Global Finance Journal 24 (2013) 171–187

Contents lists available at ScienceDirect

Global Finance Journal

journal homepage: www.elsevier.com/locate/gfj

Analysis of DJIA, S&P 500, S&P 400, NASDAQ 100

and Russell 2000 ETFs and their influence on
price discovery
Stoyu I. Ivanov a,1, Frank J. Jones a, Janis K. Zaima b
a
San Jose State University, Department of Accounting & Finance, College of Business, San Jose, CA 95192-0066, United States
b
Menlo College, 1000 El Camino Real, Atherton, CA 94027, United States

a r t i c l e i n f o a b s t r a c t

Article history: This study examines the temporal behavior of price discovery in the spot,
Accepted 19 May 2013 ETF and futures markets of the DJIA, S&P 500, S&P 400, NASDAQ 100 and
Available online 26 October 2013 Russell 2000. We document an increasing trend in the price discovery
metrics of exchange traded funds for all indexes but the DJIA. Contrary to
JEL classification: past studies, our findings show that the spot market rather than the
G13
futures market leads the price discovery. The arbitrage process that
G14
links exchange traded funds to spot prices, and not the futures prices
Keywords:
might explain the results. This daily arbitrage that ensures exchange
ETF market traded funds prices equal net asset values appear to promote spot
Futures market market price discovery especially with the popularity of exchange
Spot market traded funds in more recent years. We additionally document that
Price discovery the temporal behavior of the exchange traded funds price discovery
metric affects differently price discovery in the spot and futures
markets across indexes.
© 2013 Elsevier Inc. All rights reserved.

1. Introduction

This study extends past research on price discovery in three ways: (1) by employing Gonzalo and
Granger (1995) and Hasbrouck (1995) information share methodologies, (2) by allowing for temporal
changes in information shares of spot, futures and ETF markets and (3) by examining five indexes, the
Dow Jones Industrial Average (DJIA), S&P 500, S&P 400, NASDAQ 100 and Russell 2000 (with ticker
symbols DIA, SPY, MDY, QQQQ and IWM, respectively). Moreover, with the rising popularity of ETFs,
this study examines whether ETFs have generated a temporal change in the price discovery in these
markets.

E-mail address: stoyu.ivanov@sjsu.edu (S.I. Ivanov).

1
Tel.: +61 408 924 3934; fax: +61 408 924 3463.

1044-0283/$ – see front matter © 2013 Elsevier Inc. All rights reserved.
http://dx.doi.org/10.1016/j.gfj.2013.10.005
172 S.I. Ivanov et al. / Global Finance Journal 24 (2013) 171–187

While earlier studies use 1990 ETF data when ETFs began trading, by the end of March 2011 there have
been 986 ETFs that managed $1.055 trillion of assets.2 We find that the rising significance of ETFs has
provided an important price discovery role in the market and has influenced the dynamics of price
discovery in the spot market. In fact, contrary to the findings of Chu, Hsieh, and Tse (1999) we find that the
spot market consistently dominates the futures and ETF markets in price discovery. The reason for the
different findings might be attributed to the extended time period of this study and the rising popularity of
ETFs. Chu et al. (1999) use one year of intradaily data starting on January 29, 1993, when the S&P 500 ETF
was introduced whereas we examine a period spanning several years — starting in 1998 and ending on
March 18, 2011.3 Therefore, the more current and extensive time period examined in this study may help
to understand the temporal dynamics of price discovery between spot, futures, and ETF prices.
Our findings support aspects of past studies, showing that in the 1990s the S&P 500 futures market
dominates. However, in January 2001 until the end of the examined period (March 2011) the leadership
role abruptly reverses to spot market domination. A possible explanation for the reversal in the dominant
market to spot rather than futures market might be due to the process of ETF creation and redemption,
which provides an arbitrage opportunity between ETFs and the spot market.4 To help understand the
reasons for the spot market dominance a more detailed explanation of the ETF creation and redemption
process is provided in a later section.
The findings of this study are robust with respect to temporal changes in the information shares of
ETFs, spot and futures contracts in the five indexes. The temporal behavior of information shares has not
been examined in past studies of financial market price discovery, and contributes to the literature. The
results of this study suggest that price discovery changes across time and, thus, care must be taken when
shorter periods are analyzed. We additionally document that the temporal behavior of the ETF price
discovery metric affects differently price discovery in the spot and futures markets across indexes.
The structure of the paper is as follows. Section 1 provides a literature review. Section 2 presents the
methodology for the paper. Section 3 describes the data used in the paper. Section 4 provides the basic
results and Section 5 provides possible reasons for the findings. Section 6 discusses the temporal dynamics
of ETF information shares and Section 7 offers concluding remarks.

2. Literature review

Historically the competition for price leadership has been between the futures and spot markets. The
theoretical rationale behind the leverage, trading cost, uptick rule and market-wide information share
hypotheses all suggest that the informed traders should converge and trade in the futures market, thereby
making it the dominant market. In the late 1990s ETFs were introduced and gained popularity, implying
that their importance may create a potential shift in the price discovery processes in their respective
markets (Bernstein, 2009). Finance research shifted to re-examining the old price discovery leadership
role provided by the futures markets in light of the new trading vehicle. Even with the introduction of ETFs
these more recent studies find that the futures market provides the price leadership relative to the spot
and ETF markets (Hasbrouck, 2003; Chu et al., 1999; Fung, Liu, and Tse, 2008).
Chu et al. (1999) use intradaily data and the Gonzalo and Granger (1995) methodology, and they find
that the S&P 500 futures market leads the price discovery function as the S&P 500 spot and ETF markets
adjust to changes in the futures market. Chou and Chung (2006) use intradaily data and apply the
Hasbrouck (1995) methodology on the DJIA, S&P 500 and NASDAQ100 ETF and futures markets. They
exclude the spot market from their study, and cover only a six-month period from October 29, 2000 to
April 28, 2001, focusing on the period of decimalization in US stock markets. They document that before
decimalization the futures market was dominant for the three indexes, but after decimalization a shift in
price discovery dominance occurred from futures to the ETF market.

2
Source: Investment Company Institute (www.ici.org).
3
At the time of the Chu et al.'s (1999) study, the S&P 500 ETF was in its infancy and there were no other ETFs. Additionally, they
examined S&P 500 ETFs for one year from the date of its introduction on January 29, 1993 to December 31, 1993.
4
Financial ETFs are unique in this respect, because there are commodity ETFs available for trading which track the performance of
a futures contract, not the spot index market.
 S.I. Ivanov et al. / Global Finance Journal 24 (2013) 171–187 173

