthan the realized loss over ten days on the portfolio with probability0.99. This is a more common regulatorystandard but is typically computed by simply adjusting the one-day VaR as will be discussed below. The loss figures assume that the portfolio is unchanged over the ten-day period which may be counterfactual.
VIX --GARCHM
FIGURE10. IMPLIED VOLATILITIES AND GARCH VOLATILITIES
tradedoptions depends directly on the volatility of the underlyingasset. A carefully constructed portfolio of options with different strikes will have a value that measures the option market estimate of future volatility under ratherweak assumptions.This calculationis now performed by the CBOE for S&P 500 options and is reported as the VIX. Two assumptions that underlie this index are worth mentioning. The price process should be continuous and there should be no risk premiaon volatility shocks. If these assumptions are good approximations, then implied volatilities can be compared with ARCH volatilities. Because the VIX represents the volatility of one-monthoptions, the TARCH volatilities must be forecast out to one month. The results are plotted in Figure 10. The general pattern is quite similar, although the TARCH is a little lower than the VIX. These differences can be attributedto two sources. First, the option pricing relation is not quite correct for this situationand does not allow for volatility risk premia or nonnormal returns. These adjustmentswould lead to higher options prices and consequentlyimplied volatilities that were too high. Second, the basic ARCH models have very limited informationsets. They do not use information on earnings, wars, elections, etc. Hence the volatility forecasts by traders should be generally superior;differences could be due to long-lasting informationevents. This extended example illustrates many of the featuresof ARCH/GARCHmodels and how they can be used to study volatility processes. We turn now to financialpractice and describe
VOL.94 NO. 3
415
To calculate the VaR of a trading unit or a firmas a whole, it is necessaryto have variances and covariances,or equivalentlyvolatilities and correlations,among all assets held in the portfolio. Typically, the assets are viewed as responding primarilyto one or more risk factors that are modeled directly. RiskmetricsTM, for uses about 400 risk factors. example, global BARRA uses industry risk factors as well as risk factors based on firm characteristicsand otherfactors.A diversifiedU.S. equity portfolio would have risks determinedprimarilyby the aggregate market index such as the S&P 500. We will carry forwardthe example of the previous section to calculatethe VaR of a portfolio that mimics the S&P. The one-day 99-percentVaR of the S&P can be estimatedusing ARCH. Fromhistoricaldata, the best model is estimated, and then the standard deviation is calculated for the following day. In the case of S&P on November 24, this forecast standarddeviation is 0.0076. To convert this into VaR we must make an assumption about the distributionof returns.If normalityis assumed, the 1 percent point is -2.33 standard deviations from zero. Thus the value at risk is 2.33 times the standarddeviation or in the case of November 24, it is 1.77 percent. We can be 99 percent sure that we will not lose more than 1.77 percentof portfoliovalue on November24. In fact the marketwent up on the 24th so there were no losses. The assumptionof normalityis highly questionable. We observed that financial returns have a surprisingnumberof large returns.If we divide the returnsby the TARCH standarddeviations, the result will have a constant volatility of one but will have a nonnormal distribution.The kurtosis of these "devolatized returns," or "standardizedresiduals," is 6.5, which is much less than the unconditionalkurtosis, but is still well above normal.From these devolatized returns,we can find the 1-percent quantileand use this to give a betteridea of the VaR. It turnsout to be 2.65 standarddeviations below the mean. Thus our portfolio is riskier than we thought using the normal approximation. The one-day 99-percent VaR is now estimated to be 2 percent. A ten-day value at risk is often required by regulatory agencies and is frequently used internally as well. Of course, the amount a
