An alternate measure of trading, Working’s 1960 speculation index SPI, was also considered, to examine the impact of excessive speculation activity on cash market
volatility. Specifically, SPI 5 1 1 NCLCML 1 CMS if CML CMS, and SPI 5 1 1
NCSCML 1 CMS if CML , CMS, where NCL and NCS are noncommercial long and
short contracts, respectively, and CML and CMS are the commercial long and short contracts, respectively. The index reflects speculative activity over and above that which
is required to fill hedging imbalances. Following Ward and Behr 1983, we also constructed the upper and lower levels of this index. The upper level of the index, USPI,
was constructed by assuming that the nonreporting small traders are speculators. The lower level, LSPI, was constructed by assuming that all small traders are hedgers. The
examination of such indexes has regulatory implications, as regulators are mainly con- cerned about speculative activity which does not serve to address the imbalances in long
and short hedging contracts.
8
IV. Empirical Results
Table 1 reports monthly sample statistics and correlations pertaining to the variables employed in the regression analysis. The Augmented Dickey Fuller ADF test statistics
presented in Panel A were consistently significant for the volatility measures and the speculative indexes, rejecting the null of nonstationarity in these series. On the other hand,
the ADF statistics failed to reject the null of nonstationarity for the level SPEC speculator commitment, OI open interest, N speculator number and CSIZE average contract
size series. The ADF statistics were significant for the first differences of these series, indicating the presence of a unit root.
The correlations reported in Panel B indicate a positive contemporaneous relationship between the change first difference in SPEC and OI and the four volatility measures.
However, note that the correlation coefficients for DOI are much greater than those for
DSPEC. These coefficients provide some preliminary indications that speculative posi- tions are not more destabilizing than the nonspeculative positions. The correlation coef-
ficients for the remainder of the series were weak, with the exception of DCSIZE for which
the coefficients were notable for three of four volatility measures.
9
Table 2 presents the results from regression equation 1 when employing monthly data. The table also presents statistics from tests for the significance of the lagged
coefficients in the regression. The results indicate that the volatility measures were positively impacted by prior levels of volatility, consistent with the popular notion of
volatility clustering. The s
t 21
coefficient was significant for all regressions, and the test statistic for the sum of the lagged s variables was consistently significant at the 1 level.
There was also evidence of a positive, contemporaneous relationship between changes in open interest and volatility. The
DOI
t
coefficient was significant in all four volatility equations. On the other hand, there was relatively little evidence that speculative com-
mitments are positively related to stock market volatility. The DSPEC, coefficient was
8
For instance, the CFTC’s approval of the trading of any contract is contingent on its passing an economic purpose test. The test requires that the proposed contract either serve in price discovery, or be used for risk
transfer. The latter relates to the transfer of risk from hedgers to speculators [also, see Johnson 1989].
9
All measures of volatility were positively correlated, with the highest coefficient of .98 for the intraday– ADJR pairing. The jump–volatility measure was also closely related with the three other measures, with
coefficients of over .73.
328 A. Chatrath et al.
significant for the intraday measure of volatility alone. Moreover, the lagged coefficients for
DOI and DSPEC were consistently insignificant. Thus, there is little evidence that price variability is any more sensitive to speculation shocks than to shocks in the position
of other traders. The DSPEC
t 11
coefficient was significant for the interday and jump measures, providing some indication that speculative positions increase following in-
creased volatility. We further examined the result that volatility is no more influenced by shocks in
speculative commitments than by shocks in other commitments. We traced the impulse responses of SP 500 volatility to shocks in, alternately, the monthly number of hedging
contracts and the monthly number of speculative contracts. The response functions were constructed from pairs of vector autoregressive equations with only lagged regressors
with lags of 1 and 2 permitted in the regressions. The responses, presented in Figure 1,
Table 1. Summary Statistics 186 –395
Mean Std.Dev
ADF
a
ADF D
Volatility Measures
b
Interday 0.0063
0.0038 23.79
2 Intraday
0.0104 0.0055
23.32 2
ADJR 0.0105
0.0055 24.34
2 Jumps
0.1087 0.1182
22.89 2
Futures Activity
Measures
c
SPEC 26415
11431 22.19
24.10 OI
148870 38273
21.57 25.31
N 32.550
11.29 22.54
25.74 CSIZE
899.10 199.50
21.28 26.69
SPI 1.0372
0.0213 26.07
2 USPI
1.2326 0.0669
27.03 2
LSPI 1.0260
0.0130 25.82
2 Correlations
with Volatility
Interday Intraday
ADJR Jumps
DSPEC .11
.10 .10
.07 DNONSPEC
.19 .26
.26 .17
DN .02
.00 .01
.07 DCSIZE
.15 .16
.15 .01
SPI .01
.00 .01
.05 USPI
.06 .07
.07 .12
LSPI .01
2.01 2.01
2.03
a
ADF and ADF D are the Augmented Dickey Fuller statistic models with trend which tests the null hypothesis of
nonstationarity of the level series and first differenced series, respectively [see Engle and Granger 1987]. The critical values are from Engle and Yoo 1987.
