Content uploaded by Ritesh Kumar Dubey
Author content
All content in this area was uploaded by Ritesh Kumar Dubey on Apr 21, 2022
Content may be subject to copyright.
Vol.:(0123456789)
Asia-Pacific Financial Markets
https://doi.org/10.1007/s10690-021-09353-5
1 3
ORIGINAL RESEARCH
Algorithmic Trading Efficiency andits Impact
onMarket‑Quality
RiteshKumarDubey1 · A.SarathBabu2· RajneeshRanjanJha3·
UrvashiVarma4
Accepted: 25 September 2021
© The Author(s), under exclusive licence to Springer Japan KK, part of Springer Nature 2021
Abstract
Algorithmic Trading (AT) has been despised by retail traders and market regulators
for its speed. AT has taken the hit for creating un-intended volatility and hampering
the market quality due to skepticism of quote-stuffing and front-running, however in
reality the evidence pertaining to ill impacts of AT are yet to be found. This study
takes a step in the direction to decriminalize algorithmic trading and give AT it’s due
towards improvement in market quality. This study uses direct identification of AT
from Indian Stock Market (National Stock Exchange, NSE) and uses Order-to-Trade
Ratio (OTR) as a measure of AT efficiency and paves the way for regulators and
traders to come forward and appreciate the positive impact of AT on market quality.
Keywords Algorithmic trading· Algorithmic trading efficiency· High frequency
trading· HFT· Market quality· Emerging markets· Market microstructure· Order-
to-trade ratio
JEL Classification G10· G14· G15
* Ritesh Kumar Dubey
drriteshdubey84@gmail.com
A. Sarath Babu
sarathbabu@imthyderabad.edu.in
Rajneesh Ranjan Jha
rajneeshranjan.jha@ibsindia.org
Urvashi Varma
uvarma@amity.edu
1 GITAM Institute ofManagement, GITAM Deemed tobe University, Visakhapatnam,
AndhraPradesh530045, India
2 Institute ofManagement Technology (IMT), Hyderabad, Telangana501218, India
3 IBS Hyderabad, IFHE University, Hyderabad, Telangana501203, India
4 Amity Business School (ABS), Noida, UttarPradesh201313, India
R.K.Dubey et al.
1 3
1 Introduction
The transaction velocity of Algorithmic Trading (AT) has been the strength for argu-
ments pertaining to wide acceptance of AT in various markets. The sophistication of
algorithmic trading systems and advancements in the trading infrastructure has led
to the incredible capability of AT to revive its orders in a very short span of time.
This has led to unprecedented growth in the number of orders being placed, modi-
fied or cancelled by AT. Dubey etal. (2017) indicate the overwhelming nature of
share of AT responsible for order entry, order modification and order cancellation.
It is quite possible that these high number of orders being placed in the market may
lead to quote-stuffing1 and front-running2 and may dampen the expected benefits
of AT for the market quality. Hinz and Yi (2017) suggest the usage of an algorith-
mic approach towards optmal asset allocation. It is also observed that many market
regulators come up with stringent regulations on orders to trade or execution ratios
(OTR) in order to avoid quote stuffing or piling up of orders without leading them to
an actual trade. In some markets, there are monetary and trade prohibiting penalties
for breaching the desired level of order-to-trade ratios in order to keep a check/con-
trol over market quality. On the other hand, one may argue that relentless placement
of orders is actually undermining the ability or efficiency of algorithmic traders.
Nawn and Banerjee (2019b) suggest that proprietary algorithmic traders are more
informative and examine the role of algorithmic traders in the price discovery pro-
cess. Saito et al. (2018), Kelejian and Mukerji (2016), Hendershott et al. (2011),
Aggarwal and Thomas (2014, Groth (2011), Upson and Van Ness (2017) provide
inconclusive evidence on algorithmic trading’s impact on the market quality. The
reason for inconclusive or unreliable evidence can be attributed primarily due to
indirect/proxy measurement of AT. The existing literature explores the impact of AT
on market efficiency but doesn’t focus much on the efficiency of AT. There is limited
research on algorithmic trading efficiency in the extant literature. We are the first to
measure algorithmic trading efficiency with our inverse of OTR measure (1/OTR).
Our study is one of the few amongst existing literature, which identifies algorithmic
trading precisely due to the regulatory framework prevalent in the Indian Stock Mar-
ket. Lesmond (2005), Lee (2011), Lang etal. (2012) characterize emerging markets
as volatile, illiquid and informationally inefficient. Subrahmanyam (2013) also opine
that AT is often viewed as a threat to financial market stability. Therefore, with the
rapid economic growth in emerging markets and their increasing popularity among
investors (Kang & Zhang, 2014), it becomes imperative to study the algorithmic
trading efficiency and its impact on the market quality. Another prominent reason
to examine the algorithmic trading efficiency is the inherent logic for adoption of
newer technologies like AT by the emerging markets to enhance the overall market
quality by improving liquidity and price discovery and tackling the volatility.
1 Quote-stuffing in high frequency literature is described as placing orders at a very high frequency with
no intention of trading on those orders and then immediately cancelling them.
2 Front-running strategy intends to executing an order or part of the order by taking the advantage of
speed and driving the price in a direction which profits them. This strategy may be based on insider
information or information just ahead of the order to be placed in the market.
1 3
Algorithmic Trading Efficiency andits Impact on…
Similar to AT Intensity, order-to-trade ratio can be regarded as a measure of AT
activity as more than 99% of orders and more than 75% of trades are contributed
by AT (Dubey etal., 2017). Thus, order-to-trade ratio is also expected to have an
impact on the liquidity volatility and price discovery. One can argue that algo trad-
ing can be deemed efficient only if they convert their orders into trades. Because,
any price related or other information that AT orders are trying to incorporate in
the market, will be incorporated only if trades happen or else it will be just a case of
quote stuffing or piling up of the orders. Therefore, our measure of orders to trade
ratio (OTR) holds a good logical stronghold as a parameter to ascertain the algorith-
mic trading efficiency.
In this study, we have the advantage of direct identification of AT due to Securi-
ties and Exchange Board of India (SEBI). The Indian market regulator mandates for
clear identification of AT and tagging them with unique identifier to create an audit
trail. We use tick by tick data for orders and trades stamped close to micro second
(1/65636th fraction of a second). We use Order-to-trade Ratio (OTR3) and devise 1/
OTR as a measure of algorithmic trading efficiency to assess the number of actual
trades happening per submitted order by AT. We also examine the impact of algo-
rithmic trading efficiency on market quality characteristics namely liquidity, volatil-
ity and price discovery.
The rest of the paper is organized as: Sect.2 discusses the acceptance of AT in
India. Section3 discusses literature on AT, AT efficiency, research gaps and hypoth-
esis development. Section4 describes the data, variables and methodology used in
this study. Section5 illustrates the results, interpretations and discussion and; Sec-
tion6 provides concluding remarks on AT efficiency, its relevance and its impact on
market quality.
2 The Rise ofATinIndia
Algorithmic Trading (AT) in India was launched by Credit Suisse’s Advanced
Execution Services (AES) on 22nd June 2009. The momentum for AT in India
was gained by the allowance of co-location4 facilities by NSE in June 2010. Co-
location facility allows the broker member servers’ to be placed side by side along
with the exchange’s server in order to reduce latency. Dubey (2016) reports that by
August 2013 in Nifty 50 stocks more than 95% (2.04 bn) of the orders were being
placed using AT and more than 75% (48.4mn) of the trades were also AT (by fre-
quency, Total Orders: 2.14 bn, Total Trades: 64.1mn). Dubey, Chauhan, Syamala
(2017) indicate the rising participation of AT in the Indian market and observe that
by August, 2013 AT contributed to 96% of the overall orders being placed and it
also accounted for more than 75% of the trades on NSE for the NIFTY50 stocks. In
less than 4years of AT’s introduction in India, we observe a significant rise of AT
3 Order-to-trade ratio is defined as the number of orders placed (Entry + Cancel + Modify) in the market
divided by the number of trades that got executed.
4 Co-location at exchange premises is the mechanism used by exchanges to achieve lower latencies and
faster execution.
R.K.Dubey et al.
1 3
(Dubey, 2016) whereas Hendershott etal. (2011) report similar rise in the Ameri-
can market more than a decade of introduction of AT. The initial evidence of AT
participation in equity orders and trades is overwhelming for the Indian market and
definitely requires research on its impact on market quality and its efficiency.
3 Literature Review andHypothesis Development
We do not find any literature that directly relates to OTR’s impact on liquidity vola-
tility and price discovery. However, Hendershott and Riordan (2013) suggest that
AT consumes liquidity when it is cheap (i.e., when the bid-ask quotes are narrow)
and supply liquidity when it is expensive. They argue that when spreads are narrow
ATs are less likely to submit any new orders, less likely to even cancel their orders,
and more likely to initiate the trades. Wu and Siwasarit (2019) indicate the non-
spontaneous creation of market efficiency and they also suggest Order Imbalance
(OI) indicator as a measure to identify price discovery especially in emerging mar-
kets. In congruence with the above findings of the authors, one can expect that, the
OTR to decline when spreads are large and increase when spreads are narrow.
Dubey, Chauhan and Syamala, (2017) find that during August 2013 the order-
to-trade ratio of algo and non-algo orders (in the Nifty 50 stocks) were 42.20 and
6.08 respectively. The order-to-trade ratio increased for algo orders by 17% (36.21 in
September 2012–42.20 in August 2013) in a year’s span. Whereas during the same
period the orders to trade ratio of non-algo orders declined by 2% (6.19 in Septem-
ber 2012–6.08 in August 2013). They argue that, the decline in orders to trade ratio
for non-algo traders may indicate that AT is doing the quote stuffing to create fake
liquidity and then taking advantage of its speed and gets its orders traded leaving
most of the non-algo orders to remain untraded. The strategies of quote-stuffing
and front-running are expected to hamper the true price discovery of the securi-
ties. Authors suggest that the declining order-to-trade ratio is a cause of concern for
financial market regulators, who aim to create efficient markets and a level playing
field for all market participants. Furthermore, the government and economic poli-
cymakers should keep a tab on the increasing dominance of AT, which might inflict
the decline of non-algorithmic/manual traders and eventually eliminate them in the
long run.
By mere realization that the advent of AT was for improving upon the trading
efficiency and thereby improving upon the market efficiency may not be sufficient.
Therefore, we are the first to use OTR as a measure of algorithmic trading efficiency
and empirically examine the impact of OTR on liquidity, volatility and price discov-
ery measures. Based on the above discussion we propose the following hypothesis
to test the impact of algorithmic trading efficiency on market quality dimensions:
(Note: Higher OTR refers to lower algorithmic trading efficiency (1/OTR), therefore
higher OTR is expected to create liquidity crunch in markets, which is undesirable).