portfolio can lose in ten days is a lot greater than it can lose in one day. But how much greater is it? If volatilities were constant, then the answer would be simple; it would be the square root of ten times as great. Since the ten-day variance is ten times the one-day variance, the ten-day volatility multiplier would be the square root of ten. We would take the one-day standarddeviation and multiply it by 3.16 and then with normality we would multiply this by 2.33, giving 7.36 times the standard deviation. This is the conventional solution in industry practice. For November 24, the ten-day 99-percent VaR is 5.6 percent of portfolio value. However, this result misses two important features of dynamic volatility models. First, it makes a difference whether the currentvolatilities are low or high relative to the long-run average, so that they are forecast to increase or decrease over the next ten days. Since the volatility is relatively low in November, the TARCH model will forecast an increase over the next ten days. In this case, this effect is not very big as the standarddeviation is forecast to increaseto 0.0077 from 0.0076 over the ten-day period. More interestingis the effect of asymmetryin variance for multi-periodreturns.Even though each period has a symmetric distribution,the multi-period return distributionwill be asymmetric. This effect is simple to understandbut has not been widely recognized. It is easily illustratedwith a two-step binomial tree, Figure 11, as used in elementaryoption pricingmodels. In the first period, the asset price can either increase or decrease and each outcome is equally likely. In the second period, the variance will depend upon whether the price went up or down. If it went up, then the variancewill be lower so that the binomial branches will be relativelyclose together.If the price went down, the variancewill be higher so that the outcomes will be further apart. After two periods, there are four outcomes that are equally likely. The distributionis quite skewed, since the bad outcome is far worse than if the variancehad been constant. To calculate the VaR in this setting, a simulation is needed. The TARCH model is simulated for ten days using normal random variables and starting from the values of
416
JUNE 2004
Low
variance
High variance
FIGURE 11. TWO-PERIOD BINOMIAL TREE WITH ASYMMETRIC VOLATILITY
November 21.4 This was done 10,000 times and then the worst outcomes were sortedto find the Value at Risk correspondingto the 1-percent quantile. The answer was 7.89 times the standarddeviation. This VaR is substantiallylarger than the value assuming constant volatility. the simuTo avoid the normalityassumption, lation can also be done using the empiricaldistribution of the standardizedresiduals. This each draw simulationis often called a bootstrap; is equallylikely to be any of the randomvariables residuals. The Ocof the standardized observation couldbe drawnonce tober1987 crashobservation or even twice in some simulationsbut not in deviation others.Theresultis a standard multiplier of 8.52 thatshouldbe used to calculateVaR. For our case, the November 24 ten-day 99-percent VaR is 6.5 percentof portfoliovalue. VII. FinancialPractice-Valuing Options Another importantarea of financial practice is valuation and management of derivatives
4 In the example here, the simulationwas startedat the unconditionalvariance so that the time aggregationeffect the meanwas takento be couldbe examinedalone.In addition, zero but this makes little differenceover such shorthorizons.
such as options. These are typically valued theoretically assuming some particularprocess for the underlying asset and then marketprices of the derivatives are used to infer the parameters of the underlyingprocess. This strategyis often free pricing."It is inadequate called "arbitrage for some of the tasks of financial analysis. It cannot determine the risk of a derivative position since each new market price may correspond to a different set of parametersand it is of theseparameter the size andfrequency changes that signify risk. For the same reason, it is difficult to find optimal hedging strategies. Finally, there is no way to determinethe price of a new issue or to determine whether some derivatives are tradingat discounts or premiums. A companion analysis that is frequentlycarried out by derivatives traders is to develop fundamentalpricing models that determinethe price for a derivative based on the appropriate of the underlyingasset. observed characteristics These models could include measuresof trading cost, hedging cost, and risk in managing the options portfolio. In this section, a simple simulation-based option pricingmodel will be employed to illustrate the use of ARCH models in this type of fundamental analysis. The example will be the pricing of put options on the S&P 500 that have ten tradingdays left to maturity. A put option gives the owner the right to sell an asset at a particularprice, called the strike price, at maturity. Thus if the asset price is below the strike,he can make money by selling at the strikeand buying at the marketprice. The profit is the difference between these prices. If, however, the market price is above the strike, then there is no value in the option. If the investorholds the underlyingasset in a portfolio and buys a put option, he is guaranteedto have at least the strikeprice at the maturitydate. This is why these options can be thoughtof as insurance contracts. The simulationworks just as in the previous section. The TARCH model is simulated from the end of the sample period, 10,000 times. The bootstrapapproachis taken so that nonnormality is already incorporatedin the simulation. This simulation should be of the "risk-neutral" in which assets i.e., the distribution distribution, are priced at their discounted expected values. The risk-neutral distribution differs from the
VOL.94 NO. 3 60 5040Fa-
n 1 10,;oQ
u.
417
+ +
aL
0.164 0.160
3020 10+
Q. 0.15&
++
0.152 0.148
+4+
O1 920
o 960
1,000 K
1,040
1,080
n U. I I'l'. -- 1AA 0 920 90 960
1,000 K
1,040
1,080
FIGURE 12. PUT PRICES FROM GARCH
SIMULATION
VOLATILITIES GARCH FROM FIGURE13. IMPLIED SIMULATION
empirical distribution in subtle ways so that there is an explicit risk premiumin the empirical distributionwhich is not needed in the risk neutral. In some models such as the BlackScholes, it is sufficientto adjustthe mean to be the risk-free rate. In the example, we take this route. The distributionis simulatedwith a mean of zero, which is taken to be the risk-free rate. As will be discussed below, this may not be a sufficient adjustment to risk-neutralize the distribution. From the simulation,we have 10,000 equally likely outcomes for ten days in the future. For each of these outcomes we can compute the value of a particular put option. Since these are equally likely and since the riskless rate is taken to be zero, the fair value of the put option is the average of these values. This can be done for put options with different strikes. The result is plotted in Figure 12. The S&P is assumed to begin at 1,000 so a put option with a strike of 990 protects this value for ten days. This put option should sell for $11. To protect the portfolio at its currentvalue would cost $15 and to be certainthat it was at least worth 1,010 would cost $21. The VaR calculated in the previous section was $65 for the ten-day horizon. To protect the portfolio at this point would cost around$2. These put prices have the expected shape; they are monotonically increasing and convex. However, these put prices are clearly differ-
ent from those generatedby the Black-Scholes model. This is easily seen by calculating the implied volatility for each of these put options. The result is shown in Figure 13. The implied volatilities are higher for the out-of-the-money puts thanthey are for the at-the-moneyputs;and the in-the-moneyput volatilities are even lower. If the put prices were generatedby the BlackScholes model, these implied volatilities would all be the same. This plot of implied volatilities against strike is a familiar feature for options traders.The downwardslope is called a "volatility skew" and correspondsto a skewed distribution of the underlyingassets. This feature is very pronouncedfor index options, less so for individual equity options, and virtually nonexistent for currencies, where it is called a that this is a consequence "smile."It is apparent of the asymmetric volatility model and correspondingly,the asymmetryis not found for currencies and is weaker for individual equity options than for indices. This feature of options prices is strong confirmationof asymmetricvolatility models. Unfortunately,the story is more complicated than this. The actualoptions skew is generally somewhat steeper than that generatedby asymmetric ARCH models. This calls into question the risk neutralizationadopted in the simulation. There is now increasing evidence that investors are
418
JUNE 2004
particularlyworried about big losses and are willing to pay extra premiums to avoid them. This makes the skew even steeper.The required risk neutralizationhas been studied by several authors such as Jens C. Jackwerth (2000), Joshua V. Rosenberg and Engle (2002), and David S. Bates (2003). VIII. New Frontiers It has now been more than 20 years since the ARCH paper appeared.The developments and applications have been fantastic and well beyond anyone's most optimistic forecasts. But what can we expect in the future?What are the next frontiers? There appearto be two importantfrontiersof researchthat are receiving a great deal of attention and have importantpromise for applications. These are high-frequency volatility models and high-dimension multivariatemodels. I will give a shortdescriptionof some of the promising developments in these areas. Merton was perhapsthe first to point out the benefits of high-frequency data for volatility measurement. By examining the behavior of stock prices on a finer and finer time scale, better and better measures of volatility can be achieved. This is particularly convenient if volatility is only slowly changing so that dynamic considerations can be ignored. Andersen and Bollerslev (1998a) pointed out that intra-daily data could be used to measurethe performance of daily volatility models. Andersen et al. (2003) and Engle (2002b) suggest how intradaily data can be used to form better daily volatility forecasts. However, the most interesting question is how to use high-frequencydata to form highfrequency volatility forecasts. As higher and higher frequencyobservationsare used, there is apparentlya limit where every transaction is observedand used. Engle (2000) calls such data ultra high frequency data. These transactions occur at irregularintervals ratherthan equally spaced times. In principle, one can design a volatility estimatorthat would update the volatility every time a tradewas recorded.However, even the absence of a tradecould be information useful for updatingthe volatility so even more frequentupdatingcould be done. Since the time