b
Interday volatility given by the monthly average of uP
t
2 P
t 21
uP
t 21
, where P
t
is the daily closing cash price; Intraday volatility given by the monthly average of P
t H
2 P
t L
1
⁄
2
P
t H
1 P
t L
, where P
H
and P
L
are the intraday high price and low price, respectively; ADJR is the monthly average of the adjusted price range, and Jumps is the monthly frequency of price jumps.
c
SPEC is the commitment of large speculators; NONSPEC is the commitments of all traders other than large speculators; N is the monthly number of reporting speculators; CSIZE is the monthly average number of contracts held by speculators; SPI
represents Working’s 1960 speculative index, and USPI represents the upper and lower levels of this index as formulated in Ward and Behr 1983.
and represent significance levels of .05, .01, respectively.
Speculative Activity and Stock Market Volatility 329
indicate the prolonged influence on volatility to relatively major increments in commit- ments. As the responses were similar for all volatility measures, only the responses for
intraday volatility are presented. The response functions suggest that the reaction of the SP 500 index to speculation shocks is smaller than to hedging shocks. Moreover, the
responses to hedging shocks persist for longer periods of time relative to speculative
Table 2. Regression Results 186 –395 Cash Volatility, Speculative and Non-Speculative Commitments
Interday Intraday
ADJR Jumps
a 0.003
0.005 0.005
0.027 7.13
6.34 6.20
2.28 s
t 21
0.246 0.329
0.313 0.422
3.50 4.71
4.44 4.37
s
t 22
0.239 0.132
0.151 0.295
2.50 2.42
2.97 2.53
DSPEC
t 12
20.25e 0.16e
27
0.49e
27
20.28e
27
20.39 0.22
0.68 20.09
DSPEC
t 11
0.70e
27
0.61e
27
0.85e
27
0.49e
26
1.71 1.01
1.44 1.66
DSPEC
t
0.87e
2726
0.10e
26
0.19e
25
0.19e
25
1.48 1.72
1.63 1.48
DSPEC
t 21
20.65e
28
20.10e
27
20.92e
28
0.36e
26
20.12 20.14
20.13 1.26
DSPEC
t 22
0.73e
27
0.12e
26
0.11e
26
0.34e
27
1.46 1.17
0.98 0.02
DOI
t 12
20.12e
27
20.20e
27
20.24e
27
20.15e
27
20.58 20.70
20.82 20.04
DOI
t 11
20.91e
27
0.45e
28
0.30e
29
20.57e
26
21.08 0.23
0.16 21.26
DOI
t
0.31e
26
0.53e
27
0.51e
27
0.83e
26
1.99 1.72
1.66 1.97
DOI
t 21
0.20e
27
0.15e
27
0.13e
27
0.23e
26
1.60 0.55
0.49 0.57
DOI
t 22
0.46e
28
0.49e
28
0.31e
25
0.27e
26
0.67 0.21
0.13 0.72
Lagged s
1
0.485 0.461
0.465 0.718
8.45 6.84
6.33 5.31
Lagged DSPEC
0.67e
27
0.11e
26
0.10e
26
0.40e
26
0.86 1.02
0.97 1.46
Lagged DOI
0.25e
27
0.20e
27
0.16e
27
0.52e
26
0.70 0.17
0.34 0.73
Adjusted R
2
.14 .16
.15 .38
Durbin h 0.11
0.34 0.24
0.19
The coefficients and statistics relate to the regression: s
t
5 a 1
O
i 51
I
b
i
s
t 2i
1
O
j 5 22
J
g
j
DSPEC
t 2j
1
O
k 5 22
K
j
k
DOI
t 2k
1 e
t
, where s
t
is alternately: 1 interday volatility, given by the monthly average of uP
t
2 P
t 21
uP
t 21
, where P
t
is the daily closing cash price; 2 intraday volatility, given by the monthly average of P
t H
2 P
t L
1
⁄
2
P
t H
1 P
t L
, where P
H
and P
L
are the intraday high price and low price, respectively; 3 ADJR, which is the monthly average of the adjusted price range; and 4 Jumps, which
is the monthly frequency of price jumps; SPEC
t
and OI
t
are the commitment of speculators and open interest in month t, respectively; figures in parentheses are t statistics; Durbin h is an asymptotically normal statistic which tests the null of no
autocorrelation. , , represent significance levels of .10, .05, and .01, respectively.
330 A. Chatrath et al.
shocks. These findings are consistent with the notion that speculation activity is, in relative terms, not a destabilizing agent for cash markets.
Table 3 reports the results from the solution to regression equation 2. The leading DN
coefficient was significant for the interday volatility measure alone. However, the leading DSIZE coefficients were positive for three of four regressions, providing some evidence
that speculative interest increases declines following rising falling volatility. The DN
t
coefficient was insignificant for all regressions, indicating a lack of contemporaneous relationship between cash market volatility and the change in the number of speculators
in index futures. However, the DSIZE
t
coefficient was positive and significant for all but the jumps measure of volatility. Thus, there seems to be a positive contemporaneous
relationship between the average size of speculator holdings and normal cash market volatility. On the other hand, the lagged
DN coefficients and the lagged DSIZE coefficients were insignificant for all but the jumps variable. Further, the sum of the lagged
DN and sum of the lagged
DSIZE coefficients, as well as the joint sums of these coefficients, were insignificant for all but the jumps variable. In other words, as inferred from Table 2, there
seems to be no causal influence running from speculative activity to normal cash market volatility. However, there seems a positive bi-directional influence running from specu-
lation to the frequency of jumps in the cash market.
We also ran equations 1 and 2, employing weekly data for the shorter 101992–3 1995 interval. The deployment of this alternate data did not qualitatively change the
results, and in the interest of brevity, we only present the results pertaining to the solution of equation 2. These results, presented in Table 4, further support the notion that
speculation does not lead to increased volatility. The sum of lagged DN andor DSIZE
Figure 1. Impulse responses of SP 500 intraday volatility to one standard deviation shock in trader commitments.
Speculative Activity and Stock Market Volatility 331
coefficients were consistently nonpositive. In fact, there were instances when these coefficients were significantly negative the regressions involving interday and ADJR
measures of volatility. The DN
t
coefficients were consistently insignificant, and the DSIZE
t
coefficients were significant for only two of four regressions. It is also notable that
Table 3. Regression Results 186 –395 Cash Volatility, Number of Speculators, and Contract Size
Interday Intraday
ADJR Jumps
a 0.003
0.006 0.006
0.030 8.01
6.40 6.38
2.45 s
t 21
0.267 0.322
0.307 0.397
5.65 5.52
4.82 4.43
s
t 22
0.194 0.082
0.093 0.296
2.34 1.63
1.91 2.52
DN
t 12
0.29e
27
0.39e
27
0.50e
27
0.21e
25
0.21 0.21
0.28 0.81
DN
t 11
0.17e
26
0.14e
26
0.17e
26
0.25e
25
1.96 1.23
1.49 1.68
DN
t
0.17e
26
0.29e
26
0.23e
26
0.23e
25
1.37 1.51
1.30 0.99
DN
t 21
0.77e
27
0.10e
26
0.68e
27
0.33e
25
0.69 0.69
0.48 1.03
DN
t 22
0.78e
27
0.14e
26
0.96e
27
0.34e
25
0.94 1.25
0.92 0.74
DSIZE
t 12
20.28e
26
0.31e
25
0.14e
25
0.52e
24
20.12 1.56
0.42 0.82
DSIZE
t 11
0.49e
25
0.26e
26
0.37e
25
0.59e
23
1.78 0.07
1.80 1.01
DSIZE
t
0.45e
25
0.69e
25
0.65e
25
0.70e
24
2.06 1.85
1.75 1.28
DSIZE
t 21
0.29e
25
0.24e
25
0.18e
26
0.11e
23
1.55 1.09
0.84 1.91
DSIZE
t 22
0.13e
25
0.30e
25
0.28e
25
0.10e
23
0.78 1.47
1.49 1.73
Lagged s
1
0.461 0.404
0.399 0.694
7.23 5.13
4.87 5.29
Lagged DN
0.16e
26
0.25e
26
0.16e
26
0.67e
25
0.86 1.00
0.72 1.21
Lagged DSIZE
0.42e
25
0.54e
25
0.47e
25
0.21e
23
1.36 1.49
1.36 3.04
Lagged DN1DSIZE
0.44e
25
0.56e
25
0.48e
25
0.22
24
1.37 1.50
1.36 3.19
Adjusted R
2
.10 .08
.06 .32
Durbin h 20.24
20.04 20.10
0.47
The coefficients and statistics relate to the regression: s
t
5 a 1
O
i 51
I
b
i
s
t 2i
1
O
j 5 22
J
g
j
DN
t 2j
1
O
k 5 22
K
j
k
DHSIZE
t 2k
1 e
t
, where s
t
is alternately: 1 interday volatility, given by the monthly average of uP
t
2 P
t 21
uP
t 21
, where P
t
is the daily closing cash price; 2 intraday volatility, given by the monthly average of P
t H
2 P
t L
1
⁄
2
P
t H
1 P
t L
, where P
H
, and P
L
are the intraday high price and low price, respectively; 3 ADJR, which is the monthly average of the adjusted price range; and 4 Jumps, which
is the monthly frequency of price jumps; N
t
and HSIZE
t
represent the number of speculators and the number of contracts held by speculators in month t; figures in parentheses are t statistics; Durbin h is an asymptotically normal statistic which tests the
null of no autocorrelation; , , represent significance levels of .10, .05, .01, respectively.
332 A. Chatrath et al.
the coefficients for lagged volatility were strikingly weaker for the more recent period. This result suggests lesser volatility clustering for the more recent interval.
Table 5 reports the regression results that involve the speculative indexes. Given that the results pertaining to the weekly data are similar to that for the monthly data, only the
results for the monthly data are presented. Further, given that the results involving SPI are almost identical to those involving LSPI, we only report the results for regressions
involving SPI and USPI. The results continue to support a positive relationship between past and contemporaneous levels of volatility. However, there is little evidence of a
positive and contemporaneous relationship between the measures of volatility and level of speculation, as was indicated in Tables 2, 3 and 4. Moreover, neither the lagged USPI
coefficients, nor the sum of these coefficients were significant. Thus, the level of excess speculation does not seem to have had a causal influence on the variability of cash
markets, measured either in terms of normal volatility, or in terms if nonnormal jump volatility. This finding is especially important to regulators, since their concerns are
directed more to speculation which does not serve to fill hedging imbalances. In other words, there are indications that the positive influence of speculation on jumps found in
Table 3 arises mainly due to the interaction of large speculators and hedgers.
Over all, the regression results in Tables 2–5, and Figure 1 can be summarized as follows. First, the evidence suggests autoregressive patterns in the volatility measures.
Second, there is weak evidence of a positive, contemporaneous relationship between the commitments of speculators and cash market volatility. Third, there is no evidence to
indicate that the number and average holdings of speculators increase interday, intraday, or ADJR spot market volatility. Furthermore, there is no evidence that excessive specu-
lation causes cash market volatility, however measured. Fourth, speculative positions tend to increase following increased volatility. In other words, speculators may be reacting to
volatility, rather than causing it. Finally, the impulse responses indicated that speculative contracts have a smaller and less lasting influence on spot volatility than hedging
contracts.
10
V. Conclusions