Hypothesis 1: Order-to-trade ratio is inversely related to liquidity
1 3
Algorithmic Trading Efficiency andits Impact on…
Hypothesis 1a: For the Nifty50 stocks, quoted spread declines with the decline in
order-to-trade ratio.
Hypothesis 1b: For the Nifty50 stocks, proportional quoted spread declines with the
decline in order-to-trade ratio.
Hypothesis 1c: For the Nifty50 stocks, proportional effective spread declines with
the decline in order-to-trade ratio.
Hypothesis 1d: For the Nifty50 stocks, proportional realized spread declines with
the decline in order-to-trade ratio.
Hypothesis 1e: For the Nifty50 stocks, order depth increases with the decline in
order-to-trade ratio.
Hypothesis 1f: For the Nifty50 stocks, order imbalance declines with the decline in
order-to-trade ratio.
Market regulators worry about volatility as much as liquidity (Groth, 2011).
Nawn and Banerjee (2019b) suggest that AT does not reduce the liquidity supply
during volatile markets, thus indicating standard or enhanced participation of AT
during volatile periods. Market regulators (NSE, BSE, Eurex, and others) have often
developed strict regulations for AT to adhere to the benchmark orders-to-trade or
execution ratios (OTR) to avoid monetary penalties. AT’s ability to quickly gather
new information and incorporate the same in the orders may lead to subsequent
quote stuffing or piling up orders on one side without leading them to actual trade.
Quote-stuffing will drive the prices to extremes within a fraction of seconds, thereby
creating a volatile market that will hamper the liquidity and contribute to adverse
selection. In some markets, the regulators impose monetary and trade prohibiting
penalties for breaching the desired levels of order-to-trade ratios to check market
volatility. Therefore, we propose that if AT is efficient, it should not contribute to
impending volatility in the market. Hence, we propose the following hypothesis:
(Note: Higher OTR will mean lower algorithmic trading efficiency (1/OTR)).
Hypothesis 2: Order-to-trade ratio is directly proportional to volatility
Hypothesis 2a: For the Nifty50 stocks, the volatility increases with increase in
order-to-trade ratio.
The arguments of piling orders on one side of the trade create skepticism about
the relationship between the order-to-trade ratio and price discovery (Dubey etal.,
2017; Egginton, Van Ness, and Van Ness 2016). Therefore, with increasing order-
to-trade ratio, it is quite possible that the scope of adverse selection also increases.
Thereby, higher order-to-trade ratio may impact the price discovery adversely. The
price efficiency or market efficiency is also a major concern for the market regulators
R.K.Dubey et al.
1 3
and exchanges. And, that is why they often penalize for higher order-to-trade ratios
which essentially leads to various market imperfections. Therefore, from the above
discussion we formulate the following hypothesis: (Note: Higher OTR will mean
lower algorithmic trading efficiency (1/OTR)).
Hypothesis 3: Order-to-trade ratio is inversely related to price discovery
Hypothesis 3a: For the Nifty50 stocks, the adverse selection increases with order-
to-trade ratio.
4 Data, Variable Construction, andMethodology
We examine Nifty 50 stocks listed on the National Stock Exchange of India (NSE)
as they are most actively traded and are a representative sample for the various
industries and sectors of the economy. Nifty50 captures more than 65% of NSE’s
float-adjusted market capitalization and is a true reflection of the Indian stock mar-
ket. “The NIFTY 50 covers major sectors (23) of the Indian economy and offers
investment managers exposure to the Indian market in one efficient portfolio5”. Hen-
dershott etal. (2011) suggest that AT is more active in liquid stocks, and therefore
Nifty50 stocks are a suitable choice for the study. Furthermore, from our analysis
in the current study, it is evident and coherent with the existing literature (Dubey
etal., 2017) that more than 95% of the orders in Nifty50 are AT orders. The data
pertaining to algorithmic trading is obtained from NSE DOTEX from their Order
Level Historical Dataset for Sep, 2012–Aug, 2013 (12 Months, 248 trading days).
We clean the dataset by omitting the stocks which has inconsistent (missing obser-
vation and data) data even on a single trading day during the data sample. Finally,
we arrive at a dataset of 49 NSE Nifty Stocks.
The NSE DOTEX dataset contains standard details of tick-by-tick order level
data, trade level data pertaining to CashWith our data set we also/Capital Market
segment for all stocks listed on exchange (NSE). The dataset comes with a unique
identification flag for AT (0 for Algo, 1 for Non-Algo). The dataset captures high
frequency trade and order level information up to the 65536th fraction of a second.
The order level data contains information on 17 different variables and the trade
level data provides information on 14 different variables as indicated in Appendix 1.
Our dataset time frame coincides with Nawn and Banerjee (2019b),6 Dubey etal.,
(2017) and it also overlaps with Nawn and Banerjee (2019b)7 and therefore, this
study would also contribute by comparing and validating the findings of the existing
5 NIFTY 50 Index Methodology. (2019). https:// www1. nsein dia. com/ conte nt/ indic es/ Method_ Nifty_ 50.
pdf
6 Nawn, S., & Banerjee, A. (2019b). Do the limit orders of proprietary and agency algorithmic traders
discover or obscure security prices? Journal of Empirical Finance, 53, 109–125 (Data Period: November
2012 and December 2012).
7 Nawn, S., & Banerjee, A. (2019b). Do proprietary algorithmic traders withdraw liquidity during mar-
ket stress? Financial Management, 48(2), 641–676. (Data Period: January 2013–December 2013).
1 3
Algorithmic Trading Efficiency andits Impact on…
literature. With our data set we also resolve the criticism of small sample in exist-
ing literature (Groth, 2011 (5 days), Brogaard et al., 2014 (5 days), Aggarwal &
Thomas, 2014 (59 days), Dubey et al., 2017 (40 days) by incorporating a longer
duration of 125 trading days. The longer duration of data sample will enhance the
reliability of our results and especially because we have a direct identification of
AT. SEBI’s circular dated 30th March 2012 makes it mandatory for “all algorith-
mic orders to be tagged with a unique identifier provided by the stock exchange in
order to establish audit trail”. This circular mandates for a clear identification of AT
and also defines AT for the first time. Hence, our data set for the period 1st Sep-
tember 2012–31st August 2013 captures the unambiguous and continuous data on
AT. Given the clear flag for AT, the concerns regarding the reliability and validity
of proxy (as a measure of AT) and quantification of its impact on liquidity is elimi-
nated, unlike the other studies on algorithmic trading in the past. For the firm level
data, we use Centre for Monitoring Indian Economy (CMIE) Prowess. We collect
the data for control variables such as market capitalization, market to book ratio,
share turnover, and price of the stocks.
We define order-to-trade ratio (
OTRit
) as the number of orders placed
(Entry + Cancel + Modify) in the market divided by the number of trades that got
executed for the stock i in the time interval t. We interpret 1/
OTRit
as measure of
algorithmic trading efficiency and therefore lower
OTRit
will be signifying higher
algo trading efficiency and vice-versa. We use liquidity volatility and price dis-
covery measures which are extensively used in literature and are also regarded as
benchmark measures. We use the measures, which will be suitable for short-term
(intra-day / high frequency) liquidity measurements (refer Table 1and Appendix2for
details on various measures and variables). We use quoted spread, proportional
quoted spread, proportional realized spread, proportional effective spread, depth
and order imbalance as various liquidity measures. We use the Parkinson (1980)
HI-LO volatility measure which is used extensively in high frequency literature and
we also use frequency of trades as another volatility measure as suggested by Jones
etal., (1994) and Chordia etal., (2000). We use adverse selection, the number of
trades (trading frequency) and trading volume as measures of informed trading,
which are extensively used in literature and are also regarded as benchmark meas-
ures. We use these high-frequency measures as the speed of the AT is very high and
the transactions can happen in fractions of seconds and the impact of AT is expected
to be observed in short time intervals (1s/30Second/1min). These liquidity meas-
ures are also widely accepted in existing high frequency and algorithmic trading
literature (Hendershott, etal., 2011; Aggarwal & Thomas, 2014; Kang & Zhang,
2014; Goyenko etal., 2009). Since we have continuous orders data and trades data,
we construct the 1ms matched limit order book (LOB) which contains information
on all outstanding buy and sell orders for the Nifty50 stocks. The construction of the
limit order book helps us in estimation of the various market quality measures.
However, to establish the algo trading efficiency, we need to investigate if orders
to trade ratio has any impact on liquidity, volatility and price discovery measures.
We expect that with increase in order-to-trade ratio, liquidity, volatility and price
discovery benefits decline. Therefore, we expect higher spreads (lower liquidity),
higher volatility and higher price impacts (poor price discovery and higher adverse
R.K.Dubey et al.
1 3
Table 1 Summary Statistics (Liquidity, Volatility and Price Discovery Measures, Quintile Wise)
This table presents summary statistics for the 49 constituents of the NSE Nifty 50 index between March 1, 2013 and August 31, 2013. The dataset contains order level data
from National Stock Exchange provided by NSE DOTEX along with market capitalization, share turnover, market to book ratio and adjusted closing prices data obtained
from CMIE Prowess database. The dataset is sorted into quintiles based on market capitalization, where quintile 5 contains largest-cap stocks. All the variables are 99%
winsorized.