at which trades arrive is random, the formulation of ultra high frequency volatility models requiresa model of the arrivalprocess of trades. Engle and Jeffrey R. Russell (1998) proposethe Autoregressive Conditional Duration or ACD model for this task. It is a close relative of ARCH models designed to detect clustering of trades or other economic events; it uses this informationto forecastthe arrivalprobabilityof the next event. Many investigators in empirical market miare now studying aspects of financrostructure cial marketsthat are relevantto this problem.It turns out that when trades are clustered, the volatility is higher. Tradesthemselves carryinformation that will move prices. A large or medium-size buyer will raise prices, at least partly because market participantsbelieve he could have importantinformationthat the stock is undervalued.This effect is called price impact and is a centralcomponentof liquidityrisk, and a key feature of volatility for ultra high frequency data. It is also a central concern for traderswho do not want to tradewhen they will have a big impact on prices, particularlyif this is just a temporaryimpact. As financialmarkets become ever more computer driven, the speed and frequencyof tradingwill increase.Methods to use this informationto better understandthe volatility and stability of these marketswill be ever more important. The other frontierthat I believe will see substantial development and application is highdimension systems. In this presentation,I have focused on the volatility of a single asset. For most financialapplications,there are thousands of assets. Not only do we need models of their volatilities but also of their correlations.Ever since the original ARCH model was published there have been many approachesproposed for multivariatesystems. However, the best method to do this has not yet been discovered. As the number of assets increase, the models become extremely difficult to accurately specify and estimate. Essentially there are too many possibilities. There are few published examples of models with more than five assets. The most successful model for these cases is the constant conditional correlationmodel, CCC, of Bollerslev (1990). This estimatorachieves its performance by assuming that the conditional correlationsare constant. This allows the vari-
VOL.94 NO. 3
419
ances and covariances to change but not the correlations. A generalizationof this approachis the Dynamic ConditionalCorrelation,DCC, model of Engle (2002a). This model introduces a small numberof parameters to model the correlations, of the number of assets. The volatilregardless ities are modeled with univariatespecifications. In this way, large covariance matrices can be forecast.The investigatorfirstestimatesthe volatilities one at a time, and then estimates the correlationsjointly with a small numberof additional parameters. Preliminary research on this class of models is promising.Systems of up to 100 assets have been modeled with good results. Applications to risk management and asset allocation follow immediately. Many researchersare alreadydeveloping relatedmodels that could have even better performance.It is safe to predictthat in the next several years, we will have a set of useful methods for modeling the volatilities and correlationsof large systems of assets. REFERENCES Andersen,Torben G. and Bollerslev,Tim. "Answering the Skeptics: Yes, StandardVolatility Models Do Provide Accurate Forecasts." International Economic Review, November 1998a, 39(4), pp. 885-905. . "DeutscheMark-DollarVolatility: Intraday Activity Patterns, Macroeconomic Announcements,and Longer Run Dependencies." Journal of Finance, February 1998b, 53(1), pp. 219-65.
Andersen, Torben G.; Bollerslev, Tim; Diebold, Francis X. and Labys, Paul. "Modeling and
. "Modelling the Coherence in ShortRun Nominal Exchange Rates: A Multivariate Generalized Arch Model." Review of Economics and Statistics, August 1990, 72(3), pp. 498-505.
Bollerslev, Tim; Chou, Ray-Yeutien (Ray) and Kroner, Kenneth F. "Arch Modeling in Fi-
nance: A Review of the Theory and Empirical Evidence." Journal of Econometrics, April-May 1992, 52(1-2), pp. 5-59.
Bollerslev, Tim; Engle, Robert F. and Nelson,
Daniel. "Arch Models," in Robert F. Engle and Daniel L. McFadden,eds., Handbookof econometrics, Vol. 4. Amsterdam: NorthHolland, 1994, pp. 2959-3038.
Bollerslev, Tim and Rossi, Peter E. "Dan Nelson