Fung et al. (2008) use intradaily data and examine the spot, futures and exchange traded fund markets
for the Dow Jones Industrial Average index. They use the Johansen cointegration methodology to estimate
a vector error correction model and determine that the futures market leads both the spot and ETF market.
However, a more recent study by Tse, Bandyopadhyay, and Shen (2006) detects a shift towards price
discovery predominantly in the ETF market, which supports the Bernstein (2009) claim. Tse et al. (2006)
examine the price discovery among the DJIA ETF, floor-traded regular futures contract and electronically-
traded mini futures contract using intradaily trades and quotes in each market. They use the Hasbrouck (1995)
methodology to determine the dominant force in the price discovery and find that the DJIA e-mini futures
market contributes most to price discovery and, thus, dominates. They also replicate the Hasbrouck (1995)
study by examining the S&P 500 ETF, regular and e-mini contracts. In contrast to the findings of Hasbrouck
(1995), Tse, Bandyopadhyay and Shen find that the S&P 500 ETF is the dominant market, not the e-mini futures
market, thus providing conflicting results across the two examined indexes.
Other studies which employ the Hasbrouck (1995) methodology are Kumar and Tse (2009) in the area
of Indian single-stock futures, Tse and Martinez (2007) in the area of international iShare ETFs, Tse and
Xiang (2005) in the area of energy E-mini futures and Fang and Sanger (2012) in the area of S&P 500 ETFs.
Fang and Sanger (2012) find that almost 50% of the price discovery in S&P 500 ETFs occurs in the ETF
market and that ETFs are driven by private information as much as the individual index stocks. They use
intradaily data for the period September 1, 2006 to December 1, 2009. Fang and Sanger (2012) examine
only the cash and ETF markets and do not consider the futures market in their study.
Hasbrouck (2003) considers the S&P 500, the NASDAQ 100 and the S&P 400 MidCap indexes with
intradaily data over the three month period from March 1, 2000 to May 31, 2000. Both floor-traded and
electronically-traded, small denomination (E-mini) futures contracts, ETFs and sector ETFs (which
decompose the S&P 500 into component industries) are considered. Using the Hasbrouck (1995)
information share approach he shows that the S&P 500 and NASDAQ 100 indexes' price discovery are
dominated by the E-mini futures market. The contribution of ETFs is small. Although the results for the
S&P MidCap 400 index are less clear, the ETFs dominate the price discovery. The sector ETFs contribute
little to the S&P 500 index, suggesting that little information is being generated in the sector funds.
We contribute to the literature by examining the temporal behavior of the spot, futures and ETF
markets' information shares. This allows us to determine if structural changes in the dominance of price
discovery has occurred.

3. Methodology

Gonzalo and Granger (1995) and Hasbrouck (1995) methodologies call for the identification of a
Vector Error Correction Model (VECM) of the long term equilibrium relation among the cointegrated
series of index, ETF and futures prices.
The following VECM specification is used:

′
X
k
Δpt ¼ αβ pt−1 þ Γ i Δpt−i þ ut ; ð1Þ
i¼1

where pt is the 3 × 1 vector of futures, ETF and index prices and ut is the error term, and Δ is the first
difference operator.
The Gonzalo and Granger (1995) methodology defines the common factor based on the estimated
VECM as Wt:

′
Wt ¼ α ⊥ pt : ð2Þ

They identify it as an observable linear combination of the original futures, ETF and index prices, and a
cointegrating relation Zt:

′
Zt ¼ β pt : ð3Þ
174 S.I. Ivanov et al. / Global Finance Journal 24 (2013) 171–187

Based on the common factor, Wt, and the transitory (or cointegration) relation, Zt, they derive the
permanent-transitory decomposition (PT) of the vector of futures, ETF and spot index prices as:

pt ¼ A1 Wt þ A2 Zt ; ð4Þ

where:
 −1
′
A1 ¼ β⊥ α ⊥ β⊥ ; ð5Þ
and
 −1
′
A2 ¼ α β α : ð6Þ

The price discovery measures based on this method are calculated as the proportions of the futures, ETF
and spot index prices in the common permanent component on a monthly basis. In other words this is a
vector of coefficients that add up to one because of normalization as discussed in Chu et al. (1999).
Hasbrouck (1995) shows that the VECM can be represented as a vector moving average:

Δpt ¼ AðLÞut ; ð7Þ

with A(L) the matrix polynomial in the lag operator, L, with elements the parameters aij, and

2 2 2 2 2 ! ′
σ w ¼ σ 11 σ 1 þ σ 12 σ 2 ¼ a Ω!
a ; ð8Þ

where Ω is the covariance matrix also obtained from the VECM. The parameters aij, are the impulse
responses which are interpreted as the impact of one standard deviation change in one variable on the
other variables. The information share is computed on a monthly basis for each index as follows:

a2ij σ 2i
Ii ¼ ; ð9Þ
σ 2w

where aij are obtained from the vector moving average representation of the VECM as shown by
Hasbrouck (1995).
After computing the two ETF price discovery metrics we examine their temporal behavior and relation
with the respective futures and spot markets' price discovery measures. However, Figuerola-Ferretti and
Gonzalo (2010) argue that a limitation of the Hasbrouck (1995) methodology is that it produces upper and
lower bounds for the computed information share. When high frequency data are used and the
appropriate lag length is selected, Hasbrouck (1995) points out that the upper and lower bounds of the
information share converge. Despite convergence, Sapp (2002) shows that a closed form solution for the
statistical properties of the Hasbrouck (1995) information share measure does not exist. Thus, parametric
models cannot be used to examine their properties. Sapp (2002) uses nonparametric methods in his
information share analysis. In contrast, the Gonzalo–Granger price discovery measure statistical properties
have a closed form solution and as such are suitable to examine across time. Therefore, we apply both the
Gonzalo–Granger and Hasbrouck methodologies to test the robustness of price discovery by utilizing
parametric and non-parametric models.
Considering the existence of closed form solution on the statistical properties of the Gonzalo–Granger
price discovery metrics a simple ordinary least squares (OLS) model can be used to examine the
relationship among the three series. The OLS models are as follows:

PT2;t ¼ γ 0 þ γ1 PT1;t−1 þ lt ; ð10Þ

PT3;t ¼ ψ0 þ ψ1 PT1;t−1 þ mt ; ð11Þ

 S.I. Ivanov et al. / Global Finance Journal 24 (2013) 171–187 175

Table 1
This table reports each index, futures contract and ETF ticker symbol, name and the date for the first available data point.

Index Futures ETF

Ticker Description Start Ticker Description Start Ticker Description Start date
date date

INDU Dow Jones 4/1/1993 YM E-mini Dow Futures 4/5/2002 DIA Dow Diamonds 7/1/1998
Industrial Continuous Trust
Average
NDX NASDAQ 100 1/2/1997 NQ E-mini NASDAQ 100 7/1/1999 QQQQ NASDAQ 100 3/10/1999
index continuous Trust
INX S&P 500 index 2/1/1983 ES E-mini S&P 500 9/11/1997 SPY S&P depository 7/1/1998
continuous receipts
RUT Russell 2000 1/2/1997 TF E-mini Russell 10/25/2001 IWM iShares Russell 6/28/2000
index 2000 continuous 2000 index fund
IDX S&P 400 1/2/1998 EMD E-mini S&P MidCap 400 1/28/2002 MDY S&P MidCap 2/5/1999
MidCap continuous depository receipts
index
Ending date 03/18/2011
For all

with PT1,t the price discovery metric of the ETF, PT2,t the price discovery metric of the index and PT3,t the
price discovery metric of the futures contract and lt and mt the error terms. However, we document that all
the price discovery measures are integrated and, thus, the price discovery measures need to be modified
by taking the rate of change of price discovery measures or the logarithm of these measures before
applying the OLS methodology.

4. Data

We use the DJIA, S&P 500, S&P 400, NASDAQ 100 and Russell 2000 spot index levels, futures contracts
and ETFs on an intradaily basis.5 The data are obtained from pitrading.com. A detailed description of the
indexes, futures contracts and ETFs is provided in Table 1.
All ETFs are unit investment trusts with the exception of the Russell 2000 (IWM), which is an open end
fund. All futures contracts are traded on the CME Globex with the exception of the Russell 2000 futures.
The dataset for indexes, futures contracts and ETFs start at different dates. For example, the DJIA data start
on April 1, 1993, the E-mini Dow Futures Continuous Contract data start on April 5, 2002 and the
Diamonds ETF data start on July 1, 1998. All data end on March 18, 2011. Therefore, for the individual
indexes, futures contracts and ETFs each data range varies. Table 2 provides summary statistics of index
levels, futures prices and ETF prices.
When information shares are computed for the index, futures and ETF series in Table 2, each data is
merged such that the start and ending dates are the same. The common start date will be determined by the
shortest series of the three. Thus, in the case of the DJIA, the common period extends from April 5, 2002 to
March 18, 2011, where its start date is determined by the data availability of the futures contract. The time
period for the NASDAQ 100 index extends from period July 1, 1999 to March 18, 2011 constrained by the data
availability of the futures contract.
The dataset for the S&P 500 index overlaps between July 1, 1998 and March 18, 2011 due to the data
availability for the ETF. The data for the Russell 2000 index extend from October 25, 2001 to March 18,
2011 and the S&P 400 index dataset have common dates from January 28, 2002 to March 18, 2011, both
constrained by the shorter futures contract datasets.