# observations: 49 * 125 * 22,500 (stock * day * seconds)
Variable Description Source Mean Mean Q1 Mean Q2 Mean Q3 Mean Q4 Mean Q5 Symbol
qspreadit
Quoted half spread DOTEX 2.02 1.55 1.98 2.23 2.12 2.14 QS
pqspreadit
Proportional quoted spread DOTEX 0.37 0.41 0.47 0.33 0.36 0.33 PQS
pespreadit
Proportional effective half spread (× 10–6) DOTEX 0.00 − 39.69 − 87.57 102.78 − 88.97 222.43 PES
prspreadit+30
Proportional realized half spread, 30 Seconds (× 10–6) DOTEX 0.00 − 4.58 21.36 6.90 − 3.99 311.95 PRS_30Sec
prspreadit+60
Proportional realized half spread, 1 Minute (× 10–6) DOTEX 0.00 13.19 14.17 10.73 − 11.04 321.35 PRS_1Min
depthit
Quoted order depth DOTEX 91.62 82.87 79.99 72.53 88.14 112.67 DEPTH
OIit
Quoted order imbalance DOTEX − 10.38 − 10.42 − 5.07 − 10.69 − 10.95 − 13.31 OI
Volatilityit
Parkinson 30 Second HI-LO volatility (× 10–4) DOTEX 0.00 1.11 1.09 1.98 0.86 2.57 VOL
numtradesit
Number of trades DOTEX 26.70 25.08 30.06 19.16 23.30 30.22 NUMTRD
Adv_Selectionit+30
Adverse selection component half spread, 30 Sec (× 10–4) DOTEX 0.00 0.25 0.96 12.11 − 0.82 4.79 AdvSel_30
Adv_Selectionit+60
Adverse selection component half spread, 1 Min (× 10–4) DOTEX 0.00 0.12 0.78 11.16 − 0.86 4.73 AdvSel_60
Trd _Fre qit
Trade frequency DOTEX 26.70 25.08 30.06 19.16 23.30 30.22 TrdFreq
Trd _Vo lit
Volume of trades (× 104) DOTEX 67.76 56.70 69.49 43.50 58.62 86.72 TrdVol
AT_Intensityit
Algorithmic trading intensity DOTEX 69.74 64.33 63.13 78.10 72.38 74.13 AT_Intensity
OTRit
Order-to-trade Ratio DOTEX 22.92 21.64 18.96 23.14 24.94 24.87 OTR
MBRatioit
Log of Market to book ratio (daily) Prowess 0.97 0.52 0.91 1.16 0.82 1.23 MBRatio
MCapit
Log of Market capitalization (Rs. Billion, daily) Prowess 13.38 12.55 13.05 13.23 13.26 14.13 MCap
ShrTOit
Log of Share turnover (daily) Prowess 14.91 15.04 15.14 14.08 14.63 15.20 ShrTO
PInverseit
Log of Inverse of closing price (INR, daily) Prowess − 5.82 − 5.89 − 5.57 − 6.20 − 5.79 − 5.82 PInverse
1 3
Algorithmic Trading Efficiency andits Impact on…
selection) with increase in OTR. The algo efficiency can be interpreted if it happens
otherwise, which means decline in OTR leads to lower spreads, lower volatility and
lower price impact or adverse selection.
We investigate the relationship between OTR, liquidity, volatility and price dis-
covery measures using the following regression equations:
where i = 1, 2, 3, …. 49 and t = 1, 2, 3, …. 22,500 for each of the 125 trading days.
Lit
denotes each of the liquidity measures,
Vit
denotes each of the volatility meas-
ures and
Pit
denotes the price discovery measures for the given stock i at the given
second t on the trading day.
We run the regression analysis at 5 different levels with OTR being regressed on
1) The overall sample, 2) Quintile wise, 3) Small, Mid and Large cap categoriza-
tion wise, 4) Fama–MacBeth regression (date-wise) and 5) Fama–MacBeth regres-
sion (stock-wise). Taking a cue from the existing literature (Hendershott, et al.,
2011; Aggarwal & Thomas, 2014; Kang & Zhang, 2014; Goyenko et.al., 2009) we
use control variables including log of market capitalization (
MCapit)
, log of mar-
ket to book ratio (
MBRatioit)
, log of share turnover (
ShrTOit)
and log of 1/price
(
PInverseit)
in our regression model to control for their impact on the liquidity, vola-
tility and price discovery measures. We also control for the stock fixed effects and
time fixed effects.
5 Results andDiscussion
We investigate the impact of OTR on the various liquidity, volatility and price
discovery measures. We create these measures from our continuous order book
matched from the orders data and trades data and our dataset details information
matched up to milliseconds’ interval. For our analysis, we also sort the stocks into
quintiles (quintile 1 corresponds to small cap stocks and quintile 5 corresponds to
large cap stocks) based on market capitalization and we also create another segrega-
tion and categorize them into large cap, mid cap and small cap stocks. The market
capitalization of the stocks and the categorization are indicated in APPENDIX 3
and 4. In the Table1 we report the quintile wise mean and in Table2 we report the
descriptive statistics of all the variables used for our analysis. The descriptive statis-
tics indicate that mean of most of the variables is highest for the quintile Q5 (large
cap). We also indicate significant correlation between all the variables in Table3.
From Table4, Table5 and Table6, we find that the coefficient
𝜷
is significant at 1%
for each of the equations corresponding to the liquidity, volatility and price discov-
ery measures respectively.
(1)
Lit
=
𝛼it
+
𝛾it
+
𝛽OTRit
+
𝛿1MCapit
+
𝛿2MBRatioit
+
𝛿3ShrTOit
+
𝛿4PInverseit
+
𝜀it
(2)
Vit
=
𝛼it
+
𝛾it
+
𝛽OTRit
+
𝛿1MCapit
+
𝛿2MBRatioit
+
𝛿3ShrTOit
+
𝛿4PInverseit
+
𝜀it
(3)
Pit
=
𝛼it
+
𝛾it
+
𝛽OTRit
+
𝛿1MCapit
+
𝛿2MBRatioit
+
𝛿3ShrTOit
+
𝛿4PInverseit
+
𝜀it
R.K.Dubey et al.
1 3
The significant
𝜷
essentially confirms our primary hypothesis that OTR impacts
liquidity, volatility and price discovery. Next, we examine the algorithmic trading
efficiency, where higher OTR is being treated as lower efficiency and lower OTR
is deemed as higher efficiency. Therefore, we investigate for the directions (sign) of
the
𝜷
coefficients. We expect that with rise (increase) of OTR, the liquidity (spread),
volatility and adverse selection should increase and hence show positive ( +) sign.
The negative (−) sign would indicate higher the OTR, lower will be the expected
liquidity (spread), volatility and price discovery measures.
From the Table4 it can be observed that all the signs of the coefficient
𝜷
are
as expected ( +) except for proportional quoted spread and order imbalance. The
𝜷
coefficient of
OTRit
for quoted spread is observed to be significant and is 44.0444
(bps). This indicates that with a unit rise/decline in OTR the quoted spread also
rises/declines by 44.0444 basis points (bps). Similarly, the
𝜷
coefficient of
OTRit
for
proportional effective spread, proportional realized spread (30-s, 1-Min) and depth
are observed to be 0.0339 bps 0.0361 bps 0.0362 bps and 7448.829 bps respectively.
The findings indicate that with increase in algorithmic trading efficiency (decline
in OTR) the liquidity improves. The negative coefficient of order imbalance can be
interpreted as the basic nature of AT where it is not essentially piling up the orders
for only one side of the trade. From the Table5, we observe that a unit rise/decline
in OTR leads to 0.0063 basis points rise or decline in the (Parkinson’s Hi-Lo)
volatility. It indicates that with rise in algorithmic trading efficiency the volatility
Table 2 Descriptive Statistics
(Liquidity, Volatility and Price
Discovery Measures)
#Observations: 49*125*22,500 (stock * trading days * seconds)
(@Rounding Off to Nearest Integer, $- sign indicates buy side domi-
nance)
Variables Mean@Std. Dev Minimum@Maximum@
QS 2 2.65 0 10
PQS 0 0.52 0 39
PES 0 0.16 − 416 417
PRS_30Sec 0 0.04 − 15 16
PRS_1Min 0 0.04 − 15 16
Depth 92 118.30 1 1985
OI$ − 10 88.43 − 199 200
Vol_30Sec 0 0.01 0 4
NumTrd 27 67.28 1 3618
AdvSel_30Sec 0 0.24 − 1 419
AdvSel_1min 0 0.23 − 1 418
TrdFreq 27 67.28 1 3618
TrdVol 6776 × 1024791 × 10351689 × 107
MBRatio 1 0.91 0 4
ShrTO 15 1.06 8 18
PInverse − 6 0.83 − 7 − 3
AT_Intensity 70 4.71 35 79
OTR 23 51.89 0 3313
1 3
Algorithmic Trading Efficiency andits Impact on…
Table 3 Pairwise correlations of all variables for Nifty 50 stocks
All the correlation coefficients are found to be significant at 1%, i.e., ***)
This table represents the correlation statistic of the liquidity, volatility, and price discovery variables used in our study for the Nifty 50 Stocks
#Observations: 49*125*22,500 (stock*trading days*seconds)
_NAME_ QS PQS PES PRS_30Sec PRS_1Min Depth OI Vol_30Sec NumTrd AdvSel_30Sec AdvSel_1min TrdFreq TrdVol OTR
QS 1.000
PQS 0.454 l.000
PES 0.000 0.004 1.000
PRS_30Sec 0.002 − 0.001 0.099 l.000
PRS_1Min 0.002 − 0.001 0.124 0.568 1.000
Depth 0.175 − 0.093 0.001 0.001 0.001 1.000
OI − 0.049 0.011 0.000 − 0.001 0.000 − 0.071 1.000
Vol_30Sec 0.008 − 0.001 0.044 0.233 0.234 0.005 − 0.001 1.000
NumTrd − 0.016 − 0.010 0.000 − 0.002 − 0.001 0.263 0.003 − 0.001 l.000
AdvSel_30Sec 0.002 0.027 0.283 − 0.013 − 0.004 0.000 − 0.001 0.000 0.000 l.000
AdvSel_1min 0.002 0.028 0.391 − 0.009 − 0.006 0.000 − 0.001 − 0.001 0.000 0.988 1.000
TrdFreq − 0.016 − 0.010 0.000 − 0.002 − 0.001 0.263 0.003 − 0.001 l.000 0.000 0.000 l.000
TrdVol 0.018 − 0.012 0.000 − 0.001 0.000 0.169 − 0.006 0.001 0.326 0.000 0.000 0.326 1.000
OTR 0.137 − 0.035 0.001 0.006 0.006 0.338 − 0.046 0.004 − 0.124 0.000 0.000 − 0.124 − 0.038 1.000
R.K.Dubey et al.
1 3
declines, which is a good sign for all the market participants. The other frequency-
based measure of volatility shows a significant but contradictory indication. We find
that a unit rise in the OTR leads to 1480.890 basis points decline in the frequency of
the trades. This may be a reason to worry as it may indicate the quote stuffing aspect
of the algo orders. The observed relationship between the price discovery meas-
ures and algorithmic trading efficiency seems to be working out perfectly for the
adverse selection/price impact measures but it again shows an unexpected sign for
the coefficients for the frequency-based price discovery measures (trade frequency
and trade volume). Both, 30s and 1-min adverse selection measures show positive
impact of the algorithmic trading efficiency. From the Table 6, we find that a unit
rise or decline in the algorithmic trading efficiency measure (OTR) leads to 0.0043
basis points rise or decline in 30 s’ adverse selection and 0.0069 basis points rise
or decline in 1-min adverse selection. Thus, higher algorithmic trading efficiency
(lower OTR) results in lower price impacts which is clearly a good sign.
