Profitability of Trading Strategies on High-Frequency data, with

Trading Costs

Tirthankar Patnaik∗ Susan Thomas

April 28, 2004

Abstract

This paper tests for the profitability of contrarian and momentum trading strategies in the Indian equity
markets, in the period 1996-2002, explicitly accounting for transaction costs. Using raw intraday data for order-
book from the National Stock Exchange (NSE), we calculate the actual prices at which trades would take place
for an orider of INR 10,000, using the price impact of each of the stocks studied. We find that for the Indian
market, momentum strategies are profitable in the short-run, but profits fall as the formation period increases.
Our methodology eliminates market-microstructure biases, like the bid-ask bounce. We find that profits persist
in some strategies even after these effects are taken care of.
Keywords: Contrarian, and momentum trading strategies, Transaction costs, High Frequency Data, Market
Liquidity

∗ Research Student, IGIDR. [email protected]

Contents
1 Introduction 1

2 Methodology 2
2.1 Portfolio Selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.2 Trading costs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.3 Trading strategies: Loser-Winner analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.4 Trading Strategies: Size-effect, and other causes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.5 Decomposition of MIS profits. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

3 Data 10

4 Results 12
4.1 Analyzing causes of profits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

5 Conclusion 17

ii
1 Introduction
Expand

The profitability of contrarian and momentum strategies has been an issue of continuing debate in the Efficiency
Market Hypothesis (EMH) literature. Profitability of trading strategies implies predictability of stock returns, and
thus goes against the EMH, which, if true, would imply that no trading strategy can be profitable for sustained
periods.

However, trading strategies have been found to be profitable in numerous markets over the years, even after
adjusting for microstructure and time-varying risk, calendar and size effects.1 Contrarian and momentum strategies
have long been studied in the literature, starting from Bondt and Thaler (1985). 2 Include a s
stantial sur
If changes in stock prices are persistent, then a strategy which advises going long when stock prices are high, and
of the litt. h
going short when they are low, in other words, buying winners, and selling losers, will prove to be profitable. Such
strategies are called momentum strategies, in that they aim to capitalize on the persistence of returns. On the
other hand, if there’s significant mean-reversion in stock prices, then contrarian strategy would be profitable. This
strategy would take a contrarian view to the market, on the assumption that current losers would become winners
later.

Indian Equity Markets

India, has a higher trading and turnover ratio than most other emerging stock markets. Since the liberalization
process was started in 1992, trading has boomed, and investor participation in the stock markets has increased.
From an average of 60,000 trades per day in 1996, NSE, recorded almost a million trades each day on an average
in 2002. Market capitalization has increased ten-fold in the last decade.3 However, the primary market has been
on a decline, since the mid-ninties. A spate of financial scams, and crises, together with declining interest rates in
recent years have led to most firms forsaking IPOs in favour of cheaper debt-financing.4 This means there’s scant
fresh capital in the equity market.. Investor composition has changed from the past, when a few brokers dominated Cite Saura
the market, to a large proportion now being retail investors, and Financial Institutional Investors (FII)s. The total thesis
number of investors in the capital markets has gone up from about nine million in the early 1990’s to about 20 million
by 1997. This growth is higher among rural households than urban, and there’s an increasing institutionalization
in the markets. There has also been a change in the market microstructure of stock exchanges in India, from open-
outcry systems to continuous limit-order book markets, like the NSE Exchange for Automated Trading (NEAT)
system of the NSE, and the BSE Online Trading (BOLT) system of the Bombay Stock Exchange (BSE). Banks
have been allowed to finance margin trading, and short-selling has been allowed to a limited extent NSE (2002, pp.
147). These systems have facilitated higher trading intensities than was possible earlier.

In this paper, we examine if CIS/MIS trading strategies have been profitable in the Indian markets in the last several
years, examine their causes, and also how profitability has changes over variations in the return horizon.As Conrad
and Kaul (1998) point out, the fact two “diametrically opposite” strategies have been found to be successful
in the literature seems ’perplexing’, albeit over different return horizons. We evaluate a set of contrarian, and
momentum trading strategies on intraday stocks buy/sell prices, and calculate their profits. Since transaction costs
1 Bondt and Thaler (1985) find that loser portfolios of 35 stocks outperform the market by 19.6% on an average, three years after
portfolio formation, and loser’s outperform winners by about 24% in that period.
2 See Conrad and Kaul (1998), Hameed and Kusnadi (2002), Jegadeesh and Titman (2002) and Kang et al. (2002), among others.
3 NSE (2002, pp. 18)
4 The composition of debt-equity financing has changed from 27.61-72.39 in 1995-96 to about 83.12-16.88 in 2001-02.NSE (2002, pp.

59)

1
are important in deciding if gains from a trading strategy are tanglible, profits from trading strategies are not
scale-invariant. As trade sizes increase, costs to trade rise as well, eliminating the gains to a trading strategy.

Analyses of the profitability of trading strategies do not include costs of trading that are non-proportional. We have
considered the price impact of a trade in order to explicitly incorporate trading costs in this paper. Statistically
significant returns from our strategies are thus economically viable. We have used the limit-order book of the NSE
to calculate exact buy/sell prices for a given size of trade, and used them in evaluating our strategies.

We next try and find out if the causes behind the profitability of contrarian, and momentum trading strategies. We
implement zero-cost momentum trading strategies, and determine their profit components. Initially we find out if
there’s a size effect in the profitability of our strategies. We then determine if profits are due to common-factor
realizations, or due to stock-specific causes. Profits from the strategies tested are then decomposed into components
due to cross-covariances, serial-covariances, and mean variation in expected returns (Lo and MacKinlay (1990)).

This paper is structured as follows: In section 2, we provide the details of the procedure we’ve followed in this
paper. This includes discussion on Loser-Winner analysis, trading costs, and the decomposition of momentum
profits. Section 3 explains the data we’ve used, and intial exploratory analysis on it. Section 4 discusses the results,
and key findings. Section 5 concludes.

2 Methodology

Various causes have been ascribed to the profitability of contrarian and momentum trading strategies. These
range from hypotheses abut investor behaviour, like overreaction, underreaction, herd behaviour, etc to temporary
departures from market efficiency, and unaccounted factors in asset-pricing models.

If investors over-react to firm-specific news about earnings, and dividends, then prices would always tend to correct
themselves to their fundamentals, after the intial euphoria is over. A strategy that goes contrarian to the market
view, in other words, buying losers, and selling winners, would be profitable in such a case. Bondt and Thaler
(1985) find that loser portfolios outperfomed the market about 19.6% on average on monthly returns in the period
1965-82. On the other hand, if investors underreact to news, then winners today will remain winners tomorrow,
and may improve performance. The ’price-ratio’ and the ’price-earnings’ anomalies belong to this category. In the
price-earning anomaly, firms with low P/E ratios are undervalued.5 Once future earnings are higher than expected,
the stock price adjusts, but the market takes longer. Basu (1977) shows that firms with extremely small P/E ratios
earn higher risk-adjusted returns than high P/Es. 6 Another strand of the literature attributes this phenomenon
to the change in ’Winner’ and ’Loser’ stocks’ risk premia over time. While returns to losers are higher, so is their
risk. And while winners ’lose’, their risk goes down. Accounting for time-varying risk premia could account for
CIS/MIS profits.7 .