Journal of Business and EcoRemembered." nomic Statistics, October 1995, 13(4), pp. 361-64.
Burns, Pat; Engle, Robert F. and Mezrich, Jo-
and Volatilities of Asynseph. "Correlations chronous Data." Journal of Derivatives, Summer 1998, 5(4), pp. 7-18.
Chou, Ray-Yeutien (Ray); Engle, Robert F. and
Forecasting Realized Volatility." Econometrica, March 2003, 71(2), pp. 579-625. Bates, David S. "EmpiricalOption Pricing: A Retrospection." Journal of Econometrics, September-October 2003, 116(1-2), pp. 387-404.
Black, Fisher and Scholes, Myron. "The Valua-
tion of OptionContractsand a Test of Market Efficiency." Journal of Finance, May 1972, 27(2), pp. 399-417. Bollerslev, Tim. "Generalized Autoregressive Conditional Heteroskedasticity."Journal of Econometrics,April 1986, 31(3), pp. 307-27.
Kane, Alex. "MeasuringRisk-Aversion from Excess Returnson a Stock Index."Journal of Econometrics,April-May 1992, 52(1-2), pp. 201-24. StochasticProClark,PeterK. "A Subordinated cess Model with Finite Variance for Speculative Prices." Econometrica, January 1973, 41(1), pp. 135-56. Engle, Robert F. "Autoregressive Conditional Heteroskedasticity with Estimates of the Variance of U.K. Inflation."Econometrica, July 1982, 50(4), pp. 987-1008. . "Estimates of the Variance of U.S. InflationBased Upon the Arch Model."Journal of Money, Credit, and Banking, August 1983, 15(3), pp. 286-301. . "The Econometrics of Ultra-HighData."Econometrica, 2000, January Frequency 68(1), pp. 1-22. . "Dynamic Conditional Correlation: A Simple Class of Multivariate GeneralHeteroskedasConditional ized Autoregressive Business and Models." Journal of ticity Economic Statistics, July 2002a, 20(3), pp. 339-50. . "New Frontiers for Arch Models." Journal of Applied Econometrics, September-October 2002b, 17(2), pp. 425-46.
420
JUNE 2004
Engle, Robert F. and Ishida, Isao. "Forecasting Variance of Variance:The Square-Root,the Affine, and the Cev Garch Models." Department of Finance working papers, New York University, 2002. Engle, Robert F.; Ito, Takatoshiand Lin, WenLing. "Meteor-Showers or Heat WavesHeteroskedastic Intradaily Volatility in the Foreign-Exchange Market." Econometrica, May 1990a, 58(3), pp. 525-42. Engle, Robert F.; Lilien, David M. and Robins, Russel P. "Estimating Time-Varying Risk Premia in the Term Structure:The Arch-M Model." Econometrica, March 1987, 55(2), pp. 391-407. Engle, Robert F.; Ng, Victor K. and Rothschild, Michael. "Asset Pricing with a Factor-Arch Covariance Structure: Empirical Estimates for TreasuryBills." Journal of Econometrics, July-August 1990b, 45(1-2), pp. 213-37. Engle,RobertF. and Russell,JeffreyR. "Autoregressive ConditionalDuration:A New Model for Irregularly Spaced Transaction Data." Econometrica, September 1998, 66(5), pp. 1127-62. French, Kenneth R.; Schwert, G. William and Stambaugh,Robert F. "Expected Stock Returns and Volatility." Journal of Financial Economics, September1987, 19(1), pp. 3-29. Friedman,Milton."Nobel Lecture:Inflationand Unemployment."Journal of Political Economy, June 1977, 85(3), pp. 451-72. Glosten, Lawrence R.; Jagannathan,Ravi and Runkle, David E. "On the Relation between the Expected Value and the Volatility of the Nominal Excess Return on Stocks." Journal of Finance, December 1993, 48(5), pp. 1779-801. Hamao,Yasushi;Masulis,Ron W. and Ng, Victor K. "Correlations in Price Changesand Volatilacross International StockMarkets." Review ity of Financial Studies, Summer 1990, 3(2), pp. 281-307.
Harvey,AndrewC.; Ruiz, Esterh and Shephard, Stochastic VarianceModNeil. "Multivariate els." Review of Economic Studies, April 1994, 61(2), pp. 247-64. Jackwerth,Jens C. "RecoveringRisk Aversion from Option Prices and Realized Returns." Review of Financial Studies, Summer 2000, 13(2), pp. 433-51. Markowitz, Harry M. "Portfolio Selection." Journal of Finance, March 1952, 7(1), pp. 77-91. Merton,RobertC. "Theoryof Rational Options Pricing." Bell Journal of Economics and ManagementScience, Spring 1973, 4(1), pp. 141-83. . "On Estimating the Expected Return on the Market: An Exploratory Investigation." Journal of Financial Economics, December 1980, 8(4), pp. 323-61. Nelson,DanielB. "ConditionalHeteroskedasticity in Asset Returns: A New Approach." Econometrica, March 1991, 59(2), pp. 34770. Rosenberg, Joshua V. and Engle, Robert F. "EmpiricalPricing Kernels." Journal of Financial Economics, June 2002, 64(3), pp. 341-72. Sharpe,William."CapitalAsset Prices: A Theory of MarketEquilibriumunder Conditions of Risk." Journal of Finance, September 1964, 19(3), pp. 425-42. Taylor, StephenJ. Modeling financial time series. New York: John Wiley, 1986. . "Modeling Stochastic Volatility: A Review and ComparativeStudy."Mathematical Finance, April 1994, 4(2), pp. 183-204. Tobin,James."LiquidityPreferenceas Behavior TowardsRisk."Review of Economic Studies, February1958, 25(2), pp. 65-86. Zakoian, Jean Michael. "ThresholdHeteroskedastic Models." Journal of Economic Dynamics and Control, September 1994, 18(5), pp. 931-55.

Engle (2004)

Uploaded by

Copyright:

Available Formats

Engle (2004)

Uploaded by

Document Information

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

Engle (2004)

Uploaded by

Copyright:

Available Formats

American Economic Association

Risk and Volatility: Econometric Models and FinancialPracticet

FIGURE 1. S&P 500 DAILY PRICES AND RETURNS FROM

2. S&P 500 DAILY BEFORE1987 FIGURE

2003 1963 TO NOVEMBER JANUARY

3. S&P 500 1988 TO 2000 FIGURE

o 1998 1999 -2000 2001 2002 2003 - SPRETURNS

4. S&P 500 1998 TO2003 FIGURE

412 0.25 0.2 0.15 0.1

mSP Returns SQ Returns

FIGURE 7. HISTORICAL VOLATILITIES WITH VARIOUS WINDOWS

1.4 1.2 1.0 0.80.60.10 ,,

THREE INTERVALS: FIGURE 9. GARCH CONFIDENCE DEVIATIONS STANDARD

FIGURE 8. GARCH VOLATILITIES

FIGURE10. IMPLIED VOLATILITIES AND GARCH VOLATILITIES

VOL.94 NO. 3 60 5040Fa-

n U. I I'l'. -- 1AA 0 920 90 960

FIGURE 12. PUT PRICES FROM GARCH

VOLATILITIES GARCH FROM FIGURE13. IMPLIED SIMULATION

You might also like