5
Although the Russell 2000 futures originally traded on the CME Globex; however, since September 2008 it has moved to the
Intercontinental Exchange (ICE).
176 S.I. Ivanov et al. / Global Finance Journal 24 (2013) 171–187

Table 2
This table provides summary statistics for our sample securities. The sample periods are: April 5, 2002 until March 18, 2011 for the
DJIA index, July 1, 1999 until March 18, 2011 for the NASDAQ 100 index, July 1, 1998 until March 18, 2011 for the S&P 500 index,
October 25, 2001 until March 18, 2011 for the Russell 2000 index and January 28, 2002 until March 18, 2011 for the S&P 400 index.
DIA, QQQQ, SPY, IWM, and MDY are the DJIA, NASDAQ 100, S&P 500, Russell 2000 and S&P 400 ETF ticker symbols, respectively. ADF
is Augmented Dickey Fuller Unit Root Test and PP is Phillips–Perron Unit Root Test.

Dow Jones Industrial NASDAQ 100 S&P 500 index

Prices DJIA DIA DJIA futures NDX QQQQ NDX futures S&P 500

N 864,579 864,579 864,579 1,135,823 1,135,823 1,135,823 1,215,770

Mean 10541.28 105.40 104.77 1816.31 44.97 50.43 1190.44
Median 10498.73 104.97 106.74 1673.68 41.28 45.32 1190.45
Std. dev. 1535.27 15.34 15.06 680.81 17.04 19.54 183.64
Min 6470.91 64.79 59.38 795.25 19.78 25.18 666.79
Max 14196.88 141.93 138.02 4816.35 120.48 134.97 1576.03
ADF 0.6497 0.6199 0.6718 0.7539 0.7240 0.7850 0.5483
PP 0.6790 0.6740 0.7150 0.7550 0.7550 0.7930 0.5190

H0: Rank = Trace Critical value H0: Rank = r Trace Critical value H0: Rank = r
r (5%) (5%)

Johansen 0 3644.24 34.80 0 4714.16 34.80 0

Test 1 9.25 19.99 1 18.86 19.99 1
2 2.38 9.13 2 3.67 9.13 2

S&P 500 index Russell 2000 index S&P 400 index

SPY S&P 500 futures Russell 2000 IWM Russell 2000 futures S&P 400 MDY S&P 400 futures

The summary statistics on the mean and median are reported as we later conduct Gonzalo–Granger
and Hasbrouck price discovery using the two measures. Moreover, if the mean and median are close for
each of the indexes it may imply normality. We find that the spot indexes exhibit similar mean and
median except for NASDAQ 100. For example, the mean for the DJIA is 10,541.28 and its median is
10,498.73. Likewise, the mean and median are close for the S&P 500 index (1190.44 and 1190.45,
respectively), Russell 2000 (632.67 and 641.77, respectively), and for S&P 400 index (1648.61 and
1671.63, respectively). Similar results exist for the ETFs and futures. The DJIA ETF mean is 105.40 and its
median is 104.97 while the DJIA futures mean is 104.77 and its median is 106.74. However, the mean and
median for QQQQ, NDX, MDY, and the S&P 400 futures are not as close. The QQQQ mean equals 44.97
while its median is 41.28, and the NDX futures mean equals 50.43 and its median is 45.32. Likewise the
MDY or the S&P 400 ETF equals 128.11 while its median is 133.22. Additionally, its futures mean is 724.96
while its median is 752.50. Since the closeness of the mean and median does not test for normality of the
price discovery metric we conduct statistical tests of normality by using the Jarque–Bera and the
Kolmogorov–Smirnov tests in the next section.

5. Basic results

Studies of information shares follow a stringent cointegration methodology by first testing the time
series of each market for stationarity. After it is confirmed that stationarity is not present, we test for
cointegration and a VECM is developed for the long-term relation between the series. Once these
conditions are met, information shares can be computed. Table 2 reports the results of the Augmented
Dickey Fuller and Phillips–Perron Unit Root Tests on indexes, ETFs and futures contracts series and the
Johansen cointegration test. The non-stationarity test results fail to reject non-stationarity for all series.
The cointegration test rejects no cointegration among the indexes, futures contracts and ETF series and
fails to reject the existence of at least one and at most two cointegrating vectors. Note that these tests are
conducted over the respective entire sample periods for each series whereas the price discovery analysis
is conducted on a monthly basis.
Table 3 provides summary statistics of the index, futures and ETF price discovery measures for each
index. Table 3 panel A presents the Gonzalo and Granger (1995) permanent-transitory component
 S.I. Ivanov et al. / Global Finance Journal 24 (2013) 171–187 177

Table 2 (continued)
S&P 500 index Russell 2000 index S&P 400 index

SPY S&P 500 futures Russell 2000 IWM Russell 2000 futures S&P 400 MDY S&P 400 futures

1,215,770 1,215,770 811,882 811,882 811,882 734,731 734,731 734,731

119.12 125.83 632.67 63.07 63.93 1648.61 128.11 724.96
119.17 126.95 641.77 63.98 65.62 1671.63 133.22 752.50
18.33 21.23 124.51 12.36 11.94 312.21 25.31 129.76
67.12 62.58 324.98 32.30 31.98 795.67 67.85 378.10
157.50 173.75 856.48 85.19 84.93 2403.27 178.81 981.20
0.3758 0.5375 0.7389 0.5797 0.6262 0.4478 0.7947 0.8314
0.5050 0.5950 0.6440 0.6090 0.6430 0.5110 0.7930 0.8290

Trace Critical value H0: Rank = Trace Critical value (5%) H0: Rank = Trace Critical value
(5%) r r (5%)

8823.34 34.80 0 2820.47 34.80 0 41.06 34.80

13.71 19.99 1 10.51 19.99 1 7.73 19.99
6.29 9.13 2 4.63 9.13 2 2.16 9.13

decompositions and Table 3 panel B reports the Hasbrouck (1995) information shares for each index.
Contrary to the findings of Chu et al. (1999), the Gonzalo and Granger (1995) permanent-transitory
component decomposition summary statistics in Table 3 panel A show that the spot market dominates
price discovery among the three market venues for all indexes. The spot market has an average
information share of 0.706 for the DJIA, 0.629 for the NASDAQ 100 index, 0.393 for the S&P 500 index,
0.376 for the Russell 2000 index and 0.554 for the S&P 400 index.
The futures market is second to the spot market in price discovery for all but the Russell 2000 index when
the mean and median monthly information shares are examined. When each series is divided in half, all ETF
price discovery measures are higher in the second half with the exception of the DJIA index ETF (DIA) which
experiences a decrease in its ETF price discovery measure. All futures contract permanent-transitory component
decompositions decrease in the second half of each series. Likewise, all five indexes' spot permanent-transitory
component decompositions increase in their second half. These findings suggest that the spot and ETF have
increased their price discovery in the examined period and subsequently reduced the price discovery
contribution of the futures market.
There is consistency in the findings of this study and the results presented by Chou and Chung (2006)
who find improvement in the pricing efficiency of ETFs after decimalization. However, the influence of
decimalization on all ETFs cannot be tested because the data for all but SPY and QQQQ start after
decimalization has been initiated in August of 2000. Moreover, the limited observations for SPY and QQQQ
prior to decimalization make it difficult to draw meaningful conclusions.
For comparison purposes only, Table 3 panel B presents the Hasbrouck (1995) information shares and
indicates that for all indexes the spot market dominates price discovery among the three market venues.
Again, the results are contrary to the findings of Chu et al. (1999) when the Hasbrouck methodology is
utilized and attests to the robustness of the findings in this study. The spot market has a mean information
share of 0.631 for the DJIA, 0.740 for the NASDAQ 100 index, 0.474 for the S&P 500 index, 0.552 for the
Russell 2000 index and 0.458 for the S&P 400 index. The futures market is second to the spot market in the
price discovery for the DJIA, NASDAQ 100 and S&P 500 indexes, whereas the ETF market is second to the
spot market in the price discovery for the Russell 2000 and S&P 400 indexes when mean monthly
information shares are examined. However, if the medians are examined, the futures markets of the DJIA
and NASDAQ 100 indexes dominate the ETF market. The medians of the Russell 2000 and S&P 400 ETF
 178
Table 3
This table provides summary of monthly price discovery measures. The sample periods are: April 5, 2002 until March 18, 2011 for the DJIA index, July 1, 1999 until March 18, 2011 for the NASDAQ
100 index, July 1, 1998 until March 18, 2011 for the S&P 500 index, October 25, 2001 until March 18, 2011 for the Russell 2000 index and January 28, 2002 until March 18, 2011 for the S&P 400
index. DIA, QQQQ, SPY, IWM, and MDY are the DJIA, NASDAQ 100, S&P 500, Russell 2000 and S&P 400 ETF ticker symbols, respectively.

Dow Jones Industrial NASDAQ 100 index S&P 500 index Russell 2000 index S&P 400 index
Average

DJIA DIA DJIA NDX QQQQ NDX S&P SPY S&P 500 Russell IWM Russell 2000 S&P MDY S&P 400
futures futures 500 futures 2000 futures 400 futures

N 108 108 108 141 141 141 153 153 153 114 114 114 111 111 111

A. Gonzalo and Granger (1995) permanent-transitory components

S.I. Ivanov et al. / Global Finance Journal 24 (2013) 171–187

Mean 0.706 0.110 0.185 0.629 0.111 0.260 0.393 0.242 0.365 0.376 0.371 0.252 0.554 0.186 0.259
First half mean 0.578 0.136 0.286 0.518 0.075 0.407 0.240 0.183 0.577 0.274 0.292 0.434 0.491 0.109 0.400
Second half mean 0.828 0.084 0.088 0.732 0.147 0.121 0.538 0.300 0.163 0.471 0.450 0.078 0.610 0.262 0.128
Median 0.660 0.111 0.193 0.629 0.119 0.259 0.381 0.256 0.290 0.385 0.431 0.165 0.532 0.167 0.268
First half median 0.576 0.123 0.282 0.557 0.081 0.335 0.309 0.150 0.484 0.229 0.269 0.451 0.488 0.088 0.420
Second half median 0.865 0.067 0.067 0.702 0.167 0.091 0.578 0.284 0.129 0.482 0.450 0.063 0.595 0.268 0.095
Std. dev. 0.149 0.052 0.118 0.164 0.085 0.190 0.215 0.119 0.285 0.145 0.164 0.218 0.126 0.111 0.160
Skewness 0.442 0.681 0.020 0.045 0.390 0.020 0.012 −0.235 0.014 0.132 −0.886 0.019 0.322 0.486 0.043
Kurtosis 0.916 0.144 0.399 0.950 −0.852 0.952 0.746 −0.648 0.982 0.744 −0.372 0.684 0.829 −0.711 0.526
Jarque–Bera normality test b.0001 b.0001 b.0001 b.0001 b.0001 b.0001 b.0001 b.0001 b.0001 0.0710 b.0001 b.0001 0.0043 b.0001 0.0314
Kolmogorov–Smirnov normality test b.0001 b.0001 b.0001 b.0001 b.0001 b.0001 b.0001 b.0001 b.0001 b.0001 b.0001 b.0001 b.0001 0.0194 b.0001
(p-value)
ARCH Test (Null: No ARCH) 0.0066 0.8616 0.0996 b.0001 0.0370 b.0001 0.0062 0.9085 0.0046 0.5115 0.5630 0.4277 0.7886 0.5250 0.0002
Ljung–Box Q2(12) (Null: No ARCH) 0.4874 0.0016 0.2269 0.1013 0.1976 0.0412 0.3054 0.0121 0.0501 0.0104 0.1384 0.2299 b.0001 0.0006 0.0221
ADF Zero mean model 0.8272 0.1584 0.0685 0.8158 0.2135 0.0579 0.6924 0.4006 0.0455 0.3760 0.5963 0.3087 0.5693 0.3266 0.3867
ADF single mean model 0.6651 0.0012 0.3991 0.4591 0.2391 0.4486 0.4221 0.1193 0.4351 0.3493 0.1613 0.8504 0.2033 0.3014 0.7758
ADF trend model 0.0338 b.0001 0.0178 0.0082 0.6495 0.1063 0.0132 0.2905 0.2706 0.0022 0.7092 0.1297 0.0394 0.5482 0.0162
PP Zero mean model 0.8210 0.1160 0.0580 0.8350 0.2000 0.0730 0.7040 0.3740 0.0350 0.3200 0.5880 0.3410 0.6030 0.3070 0.3890
PP single mean model 0.4020 0.0010 0.3260 0.4330 0.1940 0.4850 0.4250 0.0490 0.3990 0.2130 0.1730 0.8160 0.1870 0.2870 0.7290
PP trend model 0.0010 0.0010 0.0010 0.0010 0.5670 0.0040 0.0050 0.1140 0.1550 0.0010 0.7420 0.0220 0.0320 0.5170 0.0080

B. Hasbrouck (1995) information shares

Mean 0.631 0.122 0.247 0.740 0.101 0.159 0.474 0.215 0.310 0.552 0.229 0.220 0.458 0.359 0.182
First half mean 0.661 0.143 0.196 0.739 0.075 0.186 0.422 0.202 0.376 0.517 0.235 0.248 0.481 0.313 0.207
Second half mean 0.596 0.102 0.302 0.738 0.127 0.136 0.530 0.228 0.242 0.581 0.224 0.195 0.433 0.407 0.160
Median 0.648 0.088 0.216 0.784 0.079 0.107 0.544 0.194 0.196 0.588 0.243 0.159 0.427 0.257 0.110
First half median 0.653 0.118 0.198 0.779 0.065 0.147 0.478 0.156 0.221 0.524 0.241 0.236 0.444 0.232 0.129
Second half median 0.601 0.068 0.260 0.813 0.109 0.067 0.565 0.215 0.170 0.676 0.248 0.039 0.376 0.385 0.068
Std. dev. 0.205 0.113 0.223 0.188 0.104 0.175 0.297 0.169 0.303 0.201 0.125 0.237 0.335 0.313 0.202
Skewness 0.013 1.381 0.000 0.000 3.047 0.000 0.000 1.181 0.000 0.000 0.27 0.000 0.000 0.55 0.000
Kurtosis 0.979 1.731 0.984 0.999 13.462 0.824 0.996 1.753 0.993 0.880 0.073 0.956 0.999 −0.988 0.923
 S.I. Ivanov et al. / Global Finance Journal 24 (2013) 171–187 179