Irrational behaviour, and herding are some of the other reported causes.

In a portfolio, if the returns of two stock are serially independent, but are positively cross-correlated, then a
Contrarian Investment Strategy (CIS) would still be profitable. Lo and MacKinlay (1990) find this to be the major
5 since
after a low earnings the outlook on the firm is gloomy.
6 Some behavioural studies attribute overreaction to the presence of quasi-rational agents in the economy, who do not revise their
expectations according to the Bayes’ rule, and of agents with heterogenous expectations. See Bondt and Thaler (1985).Lehmann (1990)
avers that due to the cognitive misperception of investors, there are ’fads’ that result in return-reversals, and predictibility.
7 See Chan (1988) for more details.

2
cause of CIS profits, rather than investor overreaction. Further they find that large stocks lead smaller stocks in
assimilation of news, and that this lead-lag related size-effect is the major cause of the positive cross-correlation
across stocks. Jegadeesh and Titman (1995) however find that it’s not the lead-lag structure that causes contrarian
and momentum profits, but overreactions to certain common factors in the economy. They test their hypothesis
using a single-factor asset-pricing model.

Since stocks with a large market capitalization would tend to be more liquid, one would expect that ”large” stocks
would ’lead’ smaller stocks, in the assimilation of news. The lead-lag structure in correlations of stock returns has
a definite link with the size effect. But it’s interesting to note that the direction of the linkage could be either way.
For instance, it’s been found that small stocks lead larger stocks in some markets.

At the transaction level, prices fluctuate between the lowest ask, and the highest bid. This leads to the familiar
negative first-order correlation in stock returns at that level. Profits of a trading strategy that tends to capitalize
on this phenomenon would be spurious.

2.1 Portfolio Selection

Trading strategies are usually tested on portfolios. Portfolio selection can be implemented in terms of what’s called
the F/S/H framework. Stocks are ranked into groups on the basis of their mean returns during a formation period
F in the past. The top group of stocks constitutes the ’winner’ portfolio, and the bottom group, the ’loser’ portfolio.
After a ’skip’period S, buy/sell positions are taken depending on the strategy employed, and are held for a holding
period H. And returns on these positions are calculated on evaluation during F and holding period H. This cycle
is repeated till the end of the sampling period.

Portfolios are typically equally weighted. Ranks of stocks within a portfolio have no bearing on the performance of
a strategy. (See Kang et al. (2002); Korajczyk and Sadka (2002) for more details. For the analysis formation and
holding periods of 25, 50, 75, 150, and 250 days, and skip period of a day. Our parameters for the strategy could
be summarised as follows

Formation Skip Holding

25 1 25
50 1 50
75 1 75
150 1 150
250 1 250

Table 1: Formation and Holding periods used as the strategy parameters.

Hence our strategies are based on sets (25/1/25) to (250/1/250) (25). For instance, a (100/1/25) returns strategy
would be implemented by selecting stocks on the basis of their mean returns over the last 100 days. After a skip
period of 1 day, either buy/sell positions would be taken, and held for 25 days, and reversed thereafter, and the
cycle repeated. A key point to be noted in this process is that the composition of the portfolios may easily change
during the implementation of the strategy over the entire period, since it’s reviewed every 100+25 days.

Choice of overlapping vs. non-overlapping data

When returns are calculated, we pay special attention to trading costs as follows.

3
2.2 Trading costs

Profits in a trading strategy are usually assumed to be scale-invariant. However, today we know from the market
microstructure literature that transaction costs play a very important role in determining whether profits from
these trading strategies translate into tangible economic profits, and are not just ’paper-profits’. Earlier analyses
of profits assume that transactions take place at the bid-ask, or the best price prevalent in the market. However,
returns to trading strategies depend on transaction costs, and thus the size of the trades involved. In this section
we present a short sketch of the transaction costs involved in trading, with paricular reference to India.

Transaction costs could be explicit, implicit and also include missed opportunities to trade (Harris (2003)). Explicit
trading costs comprise the brokerage fees that traders pay to trade, payments to the exchange, commissions paid
by traders to brokers, stamp duties, registration fees, and capital gains taxes. Aside from some fixed costs explicit
trading costs are usually proportional to the size of the trade involved. For instance, a buy-sell round trip at the
ECN ICICI-Direct is about 1.77%. Capital gains tax in India is about 30% if the gains are realized within a year.8
Fixed costs include the INR 5,000 that every broker has to pay to the exchange every year, if his turnover in the
year is less than INR 10 million. For larger volumes, the registration fee becomes proportional, and equals 1 basis
point of the value over INR 10 million. Traders pay transaction charges to the exchange at 0.004% of the turnover.
Brokerages range from 0.15% to 2.5%, pegged by the Securities and Exchange Board of India (SEBI). In addition,
the brokers also pay stamp duty, turnover fee, servie tax, and stamp duty. They also pay service charges to the
depositories.9 However, explicit transaction costs have been steadily reduced over the years by the SEBI10 .

Implicit costs are intangible costs that traders pay when they trade. These include the price impact of trades,
also called the impact cost, and the bid-ask spread. For instance, a buyer may have to pay more than the price
prevailing in the market, if the order is large, and a seller may realize less than the market price. Price impacts are
usually calculated as being over and above a specified benchmark price, or estimated using econometric models.
While costs rise with the size of trades, they’re non-proportional.

Econometric models estimate the spread that exists between the trading price, and the benchmark, and it’s com-
ponents, based on the absolute transaction price, the price reversals caused when traders trade, and order flows.
These include Roll’s serial covariance estimator (Roll (1984)), the Glosten-Harris (Glosten and Harris (1988)), and
the Glosten-Milgrom (Glosten and Milgrom (1985)) non-proportional trading cost models.See O’Hara (1995) for a
detailed survey. For a trade of size Q, price impacts IC based on pre-specified benchmark prices P ∗ are usually
calculated as 
+1 Buy Trade
IC = Q ∗ sgn(trade) ∗ (P − P ∗ ), sgn(trade) = (1)
−1 Sell Trade

There are various benchmarks used in practice. The gap between the trade price, and the benchmark gives the
spread in the trade. The effective spread is calculated as twice the gap between the trade price, and the midquote
at the time-of-trade. The realized spread is twice the gap between the trade price, and the price after a specified
time interval, usually the closing price. Sometimes opening prices are also used as the benchmark11 In other
cases, the volume-weighted average price is the preferred benchmark. The sign of the trade, (buy/sell), is done
8 Currently equity held for periods longer than one year is considered investment and is exempted from capital gains tax.
9 More than 99} of trading takes place in the demat form.
10 See NSE (2002) for more details.
11 These are called ’Perold’s implementation shortfalls’.

4
using the Lee and Ready algorithm (Lee and Ready (1991)).12 Other than these, there’s the opportunity cost of
missing otherwise profitable trades. These occur when traders don’t execute their trades on time. Costs of trading
differ across agents in the market. For instance, the average transaction cost for retail investors is higher than for
institutional investors, as the latter don’t pay mark-to-margins. Further, among the institutional investors, mutual
funds, and financial institutions pay lower brokerages than foreign institutional investors (NSE (2002, pp. 135)).

While proportional costs have been included in studying trading profits since Keim (1983), Lehmann (1990), and
many others, the studies on the effect of non-proportional costs on trading strategies are relatively recent in nature.c
Ball et al. (1995), Grundy and Martin (2001), and others have shown that microstructure effects could significantly Add s
reduce the profits of a strategy. Korajczyk and Sadka (2002) have tried to quantify the limits to which gains could more his
be made from momentum-based strategies before transaction costs first make profits statistically significant, and here.
then drive them down to zero. They consider proportional and non-proportional models of price imapct in their
paper.

In this paper, we consider control for one measure of non-proportional trading costs, the price impact while studying
trading profits. We calculate the price impact of a trade in a give stock, using the entire market by price for that
stock. We use snapshots of the entire limit-order book of the NSE for this purpose. In this way, we do not have
to use the Lee and Ready algorithm to identify the trades. We don’t also have to use any econometric model to
estimate the price impact, since we have the actual values.

Impact cost is the price that a buyer pays over and above the ideal buying price of stock, or the seller forfeits when
he sells the stock. We can calculate this cost for any ’reasonably-sized’ trade for a stock, and for this we have the
order-book. This cost is proportional to the size of the order, but the relationship is nonlinear. Since the buy and
sell prices in a strategy are dependent on the size of the trade, so are our profits.

For a buy-order of quantity QBuy

k,t in stock k at time t, the benchmark price at which the trade should ideally occur
is
∗ 1h Buy Buy
i
Pk,t = (max Pi,t ) + (min Pi,t ) (2)
2 i i

The total buying cost is

m
X m
X
Buy
Pk,t = qi,k,t · pBuy
i,k,t 3 qi,t = QBuy
k,t (3)
i=1 i=1

and p1,k,t > p2,k,t > · · · > pm,k,t . The price impact is for stock k at time t for a buy-order QBuy
k,t is then given by

Buy
" #
Pk,t
ICQBuy = 100 −1 (4)
k,t ∗ · QBuy
Pk,t k,t

Here pi,k,t i = 1, . . . , m, are the limit-orders on the buy side in the LOB, for stock k, at time t and qi,k,t i = 1, . . . , m
are the quantities available at those prices that progressively clear the order. The impact cost on the sell side can be
similarly calculated. We avoid market-orders, and stop-loss orders in this calculation.13 Since we have the Market
By Price (MBP) each hour, we’re able to calculate the actual buy/sell prices of a stock, for an order of INR 10,000.
12 The Lee and Ready (1991) algorithm assumes that traders trade to satisfy their demands for liquidity. Hence if the trade took

place near the bid then, then it’s a buy, and if it’s near the ask, then it’s a sell. If the trade is on the midquote, then the algorithm
checks for the last price change. There’s no way of knowing if the order gets executed over multiple trades. Further, traders may not
always trade for liquidity.
13 Potentially, these strategies could be implemented on both overlapping and non-overlapping data—we stick to overlapping data to

for it’s improved power properties.

5
Alongwith with the ’skip’ period introduced in our strategies, this ensures that we avoid potential discrepancies
that could ensue with the bid-ask bounce in intraday returns. Assuming that trading takes place once a day, the
best time to trade would then be at the point where the spreads are the lowest. It has been found that for the
NSE spreads come down significantly during the course of the day (NSE (2002)). Accordingly, we have based our
analysis on the prices calculated out the last snapshot available of the day, at 1400 hrs.

2.3 Trading strategies: Loser-Winner analysis

In this paper, we focus on four different contrarian and momentum strategies. framework, and calculate their
profits. These strategies are described below:

1. Contrarian-Loser This strategy involves buying a loser selected at the end of the formation period F,
waiting for the skip period S, and then entering into positions. At the end of the holding period H, positions
are reversed. This strategy assumes that the market has overreacted, and expects these losers, to become
winners by the end of the holding period. The returns from such a strategy are given by
PN L,Sell PN L,Buy
L i=1 Pi,t − i=1 Pi,t−H
Rt,CIS = PN L,Buy
(5)
i=1 Pi,t−H

2. Contrarian-Winner The contrarian winner strategy involves selling a winner portfolio over the formation
period, entering into positions at the start of the holding period, and then buying it back at the end of the
holding period.14 Returns for this strategy are given by
PN PN
W
− i=1 PtW,Buy + i=1 Pt−H
W,Sell
Rt,CIS = PN W,Sell
(6)
i=1 Pt−H

3. Momentum-Loser This strategy short-sells losers, and then buys them back, at the end of the holding
period. Returns for this strategy are given by
PN L,Buy PN L,Sell
L
− i=1 Pi,t + i=1 Pi,t−H
Rt,M IS = PN L,Sell
(7)
i=1 Pi,t−H

4. Momentum-Winner This strategy assumes that winners chosen over the formation period shall remain
winners, and buys them. Positions are reversed at the end of the holding period.
PN W,Sell PN W,Buy
W i=1 Pi,t − i=1 Pi,t−H
Rt,M IS = PN W,Buy
(8)
i=1 Pi,t−H

All the strategies used so far are ’costly’ in the sense that they need some initial capital to get started. Since costs
of borrowing funds are non-zero, returns from such strategies must be at least be more than the risk-free rate of
return, to make them viable. Hence we also implemented zero-cost momentum strategies, where winners are bought
and losers are sold simultaneously, and zero-cost contrarian strategies, where losers, are bought, and winners are
14 We assume that short-selling is allowed for this strategy.

6
sold simultaneously. No assumption about initial available capital is made for such strategies. Note that these are
zero-cost strictly in the case where buying and selling costs are identical. Returns from these strategies are given
by

N 
X  XN  
W,Buy W,Sell L,Sell L,Buy
Rt,ZC,Contra = −Pi,t + Pi,t−H + Pi,t − Pi,t−H (9)
i=1 i=1
N 
X  XN  
W,Sell W,Buy L,Buy L,Sell
Rt,ZC,M oment = Pi,t − Pi,t−H + Pi,t − Pi,t−H (10)
i=1 i=1

L,Sell
Here, for instance, Pi,t implies the selling price of the ith stock in the loser portfolio. Returns to zero-cost trading
strategies could also be scale-adjusted. In this case, we normalize the each of the components in the expressions
(See Eqns 5, 6) by the entering prices, like in Eqns. 1-4. Another advantage of using a zero-cost momentum strategy
is that it’s been well-studied in the literature, for establishing the causes behind profits from such strategies.

2.4 Trading Strategies: Size-effect, and other causes

Once we estimate the profits and if they’re significant, we formally examine the causes behind profitability of the
contrarian, and momentum strategies tested.

We consider a zero-cost momentum trading strategy on an arbitrage portfolio, as in Lehmann (1990). This strategy
buys ’winners’ and sells ’losers’ simultaneously. Profits could also be due to what’s called the size-effect, as described
in section 1. The size effect has been extensively documented in the literature. Stocks with relatively small market Will desc
capitalizations tend to outperform those with larger market capitalization. Therefore, to study the causes behind the size-effec
profits of momentum strategies in India, and to control for the size effect, we divide the set of stocks we study detail there.
into quintiles based on their market capitalization in the middle of the sampling period first, and then calculate
the returns of the strategy on each of the size-sorted quintiles. Within each quintile portfolio, we calculate the
deviations of the returns of stocks from the portfolio mean return. Stocks with positive deviations are then ’bought’,
in that they are assigned positive weights in the portfolio, and stocks with negative deviations are similarly ’sold’.

We test this strategy on holding periods ranging from 100 to 1000. We initially use the Jegadeesh and Titman
(1995) factor model to find out if returns from the strategy being tested come from common market, or from
stock-specific factors. We then use the Lo and MacKinlay (1990) decomposition to find out if these profits are
due to investor overreaction, underreaction, or due to the lead-lag structure of stock returns, shown as serial, and
cross-correlation.

Let Rit , ı, . . . , N be the returns of the N stocks in this portfolio, at time t. The weights are given by

1 
wit = ± Ri,t−k − Rm,t−k , i, . . . , N. (11)
N
A contrarian strategy in this setup would imply a negative sign for the weights, and would imply we sell stocks
which give higher returns than the index (Winners), and buy those which don’t (Losers). A momentum strategy
would imply precisely the opposite positions to take.

7
The scale-invariant profits from such a strategy are

N
X
πt (k) = wit (k) · Rit (12)
i=1

As we’d done in the previous section, we can view this strategy in terms of the F/S/H framework. For a trading
strategy that squares up at time t, we take positions at time t − F + H.

1 
wit (F + H) = − Ri,t−F +H − Rm,t−F +H (13)
N

Hence the value of the portfolio at time t − H is

N
X
Vt−H = wit (F + H)Pi,t−H (14)
i=1

At time t, we square off our positions,

N
X
Vt = wit (F + H)Pi,t (15)
i=1

And the portfolio profits over the period H are

N  
X Pi,t − Pi,t−H
πt = wit (F + H) (16)
i=1
Pi,t−H

The availability of the order book snapshots of this data enables us to generate actual buy/sell prices, free from
bid-ask bounce, as detailed in Section 2.2. With buy/sell prices being different (and allowing for short-selling),
portfolio returns are dependent on the sign of the weights, and are not scale-invariant. A key difference between the
portfolio formation process this time is that while the composition of the portfolio remains unchanged, the weights
attached to the components change. In effect, there are ’winners’ and ’losers’ in each portfolio. The weights of
the stocks in the portfolio are dependent on the trade being a buy or a sell. This strategy is implemented on
overlapping periods, to increase the power of our results.

Buy
PitSell − Pi,t−1
wi,t−F > 0 ⇒ rit = Buy
(17a)
Pi,t−1

PitBuy − Pi,t−1
Sell
wi,t−F ≤ 0 ⇒ rit = Sell
(17b)
Pi,t−1

Lo and MacKinlay (1990) posit that instead of investor overreaction, a majority of contrarian profits could be
attributed to the ’lead-lag’ structure between stocks, in which prices of one stock ’follow’ another. There exists
empirical evidence of the link between the lead-lag structure, and the size-effect, wherein if prices of small stocks
go up, prices of large stocks follow. Kang et al. (2002) find that in the Chinese market, small stocks follow large
stocks, and in the opposite direction. This is tested using the sample autocorrelation matrices of each of the five
size-sorted sample cross-correlation matrices. Consider first the asymmetric cross-correlation matrices ΥK of the
five size-sorted quintiles and the Equally Weighted Market Index (EWMI), given by

8
 Υk = D−1/2 E[(Zt−k − µ)(Zt − µ)0 ]D−1/2 (18)

where Zt = [R1t R2t R3t R4t R5t Rmt ]0 is the vector of returns from the size-sorted quintiles, and the equally-
weighted market index. D is given by diag[σ12 , σ22 , σ32 , σ42 , σ52 , σm
2
, ], and µ = E[Zt ]. R1t is the smallest quintile,
and R5t is the largest. The last vector is the equally weighted market index. These matrices convey the lead-lag
relations, if any, between the size-sorted quintiles. If the covariance between current returns of a small stock and
lagged returns of a large stock is more than that between the lagged returns of the small stock, and the current
returns of the large stock, then events impact the large stock before the small stock, and the smaller stock ’lags’.

2.5 Decomposition of Momentum Investment Strategy (MIS) profits.

itemize
Jegadeesh and Titman (1995) estimate a one-factor model of stock returns, to access CIS/MIS profitability, their
assumption being that the lead-lag structure in stock returns arises due to the response period of the stock prices
to common factors, and not due to over- or underreactions.

The model is given by

rit = µi + b0,i ft + b1,i ft−1 + eit (19)
15
CIS/MIS profits are then given by

E(π c ) = −σµ2 − Ω − δσf2 (20a)

E(π m ) = σµ2 + Ω + δσf2 (20b)

N
1 X
σµ2 = (µi − µ)2 (21a)
N i=1

N
1 X
Ω= cov(ei,t , ei,t−1 ) (21b)
N i=1

N
1 X
δ= (b0,i − b0 )(b1,i − b1 ) (21c)
N i=1

Here σµ2 is the cross-sectional variation in expected returns, Ω is the mean serial covariance of the idiosyncratic
component of returns, and δ denotes the response to common factors. Ω measures the reaction of stock returns to
firm-specific factors, and thus contributes +ve(-ve) to momentum(contrarian) profits. δ measures the reaction of
stock returns to common-factor realizations. We have chosen the market index NIFTY as the common factor in Will add m
this model. We first used this decomposition to get prima facie evidence of whether returns to trading strategies detail here.
are due to common factor realizations, or due to stock-specific factors. itemize

The Lo and MacKinlay (1990) decomposition of profits to trading strategies differentiates between the cross-
sectional dependence across stock returns and the serial depenence in individual stock returns.
15 (See Kang et al. (2002) for more details.)

9
MIS profits can be decomposed thus:

E [πt (k)] = −Ck − Ok + σ 2 (µ) (22)

1  0 
Ck = 2
e Γ k e − tr(Γk )
N 
N −1
Ok = − tr(Γk )
N2
N
1 X 2
σ 2 (µ) = (µi − µm )
N i=1
µm = µ0 e/N.

Here Ck depends only on the cross-covariance terms of the auto-covariance matrix Γk , Ok depends on the serial
covariance terms, and σµ2 depends on the cross-sectional variation between the unconditional means of the stocks.
For an MIS strategy to be profitable, we expect Ck , Ok < 0.

If there’s strong positive autocorrelation in the EWMI returns, and weak negative autocorrelation among individual
stock returns, then there must be significant positive autocorrelation across securities. Ck depends only on the
off-diagonal elements of the autocovariance matrix Γk , Ok depends only on the diagonal elements of Γk , and σ 2 (µ) is
a constant, independent of k, and is the cross-sectional variance of mean-returns. We ran both the decompositions
on each of the size-sorted quintiles generated. Our estimates are robust to heteroscedasticity, and autocorrelation.
This completes the methodology followed. We next describe the data used in the paper.

3 Data

We utilize snapshots of the limit order book of the NSE, which is available four times viz., at 1100, 1200, 1300,
and 1400 hours a day.16 . Details on the NSE Release A CD The NSE Release A CD comes once a month from
the NSE. It contains information and raw data about each and every trade that took place in the month, in each
stock. It contains the daily open, high, low, and close prices for all the stocks traded in the exchange, and all
the important announcements by the NSE. Alongwith this data, it contains the all the intraday values taken by
the major indices S & P CNX Nifty (NIFTY), DEFTY, and MIDCAP. It also contains snapshots of the entire
limit-order book, the market by price, about four times during the day, and the trade master, whcih contain the
stock masters for each scrip traded. We use the trades files to calculate the stock returns for the portfolios. Table
2 gives a sample of the raw trades file. The first column denotes the serial number of the trade that day. NSE
records more than 900,000 trades per day on average. The second and the third columns denote the stock, and it’s
series, where ’EQ’ denotes common stock. Timestamp records the time at which the trade occurred, to the nearest
second. This naturally gives rise to multiple timestamps per second. We discretize this data to get the returns for
the stocks we’re interested in.

The order book files provide the bid and asks for all the stocks trading at that point in time. Figure 3 gives a
sample of the order-book. Here the first column gives the order ID no. The second and the third columns indicate
the stock for which the order is given, and it’s series—EQ means a common stock. Q implies the quantity of the
limit-order submitted, and P denotes the price at which it’s to be executed. Time is the timestamp at which the
16 There have been periods in the interim where only two-three of the snapshots were available. We have thus presented our analyses

on the final snapshot of the day, at 1400.

10
 Trade Stock Series Timestamp Price Quantity
9991 BANKBARODA EQ 09:59:10 57.50 250
9992 KPIT EQ 09:59:11 179.70 190
9993 BPCL EQ 09:59:11 266.50 50
9994 GEOMETRIC EQ 09:59:11 521.70 20
9995 HUGHESSOFT EQ 09:59:11 234.05 40
9996 HINDPETRO EQ 09:59:11 278.50 400
9997 HINDPETRO EQ 09:59:11 278.50 250
9998 HINDPETRO EQ 09:59:11 278.50 4500
9999 HINDPETRO EQ 09:59:11 278.25 20
10000 HINDPETRO EQ 09:59:11 278.20 1000

Table 2: Sample trades file from the NSE Release A CD for June 11, 2002.

limit-order was submitted, or last modified. B/S denote if the order is buy or a sell. F1-F3 are flags that accompany
the order. F1 is called the day-flag, and the four characters denote (1) Day order, (2) Good till date, (3) Cancel,
(4) Immediate/Cancel. The most common values are “ynnn”. F2 denotes the quantity flags (1) Minimum fill, (2)
All or none, (3) Disclosed quantity. Usually this flag is “nnn”. F3 denotes the price flags, viz., (1) At the open
(ATO) price, (2) Market price, (3) Stop loss order.17 A typical order-book snapshot contains more than 200,000
entries. We restricted our analysis to those stocks that had non-missing orders at least 5200 times in the sample
period of about 1618 days (at the four snapshots considered) which ranged from March 1996 until October 2002.
This implied that any stock had to be traded at least 80% of times, to get selected in our portfolios. This selection
gave us 200 stocks. The list of stocks is provided in the appendix.

Order ID Stock Series Q P Time B/S F1 F2 F3 Book MinFill Disc GTD

200206115271013 RELIANCE EQ 50 227.15 11:03:31 B ynnn nnn nnn RL 0 0 .
200206115009106 RELIANCE EQ 100 228.60 09:56:27 B ynnn nnn nnn RL 0 0 .
200206115009187 RELIANCE EQ 100 233.60 09:56:27 B ynnn nnn nnn RL 0 0 .
200206115128316 RELIANCE EQ 10 234.10 10:17:47 B ynnn nnn nnn RL 0 0 .
200206115128036 RELIANCE EQ 10 237.10 10:17:44 B ynnn nnn nnn RL 0 0 .

Table 3: A sample of the limit-orders for RELIANCE at 1400 hours on June 11, 2002.

Using the snapshots available, we calculated the price impacts of all the 200 stocks under study on the buy and the
sell sides, for an order of INR 10,000. 18 for all 1618 days under study, at all times the snapshots were available.
Using these costs we were able to calculate the exact buy and sell prices, inclusive of the price impact, for a trade,
as detailed in Section 2.2. Our choice of the frequency of the data depends on the fact that correlation structure
in daily, and HF data is much stronger than in weekly, or monthly data. Further, daily data suffers the least from
spurious correlations resulting from non-synchronous trading.
17 Information is from the NSE Release A CD, for December 2002.
18 A sample portfolio with 40 stocks thus is worth about INR 400,000. We have chosen the size of the portfolio to be about double
the size of minimum lot-size required to trade on the derivatives segment of the NSE.

11
 Date Time Shares Purchase Ideal Bought Buy Sold Sell
20021001 110000 39 10000.00 254.00 39 254.10 39 253.90
20021001 120000 39 10000.00 253.57 39 253.65 39 253.50
20021001 130000 39 10000.00 252.50 39 252.60 39 252.40
20021001 140000 39 10000.00 252.45 39 252.59 39 252.40
20021003 110000 41 10000.00 243.80 41 243.85 41 243.75
20021003 120000 41 10000.00 243.32 41 243.35 41 243.30
20021003 130000 41 10000.00 242.70 41 242.75 41 242.65
20021003 140000 41 10000.00 241.38 41 241.45 41 241.30
20021004 110000 41 10000.00 239.00 41 239.10 41 238.90
20021004 120000 41 10000.00 238.70 41 238.75 41 238.65
20021004 130000 42 10000.00 236.82 42 236.90 42 236.71
20021004 140000 42 10000.00 236.88 42 236.95 42 236.80

Table 4: Sample order-book data for RELIANCE in October 2002, for a buy/sell order of Rs. 10,000.

4 Results

Here we present the results of testing the four trading strategies. Returns are arranged across multiple holding
periods, for each formation period. Table 5 provides the results for each of the ’costly’ as well as the zero-cost
strategies.

The contrarian loser (CL) strategy buys losers over the formation period, and sells them after the holding period.
This strategy is found to be profitable in the short-run, in the sense that if stocks are chosen over the last 25 days,
then the strategy gives positive returns. However, profits come down as the holding period increases. Profits from
this strategy for the shortest formation period are statistically significant. The contrarian winner strategy and the
momentum loser strategies’s returns are also provided for the purposes of illustration. If the CL strategy makes
money, one expects the ML strategy to make losses.For the shortest formation period therefore, (25) we find that
the MW strategy clearly makes money, and that the profits rise as the holding period increases. The momentum-
winner MW strategy buys winners, hoping that they’ll continue to be winners. For the shortest formation period,
the MW strategy is profitable, and profits are more than in the CL strategy, but not statistically significant.

In ideal circumstances one would have expected returns from the contrarian loser strategy to be exactly opposite the
returns from the momentum loser strategy. But since we have different buy/sell prices as the buy/sell impacts from
the trades are different, returns from these strategies do not cancel out. This makes testing the zero-cost contrarian
and momentum strategies all the more relevant. Returns from the zero-cost strategies have been calculated in two
ways, adjusted wherein the returns are normalized by the entering positions19 , to help compare with the rest of
the strategies, and the unadjusted returns. We first discuss returns from the adjusted zero-cost strategies first.
The zero-cost momentum strategy is profitable for the shortest formation period, with profits rising from 0.263 to
0.353, as the holding period rises from 25 days to 250 days. The strategy proves to be profitable in for almost
all the formation periods, from 25 to 250 days, except for the 50-day formation period. However, profits come
down as the the formation period increases, from 0.353 to about 0.031, for the holding period set at 250 days.
The CL strategy is also profitable but far lesser than the MW. The CW, and the ML strategies incur losses. The
zero-cost momentum strategy is profitable, and profits rise as positions are held for longer periods. Returns to the
scale-adjusted zero-cost strategies are statistically significant at 10%.
19 Details on the calculations of adjusted returns

12
 Form Hold CL CW ML MW ZCC ZCM
25 25 0.050 (0.033) -0.391 (0.340) -0.072 (0.034) 0.335 (0.307) -0.341 0.263
25 50 0.046 (0.032) -0.388 (0.327) -0.068 (0.034) 0.333 (0.297) -0.342 0.265
25 75 0.046 (0.036) -0.365 (0.304) -0.068 (0.038) 0.312 (0.275) -0.319 0.244
25 150 0.041 (0.034) -0.397 (0.335) -0.062 (0.036) 0.341 (0.302) -0.356 0.278
25 250 0.038 (0.040) -0.476 (0.411) -0.061 (0.042) 0.414 (0.373) -0.437 0.353
50 25 0.186 (0.169) -0.045 (0.017) -0.219 (0.182) 0.023 (0.016) 0.140 -0.196
50 50 0.183 (0.161) -0.062 (0.025) -0.216 (0.173) 0.036 (0.022) 0.120 -0.179
50 75 0.237 (0.216) -0.060 (0.026) -0.273 (0.232) 0.034 (0.023) 0.176 -0.239
50 150 0.217 (0.206) -0.051 (0.021) -0.253 (0.222) 0.026 (0.019) 0.166 -0.226
50 250 0.234 (0.232) -0.061 (0.028) -0.273 (0.251) 0.035 (0.025) 0.172 -0.238
75 25 0.016 (0.016) -0.106 (0.055) -0.037 (0.017) 0.078 (0.050) -0.089 0.041
75 50 0.019 (0.017) -0.125 (0.057) -0.040 (0.018) 0.093 (0.051) -0.105 0.053
75 75 0.006 (0.015) -0.122 (0.056) -0.027 (0.016) 0.090 (0.050) -0.115 0.063
75 150 -0.001 (0.016) -0.119 (0.060) -0.019 (0.017) 0.089 (0.054) -0.120 0.069
75 250 -0.004 (0.019) -0.139 (0.070) -0.017 (0.021) 0.105 (0.063) -0.143 0.088
150 25 0.021 (0.017) -0.060 (0.020) -0.041 (0.018) 0.037 (0.019) -0.039 -0.004
150 50 0.016 (0.017) -0.076 (0.028) -0.036 (0.019) 0.050 (0.025) -0.060 0.013
150 75 0.014 (0.017) -0.077 (0.029) -0.035 (0.018) 0.050 (0.025) -0.062 0.015
150 150 0.006 (0.017) -0.070 (0.024) -0.026 (0.018) 0.045 (0.022) -0.064 0.018
150 250 0.002 (0.020) -0.082 (0.033) -0.023 (0.021) 0.055 (0.030) -0.080 0.032
250 25 0.028 (0.020) -0.080 (0.034) -0.049 (0.021) 0.055 (0.032) -0.051 0.005
250 50 0.025 (0.020) -0.097 (0.040) -0.046 (0.021) 0.069 (0.036) -0.072 0.023
250 75 0.024 (0.024) -0.098 (0.043) -0.046 (0.025) 0.070 (0.039) -0.073 0.024
250 150 0.013 (0.021) -0.086 (0.038) -0.034 (0.022) 0.060 (0.035) -0.072 0.025
250 250 0.009 (0.026) -0.091 (0.044) -0.031 (0.028) 0.063 (0.040) -0.082 0.031

Table 5: Returns from CIS/MIS strategies evaluated on multiple F/S/H portfolios, using HF buy/sell data from
March 1996, till October 2002. Here CL ⇒ Contrarian-Loser, and so on. Figures in parentheses are standard
errors. ZCC and ZCM are the scale-adjusted zero-cost contrarian and momentum strategies respectively.

Returns from the unadjusted zero-cost momentum trading strategies are in table 6. We find that momentum
strategies are profitable at 250-day holding periods, almost independent of the formation period used.20 Profits
from these strategies are statistically significant at 10%. The magnitudes of profits to the strategies fall as the
formation period increases. Profits in the F=25 strategies are almost three-four times the profits in the F=250 set,
independent of the holding period. This shows the preponderance of the momentum effect in the market. Profits to
contrarian strategies fall as the positions are held for longer periods, but they rise for momentum strategies. Profits
for the momentum-winner strategy are highest when the formation period is 25 days. On average, momentum
profits are higher than contrarian profits, even after factoring in transaction costs. However, these profits are
proportional to the trading costs involved, which in this case are based on orders of INR 10,000 for each stock.

We thus find that momentum strategies are profitable in the Indian markets, at least in the short term, even after
adjustments for non-proportional costs like the price impact, are made. Profits rise as positions are held for longer
periods. But formation periods are to be kept small for the strategies to make money. The Indian market is driven
by momentum.

Having established that momentum strategies are profitable in our analysis, we now examine the causes behind
20 The strategy is profitable for at (25/1/250), (50/1/150), (50/1/250), (75/1/150), (75/1/250), (150/1/250), (250/1/150), and

(250/1/250).

13
their profitability. Consider the sample cross-correlation matrices Υ0 , Υ1 , and Υ2 .

These matrices give the correlation between the current and the lagged values of all the size-sorted quintiles, and
the equally-weighted market index. The correlation structure between these quintiles provides an initial idea about
the lead-lag effect, and it’s linkage with size. In the US markets, Lo and MacKinlay (1990) find that large stocks
lead smaller stocks, and in the same direction, that is, a rise in the large stocks is followed by a rise in the smaller
stocks. In the Chinese markets, small stocks lead large stocks, in the opposite direction.Kang et al. (2002).

 
100.00 81.45 77.66 71.02 65.23 86.79
 

 81.45 100.00 88.27 82.58 77.07 94.13 

 77.66 88.27 100.00 86.48 81.91 95.01 
Υ0 = 
 


 71.02 82.58 86.48 100.00 83.73 92.27 


 65.23 77.07 81.91 83.73 100.00 89.02 

86.79 94.13 95.01 92.27 89.02 100.00
Symmetric.  
8.897 11.18 10.81 9.678 9.993 11.07
 

 9.392 8.475 9.950 9.688 9.793 10.33 

 7.230 7.887 7.857 8.325 8.577 8.708 
Υ1 = 
 


 3.605 4.686 5.169 5.555 6.294 5.512 


 2.380 2.631 3.194 4.341 4.402 3.671 

7.003 7.719 8.188 8.304 8.615 8.696
Asymmetric.  
8.469 8.526 8.457 8.728 7.163 9.038
 

 8.167 6.488 6.818 7.562 5.782 7.604 

 6.136 6.498 6.287 7.303 5.742 6.972 
Υ2 = 
 


 4.902 5.236 5.491 5.678 5.334 5.816 


 3.378 3.495 3.661 5.315 3.820 4.264 

6.856 6.654 6.754 7.611 6.107 7.414
We see that the above-diagonal entries are almost always larger than the below-diagonal entries:

e.g., Corr(R5,t−1 , R1t ) < Corr(R1,t−1 , R5t )

Correlation between lagged values of smallest quintile and current values of the largest quintile is 2.38% but that
between the current values of the smallest quintile, and the lagged values of the largest quintile is 9.99%. To get
this idea clear, consider the difference matrices

 
0.000 1.786 3.579 6.074 7.613 4.071
 

 −1.79 0.000 2.063 5.002 7.162 2.611 

0
 −3.58 −2.06 0.000 3.156 5.383 0.520 
Υ 1 − Υ1 = 
 


 −6.07 −5.00 −3.16 0.000 1.953 −2.79 


 −7.61 −7.16 −5.38 −1.95 0.000 −4.94 

−4.07 −2.61 −.520 2.792 4.944 0.000

14
  
0.000 0.359 2.321 3.826 3.785 2.182
 

 −.359 0.000 0.320 2.327 2.287 0.950 

0
 −2.32 −.320 0.000 1.813 2.080 0.218 
Υ2 − Υ2 = 
 


 −3.83 −2.33 −1.81 0.000 0.020 −1.79 


 −3.78 −2.29 −2.08 −.020 0.000 −1.84 

−2.18 −.950 −.218 1.794 1.843 0.000
 
0.000 1.253 3.025 4.781 3.999 2.780
 

 −1.25 0.000 1.146 3.464 3.992 1.530 

0
 −3.03 −1.15 0.000 2.478 3.579 0.328 
Υ3 − Υ3 = 
 


 −4.78 −3.46 −2.48 0.000 2.046 −1.98 


 −4.00 −3.99 −3.58 −2.05 0.000 −3.02 

−2.78 −1.53 −.328 1.976 3.018 0.000
The above-diagonal entries are almost uniformly positive, while the below-diagonal entries are negative. This
implies large stocks have higher correlation with lagged values of small stocks, than small stocks have with lagged
values of large stocks. Large stocks thus ‘follow’ small stocks and in the same direction. In other words, a rise in
a small-cap stock may lead to a rise in a large-cap one too!21 This shows conflicting results between the Indian,
Chinese, and the American markets about the transmission of news between large, and small stocks. Also, the
fact that serial correlations are positive (since, trΥn > 0) means that the effect of bid-ask bounce on the returns is
negligible.

4.1 Analyzing causes of profits

Results from the Jegadeesh and Titman (1995) decomposition on each of the the size-sorted quintiles, in table 7
show that almost all the momentum profits are due to firm-specific factors, rather than common factor realizations,
or due to the cross-sectional variance of expected returns. Expected profits go down with size, from 7.15 for the
smallest quintile, to 4.78 for the largest quintile.

It would now be interesting to know how these profits are divided across cross-sectional and serial dependences in
the stocks in each portfolio. Also, since the contribution of the lead-lag structure δσf2 is positive, apriori we expect
that an MIS strategy would be profitable.

Results in table 8 show that the momentum strategy is clearly profitable, for all the portfolios, but profits die
down as size increases. Investor underreaction, and overreaction to large, liquid stocks is typically much lower
than for small, relatively illiquid stocks, and this is borne out by the expected profits which fall from 46.249 to
0.312, between the smallest, and the largest quintiles, for a holding period of 100. Profits also come down as the
holding period increases, from 100, to 1000. However, what’s important here is that profits come down as we trade
in portfolios comprising larger stocks. The required investment in a portfolio also falls with size, and with longer
holding periods. It’s thus cheaper to hold a portfolio of large stocks, than one with smaller stocks. Further, there
are tangible profits in holding small stocks, though as the cost of holding the portfolio rises. We find that Ck , and
Ok are significantly negative, for all the portfolios considered, and thus contribute positively to MIS profits. The
fall in MIS profits is reflected in the values of Ck , and Ok . The mean cross-sectional variation σµ2 falls with size
from 7.547 to about 0.03.
21 The fact that large stocks follow small stocks and in the same direction need not imply contrarian profitability, since the lags of

the strategies tested here (100 to 1000), are much more than that of the ACF matrices (0 to 4).

15
 Form Hold ZCM p-value
25 25 -96.81 0.069
25 50 -11.21 0.847
25 75 -8.74 0.897
25 150 74.71 0.377
25 250 266.81 0.009
50 25 -52.47 0.321
50 50 -37.44 0.528
50 75 32.65 0.632
50 150 152.28 0.082
50 250 403.81 0.000
75 25 -53.39 0.288
75 50 -54.76 0.358
75 75 30.71 0.639
75 150 190.84 0.020
75 250 358.69 0.000
150 25 -100.91 0.050
150 50 -59.13 0.310
150 75 -41.75 0.526
150 150 116.53 0.173
150 250 259.89 0.011
250 25 -157.83 0.001
250 50 -64.57 0.267
250 75 8.60 0.891
250 150 144.29 0.080
250 250 244.85 0.014

Table 6: Unadjusted Returns from zero-cost momentum strategies evaluated on multiple F/S/H portfolios, using
HF buy/sell data from March 1996, till October 2002.

From the percentage contributions, we find that cross-sectional correlation across stocks accounts for at least 50% Will fur
of MIS profits in a portfolio. This percentage ranges between 53 and 80%. Percentage contribution of Ck rises with discuss
size, and holding period, except at k=1000 for the largest portfolio. Percentage contribution of serial covariance results
ranges from 32% to about 1%, and falls with size and the number of lags. One would expect that serial correlations
in the larger stocks would be practically absent, but would be present in the smaller stocks, and this is the case
here. Though the third component, σµ2 falls across portfolios, it’s contribution remains around 16%, owing to the
falls in it’s profits. This disproves the Conrad and Kaul (1998) hypothesis that the DGPs of stocks are basically
random walks with drifts, and that profits in CIS/MIS are due to variation in this drift rates. For the largest
quintile, the serial correlation component in MIS profits nearly dies down. We get Ok = − NN−1 tr(Γk ) < 0, for
all portfolios examined. ⇒ tr(Γk ) > 0. Returns are thus +ve correlated, for the portfolio, but the correlations
become insignificant with size. We also have Ck = N12 [e0 Γk e − tr(Γk )] < 0. This implies e0 Γk e < 0. Stocks are
thus negatively cross-correlated within a portfolio. With buy/sell data, problems of bid-ask bounce, and non-
synchronous data do not arise. Profitability of MIS profits varies from 60% over a 100-day horizon in the smallest
quintiles to about 0.77% over the same horizon, in the largest.

To summarize, we find that cross-correlation across stocks accounts for almost 50% of momentum profits in the
Indian markets. Profits due to own serial-correlation, i.e., due to Ok do not account for more than 30-35% of
momentum profits, in any portfolio, over any time horizon. In other words, overreaction, while important, is not

16
 Portfolio σµ2 Ω δσf2 E(π m ) %-σµ2 %-Ω %-δσf2
0 0.000452 7.083398 0.072699 7.156550 0.006313 98.977842 1.015845
1 0.000409 4.671060 0.014889 4.686358 0.008718 99.673570 0.317712
2 0.000623 4.166040 0.019849 4.186513 0.014891 99.510983 0.474127
3 0.000349 4.074910 0.098978 4.174237 0.008372 97.620468 2.371160
4 0.000442 4.782805 0.000660 4.783908 0.009247 99.976958 0.013795

Table 7: Jegadeesh and Titman (1995) decomposition of MIS profits.

the major driver of profits.

5 Conclusion

We have tested a set of contrarian, and momentum trading strategies for the Indian equity markets in the period
1996-2002. We find that momentum strategies make statistically significant profits at long holding periods. But
profits fall as the formation period over which stocks were chosen increases. Costs to trade play an important role in
determining economic profitability.x Traditional analyses of contrarian, and momentum profits have either omitted
trading costs, or have included poportional costs alone. Prportional costs, do not provide an accurate picture of
trading profits. We have accounted for the price impact on trading while calculating our returns.

17
 Portfolio k Cˆk Oˆk σµ2 E[π̂t ] I %-Cˆk %-Oˆk %-σµ2
0 100 -23.510 -15.192 7.547 46.249 69.571 50.83 32.84 16.31
0 200 -23.122 -12.694 7.508 43.324 68.819 53.37 29.30 17.33
0 300 -22.549 -10.877 7.479 40.905 67.832 55.12 26.59 18.28
0 500 -21.541 -8.650 7.396 37.587 65.104 57.31 23.01 19.67
0 1000 -19.555 -7.996 7.152 34.702 59.374 56.35 23.04 20.60
1 100 -3.937 -2.575 1.602 8.113 54.775 48.52 31.73 19.73
1 200 -3.862 -1.997 1.596 7.455 54.194 51.79 26.79 21.41
1 300 -3.725 -1.636 1.589 6.950 53.356 53.59 23.54 22.86
1 500 -3.433 -1.046 1.568 6.048 51.231 56.76 17.29 25.93
1 1000 -2.915 -0.715 1.507 5.138 46.709 56.74 13.92 29.33
2 100 -2.118 -0.562 0.522 3.201 49.512 66.14 17.55 16.29
2 200 -2.124 -0.427 0.516 3.068 49.001 69.24 13.91 16.83
2 300 -2.004 -0.443 0.508 2.955 48.254 67.82 14.99 17.18
2 500 -1.844 -0.410 0.496 2.750 46.482 67.04 14.91 18.04
2 1000 -1.546 -0.420 0.466 2.432 42.187 63.58 17.26 19.14
3 100 -0.884 -0.204 0.148 1.236 41.118 71.54 16.50 11.94
3 200 -0.856 -0.150 0.148 1.154 40.779 74.19 12.98 12.82
3 300 -0.801 -0.104 0.142 1.048 40.272 76.44 9.96 13.58
3 500 -0.720 -0.123 0.131 0.974 39.158 73.90 12.67 13.42
3 1000 -0.556 -0.108 0.105 0.769 36.299 72.34 14.00 13.65
4 100 -0.212 -0.055 0.045 0.312 40.179 68.05 17.53 14.41
4 200 -0.214 -0.020 0.045 0.279 39.711 76.68 7.14 16.17
4 300 -0.186 -0.004 0.042 0.232 39.202 80.10 1.54 18.35
4 500 -0.174 -0.004 0.039 0.217 38.158 80.23 1.80 17.95
4 1000 -0.114 -0.037 0.030 0.181 35.167 62.93 20.24 16.81

Table 8: Lo and MacKinlay (1990) Profit Decomposition of an MIS strategy over multiple lag-lengths, and over
different size-sorted quintile portfolios. The strategy is implemented over HF buy/sell data over the period 1996-
2002.

References
Ball, R., Kothari, S. P., and Shanken, J. (1995). