Peer Reviewed

Jonathan M. Godbey [email protected] is an Assistant Professor, Department

of Finance and Business Law at James Madison University and James W.
Mahar [email protected] is an Assistant Professor, Department of Finance at St.
Bonaventure University.

Abstract
Assuming use of the correct option pricing model and an efficient market,
an option’s implied volatility is the market’s consensus forecast of future realized
volatility over the remaining life of that option. In this paper the authors examine
460 of the S&P 500 firms to demonstrate that: (1) implied volatility is a better
forecaster of realized volatility than historic volatility or GARCH models and (2)
the information content of implied volatility significantly decreases with liquidity.
Since individual equity options are American style, implied volatility estimates
were obtained from calls and puts separately, rather than only from calls or
pooled data.

1
 Introduction
An accurate forecast of unobservable volatility is a necessary component of
virtually all option pricing methodologies. Therefore, rational option market
participants will seek the best possible forecast of future realized volatility (RV)
over the life of the option. They will estimate RV from both public and private
information. Market prices are then set according to an option pricing model, the
observed parameters of that model and a volatility estimate that reflects the
aggregated forecast of market participants. By setting the result of a particular
option pricing model equal to the market price the implied volatility (IV) of that
model is obtained. Merton (1973) shows that if there is efficiency in the options
market and participants use the correct model, then IV should be the best
estimate of RV.

While the widely accepted hypothesis that IV is the market’s best forecast of
RV is theoretically appealing, empirical testing has proven to be difficult. The joint
hypothesis of market efficiency and a correct option-pricing model along with
other complexities prevent empirical tests of the information content of IV from
being conclusive. Nevertheless, the empirical tests are not without merit. With the
unique circumstances of each study in mind, one can develop a better
understanding of the information content of IV as well as that of historic volatility
(HV) and GARCH forecasts (GAR).

Previous research into the predictive ability of IV for individual stocks used
much smaller samples. Lamourex and Lastrapes (1993) examine 10 firms from
April 1982 to March 1984. They extract an IV from the Hull and White (1987)
model and show that it is biased but a better predictor of RV than HV or GARCH.
They further find that HV improves the forecast of IV alone. Mayhew and Stivers
(2003) examine 50 firms from 1988 to 1995. IV is obtained using a binomial tree
procedure described by Whaley (1993). They find that IV “reliably outperforms
GARCH and subsumes all information in return shocks beyond the first lag.”
They show that for lower volume options IV loses some of its predictive power
and may be inferior to approaches relying only on past data.

We use IV estimates from 460 individual stocks from the S&P 500 from
October 1, 2001 to September 13, 2002 to test the ability of IV to predict RV. We
then compare the predictive power of IV to that of HV and GAR. Because
previous research suggests that volume is an important factor in the information
content of IV, we rank firms on their option volumes and test the relative
predictive power of IV, HV and GAR across volume quintiles. Unlike previous
research, the size of our sample is large enough to allow us to conclude that the
results are not limited to selected stocks.

2
 We find that IV, HV and GAR all provide useful information in forecasting
realized volatility. However, the information content of the variables is not the
same. Both alone and in more inclusive models, IV has more predictive ability
than HV or GAR. Moreover, when all variables are examined simultaneously,
only IV is significant for all quintiles. These findings are consistent with the
hypothesized view that IV has higher information content than HV or GAR.
Equally interestingly, we show that while implied volatility is a significant predictor
of realized volatility, its predictive ability increases with option market volume.
This last finding may have important implications for options that trade
infrequently.

Our work is unique in that we look at implied volatility from calls and puts
separately while previous research relied solely on calls or an index of both.
Because individual equity options are American options, the implied volatility
based on puts and calls may be different. By keeping the data separate, we are
able to show that the implied volatility estimate from each type of option has the
similar amount of information and this information content is positively related to
the liquidity of the options.

In the next section of this paper we provide a review of relevant literature.

Following it is a section which describes the data and tests the methodology. It is
followed by a section which contains the empirical findings and a section which
concludes the paper.

Literature Review
Poon and Granger (2003) provide an extensive review of the literature related
to forecasting volatility. They divide the existing research into two general
categories: (1) papers using historical data only and (2) papers using IV alone or
in addition to historical data. In general, the latter studies have found that IV
contains a significant amount of information and that it is often superior to models
that rely on historical information alone.

Since it is reasonable to assume that different markets have differing degrees

of efficiency, the forecasting power of IV for one asset class does not necessarily
mean that IV will have equivalent capabilities in another. While the testing
methodologies may be similar, the results of the IV tests should be considered
according to asset class. Following Poon and Granger (2003) these classes are:
individual stocks, market indices, currencies, and other assets.

Latane and Rendleman (1976) use weekly data from options on 24 individual
stocks from October 1973 to June 1974 to find that IV provides a more accurate
forecast than historic volatility (HV). Chiras and Manaster (1978), Schmalensee
and Trippi (1978), and Beckers (1981) support Latane and Rendleman’s findings
for individual stocks. All of these studies use the Black and Scholes (1973) model
to find IV. These studies are not conclusive for several reasons: the options

3
market underwent a structural change after the crash of October 1987 (see
Rubinstein (1994)); they assume constant volatility; they do not adjust for
dividends, or they do not adjust for the possibility of early exercise; and with the
exception of Beckers (1981), these studies did not control for the
nonsynchronous data.

Lamoureux and Lastrapes (1993) and Mayhew and Stivers (2003) are the
only major studies that we are aware of to examine IV’s predictive power for
individual stocks when compared to conditional heteroskedasticity models. Both
studies find that IV is a better predictor of RV than HV or GAR. Mayhew and
Stivers (2003) provide the strongest support for IV. They show that IV “captures
most or all of the relevant information in past return shocks, at least for stocks
with actively traded options.” They further show that the predictive power of IV
deteriorates with option volume.

Donaldson and Kamstra (2005) compare the predictive ability of implied

volatility extracted from call options on the S&P 500 to ARCH models. They find
that for high volume periods IV is more informative than ARCH models, but that
ARCH models are more informative in low volume periods. Regardless of option
volume, they find that IV and ARCH models provide additional information.

In this paper we examine the information content of IV, GARCH based forecasts,
and HV for individual stocks. Our sample is significantly larger then previous studies.
We examine approximately 460 firms from the S&P 500. Additionally, we examine
the differences in the information content of implied volatilities across both puts and
calls. To the best of our knowledge, this has not been done in the academic
literature. It is important not only because we are dealing with American options, but
also due to the previous findings of Mayhew and Stivers (2003) and Donaldson and
Kamstra (2005) that report that liquidity is an important determinant in the
information content of the option. Given that there may be liquidity differences
across markets, puts are analyzed separately.

Data and Methodology

Daily IV for both calls and puts from October 1, 2001 to September 13, 2002
are obtained from iVolatility.com for 460 of the 500 firms in the S&P 500 as of
October 1, 2001. IV is computed using the Black-Scholes model and four at-the-
money options of differing expirations. The results are then normalized to give
estimates of IV for an option with 30 calendar days to maturity. Adjustments are
made for dividends and “a proprietary weighting technique factoring the delta and
vega of each option participating” in the calculations.

Dividend adjusted stock prices are obtained from by CRSP. To approximate 30

calendar days, prices for 22 consecutive trading days are used to find daily RV and
HV. The calculation for annualized volatility of daily returns is as follows:

4
 [
σ = 252 ∑ (rt − r ) / ( N − 1)
2
](
1/ 2)
(1)

where rt = ln(St/St-1) and r is the mean return. St is the dividend adjusted stock
price on day t. Consistent with iVolatility.com’s methodology an average of 252
trading days per year is assumed.

GAR forecasts were obtained from a GARCH(1,1) model as developed by

Bollerslev (1986). The equation for GARCH(1,1) is:

σ n2 = γVL + αu n2−1 + βσ n2−1 (2)

where VL is the long-run average variance rate, ut = ln(St/St-1), σn = the estimate

of volatility for day n and γ + α + β = 1.0. The maximum likelihood method was
used to estimate γ, α, and β by using all available data in the estimation period.
The GAR forecast is given by:

[ ] (
E σ n2+ k = V L + (α + β ) k σ n2 − V L ) (3)

Since the parameters are estimated using the entire data set (as in Jorion
(1995) and Szakmary et. al. (2003)) the GAR forecasts benefit from information
that would not have been available. Hence, our GAR forecasts have a slight
advantage over those employed in practice.

We test the hypothesis that (1) IV is an unbiased and efficient estimator of

RV, (2) HV is an unbiased and efficient estimator of RV, (3) GAR is an unbiased
and efficient estimator of RV, and (4) IV contains all information present in HV
and GAR. Our tests follow the methodology of Canina and Figlewski (1993) and
Szakmary, Ors, Kim and Davidson (2002). Specifically, we estimate the following
regressions:

RVt = α1+ β1*IVt + εt (4)

RVt = α2+ β2*HVt + εt (5)

RVt = α3+ β3*GARt + εt (6)

RVt = α+ β1*IVt + β2*HVt + εt (7)

RVt = α+ β1*IVt + β3*GARt + εt (8)

RVt = α+ β1*IVt + β2*HVt + β3*GARt + εt (9)

If IV is the unbiased, efficient estimator that theory predicts then α1 and β1 in

equation (4) will be zero and one, respectively. If HV is the unbiased, efficient
estimator that theory predicts then α2 and β2 in equation (5) will be zero and

5
one, respectively. If GAR is the unbiased, efficient estimator that theory predicts
then α3 and β3 in equation (6) will be zero and one, respectively. If IV contains
all information present in HV and GAR, then the coefficients of HV and GAR
should be zero in equations (7) - (9), while β1 should be significantly different
from zero.

To examine the effects of liquidity on the performance of IV, HV and GAR, the
stocks are ranked by option volume and grouped into quintiles. This was done
separately for call option volume and put option volume. Greater liquidity may be
important for a two different reasons. First, it suggests a more cognitively diverse
group of investors for a particular stock option. Each investor will bring private
information and error into her estimate of the option’s price and the resulting IV.
Assuming uncorrelated errors, a greater number of investors suggest more
information and a better forecast of RV. Second, institutional investors are able to
trade more readily in liquid stocks. To the degree that institutional investors may
be more rational or have better information than individual investors, it can be
expected that IV is a better forecaster in liquid markets.

Unit Root tests (the augmented Dickey-Fuller method) were performed to test
the IV time series for stationarity. In all cases the null hypothesis of a unit root
could be rejected. Therefore, the sample is assumed to be stationary over the
time period studied which allows the use OLS regression for our analysis.

Empirical Findings
We find that (1) IV is a better predictor of RV than HV or GAR, (2) the
information content of IV decreases with liquidity of the options, (3) HV and GAR
contain significant information about RV not included in IV, (4) neither IV, HV nor
GAR appear to be unbiased and efficient estimators of RV, and (5) IV derived
from calls and puts produce near identical results. For ease of discussion, we
focus on results from call IV. Corresponding and near identical results for put IV
are reported in the associated tables.

Tables 1 – 3 (below) summarize the results from the regressions in

equations (4) – (6). Regressions are run for each individual stock. The results
are then grouped by option volume quintile. Median values for the regression
coefficients are reported along with the percentage of stocks where each
coefficient is statistically significant at the 5 percent level.

6
 Table 1
RVt = α1 + β1*IVt + εt
Panel A: Call Option Data
Volume α1 t-value β1 t-value r2 α1% β1%
1 high 0.102 2.617 0.743 8.180 0.245 0.674 0.946
2 0.107 3.009 0.647 7.713 0.220 0.723 0.936
3 0.122 3.625 0.607 6.735 0.178 0.722 0.867
4 0.095 3.176 0.625 5.952 0.145 0.659 0.868
5 low 0.165 4.320 0.426 4.700 0.095 0.837 0.739
All Firms 0.120 3.455 0.614 6.689 0.176 0.723 0.871
Panel B: Put Option Data
Volume α1 t-value β1 t-value r2 α1% β1%
1 high 0.094 2.812 0.747 8.574 0.259 0.546 0.938
2 0.084 2.553 0.663 7.681 0.218 0.593 0.934
3 0.127 3.550 0.596 7.034 0.189 0.652 0.876
4 0.134 3.687 0.548 5.735 0.133 0.692 0.846
5 low 0.141 4.190 0.482 4.711 0.098 0.753 0.774
All Firms 0.115 3.336 0.617 6.721 0.176 0.723 0.875

RV is regressed on IV for each individual firm. The firms are then grouped into
quintiles based on the volume of the options used to calculate IV. The median
values of the regression results for each quintile are reported above. The
percentage of firms with α1 and β1 significant at the 5 percent level are given in
columns 7 and 8, respectively.

Table 2
RVt = α2 + β2*HVt + εt
Panel A: Call Option Data
Volume α2 t-value β2 t-value r2 α2% β2%
1 high 0.309 11.114 0.237 3.430 0.066 1.000 0.694
2 0.303 11.174 0.209 3.049 0.052 1.000 0.618
3 0.295 11.380 0.158 2.274 0.047 0.989 0.553
4 0.241 10.696 0.249 3.747 0.059 1.000 0.650
5 low 0.241 10.354 0.247 3.861 0.069 1.000 0.726
All Firms 0.275 11.004 0.231 3.369 0.057 1.000 0.647
Panel B: Put Option Data
Volume α2 t-value β2 t-value r2 α2% β2%
1 high 0.300 11.020 0.230 3.340 0.059 1.000 0.660
2 0.285 11.090 0.218 3.211 0.064 0.989 0.626
3 0.314 11.353 0.172 2.847 0.047 0.989 0.573

7
 4 0.247 10.801 0.223 3.319 0.055 1.000 0.659
5 low 0.226 10.394 0.248 3.675 0.064 1.000 0.720
All Firms 0.275 10.975 0.231 3.353 0.057 0.999 0.649

RV is regressed on HV for each individual firm. Consistent with Table 1, the

firms are then grouped into quintiles based on the volume of the options used to
calculate IV. The median values of the regression results for each quintile are
reported above. The percentage of firms with α2 and β2 significant at the 5
percent level are given in columns 7 and 8, respectively.

Table 3
RVt = α3 + β3*GAR+ εt
Panel A: Call Option Data
Volume α3 t-value β3 t-value r2 α3% β3%
1 high 0.017 0.207 0.833 3.330 0.052 0.587 0.701
2 -0.006 0.186 0.923 3.247 0.057 0.511 0.630
3 0.010 0.231 0.828 2.838 0.037 0.511 0.620
4 -0.053 -0.708 1.162 3.547 0.059 0.554 0.761
5 low 0.030 -0.455 1.001 3.615 0.057 0.587 0.728
All -0.028 -0.139 0.967 3.289 0.055 0.546 0.667
Firms
Panel B: Put Option Data
Volume α3 t-value β3 t-value r2 α3% β3%
1 high 0.024 0.309 0.828 3.248 0.052 0.592 0.684
2 -0.009 -0.089 0.927 3.428 0.100 0.516 0.630
3 0.010 0.203 0.827 2.957 0.071 0.517 0.629
4 -0.053 -0.745 1.160 3.547 0.086 0.560 0.747
5 low -0.029 -0.452 0.986 3.581 0.056 0.598 0.728
All -0.017 -0.138 0.940 3.263 0.055 0.558 0.685
Firms

RV is regressed on GAR for each individual firm. Consistent with Table 1, the
firms are then grouped into quintiles based on the volume of the options used to
calculate IV. The median values of the regression results for each quintile are
reported above. The percentage of firms with α3 and β3 significant at the 5
percent level are given in columns 7 and 8, respectively.

If IV is an efficient and unbiased estimator of RV then α1 and β1 in

equation (2) will be zero and one, respectively. As Table 1 indicates, this is not
the case. The median α1 is 0.120, while the median β1 is 0.614. For 72.3 percent
of the firms α1 is significantly different from zero, while β1 is statistically
significant in nearly 90 percent of the stocks. Panel B of Table 1 shows that the

8
information content of IV is essentially the same for both puts and calls. In each
case the median r2 is 0.176.

Our next finding is that HV is not an efficient and unbiased estimator of

RV. As shown in Table 2, the median α2 is 0.275 while the median β2 is 0.231.
We find α2 is different from zero for all firms, while β2 is statistically significant for
approximately 65 percent of the stocks. HV explains only 5.7 percent of variation.

Results for GAR regression suggest that GAR is less biased than our
other variables as estimator. Table 3 shows that the median α3 is -0.028 and is
significantly different from zero for only 55 percent of firms. A median β3 of 0.451
suggests that it is not an efficient estimator and is statistically significant for
approximately 68 percent of stocks. Similar to HV, GAR explains only 5.5 percent
of variation.

As is mentioned above, we separate our sample into quintiles based on

the volume of option trading to examine the importance of volume on information
content. As Figure 1 (below) shows, the predictive ability (the information
content) of IV is positively related to option volume. However, we do not find such
a relationship for HV or GAR across volume quintiles. The median r2 of the IV
regression falls with every quintile as do the percentage of firms where β1 is
significant. As shown in Tables 2 and 3, liquidity has no discernable effect on the
information content for HV or GAR.

Figure I

Explained variation (R^2)

0.3

0.25

0.2
R-Squared

0.15

0.1

0.05

0
1 2 3 4 5
Volume Quintile

This relation between IV and liquidity is consistent with the findings of

Mayhew and Stivers (2003), who report a degradation of IV’s information content
based on firm size within the 50 stocks that they examined. However, it is

9
important to note that unlike Mayhew and Stivers, we use actual volume
numbers, rather than size as a proxy for market liquidity. That information content
drops with volume is also consistent with research by Donaldson and Kamstra
(2005), who report that the information content of index options is reduced in
times of low volume.

Tables 4 and 5 (below) report the results from the multiple regressions in
equations (7) and (8). IV is shown to subsume some of the information content of
HV and GAR. However, as volume decreases the contribution of HV and GAR
increases. Again, this result is consistent with the findings of Donaldson and
Kamstra (2005). As Table 4 (below) shows, when RV is regressed on IV and HV,
β2 is significant for 21.4 percent of the firms, as opposed to 64.7 percent when IV
is not included. For quintile 1, β2 is significant for only 13.0 percent of the firms.
This percentage increases steadily across quintiles to 34.8 percent for quintile 5.
Table 5 (below) shows similar results for β3. For quintile 1, β3 is significant for
only 13.7 percent of the firms.This percentage increases to 38.3 percent for
quintile 5. Consistent with Table 1, Table 4 and Table 5 show that the percentage
of firms with β1 significant decreases across volume quintiles.

Table 4
RVt = α + β1*IVt + β2*HVt + εt
Panel A: Call Option Data
Volume α t-value β1 t-value β2 t-value r2 α% β1% β2%
1 high 0.099 2.855 0.824 7.653 -0.134 -1.878 0.317 0.674 0.924 0.130
2 0.097 3.065 0.877 7.885 -0.217 -3.094 0.305 0.713 0.883 0.128
3 0.129 3.516 0.713 6.022 -0.157 -2.157 0.243 0.722 0.800 0.178
4 0.084 2.831 0.661 4.606 -0.009 -0.115 0.199 0.637 0.780 0.286
5 low 0.164 4.620 0.460 2.859 0.042 0.460 0.157 0.815 0.598 0.348
All Firms 0.128 3.350 0.740 6.007 -0.112 -1.482 0.240 0.712 0.797 0.214
Panel B: Put Option Data
Volume α t-value β1 t-value β2 t-value r2 α% β1% β2%
1 high 0.088 2.814 0.855 8.155 -0.174 -2.223 0.336 0.688 0.896 0.149
2 0.069 2.164 0.891 7.400 -0.211 -2.252 0.308 0.637 0.902 0.154
3 0.126 3.468 0.722 6.535 -0.225 -2.983 0.235 0.719 0.820 0.169
4 0.135 3.491 0.615 4.562 -0.022 -0.202 0.178 0.747 0.725 0.253
5 low 0.161 4.226 0.461 2.989 0.084 1.026 0.172 0.787 0.606 0.383
All Firms 0.119 3.157 0.734 6.113 -0.130 -1.722 0.240 0.716 0.790 0.221

RV is regressed on IV and HV for each individual firm. Consistent with Table 1,

the firms are then grouped into quintiles based on the volume of the options used
to calculate IV. The median values of the regression results for each quintile are
reported above. The percentage of firms with α, β1 and β2 significant at the 5
percent level are given in columns 9-11.

10
11
 Table 5
RVt = α + β1*IVt + β3*GARt + εt
Panel A: Call Option Data
Volume α t-value β1 t-value β3 t-value r2 α% β1% β3%
1 high 0.194 1.774 0.828 7.571 -0.436 -1.763 0.316 0.674 0.916 0.137
2 0.240 1.988 0.765 6.469 -0.403 -1.519 0.267 0.634 0.903 0.129
3 0.225 1.847 0.677 5.985 -0.353 -1.360 0.211 0.544 0.800 0.200
4 0.073 0.890 0.518 4.284 0.150 0.610 0.176 0.438 0.798 0.200
5 low 0.062 0.977 0.431 3.528 0.227 0.905 0.156 0.500 0.564 0.383
All 0.146 1.445 0.665 5.705 -0.233 -0.663 0.214 0.559 0.796 0.228
Firms
Panel B: Put Option Data
Volume α t-value β1 t-value β3 t-value r2 α% β1% β3%
1 high 0.210 1.972 0.817 7.705 -0.483 -1.825 0.301 0.327 0.908 0.173
2 0.230 2.026 0.783 6.723 -0.393 -1.556 0.275 0.373 0.923 0.165
3 0.164 1.813 0.650 5.613 -0.375 -1.345 0.192 0.416 0.809 0.169
4 0.068 0.793 0.494 4.167 0.184 0.762 0.179 0.538 0.703 0.307
5 low 0.058 0.867 0.380 2.950 0.265 0.915 0.154 .0489 0.598 0.391
All Firms 0.149 1.530 0.665 5.750 -0.221 -0.743 0.212 0.429 0.788 0.241

RV is regressed on IV and HV for each individual firm. Consistent with Table 1,

the firms are then grouped into quintiles based on the volume of the options used
to calculate IV. The median values of the regression results for each quintile are
reported above. The percentage of firms with α, β1 and β3 significant at the 5
percent level are given in columns 9 - 11.

Table 6 (below) shows for approximately 77 percent of the firms, r2 is

higher for IV as the independent variable than it is for HV as the independent
variable. Likewise, as shown in Table 7 (below), r2 is higher for IV as the
independent variable than it is for GAR as the independent variable in about 78
percent of the overall sample. This result is consistent with the hypothesis that IV
contains more information than just the historical volatility or GARCH models and
is consistent with previous research.

12
 Table 6
Predictive Power of IV versus HV

Volume Calls Puts

1 high 89.13% 88.86%
2 85.87% 85.71%
3 79.35% 79.77%
4 69.57% 72.53%
5 low 61.96% 59.13%
Total 77.17% 77.22%

The number and percentage of firms by option volume quintile (n=92) where IV
offers superior information (as defined as having a higher adjusted r-squared) in
forecasting RV. When sorted by volume, 13 firms were dropped due to a lack of
volume data. The last row includes the dropped firms.

Table 7
Predictive Power of IV versus GAR

Volume Calls Puts

1 high 89.1% 93.5%
2 89.1% 91.3%
3 80.4% 79.3%
4 76.1% 76.1%
5 low 58.7% 60.9%
Total 78.5% 80.2%

The number and percentage of firms by option volume quintile (n=92), where IV
offers superior information (as defined as having a higher adjusted r-squared) in
forecasting RV. When sorted by volume, 13 firms were dropped due to a lack of
volume data. The last row includes the dropped firms.

Conclusion
We use data from 460 of the S&P 500 firms to show that implied volatility
is a better predictor of future realized volatility than is historical volatility or
GARCH models. We further show that this predictive ability of the implied
volatility decreases as option volumes decreases. This is true for both puts and
calls. These results indicate that the market for individual equity options is
efficient to some degree. They also show that the market’s efficiency tends to
decline with volume.

13
 Our results have important implications for event studies which use IV to
examine changes in firm-specific risk. First, consistent with theory, IV does have
significant information content. Therefore, changes in IV signal changes in the
market’s perception of changes in future return volatility. Second, changes in IV
for stocks with active options give a more complete view of the market’s
perception of changes in future return volatility. Therefore, event studies on more
active stocks should produce more robust results than those on illiquid firms.

References
Beckers, S., 1981, Standard deviations implied in option prices as predictors of
future stock price variability, Journal of Banking and Finance 5, 297-308.

Black, F. and M. Scholes, 1973, The pricing of options and corporate liabilities,
Journal of Political Economy, 81: 637-654.

Bollerslev, T., 1986, Generalized Autoregressive Conditional Heteroscedasticity,

Journal of Econometrics, 307-27.

Canina, L. and S. Figlewski, 1993, The informational content of implied volatility,

The Review of Financial Studies 6:3, 659-681.

Chiras, D.P., and S. Manaster, 1978, The information content of option prices
and a test of market efficiency, Journal of Financial Economics 6, 213-234.

Christensen, B.J., C.S. Hansen, 2002, New evidence on the implied realized
volatility relation, European Journal of Finance 7, 187-205.

Donaldson, R.G. and M. Kamstra, 2005, Volatility forecasts, trading volume and
the ARCH vs. option-implied tradeoff, Journal of Financial Research,
forthcoming.

Jorion, P. 1995, Predicting volatility in the foreign exchange market, Journal of

Finance 1, 507-528.

Lamoureux, C. and W. Lastrapes, 1993, Forecasting stock-return variance:

toward an understanding of stochastic implied volatilities, The Review of
Financial Studies 6, 293-326.

Latane, H.A., and R.J. Rendleman, Jr., 1976, Standard deviations of stock
price ratios implied in option prices, Journal of Finance 31, 369-381.

Mayhew, S. and C. Stivers, 2003, Stock return dynamics, option volume, and
the information content of implied volatility, Journal of Futures Markets 23,
615-646.

14
Merton, R., 1973, Theory of rational option pricing, Bell Journal of Economics
and Management Science, 4: 141-183.

Poon, S. and C.W.J. Granger, 2003, Forecasting volatility in financial markets: a

review, Journal of Economic Literature, 478-539.

Rubenstein, M, 1994, Implied binomial trees, Journal of Finance, 771-818.

Schmalensee, R. and R.R. Trippi, 1978, Common stock volatility expectations

implied by option premia, Journal of Finance 33, 129-147.

Szakmary, A., E. Ors, J.K. Kim and W.N. Davidson III, 2003, The predictive
power of implied volatility: evidence from 35 futures markets, Journal of
Banking and Finance 27, 2151-2175.

Whaley, R.E., 1993, Derivatives on market volatility: hedging tools long overdue,
Journal of Derivatives 1, 71-84.

Note: The graphic containing the title of this paper was designed by Carole E.
Scott.

15