DJIA NASDAQ 100 S&P 500 Russell 2000 S&P 400

Fig. 1. Monthly index, ETF and futures Gonzalo and Granger (1995) permanent-transitory components. Note: Green is futures
contract monthly permanent-transitory component, red is ETF monthly and blue is index monthly share.

information shares are higher than the medians of their futures markets. However, the medians for the
ETFs (0.194) and futures (0.196) are almost equal for the S&P 500 index.
Next, the parametric (Jarque–Bera) and non-parametric (Kolmogorov–Smirnov) normality tests are
applied to the Gonzalo and Granger (1995) permanent-transitory component series and presented in Table 3
panel A. Both test the null hypothesis that the distribution is normal. The test statistics for ETFs reject the null
hypothesis that the permanent-transitory components are normally distributed. Additionally, ARCH
(Autoregressive Conditional Heteroskedasticity) and Ljung–Box tests are used to determine whether the
permanent-transitory component series can be described as white noise process and whether ARCH effects
are present. The ARCH tests the null hypothesis that no ARCH effects exist in the series, and that it is an
independently distributed series. The Ljung–Box Q2(12) statistic tests the null hypothesis that the series is
independently distributed. The ARCH and Ljung–Box Q2(12) statistic results presented in Table 3 panel A
suggest the presence of linear dependence and ARCH effects for all series with the exception of the Russell
2000 ETF and futures contract. Therefore, special attention needs to be given in examining the temporal
dynamics of the price discovery measures.
This cautionary note is reinforced by the non-stationarity test results which are presented in Table 3
panel A. The Augmented Dickey Fuller and Phillips–Perron non-stationarity tests are used in the study and
find that there is a unit root and, thus, non-stationarity in the indexes. The Augmented Dickey Fuller Test
uses standard five lags in the process.
The permanent-transitory components of all the indexes have unit roots. A graphical representation of
the dynamics of price discovery across time is provided in Figs. 1 through 6 which further confirms the
presence of trends in the permanent-transitory components and information shares. Figs. 1, 2 and 3
display the monthly Gonzalo and Granger (1995) permanent-transitory components whereas Figs. 4, 5
and 6 illustrate the monthly Hasbrouck (1995) information shares. The Hasbrouck (1995) information
shares are provided for comparison purposes only.
Fig. 1 plots all three of the index, ETF and futures permanent-transitory components as they must add
up to one. As shown, the spot market has the dominant price discovery. But there are temporal changes in
this dominance. Fig. 1 indicates that significant fluctuation in the dominant market has occurred across
time. Fig. 2 plots only the Gonzalo–Granger permanent-transitory components for the spot versus the
futures. Clearly the futures contract price discovery has a downward trend whereas the spot market has an
upward trend.
In Fig. 2 it is easy to identify that the futures market, which theoretically is expected to be the dominant
price discovery market, is initially dominant only for the NASDAQ 100, S&P 500 and Russell 2000 indexes.
In the last half of the period, however, the spot market is dominant for all the indexes. Fig. 3 plots the
permanent-transitory component of only the ETF. The ETF has a moderate upward trend for four indexes,
NASDAQ 100, S&P 500, Russell 2000, and the S&P 400, but not the DJIA. However, the upward trend does
not continue for each of the four indexes. The QQQQ price discovery shows a sharp increase in early 2000
only to experience an abrupt decline in April 2009. Thereafter, it stabilizes albeit at a much lower level.
Similar significant temporal changes occur for the S&P 500, the Russell 2000, and the S&P 400 indexes
where the reversal in its upward trend ends earlier than 2009. These results demonstrate that caution
must be taken when examining price discovery even over short time periods.
Figs. 4–6 repeat the analysis of Figs. 1–3, respectively, but using the Hasbrouck (1995) information
shares rather than the Gonzalo and Granger (1995) permanent-transitory components. These figures
represent analysis based on the Hasbrouck methodology. For comparison, Fig. 4 plots the index, ETF and
180 S.I. Ivanov et al. / Global Finance Journal 24 (2013) 171–187

DJIA NASDAQ 100 S&P 500 Russell 2000 S&P 400

Fig. 2. Index and futures monthly Gonzalo and Granger (1995) permanent-transitory components. Note: Green is futures contract
monthly permanent-component and blue is index monthly share.

futures information shares based on the Hasbrouck (1995) information shares. For all the indexes, the spot
market dominates and the degree of dominance increases over time.
Fig. 4 also indicates the importance of examining multiple indexes rather than just one. For example,
the S&P 500 futures contract dominates initially, supporting the study by Chu et al. (1999).
However, beginning in January 2001, in contrast to Chu et al. (1999), an abrupt switch to spot market
price discovery dominance emerges. Fig. 5 highlights this point. Clearly, the spot market information
shares dominate the futures market information shares with an exception in the beginning of the period
for the S&P 500 index.
However, Fig. 5 also illustrates that other indexes do not follow the same pattern as the S&P 500. The
spot and futures information shares for DJIA and the NASDAQ 100 show that the spot market dominates
most of the period except for around 2007–2008 where they appear about equal. The spot and futures
information shares for the Russell 2000 display a strong spot market leadership except in four or five years
around mid-2000 to 2011. In contrast, S&P 400 information share data exhibit strong spot market
dominance initially then a decline below the futures in 2011.
Fig. 6 plots the monthly information shares of ETFs for all the indexes, and illustrates that the pattern of
ETF information shares varies over time and across the indexes. These results suggest that a more detailed
analysis of the temporal dynamics of the information shares is necessary for each of the five indexes.
In summary the figures show that over time the role of the futures market has diminished, and the spot
market dominates the price discovery in more recent years. We suggest that the increased popularity of
ETFs as an investment vehicle and their creation and redemption arbitrage mechanism that engages the
spot market may be the reason for the changing price discovery.

6. The relationship between ETF and the spot markets

The illustrations from Figs. 1–3 and 4–6 provide evidence that the introduction of ETFs reduced the
dominant price discovery role of the futures market, and increased the role of the spot market. As the ETF
market developed and grew in popularity we propose that the ETF arbitrage mechanism that links the spot
market and ETF contributed to the dominance of the spot price discovery leadership. The creation–
redemption process used in the ETF arbitrage mechanism introduced a linkage between the spot market
for stocks and the ETF, and not the futures market.
The ETF trading and the subsequent arbitrage mechanism for an investor buying an ETF works in the
following way.6 An investor buying ETF shares from another investor on a securities exchange pays the
price of the ETF on the exchange. However, the price the investor pays is not the NAV of the ETF (net asset
value, which is the value of the portfolio of securities underlying the ETF). Such buying could cause the
price of the ETF to be greater than the price of the NAV. Selling the ETF could cause the price of the ETF to
be less than the price of the NAV even though they should be identical.
To maintain equality, or near equality, of the NAV of the securities in the portfolio and the share price of
the ETF an agent, also known as the Authorized Participant (AP), is commissioned by the sponsor of the
ETF to arbitrage between the price of the ETF and the value of the underlying portfolio. In this case when
an investor is buying an ETF, the agent would buy the (lower-priced) stocks in the ETF and sell the

6
For a detailed discussion of the ETF arbitrage mechanism please see Gastineau (2004), Engle and Sarkar (2006), Delcoure and
Zhong (2007) and DeFusco, Ivanov, and Karels (2011).
 S.I. Ivanov et al. / Global Finance Journal 24 (2013) 171–187 181

DJIA NASDAQ 100 S&P 500 Russell 2000 S&P 400

Fig. 3. ETF monthly Gonzalo and Granger (1995) permanent-transitory components.

DJIA NASDAQ 100 S&P 500 Russell 2000 S&P 400

1 1 1 1 1
0.5 0.5 0.5 0.5 0.5
0 0 0
0 0

200110
200301
200404
200507
200610
200801
200904
201007

200201
200303
200405
200507
200609
200711
200901
201003
199807
200003
200111
200307
200503
200611
200807
201003
200204
200306
200408
200510
200612
200802
200904
201006

199907
200101
200207
200401
200507
200701
200807
201001

Fig. 4. Monthly index, ETF and futures Hasbrouck (1995) information shares. Note: Green is futures contract monthly information
share, red is ETF monthly information share and blue is index monthly information share.

(higher-priced) ETF. These arbitrage actions by the AP allows them to capture the potential arbitrage and
keep the price of the ETF very close to (or equal to) that of the NAV.
The APs, who are typically large sophisticated institutions, restore their original position (netting only
the arbitrage profit) by dealing directly with the ETF sponsor. Additionally, only APs may create or redeem
shares of an ETF with the ETF sponsor and only in large, specified quantities called creation/redemption
units. Therefore, if the investor initially buys the ETF, the AP may create new ETF shares by providing the
sponsor with a specified basket of stocks, i.e. a creation unit, and the sponsor responds by transferring the
corresponding number of the ETF shares to the AP. Similarly, if the original investor sold an ETF, an AP
could redeem ETF shares by providing the fund with a specific number of ETF shares, a redemption unit,
and the fund will transfer to the AP the specific basket of stocks. These transfers are considered “in-kind
transfers” of assets and have no net cost or tax impact since no cash changes hands. This “in-kind” creation
and redemption process also tends to keep an ETF's market price close to its NAV. If discrepancies arise
between a fund's market price and its NAV, it opens up a profit-making arbitrage opportunity for an AP,
thereby driving the market price of a share of the ETF and the portfolio's NAV closer together.
By design, ETFs are constructed to track the performance of an index on intradaily basis, and it either
consists of the precise composition of the index or use a sampling technique to replicate the index
throughout the day. The arbitrage mechanism of creating and redeeming ETF units replicates the
performance of the underlying index as much as possible. Moreover, the process ensures the existence of a
very close and rapid link between the ETF market and the spot market for stocks. Changes in the ETF
market will be transmitted to the spot market via the creation–redemption process. Changes in the spot

DJIA NASDAQ 100 S&P 500 Russell 2000 S&P 400

1.2 1.2 1.2 1.2 1.2
1 1 1 1 1
0.8 0.8 0.8 0.8 0.8
0.6 0.6 0.6 0.6
0.6
0.4 0.4 0.4
0.4 0.4
0.2 0.2 0.2
0.2 0.2
0 0
0 0
0

Fig. 5. Index and futures monthly Hasbrouck (1995) Information shares. Note: Green is futures contract monthly information share
and Blue is index monthly information share.

DJIA NASDAQ 100 S&P 500 Russell 2000 S&P 400

Fig. 6. ETF monthly Hasbrouck (1995) information shares.

182 S.I. Ivanov et al. / Global Finance Journal 24 (2013) 171–187

Table 4
This table reports regression results for the relationship between the rate of change (ROC) for the permanent-transitory components
of ETF and the ROC of the index (or futures in Eq. (11)) price discovery measure. The equations are:

PT2;t ¼ γ0 þ γ1 PT1;t−1 þ lt ;

PT3;t ¼ ψ0 þ ψ1 PT1;t−1 þ mt ;

where PT1, t is the price discovery metric of the ETF, PT2,t is the price discovery metric of the index, and PT3,t is the price discovery
metric of the futures contract. The error term for Eq. (10a) is mt and for Eq. (11a) is lt.

ROC DJIA DJIA futures NDX* NDX futures*

Coeff p-Value Coeff p-Value Coeff p-Value Coeff p-Value

Intercept 0.020 0.038 0.035 0.265 0.094 0.244 0.011 0.664

ROC ETF −0.085 b.0001 −0.107 0.037 −0.025 0.401 0.010 0.279
R-sq 0.229 0.041 0.006 0.010
ARCH Resid b.0001 0.864 0.909 0.097
Zero mean b.0001 b.0001 b.0001 b.0001
Single mean b.0001 b.0001 b.0001 b.0001
Trend b.0001 b.0001 b.0001 b.0001
*Restricted sample *Restricted sample

SP500 SP500 futures R2000 R2000 futures SP400 SP400 futures

Coeff p-Value Coeff p-Value Coeff p-Value Coeff p-Value Coeff p-Value Coeff p-Value

market will be transmitted to the ETF market by the effect of NAV on the ETF price whereby the spot
market actively and directly responds to ETF trading. Therefore, we suggest that the arbitrage mechanism
and the creation–redemption process have allowed for an increase in the spot market role in price
discovery while at the same time have allowed for a decrease in the role of the futures market.
Finally, we conduct regression analysis to examine the relationships of ETF permanent-transitory
components to those of the index and the futures contracts. The analyses add to our understanding of the
connection between futures and spot markets to the ETF markets.

7. Temporal dynamics of ETF information shares

Our final analysis attempts to determine how the developing ETFs over time affect the spot and the
futures markets. Since we establish that the permanent-transitory components are non-stationary the
rates of change of the permanent-transitory components and the logarithms of the series are utilized in an
attempt to remove the influence of non-stationarity. The two techniques together help to remove the
autoregressive processes and non-stationarity in the error terms, but neither one alone can eliminate the
non-stationarity and ARCH effects by themselves. This might explain the lack of significance in some of the
results.
The rate of change removes the ARCH in all but the DJIA index, NDX futures, Russell 2000 and S&P 400
regressions. The logarithm of permanent-transitory components eliminates the ARCH effects in all regressions
but the S&P 500 and S&P 500 futures regressions. The rate of change helps to remove non-stationarity in the
error terms in all regressions whereas the logarithm removes non-stationarity only in the DJIA index, DJIA
futures, NDX, S&P 500, S&P 500 futures and Russell 2000 regressions.
Table 4 presents the regression results of relationships of ETF permanent-transitory component rates of
change and index and futures permanent-transitory component rates of change, as specified in Eqs. (10)
and (11). Note that the QQQQ results are based on the restricted sample of price discovery measures
because QQQQ price discovery abruptly declined after April 2009 and stabilized at the same level and as
such would not add to the analysis. Prior to that date QQQQ price discovery had been consistently
increasing. The restricted sample is based on the period prior to April 2009.
The results indicate that a positive correlation exist between the ETF rate of change of price discovery
and the S&P 500 index and the S&P 400 futures. In contrast, the ETF is negatively correlated to the DJIA
index, DJIA futures, Russell 2000 index, and the S&P 400 index. This implies that the higher the rate of
change of price discovery of the DIA ETF, the lower the rate of change of price discovery of the DJIA futures
 S.I. Ivanov et al. / Global Finance Journal 24 (2013) 171–187 183

Table 4 (continued)
SP500 SP500 futures R2000 R2000 futures SP400 SP400 futures

Coeff p-Value Coeff p-Value Coeff p-Value Coeff p-Value Coeff p-Value Coeff p-Value

0.122 0.030 0.031 0.264 0.023 0.124 0.028 0.400 0.018 0.078 0.018 0.504
0.026 b.0001 −0.001 0.568 −0.209 b.0001 0.085 0.435 −0.277 b.0001 0.201 0.049
0.203 0.002 0.145 0.006 0.308 0.035
0.300 0.962 0.016 0.586 0.016 0.758
b.0001 b.0001 b.0001 b.0001 b.0001 b.0001
b.0001 b.0001 b.0001 b.0001 b.0001 b.0001
b.0001 b.0001 b.0001 b.0001 b.0001 b.0001

contract. However, for the S&P 500, the higher the rate of change of price discovery of the SPY ETF, the
higher the rate of change of price discovery of the S&P 500 index. Similarly the higher the rate of change of
price discovery of the MDY S&P 400 ETF, the higher is the rate of change of the S&P 400 futures contract.
Therefore, the introduction of ETFs has influenced traditional price discovery indicated by the
statistically significant coefficients in the DJIA index and futures, S&P 500 index, Russell 2000 index and
S&P 400 index and futures regressions. More importantly, this shows that ETFs price discovery changes
affect the five indexes and futures price discovery differently.
The relationships are further supported by the regression results of ETF log permanent-transitory
components and index and futures log permanent-transitory components. Table 5 reports a positive
correlation between ETF and DJIA futures, NASDAQ 100 index and futures, S&P 500 index, and the S&P 400
index. It shows that the higher the log price discovery measure of the QQQQ the higher the price discovery
measure of the NASDAQ 100 index. Similarly, the higher the log price discovery measure of the NDX ETF
the higher are the NDX index and futures. Note that the QQQQ results are based on the restricted sample of
price discovery measures because QQQQ price discovery abruptly declined after April 2009 and stabilized
at the same level. The restricted sample is based on the period prior to April 2009.
Likewise, for the Russell 2000, the higher is the log price of IWM, the higher is the log price discovery
measure of the Russell 2000 index. Alternatively, the log price discovery of ETF is negatively correlated to
S&P 500 futures, Russell 2000 futures, and the S&P 400 futures. This implies that the higher the log price
discovery measure of the ETF the lower the log price discovery measure of the futures contract.
Overall, the relationships are statistically significant and the results show that each index and futures
are related differently to the respective ETF index. This adds to the evidence that one index cannot be used
as an inference to the other indexes, and a careful examination of each is necessary.
The combined results of rates of change and logarithms of price discovery measures indicate that
indeed the introduction of ETFs has affected the temporal behavior of price discovery. The temporal
relations across spot, futures and ETF markets price discovery metrics, however, seem to be very much
index specific. The temporal behavior of the S&P 500, S&P 400 and Russell 2000 ETFs price discovery is
affecting negatively the temporal behavior of the futures price discovery and positively the temporal
behavior of the respective price discovery in the spot market. The results are the opposite for the DJIA ETF
and futures and spot market. For the NDX ETF the temporal behavior of its price discovery is positively
related to the temporal behavior of the price discovery of both spot and futures markets. In support of the
univariate analyses the multivariate results indicate that the increased popularity of ETFs has had an
impact on the reduced role of the futures market as the dominant price discovery market.
184 S.I. Ivanov et al. / Global Finance Journal 24 (2013) 171–187

Table 5
This table reports regression results for the relationship between the (log) logarithms of price discovery measure of the ETF and the
(log) logarithms of the permanent-transitory components of index (or futures in Eq. (11)) measures. The equations are:

PT2;t ¼ γ0 þ γ1 PT1;t−1 þ lt ;

PT3;t ¼ ψ0 þ ψ1 PT1;t−1 þ mt ;

where PT1, t is the price discovery metric of the ETF, PT2,t is the price discovery metric of the index, and PT3,t is the price discovery
metric of the futures contract. The error term for Eq. (10) is mt and for Eq. (11) is 1t.

Logarithms DJIA DJIA futures NDX* NDX futures* SP500

Coeff p-Value Coeff p-Value Coeff p-Value Coeff p-Value Coeff

Intercept −1.066 b.0001 −0.110 0.737 −0.299 b.0001 Noint −0.149

Log ETF −0.298 b.0001 0.798 b.0001 0.106 b.0001 0.285 b.0001 0.662
R-sq 0.494 0.241 0.359 0.318 0.381
ARCH Resid 0.942 0.588 0.865 0.694 0.011
Zero mean b.0001 0.000 b.0001 0.482 0.000
Single mean 0.001 0.008 b.0001 0.675 0.003
Trend b.0001 0.001 b.0001 0.059 0.000
*Restricted sample *Restricted sample

SP500 SP500 futures R2000 R2000 futures SP400 SP400 futures

p-Value Coeff p-Value Coeff p-Value Coeff p-Value Coeff p-Value Coeff p-Value

As a final analysis, we combine the individual indexes into one dataset. Regression analyses are
conducted based on Eqs. (10a) and (11a) including dummy variables for the individual indexes.

PT2;t ¼ γ0 þ γ1 PT1;t−1 þ γ2 dDIAt þ γ 3 dQQQQ t þ γ4 dSPYt þ γ5 dIWMt þ lt ;

ð10aÞ

PT3;t ¼ ψ0 þ ψ1 PT1;t−1 þ ψ2 dDIAt þ ψ3 dQQQQ t þ ψ4 dSPYt þ ψ5 dIWMt þ mt ; ð11aÞ

where dDIA is a dummy variable with value of one if the ETF is DIA or zero otherwise, dQQQQ is a dummy
variable with value of one if the ETF is QQQQ or zero otherwise, dSPY is a dummy variable with value of
one if the ETF is SPY or zero otherwise and dIWM is a dummy variable with value of one if the ETF is IWM
or zero otherwise.
Table 6 presents the results for both rates of change and logarithms of price discovery measures. The
findings indicate that the increased popularity of ETFs has led to a declining role of the futures market as
the dominant price discovery market. The rate of change of ETF is not significantly correlated to the futures
market. Moreover, the dummy variables for the indexes are not statistically significantly correlated to the
futures rate of change. However, the rate of change is positive (+ 0.023) and statistically significant for the
index. We find that as the rate of change of the ETF price discovery rises the rate of change of the index also
increases.
It is further supported by the log regression results. Both index and futures price discovery regression
coefficients are statistically significant. However, the index log regression indicates that the log price
discovery measure of the ETF is positively correlated to the price discovery of the index (+0.147). In
contrast, the log futures contract regressions for the ETF price discovery is negatively related to the futures
contract price discovery (− 0.317). These results support the findings in the univariate analysis. All the
dummy variables are statistically significant with dDIA and dQQQQ having positive correlation to the
index log while dSPY, dIWM, and the intercept (for MDY) are negatively related. For the ETF regression on
the futures, the dummy variables, dDIA and dQQQQ are negatively related to the log of futures while dSPY
is positively correlated to the log of futures.7 In summary, the results imply that, as the popularity of ETFs

7
However, caution is necessary when interpreting the log results because the error terms have autoregressive features.
 S.I. Ivanov et al. / Global Finance Journal 24 (2013) 171–187 185

Table 5 (continued)
SP500 SP500 futures R2000 R2000 futures SP400 SP400 futures

p-Value Coeff p-Value Coeff p-Value Coeff p-Value Coeff p-Value Coeff p-Value

0.253 −2.266 b.0001 Noint −2.591 b.0001 Noint −2.829 b.0001

b.0001 −0.540 b.0001 0.550 b.0001 −0.606 b.0001 0.283 b.0001 −0.641 b.0001
0.275 0.538 0.266 0.769 0.330
b.0001 0.388 0.753 0.737 0.788
0.007 0.008 0.152 0.008 0.075
0.077 0.002 0.589 0.072 0.408
0.004 0.022 0.026 0.256 0.188

grew, it reduced the dominance of futures while increasing the role of the spot market over time.
Moreover, the findings show that each index and each futures contract are correlated differently to their
respective ETF. The reason might be due to the different level of ability to replicate an index and the
difference in replicating strategy that each ETF manager might be employing. For example, it is much
easier to replicate the composition of the 30 stocks in the DJIA but much more difficult to replicate the
exact portfolio composition of the Russell 2000.

8. Conclusion

This study extends the past studies by Chu et al. (1999), Chou and Chung (2006), Tse et al. (2006) and
Fung et al. (2008). We study the DJIA, NASDAQ 100, S&P 500, Russell 2000 and S&P 400 ETF information
shares in relation to their respective futures and underlying index information shares across time. The
Gonzalo and Granger (1995) (permanent-transitory components) and Hasbrouck (1995) (information
shares) methodologies are applied to intradaily data to determine whether the introduction of ETFs has
changed the price discovery in financial markets in recent times, and whether the market structures have
changed over time.
There is no agreement in the literature which market dominates, the futures or the spot market,
despite the numerous empirical studies in the field. This, to a certain extent, supports the strong
theoretical arguments suggesting that the futures market should provide the leadership. In theory it
is argued that informed traders are motivated to participate in the futures not the spot market.
However, since the introduction of ETFs, we find a consistently increasing role in the price discovery
process of the ETF market across time. The increased role of the ETF may explain the observed
reversal in the dominant market from the futures market to the spot market. The arbitrage
mechanism of creating and redeeming ETF units ensures that the ETF replicates the performance of
the underlying index. Because the process of ETF creation and redemption involves the spot market
and not the futures market we propose that this process helps explain the consistent increase in the
spot market price discovery and the subsequent decline in the price discovery of the futures market
over time.
We document that the impact of ETF trading has affected price discovery differently across indexes.
We find that the temporal behavior of the S&P 500, S&P 400 and Russell 2000 ETF price discovery is
affecting negatively the temporal behavior of the futures price discovery and positively the temporal
behavior of the respective price discovery in the spot market. The results are the opposite for the DJIA
186 S.I. Ivanov et al. / Global Finance Journal 24 (2013) 171–187

Table 6
This table reports regression results for the relationship between the rate of change (ROC) for the permanent-transitory components
and the ROC of ETF price discovery measure and four dummy variables representing each index, DIA, QQQQ, SPY, and IWM where
the dummy variable = 1.0 if it represents the index (DIA for example) or 0 otherwise. A second regression analysis is conducted
using the log or the logarithms of price discovery measures to replace ROC(PT2,t), ROC(PT1, t − 1), and ROC(PT3,t) in logs. The
equations are:

PT2;t ¼ γ0 þ γ1 PT1;t−1 þ γ 2 dDIAt þ γ 3 dQQQQ t þ γ 4 dSPYt þ γ5 dIWMt þ lt ;

PT3;t ¼ ψ0 þ ψ1 PT1;t−1 þ ψ2 dDIAt þ ψ3 dQQQQ t þ ψ4 dSPYt þ ψ5 dIWMt þ mt ;

where PT1, t is the price discovery metric of the ETF, PT2,t is the price discovery metric of the index, and PT3,t is the price discovery
metric of the futures contract. The error term for Eq. (10a) is lt and for Eq. (11a) is mt.

Rate of change Rate of change Logs Logs

Index p-Value Futures p-Value Index p-Value Futures p-Value

coefficient coefficient coefficient coefficient

Intercept 0.009 0.859 0.024 0.424 Intercept −0.336 b.0001 −2.215 b.0001
ROC ETF 0.023 b.0001 −0.001 0.690 LogETF 0.147 b.0001 −0.317 b.0001
dDIA −0.001 0.983 −0.001 0.977 dDIA 0.308 b.0001 −0.495 b.0001
dQQQQ 0.043 0.514 −0.003 0.944 dQQQQ 0.253 b.0001 −0.361 0.001
dSPY 0.116 0.070 0.006 0.881 dSPY −0.663 b.0001 0.317 0.003
dIWM 0.001 0.984 0.009 0.836 dIWM −0.542 b.0001 −0.023 0.839
R-sq 0.096 0.001 R-sq 0.331 0.185
Residual Residual
No ARCH 0.307 0.937 No ARCH b.0001 b.0001
Zero b.0001 b.0001 Zero b.0001 b.0001
mean mean
Single b.0001 b.0001 Single b.0001 0.001
mean mean
Trend b.0001 b.0001 Trend 0.000 0.003

ETF and futures and spot market. For the NDX ETF the temporal behavior of its price discovery is
positively related to the temporal behavior of the price discovery of both the NDX spot and NDX
futures markets.
A natural extension of this study is analyzing price discovery at even higher frequency level than the
one minute interval used in the study. There are multiple data sources today which aggregate and provide
data on millisecond interval level. A study based on the millisecond level will be of particular interest to
high frequency traders. As of the writing of this paper such high frequency data on spot, ETF and futures
instruments were not available but might be pursued in a future research.

References

Bernstein, J. (2009). Commodities ETFs: Diversification and hedging. (ETFzone.com, Published: Thursday, September 3, 2009,
Retrieved: December 22, 2009 from: http://www.etfzone.com/?template=viewarticle&article_id=971)
Chou, R. K., & Chung, H. (2006). Decimalization, trading costs, and information transmission between ETFs and index futures. Journal
of Futures Markets, 26, 131–151.
Chu, Q., Hsieh, W., & Tse, Y. (1999). Price discovery on the S&P 500 index markets: An analysis of spot index, index futures, and
SPDRs. International Review of Financial Analysis, 8, 21–34.
DeFusco, R., Ivanov, S., & Karels, G. (2011). The exchange traded funds' pricing deviation — Analysis and forecasts. Journal of
Economics and Finance, 35, 181–197.
Delcoure, N., & Zhong, M. (2007). On the premiums of iShares. Journal of Empirical Finance, 14, 168–195.
Engle, R., & Sarkar, D. (2006). Premiums–discounts and exchange traded funds. Journal of Derivatives, 13, 27–45.
Fang, Y., & Sanger, G. (2012). Index price discovery in the cash market. (Available at SSRN: http://ssrn.com/abstract=1926287 or
http://dx.doi.org/10.2139/ssrn.1926287 retrieved on May 11, 2013).
Figuerola-Ferretti, I., & Gonzalo, J. (2010). Modeling and measuring price discovery in commodity markets. Journal of Econometrics,
158, 95–107.
Fung, H., Liu, Q., & Tse, Y. (2008). An analysis of the price linkages among the DJIA index, futures, and exchange-traded fund markets.
The Review of Futures Markets, 16, 301–328.
Gastineau, G. (2004). The benchmark index ETF performance problem: A simple solution. Journal of Portfolio Management, 30,
96–103.
 S.I. Ivanov et al. / Global Finance Journal 24 (2013) 171–187 187

Gonzalo, J., & Granger, C. (1995). Estimation of common long-memory components in cointegrated systems. Journal of Business
Economics and Statistics, 13, 27–35.
Hasbrouck, J. (1995). One security, many markets: Determining the contributions to price discovery. The Journal of Finance, 50,
1175–1199.
Hasbrouck, J. (2003). Intraday price formation in U.S. equity index markets. The Journal of Finance, 58, 2375–2399.
Kumar, U., & Tse, Y. (2009). Single-stock futures: Evidence from the Indian securities market. Global Finance Journal, 20, 220–234.
Sapp, S. G. (2002). Price leadership in the spot foreign exchange market. The Journal of Financial and Quantitative Analysis, 37,
425–448.
Tse, Y., Bandyopadhyay, P., & Shen, Y. (2006). Intraday price discovery in the DJIA index markets. Journal of Business Finance &
Accounting, 33, 1572–1585.
Tse, Y., & Martinez, V. (2007). Price discovery and informational efficiency of international iShares funds. Global Finance Journal, 18,
1–15.
Tse, Y., & Xiang, J. (2005). Market quality and price discovery: Introduction of the E-mini energy futures. Global Finance Journal, 16,
164–179.