Table 4 Impact of Algorithmic Trading Efficiency on Liquidity
This table shows the impact of algorithmic trading efficiency (Order-to-trade Ratio (OTR)) on various
liquidity variables. The table regresses the various measures of the liquidity (half spreads, depth and
order imbalance) on our algorithmic trading measure. It is based on 1s observations for Nifty 50 stocks
from March, 2013 to August, 2013 which covers 2 quarters (6months or 125 trading days) of high fre-
quency data with precision of 1 jiffy (1 Second = 65,536 jiffies). The algorithmic trading is directly iden-
tified by the flag provided by the stock exchange (NSE). The specification for examining the impact is:
Lit =𝛼it +𝛾it +𝛽OTRit +𝛿1MCapit +𝛿2MBRatioit +𝛿3ShrTOit +𝛿4PInverseit +𝜀it
where
Lit
is either quoted spread, proportional quoted spread, proportional realized spread, and pro-
portional effective spread, depth or order imbalance as liquidity measures for stock i in second t of the
trading day. (The trading hours of NSE is from 9:15 am to 3:30pm which essentially means 6h 15 min
or 22,500s.)
OTRit
is the order-to-trade ratio for the stock i in second t of the trading day (where, OTR
is measured as ratio of total number of orders placed for a stock i in the time interval t to total number of
trades that took place in a stock i in the time interval t).We also use a vector of control variables includ-
ing log of market capitalization (
MCapit)
, log of market to book ratio (
MBRatioit)
, log of share turno-
ver (
ShrTOit)
and log of 1/price (
PInverseit)
. We also include stock fixed effects and time fixed effects.
t-values are also examined as they are based on standard errors that are robust to general cross section
and time-series heteroscedasticity and within group autocorrelation (Arellano and Bond, 1991). */**/***
denote significance at 10% / 5% / 1% level
# observations: 49 * 125 * 22,500 (stock * day * seconds)
P-value: < 0.0001
Liquidity Measures Coefficient on OTR Coefficients on Control Variables
OTRit
(× 10–4/bps)
MCapit
MBRatioit
ShrTOit
PInverseit
qspreadit
44.0444*** 0.4173*** − 0.2970*** − 0.5000*** − 0.5189***
pqspreadit
− 1.2904*** − 0.0707*** 0.0153*** 0.0608*** 0.0617***
pespreadit
0.0339*** 0.0001*** − 0.0001*** − 0.0001*** 0.0001***
prspreadit+30
0.0361*** 0.0002*** − 0.0001*** − 0.0001*** 0.0002***
prspreadit+60
0.0362*** 0.0002*** − 0.0001*** − 0.0001*** 0.0003***
depthit
7448.829*** 18.0798*** − 9.3260*** 18.5846*** − 25.1654***
OIit
− 599.4759*** − 3.9842*** 2.0927*** 2.2771*** 4.4114***
1 3
Algorithmic Trading Efficiency andits Impact on…
Further we investigate the relationship of algorithmic trading efficiency (1/OTR)
with liquidity, volatility and price discovery measures across the quintiles based
on market capitalization. We report our findings in Table 7, Table 8 and Table9,
where we observe a consistent relationship between OTR and liquidity, volatility
and price discovery for the largest quintiles. For the liquidity measures, we observe
(Table 7) significant improvements in liquidity measures across the various quin-
tiles with increase in algorithmic trading efficiency (lower OTR). The proportional
quoted spread seems to follow our expectation for the smaller quintiles (Q1, Q2, and
Q3); however, it contradicts our expectations for the large-cap stocks (Q4 and Q5).
Similar to our findings at the aggregate level, we observe negative coefficients for
the order imbalance, indicating the support for non-piling of orders on one side of
the trade.
Similarly, from Table 8, we find algorithmic trading efficiency to significantly
affect volatility for all the quintiles, but only the largest cap quintile (Q5) indicates
that with an increase in OTR, the volatility measure will also increase. Therefore, we
suggest that one should vouch for lower OTR. The inverse relationship is observed
between the OTR and volatility in the smaller quintiles and also between OTR and
the frequency-based measure of volatility. Similar to our findings for liquidity and
volatility, even for price discovery measures the coefficient of OTR is significant for
large cap stocks and follows the expected sign. For the smaller quintiles, it is either
insignificant or doesn’t follow the expected signs. From the Table 9 we find that
Table 5 Impact of Algorithmic Trading Efficiency on Volatility
This table shows the impact of algorithmic trading efficiency (Order-to-trade Ratio (OTR)) on various
volatility variables. The table regresses the various measures of the volatility (30-s HI-LO volatility,
number of trades) on our algorithmic trading measure. It is based on 1s observations for Nifty 50 stocks
from March, 2013 to August, 2013 which covers 2 quarters (6months or 125 trading days) of high fre-
quency data with precision of 1 jiffy (1 Second = 65,536 jiffies). The algorithmic trading is directly iden-
tified by the flag provided by the stock exchange (NSE). The specification for examining the impact is:
Vit =𝛼it +𝛾it +𝛽OTRit +𝛿1MCapit +𝛿2MBRatioit +𝛿3ShrTOit +𝛿4PInverseit +𝜀it
where
Vit
is either Parkinson’s (1980) HI-LO volatility measure capturing 30s HI and LO or number of
trades (numtradesit) for stock i in second t of the trading day. (The trading hours of NSE is from 9:15 am
to 3:30pm which essentially means 6h 15min or 22,500s.)
OTRit
is the order-to-trade ratio for the stock
i in second t of the trading day (where, OTR is measured as ratio of total number of orders placed for a
stock i in the time interval t to total number of trades that took place in a stock i in the time interval t).
We also use a vector of control variables including log of market capitalization (
MCapit)
, log of market
to book ratio (
MBRatioit)
, log of share turnover (
ShrTOit)
and log of 1/price (
PInverseit)
. We also include
stock fixed effects and time fixed effects. t-values are also examined as they are based on standard errors
that are robust to general cross section and time-series heteroscedasticity and within group autocorrela-
tion (Arellano and Bond, 1991). */**/*** denote significance at 10% / 5% / 1% level
# observations: 49 * 125 * 22,500 (stock * day * seconds)
P-value: < 0.0001
Volatility Measures Coefficient on OTR Coefficients on Control Variables
OTRit
(× 10–4/bps)
MCapit
MBRatioit
ShrTOit
PInverseit
Volatilityit
0.0063*** 0.0001*** − 0.0001*** − 0.0001*** 0.0001***
numtradesit
− 1480.890*** 1.2866*** − 1.1554*** 7.6538*** − 1.6539***
R.K.Dubey et al.
1 3
Table 6 Impact of Algorithmic Trading Efficiency on Price Discovery
This table shows the impact of algorithmic trading efficiency (Order-to-trade Ratio (OTR)) on various price discovery variables. The table regresses the various measures
of the price discovery (30s and 1min adverse selection, trade frequency and trade volume) on our algorithmic trading measure. It is based on 1s observations for Nifty 50
stocks from March, 2013 to August, 2013 which covers 2 quarters (6months or 125 trading days) of high frequency data with precision of 1 jiffies (1 Second = 65,536 jif-
fies). The algorithmic trading is directly identified by the flag provided by the stock exchange (NSE). The specification for examining the impact is:
Pit =𝛼it +𝛾it +𝛽OTRit +𝛿
1
MCapit +𝛿
2
MBRatioit +𝛿
3
ShrTOit +𝛿
4
PInverseit +𝜀it
where
Pit
is either 30–seconds or 1-min adverse selection, trade frequency and trade volume as price discovery measures for stock i in second t of the trading day. (The
trading hours of NSE is from 9:15 am to 3:30pm which essentially means 6h 15min or 22,500s.)
OTRit
is the order-to-trade ratio for the stock i in second t of the trad-
ing day (where, OTR is measured as ratio of total number of orders placed for a stock i in the time interval t to total number of trades that took place in a stock i in the
time interval t). We also use a vector of control variables including log of market capitalization (
MCapit)
, log of market to book ratio (
MBRatioit)
, log of share turnover
(
ShrTOit)
and log of 1/price (
PInverseit)
. We also include stock fixed effects and time fixed effects. t-values are also examined as they are based on standard errors that are
robust to general cross section and time-series heteroscedasticity and within group autocorrelation (Arellano and Bond, 1991). */**/*** denote significance at 10%/5%/1%
level
# observations: 49 * 125 * 22,500 (stock * day * seconds)
P-value: < 0.0001
Price Discovery Measures Coefficient on OTR Coefficients on Control Variables
OTRit
(× 10–4/bps)
MCapit
MBRatioit
ShrTOit
PInverseit
Adv_Selectionit+30
0.0043*** 0.0000*** − 0.0002*** − 0.0003*** − 0.0001***
Adv_Selectionit+60
0.0069*** 0.0000*** − 0.0002*** − 0.0003*** − 0.0001***
Trd _Fre qit
− 1480.890*** 1.2866*** − 1.1554*** 7.6538*** − 1.6539***
Trd _Vo lit
− 34,673,470*** 133,538.166*** − 55,135.187*** 245,804.451*** − 187,707.335***
1 3
Algorithmic Trading Efficiency andits Impact on…
Table 7 Impact of Algorithmic Trading Efficiency on Liquidity
This table shows the impact of algorithmic trading efficiency (Order-to-trade Ratio (OTR)) on various liquidity variables. The table regresses the various measures of the
liquidity (half spreads, depth and order imbalance) on our algorithmic trading measure. It is based on 1s observations for Nifty 50 stocks from March, 2013 to August,
2013 which covers 2 quarters (6months or 125 trading days) of high frequency data with precision of 1 jiffy (1 Second = 65,536 jiffies). The algorithmic trading is directly
identified by the flag provided by the stock exchange (NSE). The specification for examining the impact is:
Lit =𝛼it +𝛾it +𝛽OTRit +𝛿
1
MCapit +𝛿
2
MBRatioit +𝛿
3
ShrTOit +𝛿
4
PInverseit +𝜀it
where
Lit
is either quoted spread, proportional quoted spread, proportional realized spread, and proportional effective spread, depth or order imbalance as liquidity
measures for stock i in second t of the trading day. (The trading hours of NSE is from 9:15 am to 3:30pm which essentially means 6h 15min or 22,500 s.)
OTRit
is the
order-to-trade ratio for the stock i in second t of the trading day (where, OTR is measured as ratio of total number of orders placed for a stock i in the time interval t to total
number of trades that took place in a stock i in the time interval t).We also use a vector of control variables including log of market capitalization (
MCapit)
, log of market
to book ratio (
MBRatioit)
, log of share turnover (
ShrTOit)
and log of 1/price (
PInverseit)
. We also include stock fixed effects and time fixed effects. t-values are also exam-
ined as they are based on standard errors that are robust to general cross section and time-series heteroscedasticity and within group autocorrelation (Arellano and Bond,
1991). */**/*** denote significance at 10%/5%/1% level
# observations: 49 * 125 * 22,500 (stock * day * seconds)
P-value: < 0.0001
Liquidity Measures Coefficient on
OTRit
(× 10–4/bps)
Q1 Q2 Q3 Q4 Q5 Small Cap Mid Cap Large Cap
qspreadit
2.4201*** 54.0421*** 20.5341*** 45.3738*** 36.1424*** 44.8644*** 31.2054*** 47.0699***
pqspreadit
1.3529*** 2.1870*** 0.5006*** − 0.1220*** − 0.2478*** 1.7599*** − 0.3398*** − 1.0589***
pespreadit
0.0051*** 0.0028*** − 0.0144 0.0040*** 0.0751*** − 0.0030 0.0040*** 0.0757***
prspreadit+30
0.0046*** 0.0013*** 0.0050*** 0.0192*** 0.0746*** 0.0036*** 0.0020*** 0.0739***
prspreadit+60
0.0039*** 0.0006** 0.0044*** 0.0019*** 0.0748*** 0.0026*** 0.0018*** 0.0743***
depthit
5695.38*** 7484.03*** 6281.43*** 6612.88*** 7157.73*** 7310.048*** 6322.870*** 7803.621
OIit
− 618.524*** − 303.782*** − 901.094*** − 708.213*** − 484.408***
R.K.Dubey et al.
1 3
Table 8 Impact of Algorithmic Trading Efficiency on Volatility
This table shows the impact of algorithmic trading efficiency (Order-to-trade Ratio (OTR)) on various volatility variables. The table regresses the various measures of the
volatility (30-s HI-LO volatility, number of trades) on our algorithmic trading measure. It is based on 1s observations for Nifty 50 stocks from March, 2013 to August,
2013 which covers 2 quarters (6months or 125 trading days) of high frequency data with precision of 1 jiffy (1 Second = 65,536 jiffies). The algorithmic trading is directly
identified by the flag provided by the stock exchange (NSE). The specification for examining the impact is:
Vit =𝛼it +𝛾it +𝛽OTRit +𝛿1MCapit +𝛿2MBRatioit +𝛿3ShrTOit +𝛿4PInverseit +𝜀it
where
Vit
is either Parkinson’s (1980) HI-LO volatility measure capturing 30s HI and LO or number of trades (numtradesit) for stock i in second t of the trading day. (The
trading hours of NSE is from 9:15 am to 3:30pm which essentially means 6h 15min or 22,500s.)
OTRit
is the order-to-trade ratio for the stock i in second t of the trad-
ing day (where, OTR is measured as ratio of total number of orders placed for a stock i in the time interval t to total number of trades that took place in a stock i in the
time interval t). We also use a vector of control variables including log of market capitalization (
MCapit)
, log of market to book ratio (
MBRatioit)
, log of share turnover
(
ShrTOit)
and log of 1/price (
PInverseit)
. We also include stock fixed effects and time fixed effects. t-values are also examined as they are based on standard errors that are
robust to general cross section and time-series heteroscedasticity and within group autocorrelation (Arellano and Bond, 1991). */**/*** denote significance at 10% / 5% /
1% level
# observations: 49 * 125 * 22,500 (stock * day * seconds)
P-value: < 0.0001
Volatility Measures Coefficient on
OTRit
(× 10–4/bps)
Q1 Q2 Q3 Q4 Q5 Small Cap Mid Cap Large Cap
Volatilityit
− 0.0014*** − 0.0006*** − 0.0040*** − 0.0010*** 0.0118*** − 0.0016*** − 0.0012*** 0.0146***
numtradesit
− 1774.825*** − 1560.627*** − 1183.303*** − 1282.965*** − 1777.066*** − 1572.966*** − 1215.981*** − 1677.327***
# observations: 49 * 125 * 22,500 (stock * day * seconds)
P-value: < 0.0001
1 3
Algorithmic Trading Efficiency andits Impact on…
Table 9 Impact of Algorithmic Trading Efficiency on Price Discovery
This table shows the impact of algorithmic trading efficiency (Order-to-trade Ratio (OTR)) on various price discovery variables. The table regresses the various measures
of the price discovery (30s and 1min adverse selection, trade frequency and trade volume) on our algorithmic trading measure. It is based on 1s observations for Nifty 50
stocks from March, 2013 to August, 2013 which covers 2 quarters (6months or 125 trading days) of high frequency data with precision of 1 jiffy (1 Second = 65,536 jif-
fies). The algorithmic trading is directly identified by the flag provided by the stock exchange (NSE). The specification for examining the impact is:
Pit =𝛼it +𝛾it +𝛽OTRit +𝛿
1
MCapit +𝛿
2
MBRatioit +𝛿
3
ShrTOit +𝛿
4
PInverseit +𝜀it
where
Pit
is either 30–seconds or 1-min adverse selection, trade frequency and trade volume as price discovery measures for stock i in second t of the trading day. (The
trading hours of NSE is from 9:15 am to 3:30pm which essentially means 6h 15min or 22,500s.)
OTRit
is the order-to-trade ratio for the stock i in second t of the trad-
ing day (where, OTR is measured as ratio of total number of orders placed for a stock i in the time interval t to total number of trades that took place in a stock i in the
time interval t). We also use a vector of control variables including log of market capitalization (
MCapit)
, log of market to book ratio (
MBRatioit)
, log of share turnover
(
ShrTOit)
and log of 1/price (
PInverseit)
. We also include stock fixed effects and time fixed effects. t-values are also examined as they are based on standard errors that are
robust to general cross section and time-series heteroscedasticity and within group autocorrelation (Arellano and Bond, 1991). */**/*** denote significance at 10%/5%/1%
level
# observations: 49 * 125 * 22,500 (stock * day * seconds)
P-value: < 0.0001
Price Discovery
Measures
Coefficient on
OTRit
(× 10–4/bps)
Q1 Q2 Q3 Q4 Q5 Small Cap Mid Cap Large Cap
Adv_Selectionit+30
0.0002 − 0.0011** − 0.1504*** 0.0028*** 0.0265*** − 0.0492*** 0.0034*** 0.0302***
Adv_Selectionit+60
0.0005 − 0.0005* − 0.1473*** 0.0027*** 0.0302*** − 0.0471*** 0.0034*** 0.0337***
Trd _Fre qit
− 1774.825*** − 1560.627*** − 1183.303*** − 1282.965*** − 1777.066*** − 1572.966*** − 1215.981*** − 1677.327***
Trd _Vo lit
− 40,696,100 − 42,165,000*** − 25,306,400*** − 30,278,500*** − 45,269,500*** − 37,729,790*** − 29,791,360*** − 39,057,860***
R.K.Dubey et al.
1 3
for Q1 and Q2 the coefficients of OTR pertaining to adverse selection measures are
not significant at 1% level, they are either insignificant or significant at 5% or 10%
levels.
We further investigate the date-wise (daily) impact of OTR on liquidity, volatil-
ity and price discovery measures. To examine the date-wise (daily) impact we use
the Fama–MacBeth (Fama and MacBeth, 1973, Opler etal., 1999) model which
gives the average of the time series of the coefficients from daily (date-wise,
N = 125) cross section regressions. The cross-sectional regression is estimated
for each trading day. We report our findings in Table10, Table11 and Table12
where the impact of algorithmic trading efficiency (OTR) on liquidity, volatil-
ity and price discovery measures are summarized for date wise Fama–MacBeth
regressions. The findings are concurrent with our earlier aggregate level find-
ings, that algorithmic trading efficiency has significant impact on the liquidity,
Table 10 Impact of Algorithmic Trading Efficiency on Liquidity (Fama–MacBeth Regression, date –
wise)
This table shows the impact of algorithmic trading (OTR) on various liquidity variables and summa-
rizes date wise Fama–MacBeth regression results. The Fama–MacBeth (Fama and MacBeth, 1973, Opler
etal., 1999) model gives the average of the time series of the coefficients from daily (date-wise, N = 125)
cross section regressions. The cross-sectional regression is estimated each trading day. The cross-sec-
tional regression is based on 1 s observations for Nifty 50 stocks from March, 2013 to August, 2013
which covers 2 quarters (6months or 125 trading days). The model specification for the cross-sectional
regression is:
Lit =𝛼it +𝛾it +𝛽OTRit +𝛿
1
MCapit +𝛿
2
MBRatioit +𝛿
3
ShrTOit +𝛿
4
PInverseit +𝜀it
where
Lit
is either quoted spread, proportional quoted spread, proportional realized spread, and pro-
portional effective spread, depth or order imbalance as liquidity measures for stock i in second t of the
trading day. (The trading hours of NSE is from 9:15 am to 3:30pm which essentially means 6h 15 min
or 22,500s.)
OTRit
is the order-to-trade ratio for the stock i in second t of the trading day (where, OTR
is measured as ratio of total number of orders placed for a stock i in the time interval t to total number of
trades that took place in a stock i in the time interval t).We also use a vector of control variables includ-
ing log of market capitalization (
MCapit)
, log of market to book ratio (
MBRatioit)
, log of share turno-
ver (
ShrTOit)
and log of 1/price (
PInverseit)
. We also include stock fixed effects and time fixed effects.
t-values are also examined as they are based on standard errors that are robust to general cross section
and time-series heteroscedasticity and within group autocorrelation (Arellano and Bond, 1991). */**/***
denote significance at 10%/5%/1% level
# observations: 125 trading days
P-value: < 0.0001
Liquidity Measures Coefficient on OTR Coefficients on Control Variables
OTRit
(× 10–4/bps)
MCapit
MBRatioit
ShrTOit
PInverseit
qspreadit
40.8020*** 0.4503*** − 0.3039*** − 0.5962*** − 0.4898***
pqspreadit
− 0.1516 − 0.0682*** 0.0156*** 0.0695*** 0.0518***
pespreadit
0.0338*** 0.0001** − 0.0001*** − 0.0001*** 0.0001
prspreadit+30
0.0358*** 0.0002*** − 0.0001*** − 0.0001*** 0.0003***
prspreadit+60
0.0365*** 0.0002*** − 0.0001*** − 0.0002*** 0.0003***
depthit
6932.6390*** 20.8056*** − 11.6974*** 10.5047*** − 20.8295***
OIit
− 672.8880*** − 3.1176*** 2.2779*** 2.4278*** 3.0824***
1 3
Algorithmic Trading Efficiency andits Impact on…
volatility and price discovery measures and also, these measures tend to improve
with rise in algorithmic trading efficiency (lower OTR). For liquidity measures
we find that the coefficients of OTR are significant for all except proportional
quoted spread and the order imbalance shows a negative sign against the expecta-
tions. The findings clearly indicate that with rise or decline in algorithmic trading
efficiency (lower/higher OTR) the liquidity improves (spread measures decline/
increase). Similarly, the volatility measure also shows signs of improvement with
rise in algorithmic trading efficiency. Only the trade frequency based measure
shows a negative sign, which is concurrent with the previous findings and the rea-
son for the same could be the perception of traders. Since large number of orders
are being generated, so there must be some more/new information and hence trad-
ers may step back from trading till the time orders and the price stabilizes. How-
ever, when we examine the results for the price discovery measures, we find that
there is no significant impact on the adverse selection measures. So, it seems on
an average the order-to-trades ratio doesn’t play any significant role in the price
impact of the stocks on daily basis. We also find that for the frequency and vol-
ume based measures of price discovery the coefficients are significant but bear
signs opposite to our expectations. The plausible reason for the same could be
again traders staying cautious in order to avoid getting picked off.
Table 11 Impact of Algorithmic Trading Efficiency on Volatility (Fama–MacBeth Regression, date–
wise)
This table shows the impact of algorithmic trading (OTR) on various volatility variables and summa-
rizes date wise Fama–MacBeth regression results. The Fama–MacBeth (Fama and MacBeth, 1973; Opler
etal., 1999) model gives the average of the time series of the coefficients from daily (date-wise, N = 125)
cross section regressions. The cross-sectional regression is estimated each trading day. The cross-sec-
tional regression is based on 1 s observations for Nifty 50 stocks from March, 2013 to August, 2013
which covers 2 quarters (6months or 125 trading days). The model specification for the cross-sectional
regression is:
Vit =𝛼it +𝛾it +𝛽OTRit +𝛿
1
MCapit +𝛿
2
MBRatioit +𝛿
3
ShrTOit +𝛿
4
PInverseit +𝜀it
where
Vit
is either Parkinson’s (1980) HI-LO volatility measure capturing 30s HI and LO or number of
trades (numtradesit) for stock i in second t of the trading day. (The trading hours of NSE is from 9:15 am
to 3:30pm which essentially means 6h 15min or 22,500s.)
OTRit
is the order-to-trade ratio for the stock
i in second t of the trading day (where, OTR is measured as ratio of total number of orders placed for a
stock i in the time interval t to total number of trades that took place in a stock i in the time interval t).
We also use a vector of control variables including log of market capitalization (
MCapit)
, log of market
to book ratio (
MBRatioit)
, log of share turnover (
ShrTOit)
and log of 1/price (
PInverseit)
. We also include
stock fixed effects and time fixed effects. t-values are also examined as they are based on standard errors
that are robust to general cross section and time-series heteroscedasticity and within group autocorrela-
tion (Arellano and Bond, 1991). */**/*** denote significance at 10%/5%/1% level
# observations: 125 trading days
P-value: < 0.0001
Volatility Measures Coefficient on OTR Coefficients on control variables
OTRit
(× 10–4/bps)
MCapit
MBRatioit
ShrTOit
PInverseit
Volatilityit
0.0006*** 0.0001*** − 0.0001*** − 0.0001*** 0.0001***
numtradesit
− 1498.367*** 1.4937*** − 1.2325*** 6.9778*** − 1.2643***
R.K.Dubey et al.
1 3
Similar to the date-wise (daily) impact of OTR on price liquidity, volatility
and discovery measures, we examine the stock-wise impact of OTR. To exam-
ine the stock-wise impact we use the same Fama–MacBeth (Fama and MacBeth,
1973, Opler et al., 1999) model which gives the average of the time series of
the coefficients from stock-wise (N = 49) cross section regressions. The cross-
sectional regression is estimated for each stock for the entire 125 trading days and
we report our findings in Table13, Table14 and Table15. These tables summa-
rize the stock-wise Fama–MacBeth regression results. We find significant of OTR
only on liquidity and volatility measures. We do not find any significant impact of
OTR on any of the price discovery measures. The findings are similar to the other
stock wise regressions where we investigating the impact of OTR on liquidity,
volatility and price discovery. From Table13 we find that except for proportional
effective spread, the coefficients of all of liquidity measures are significant. The
signs of the coefficients are as expected except for the order imbalance and the
reason for the same could be the ability of the AT of not piling orders on any one
side. For the stock wise volatility analysis, we find from Table14 that the coeffi-
cients of OTR are significant. However, the sign of the coefficient of OTR for the
Table 12 Impact of Algorithmic Trading Efficiency on Price Discovery (Fama–MacBeth Regression,
date – wise)
This table shows the impact of algorithmic trading (OTR) on various volatility variables and summa-
rizes date wise Fama–MacBeth regression results. The Fama–MacBeth (Fama and MacBeth, 1973, Opler
etal., 1999) model gives the average of the time series of the coefficients from daily (date-wise, N = 125)
cross section regressions. The cross-sectional regression is estimated each trading day. The cross-sec-
tional regression is based on 1 s observations for Nifty 50 stocks from March, 2013 to August, 2013
which covers 2 quarters (6months or 125 trading days). The model specification for the cross-sectional
regression is:
Pit =𝛼it +𝛾it +𝛽OTRit +𝛿1MCapit +𝛿2MBRatioit +𝛿3ShrTOit +𝛿4PInverseit +𝜀it
where
Pit
is either 30 – seconds or 1-min adverse selection, trade frequency and trade volume as price
discovery measures for stock i in second t of the trading day. (The trading hours of NSE is from 9:15 am
to 3:30pm which essentially means 6h 15min or 22,500s.)
OTRit
is the order-to-trade ratio for the stock
i in second t of the trading day (where, OTR is measured as ratio of total number of orders placed for a
stock i in the time interval t to total number of trades that took place in a stock i in the time interval t).
We also use a vector of control variables including log of market capitalization (
MCapit)
, log of market
to book ratio (
MBRatioit)
, log of share turnover (
ShrTOit)
and log of 1/price (
PInverseit)
. We also include
stock fixed effects and time fixed effects. t-values are also examined as they are based on standard errors
that are robust to general cross section and time-series heteroscedasticity and within group autocorrela-
tion (Arellano and Bond, 1991). */**/*** denote significance at 10%/5%/1% level
# observations: 125 trading days
P-value: < 0.0001
Price Discovery
Measures
Coefficient on OTR Coefficients on Control Variables
OTRit
(× 10–4/bps)
MCapit
MBRatioit
ShrTOit
PInverseit
Adv_Selectionit+30
0.0003 0.0000 − 0.0003*** − 0.0003*** − 0.0002
Adv_Selectionit+60
0.0006 0.0001 − 0.0003*** − 0.0003*** − 0.0001
Trd _Fre qit
− 1498.367*** 1.4937*** − 1.2325*** 6.9778*** − 1.2643***
Trd _Vo lit
− 37,247,600*** 151,623.55*** − 88,732.15*** 200,359.35*** − 151,503.19***
1 3
Algorithmic Trading Efficiency andits Impact on…
volatility measure is negative which is not as per our expectation. So, we ascer-
tain that for the average stock wise analysis the volatility is affected adversely
by the OTR. Therefore, higher OTR results to higher volatility, though the coef-
ficient is very small (0.0000 × 10–4) but significant. The frequency-based measure
shows a negative sign, as discussed previously, which could be due to the appre-
hension of the traders. From Table15, we do not observe any significant impact
of OTR on the price discovery measures when we run the stock wise analysis.
Table 13 Impact of Algorithmic Trading Efficiency on Liquidity (Fama–MacBeth Regression, stock–
wise)
This table shows the impact of algorithmic trading (AT Intensity) on various liquidity variables and sum-
marizes stock wise Fama–MacBeth regression results. The Fama–MacBeth (Fama and MacBeth, 1973,
Opler etal., 1999) model gives the average of the time series of the coefficients from stock-wise (N = 49)
cross section regressions. The cross-sectional regression is estimated for each stock for the 125 trading
days. The cross-sectional regression is based on 1s observations for Nifty 50 stocks from March, 2013
to August, 2013 which covers 2 quarters (6months or 125 trading days). The model specification for the
cross-sectional regression is:
Lit =𝛼it +𝛾it +𝛽OTRit +𝛿
1
MCapit +𝛿
2
MBRatioit +𝛿
3
ShrTOit +𝛿
4
PInverseit +𝜀it
where
Lit
is either quoted spread, proportional quoted spread, proportional realized spread, and pro-
portional effective spread, depth or order imbalance as liquidity measures for stock i in second t of the
trading day. (The trading hours of NSE is from 9:15 am to 3:30pm which essentially means 6h 15 min
or 22,500s.)
OTRit
is the order-to-trade ratio for the stock i in second t of the trading day (where, OTR
is measured as ratio of total number of orders placed for a stock i in the time interval t to total number of
trades that took place in a stock i in the time interval t).We also use a vector of control variables includ-
ing log of market capitalization (
MCapit)
, log of market to book ratio (
MBRatioit)
, log of share turno-
ver (
ShrTOit)
and log of 1/price (
PInverseit)
. We also include stock fixed effects and time fixed effects.
t-values are also examined as they are based on standard errors that are robust to general cross section
and time-series heteroscedasticity and within group autocorrelation (Arellano and Bond, 1991). */**/***
denote significance at 10% / 5% / 1% level
# observations: 49 stocks
P-value: < 0.0001
Liquidity Measures Coefficient on OTR Coefficients on Control Variables
OTRit
(× 10–4/bps)
MCapit
MBRatioit
ShrTOit
PInverseit
qspreadit
25.907*** 5.2885 − 0.6467 − 0.0013 3.5242
pqspreadit
15.948*** 1.4785 − 0.1861 0.0113 1.4935
pespreadit
0.0003 0.1138 0.0144 0.0003 0.1269
prspreadit+30
0.0009*** − 0.0003 0.0002 − 0.0000 − 0.0000
prspreadit+60
0.0008* 0.0003 0.0001 − 0.0000 0.0005
depthit
5724.276*** 599.8073* − 23.4252 36.2352*** 559.2892*
OIit
− 730.976*** 50.7466 − 55.6486 − 3.3732*** 22.0769
R.K.Dubey et al.
1 3
6 Conclusion
The research in the field of Algorithmic Trading and its impact across markets is
in nascent stage. The existing research focuses primarily on impact of AT and the
findings are inconclusive as some report positive and some report negative impact
of AT. The reliability of these results is questioned by existing authors due to small
sample that they choose for the study and also because of their inability to directly
identify AT. In this study we have the advantage of direct identification of AT and
therefore, our findings are reliable and directly refer to implications of AT. We also
use a longer time period for our data set in order to tackle the criticism of small
sample.
The existing studies highlight the rise of algorithmic trading and also highlight
the various apprehensions (decline of manual trading, rise in volatility, creation
of fake liquidity, quote stuffing, adverse selection, etc.) pertaining to AT. These
apprehensions are relevant not only for the academicians but also for the mar-
ket participants and market regulators. These questions can be better answered if
we measure the algorithmic trading efficiency instead of just measuring the algo-
rithmic trading activity. In this study we propose a novel measure of algorith-
mic trading efficiency that is, inverse of order-to-trade ratio (1/OTR). We propose
Table 14 Impact of Algorithmic Trading Efficiency on Volatility (Fama–MacBeth Regression, stock–
wise)
This table shows the impact of algorithmic trading (AT Intensity) on various liquidity variables and sum-
marizes stock wise Fama–MacBeth regression results. The Fama–MacBeth (Fama and MacBeth, 1973,
Opler etal., 1999) model gives the average of the time series of the coefficients from stock-wise (N = 49)
cross section regressions. The cross-sectional regression is estimated for each stock for the 125 trading
days. The cross-sectional regression is based on 1s observations for Nifty 50 stocks from March, 2013
to August, 2013 which covers 2 quarters (6months or 125 trading days). The model specification for the
cross-sectional regression is:
Vit =𝛼it +𝛾it +𝛽OTRit +𝛿
1
MCapit +𝛿
2
MBRatioit +𝛿
3
ShrTOit +𝛿
4
PInverseit +𝜀it
where
Vit
is either Parkinson’s (1980) HI-LO volatility measure capturing 30s HI and LO or number of
trades (numtradesit) for stock i in second t of the trading day. (The trading hours of NSE is from 9:15 am
to 3:30pm which essentially means 6h 15min or 22,500s.)
OTRit
is the order-to-trade ratio for the stock
i in second t of the trading day (where, OTR is measured as ratio of total number of orders placed for a
stock i in the time interval t to total number of trades that took place in a stock i in the time interval t).
We also use a vector of control variables including log of market capitalization (
MCapit)
, log of market
to book ratio (
MBRatioit)
, log of share turnover (
ShrTOit)
and log of 1/price (
PInverseit)
. We also include
stock fixed effects and time fixed effects. t-values are also examined as they are based on standard errors
that are robust to general cross section and time-series heteroscedasticity and within group autocorrela-
tion (Arellano and Bond, 1991). */**/*** denote significance at 10%/5%/1% level
# observations: 49 stocks
P-value: < 0.0001
Volatility Measures Coefficient on OTR Coefficients on Control Variables
OTRit
(× 10–4/bps)
MCapit
MBRatioit
ShrTOit
PInverseit
Volatilityit
− 0.0000** 0.0001 0.0008 − 0.0000 0.0010
numtradesit
− 3845.476*** − 266.2039 − 3.6961 9.2912*** − 272.3729
1 3
Algorithmic Trading Efficiency andits Impact on…
that, if AT is efficient then its orders should lead to a greater number of trades
thereby creating liquidity and also making the prices informationally efficient.
In this study, we find evidence on the relationship between the order-to-trade
ratios and the liquidity, volatility and price discovery measures. We devise inverse
of order-to-trade ratio (1/OTR) as a measure of algorithmic trading efficiency.
Therefore, it is expected that with increase in algorithmic trading efficiency, the
liquidity, volatility and price discovery improve. From our analysis, we find that
the OTR’s impact is consistently significant on the liquidity, volatility and price
discovery measures and especially for large cap stocks. The prominence of OTR’s
impact on large cap stocks is as expected, because of higher amount of trading
happening in those stocks. From the results, we also find that with rise in algo-
rithmic trading efficiency (lower OTR) there is a significant improvement on the
liquidity and volatility on the basis of day wise and stock wise Fama–MacBeth
regressions. However, on an average OTR doesn’t significantly affect the price
discovery for the day wise and stock wise analysis.
This is one of the few studies that directly identify AT and also discusses about
the algorithmic trading efficiency. The findings have implications for existing
Table 15 Impact of Algorithmic Trading Efficiency on Price Discovery (Fama–MacBeth Regression,
stock–wise)
This table shows the impact of algorithmic trading (AT Intensity) on various liquidity variables and sum-
marizes stock wise Fama–MacBeth regression results. The Fama–MacBeth (Fama and MacBeth, 1973,
Opler etal., 1999) model gives the average of the time series of the coefficients from stock-wise (N = 49)
cross section regressions. The cross-sectional regression is estimated for each stock for the 125 trading
days. The cross-sectional regression is based on 1s observations for Nifty 50 stocks from March, 2013
to August, 2013 which covers 2 quarters (6months or 125 trading days). The model specification for the
cross-sectional regression is:
Pit =𝛼it +𝛾it +𝛽OTRit +𝛿
1
MCapit +𝛿
2
MBRatioit +𝛿
3
ShrTOit +𝛿
4
PInverseit +𝜀it
where
Pit
is either 30–seconds or 1-min adverse selection, trade frequency and trade volume as price
discovery measures for stock i in second t of the trading day. (The trading hours of NSE is from 9:15 am
to 3:30pm which essentially means 6h 15min or 22,500s.)
OTRit
is the order-to-trade ratio for the stock
i in second t of the trading day (where, OTR is measured as ratio of total number of orders placed for a
stock i in the time interval t to total number of trades that took place in a stock i in the time interval t).
We also use a vector of control variables including log of market capitalization (
MCapit)
, log of market
to book ratio (
MBRatioit)
, log of share turnover (
ShrTOit)
and log of 1/price (
PInverseit)
. We also include
stock fixed effects and time fixed effects. t-values are also examined as they are based on standard errors
that are robust to general cross section and time-series heteroscedasticity and within group autocorrela-
tion (Arellano and Bond, 1991). */**/*** denote significance at 10%/5%/1% level
# observations: 49 stocks
P-value: < 0.0001
Price Discovery Measures Coefficient on OTR Coefficients on Control Variables
OTRit
(× 10–4/bps)
MCapit
MBRatioit
ShrTOit
PInverseit
Adv_Selectionit+30
0.0003 0.0000 − 0.0003*** − 0.0003*** − 0.0002
Adv_Selectionit+60
0.0006 0.0001 − 0.0003*** − 0.0003*** − 0.0001
Trd _Fre qit
− 0.0277 0.2039 − 0.0008 0.0006 0.2030
Trd _Vo lit
− 0.0306 0.2344 0.0015 0.0008 0.2350
R.K.Dubey et al.
1 3
literature, market participants and also the market regulators. The reliable find-
ings can be used for policy framing by the policy makers and can be extended to
future research as well.
Appendix1
See Table 16.
Appendix2
See Table 17.
Table 16 Data-structure of dataset obtained from NSE DOTEX
a RM refers to regular market and PO refers to pre-open market
b 1 Second = 65,536 Jiffies
Orders Data Set Var. Constructed Trades Data Set Var. Constructed
Record Indicator (RM/PO)aRecord Indicator (RM/PO)
Segment (Cash) Segment (Cash)
Order Number Trade Number
Transaction Time (Jiffies)bOrdTime Trade Time (Jiffies) TrdTime
Buy / Sell Indicator QS, PQS Symbol (NSE)
Activity Type (Entry/Mod/Can) QS, PQS Series (EQ)
Symbol (NSE) Trade Price TrdPrice, QS
Series (EQ) Trade Quantity TrdQty
Volume Disclosed Buy Order Number
Volume Original OQty Buy Algo Indicator AT_Intensity
Limit Price OrdPrice, QS Buy Client Identity Flag
Trigger price Sell Order Number
Market Order Flag MLOrdInd Sell Algo Indicator AT_Intensity
Stop Loss Flag Sell Client Identity Flag
IO Flag IOC_Flag
Algo Indicator AT_Intensity
Client Identity Flag
1 3
Algorithmic Trading Efficiency andits Impact on…
Table 17 Definition of Variables Used in Study
Measures Definition Particulars
qspreadit
=(
P
Ait
−P
Bit)
PAit and PBit are the posted ask price, and bid price for security iat time t, respectively, and PMitis the
quote midpoint or mean of PAit and PBit
pqspreadit
=(
PAit−PBit
)
X
100
P
Mit
pespreadit
=
qit
(
Pit−Mi,t
)
M
it
Pit denotes the trade price, qit denotes the buy – sell side indicator (+ 1 for buys and -1 for sells) and Mit
is the midpoint at that time
prspreadit
=
qit
(
Pit−Mi,t+x
)
Mit
Pit denotes the trade price, qit denotes the buy – sell side indicator (+ 1 for buys and -1 for sells) and
Mi,t+x is the midpoint after x seconds / minutes of trade time
depthit
=NS
it
+NB
it
2
NSit and NBit denote the number of shares available on the sell side and buy side respectively for the
stock iat a given point of time t and OIit measure the order imbalance
OIit
=(NS
it
−NB
it
)
(NSit+NBit )
2
X
100
Volatilityit
=�
1
n∗
∑
30
t=1(log10
HI_30it
LO
_30it
)
2
HI_30it & LO_30it denotes the highest trade price for the stock i in second t and lowest trade price for
the stock i in second t respectively. Where, n denotes the number of trades happening in the time inter-
val for the given stock (i)
numtradesfreq
=N
N is the number of trades
Adv_selectionit
=pespreadit −prspreadit
M is the mid-quote and t and t + x represents the time t at which the mid-quote was observed and the time
after x seconds/minutes after the time t respectively
Advselection
=((
M
t+x
−M
t)
∕M
t)
Trd freq
=N
N is the number of trades
Trd vol
=V
V is the volume of trades
AT
= {0, 1} AT takes the value of 0 or 1 to denote the order as Algo or Non – Algo respectively
ATIntensity
=
TTV
AT,i,t
TTV
i,t
x
100
TTVAT,i,t refers to total traded volume in a stock (i) by AT in the time interval (t) and TTVi,t refers to total
traded volume in a stock (i) in the time interval (t)
OTRit
=
(
OrdEntryit+OrdCancelit +OrdModifyit
)
Trad e it
OTRit
is defined as the number of orders placed (Entry + Cancel + Modify) in the market divided by the
number of trades that got executed for the stock i in the time interval t
R.K.Dubey et al.
1 3
Appendix3
See Table 18.
Appendix4
See Table 19.
Table 18 List of stocks segregated into small cap, mid cap and large cap based on market capitalization
Large Cap Medium Cap Small Cap
Stock MktCap
(Rs. Mn)
Stock MktCap
(Rs. Mn)
Stock MktCap
(Rs. Mn)
TATAMOTORS 28,75,940.53 LT 8,54,892.47 GAIL 3,95,067.29
ONGC 27,12,090.37 TATAPOWER 8,54,198.99 DRREDDY 3,58,998.15
INDUSINDBK 26,55,411.02 AXISBANK 6,70,510.66 DLF 3,47,582.73
RELIANCE 25,44,470.32 MARUTI 5,97,894.46 LUPIN 3,37,240.47
COALINDIA 20,60,082.25 KOTAKBANK 5,90,820.11 CIPLA 2,95,113.74
HDFC 16,43,665.93 CAIRN 5,44,428.86 TCS 2,87,965.32
ITC 14,43,687.14 HCLTECH 5,16,911.41 BANKBARODA 2,78,976.99
SBIN 14,15,847.71 PNB 5,15,751.40 AMBUJACEM 2,75,137.35
HINDUNILVR 13,46,345.92 ULTRACEMCO 5,09,581.91 JPASSOCIAT 2,71,475.74
ICICIBANK 13,21,565.78 BAJAJ 5,09,141.27 POWERGRID 2,68,410.02
HINDALCO 12,80,399.85 BHARTIARTL 4,82,177.20 INFY 2,66,701.01
NTPC 12,54,959.68 MNM 4,73,827.68 BHEL 2,65,986.54
BPCL 11,52,170.63 NMDC 4,65,457.66 GRASIM 2,55,968.44
SUNPHARMA 10,53,859.98 ASIANPAINT 4,50,276.22 ACC 2,29,687.67
IDFC 2,21,388.16
TATASTEEL 2,08,949.02
HEROMOTOCO 1,94,138.72
HDFCBANK 1,89,027.58
RANBAXY 1,53,247.18
JINDALSTEL 1,47,458.10
VEDL 1,39,881.87
RELINFRA 96,977.56
Average 1,768,606.94 Average 573,990.73 Average 249,335.44
1 3
Algorithmic Trading Efficiency andits Impact on…
Table 19 List of stocks included in quintiles based on market capitalization
Q5 (Largest Cap) Q4 Q3 Q2 Q1(Smallest Cap)
Stock MktCap
(Rs. Mn)
Stock MktCap
(Rs. Mn)
Stock MktCap
(Rs. Mn)
Stock MktCap
(Rs. Mn)
Stock MktCap
(Rs. Mn)
COALINDIA 20,60,082.25 AXISBANK 6,70,510.66 ASIANPAINT 4,50,276.22 AMBUJACEM 2,75,137.35 ACC 2,29,687.67
HDFC 16,43,665.93 BPCL 11,52,170.63 BAJAJ 5,09,141.27 BANKBARODA 2,78,976.99 GRASIM 2,55,968.44
HINDUNILVR 13,46,345.92 CAIRN 5,44,428.86 BHARTIARTL 4,82,177.20 BHEL 2,65,986.54 HDFCBANK 1,89,027.58
ICICIBANK 13,21,565.78 HINDALCO 12,80,399.85 DRREDDY 3,58,998.15 CIPLA 2,95,113.74 HEROMOTOCO 1,94,138.72
INDUSINDBK 26,55,411.02 KOTAKBANK 5,90,820.11 GAIL 3,95,067.29 DLF 3,47,582.73 IDFC 2,21,388.16
ITC 14,43,687.14 LT 8,54,892.47 HCLTECH 5,16,911.41 INFY 2,66,701.01 JINDALSTEL 1,47,458.10
ONGC 27,12,090.37 MARUTI 5,97,894.46 MNM 4,73,827.68 JPASSOCIAT 2,71,475.74 RANBAXY 1,53,247.18
RELIANCE 25,44,470.32 NTPC 12,54,959.68 NMDC 4,65,457.66 LUPIN 3,37,240.47 RELINFRA 96,977.56
SBIN 14,15,847.71 SUNPHARMA 10,53,859.98 PNB 5,15,751.40 POWERGRID 2,68,410.02 TATASTEEL 2,08,949.02
TATAMOTORS 28,75,940.53 TATAPOWER 8,54,198.99 ULTRACEMCO 5,09,581.91 TCS 2,87,965.32 VEDL 1,39,881.87
Average 20,01,910.70 Average 8,85,413.57 Average 467,719.02 Average 2,89,458.99 Average 1,83,672.43
R.K.Dubey et al.
1 3
References
Aggarwal, N., & Thomas, S. (2014). The causal impact of algorithmic trading on market quality. Working
paper, IGIDR Mumbai – WP-2014–23
Brogaard, J., Hendershott, T., & Riordan, R. (2014). High-frequency trading and price discovery. Review
of Financial Studies, 27(8), 2267–2306. https:// doi. org/ 10. 1093/ rfs/ hhu032
Chordia, T., Roll, R., & Subrahmanyam, A. (2000). Commonality in liquidity. Journal of Financial Eco-
nomics, 56(1), 3–28. https:// doi. org/ 10. 1016/ S0304- 405X(99) 00057-4
Dubey, R. (2016). Algorithmic trading: Is the machine taking over our equity markets? (pp. 28–34). IUP
Publications.
Dubey, R. K., Chauhan, Y. K., & Syamala, S. R. (2017). Evidence of algorithmic trading from indian equity
market: interpreting the transaction velocity element of financialization. Research in International Busi-
ness and Finance, 42, 31–38. https:// doi. org/ 10. 1016/j. ribaf. 2017. 05. 014
Fama, E. F., & Macbeth, J. D. (1973). Risk, Return, and Equilibrium: Empirical Tests. Journal of Politi-
cal Economy, 81(3), 607–636.
Egginton, J. F., Van Ness, B. F., & Van Ness, R. A. (2016). Quote stuffing. Financial Management, 45(3),
583–608.
Goyenko, R. Y., Holden, C. W., & Trzcinka, C. A. (2009). Do liquidity measures measure liquidity? Journal
of Financial Economics, 92(2), 153–181. https:// doi. org/ 10. 1016/j. jfine co. 2008. 06. 002
Groth, S. S. (2011). Does Algorithmic Trading Increase Volatility? Empirical Evidence from the Fully-Elec-
tronic Trading Platform Xetra. InWirtschaftsinformatik(p. 112).
Hendershott, T., & Riordan, R. (2013). Algorithmic trading and the market for liquidity. Journal of Financial
and Quantitative Analysis, 48(04), 1001–1024. https:// doi. org/ 10. 1017/ S0022 10901 30004 71
Hendershott, T., Jones, C. M., & Menkveld, A. J. (2011). Does algorithmic trading improve liquidity? The
Journal of Finance, 66(1), 1–33. https:// doi. org/ 10. 1111/j. 1540- 6261. 2010. 01624.x
Hinz, J., & Yee, J. (2017). An algorithmic approach to optimal asset liquidation problems. Asia-Pacific
Financial Markets, 24, 109–129. https:// doi. org/ 10. 1007/ s10690- 017- 9226-1
Johnson, B. (2010).Algorithmic Trading & DMA: An introduction to direct access trading strategies. 4Mye-
loma Press.
Jones, C. M., Kaul, G., & Lipson, M. L. (1994). Transactions, volume, and volatility. Review of Financial
Studies, 7(4), 631–651. https:// doi. org/ 10. 1093/ rfs/7. 4. 631
Kang, W., & Zhang, H. (2014). Measuring liquidity in emerging markets. Pacific-Basin Finance Journal, 27,
49–71. https:// doi. org/ 10. 1016/j. pacfin. 2014. 02. 001
Kelejian, H. H., & Mukerji, P. (2016). Does high frequency algorithmic trading matter for non-AT investors?
Research in International Business and Finance, 37, 78–92. https:// doi. org/ 10. 1016/j. ribaf. 2015. 10. 014
Lang, M., Lins, K. V., & Maffett, M. (2012). Transparency, liquidity, and valuation: International evidence on
when transparency matters most. Journal of Accounting Research, 50(3), 729–774. https:// doi. org/ 10.
1111/j. 1475- 679X. 2012. 00442.x
Lee, K. H. (2011). The world price of liquidity risk. Journal of Financial Economics, 99(1), 136–161. https://
doi. org/ 10. 1016/j. jfine co. 2010. 08. 003
Lesmond, D. A. (2005). Liquidity of emerging markets. Journal of Financial Economics, 77(2), 411–452.
https:// doi. org/ 10. 1016/j. jfine co. 2004. 01. 005
Markowitz, H. (1952). Portfolio selection*. The Journal of Finance, 7(1), 77–91. https:// doi. org/ 10. 1111/j.
1540- 6261. 1952. tb015 25.x
Nawn, S., & Banerjee, A. (2019a). Do proprietary algorithmic traders withdraw liquidity during market
stress? Financial Management, 48(2), 641–676.
Nawn, S., & Banerjee, A. (2019b). Do the limit orders of proprietary and agency algorithmic traders discover
or obscure security prices? Journal of Empirical Finance, 53, 109–125.
Opler, T., Pinkowitz, L., Stulz, R., & Williamson, R. (1999). The determinants and implications of corpo-
rate cash holdings. Journal of financial economics, 52(1), 3–46.
Parkinson, M. (1980). The extreme value method for estimating the variance of the rate of return. Journal of
Business, 53(1), 61–65.
Saito, T., Adachi, T., Nakatsuma, T., etal. (2018). Trading and ordering patterns of market participants in
high frequency trading environment: Empirical study in the japanese stock market. Asia-Pacific Finan-
cial Markets, 25, 179–220. https:// doi. org/ 10. 1007/ s10690- 018- 9245-6
Subrahmanyam, A. (2013). Algorithmic trading, the flash crash, and coordinated circuit breakers. Borsa
Istanbul Review, 13(3), 4–9. https:// doi. org/ 10. 1016/j. bir. 2013. 10. 003
1 3
Algorithmic Trading Efficiency andits Impact on…
Upson, J., & Van Ness, R. A. (2017). Multiple markets, algorithmic trading, and market liquidity. Journal of
Financial Markets, 32, 49–68. https:// doi. org/ 10. 1016/j. finmar. 2016. 05. 004
Wu, P., & Siwasarit, W. (2020). Capturing the order imbalance with hidden markov model: A Case
of SET50 and KOSPI50. Asia-Pacific Financial Markets, 27, 115–144. https:// doi. org/ 10. 1007/
s10690- 019- 09285-1
Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps
and institutional affiliations.
A preview of this full-text is provided by Springer Nature.
Content available from Asia-Pacific Financial Markets
This content is subject to copyright. Terms and conditions apply.
