Professional Documents
Culture Documents
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 support@jstor.org.
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms
and Wiley are collaborating with JSTOR to digitize, preserve and extend access to The Journal
of Finance
(Ph + PL)/2
* Associate Professor of Finance, Oklahoma State University, and Associate Professor of Account-
ing, The University of Iowa, respectively.
1. Sharpe (13), Treynor (15) and Jensen (8).
2. Treynor et al. (16), and Altman and Schwartz (2).
3. All of the measures have been, or could be, employed to measure total common stock price
volatility. Traditional utility theory does not support usage of certain of the measures tested;
therefore, the association between volatility rankings needs further examination. See Markowitz
(11) for a discussion of the relationship of utility maximization to some of the measures tested.
4. See Pratt (12) for an examination of volatility over time.
119
where Ph is the high stock price for the period and PL is the low price for the
same period.'
B. Variance
n
Z p ( p)2 /n
Z - Pi-P I /n
(|E(di-d)2/n/d-
where di - Pi - Pi-1 for sub-periods i from 1 to n, d is the arithmetic mean
of the di's and n is the number of sub-periods.8
F. Semi-Variance
n
(p p)2/n
5. This measure has been employed by Clendenin (3) and Heins and Allison (7). Various modifi-
cations of this measure have been employed by Fisher (5), Eisenstadt (4), Alexander (1), and
Gould (6).
6. This measure has been recommended by a Bank Administration Institute study on meas
pension fund investment performance and was employed by Jensen (8, p. 400).
7. This measure has been employed by Levy (10, p. 599).
8. Levy (10, p. 610).
9. Markowitz (11, pp. 188-201).
\|E(R;)2/n
Z (Ln(l'+ Ri))2/n
Each volatility measure was applied to each firm's set of prices for every
time period. To test the degree of association of rankings among measures, the
Kendall coefficient of concordance'3 was computed for each period. The null
hypothesis of no community of order among the measures was rejected at
the .05 significance level for each time period (Table 1). This result could
have arisen because some individual measures have characteristics which are
quite similar to some of the other measures, and thus a high rank correlation
exists between these pairs.
TABLE 1
COEFFICIENT OF CONCORDANCE AMONG ALL EIGHT MEASURES FOR EACH TWENTY-SIX
WEEK PERIOD
Coefficient
of concordance Test statistic
2
X95 (49) = 66.3.
TABLE 2
NUMBER OF PERIODS IN WHICH SPEARMAN RANK CORRELATION COEFFICIENT WAS
POSITIVE AND SIGNIFICANT AT THE .05 LEVEL
B. C. D. E. F. G. H.
Modified
Quadratic
CofV Mean
Var MAD CofV Pr C S-v MQM (logarithms)
A. Range+. midrange 9 9 10 7 9 9 9
B. Variance 10 9 5 1 9 9
C. Mean absolute 9 4 1 9 9
deviation
D. Coefficient of 6 9 9 9
variation
E. Coefficient of 1 9 9
variation of
price changes
F. Semi-variance 10 9
G. Modified quadratic 10
mean
Note that the simplest measure, the range divided by the mid-range, has
correlation with other measures which is positive and significant for 62 com-
parisons of a possible 70 over the five-year period. It also has positive and
significant correlation with the two somewhat erratic measures, the semi-
variance and the coefficient of variation of price changes.15
Substantial agreement in every time period exists between four sets of
measures. These are: the range divided by the mid-range and the coefficient
of variation (each measuring relative dispersion); the variance and the mean
absolute deviation (each measuring absolute dispersion); the semi-variance
and the modified quadratic mean (each giving weight to downward dispersion
only); and the two modified quadratic mean measures suggested by Levy
(each measuring downward changes in sub-period rates of return).
If only the range divided by the mid-range, the variance, the coefficient
of variation of price changes, and the semi-variance are included, the coeffi-
cient of concordance is significant (that is, the hypothesis of no community of
order is rejected) for only six of the ten periods (Table 3). Thus, there is a
TABLE 3
COEFFICIENT OF CONCORDANCE AMONG THE RANGE DIVED BY THE MID-RANGE, THE
VARIANCE, THE COEFFICIENT OF VARIATION OF PRICE CHANGES AND THE
SEMI-VARIANCE FOR EACH TWENTY-SIX WEEK PERIOD
Coefficient
of concordance Test statistic
.95(49)= 66.3
15. When compared with some measures for some time periods, these two measures had significant
negative correlations. There were five such cases involving the semi-variance (three with the co-
efficient of variation of price changes, and one each with the variance and mean absolute deviation).
There were seven cases for the coefficient of variation of price changes (three with the semi-
variance, and one each with the range divided by the mid-range, the coefficient of variation and each
of the measures using a modified quadratic mean). All significant negative correlations involved one
or both of these two measures.
16. See Footnote 15.
the variance and the mean absolute deviation in only one period each. This
suggests that the distributions of stock prices are not symmetric. The coeffi-
cient of variation of price changes has relatively high correlation with only the
measures suggested by Levy (which also depend upon price changes over sub-
periods). This measure should be considered for short-term investments since
it does reflect changes in price from sub-period to sub-period.
To test stability over time, the volatility rankings computed above were
used to compute the coefficient of concordance for each measure among time
periods. The null hypothesis of no community of order of rankings among
periods was rejected for seven of the eight measures (Table 4). Only for the
TABLE 4
COEFFICIENT OF CONCORDANCE AMONG TIME PERIODS FOR EACH MEASURE
Coefficient
Measure of concordance Test statistic
X.95(49)- 66.3
range divided by the mid-range could the hypothesis not be rejected. The lack
of stability of the range divided by the mid-range was to be expected since
it depends on extreme values. Extreme values may vary considerably among
periods even though week-to-week fluctuations in price for a particular stock
remain relatively constant over time.
IV. CONCLIUSIONS
measures (except the range divided by the mid-range) past volatility is, over
time, a relatively good predictor of future volatility.
Since volatility has been used as a risk surrogate, this study could be
profitably extended by an experiment to determine whether any of the sug-
gested volatility measures approximate investment decision makers' subjective
rankings of stocks as to risk. A sample of investors or financial analysts could
be asked to rank a relatively small number of stocks as to risk and compare
these rankings with the objective volatility rankings as computed above. If
substantial agreement among investors seems to exist, the reporting of vola-
tility measures by investment services may be justified.
Three other extensions may also be of value. First, the separation of market
volatility from the total volatility may yield ranking results (based on residual
or firm volatility) which are significantly different from those based on total
volatility. Second, the effect of the length of the time period over which the
measures are computed may be important. The range divided by the mid-
range particularly is likely to be quite satisfactory for short periods and un-
satisfactory for longer periods. Finally, the changes in measures when daily
rather than weekly prices are used should be explored.
REFERENCES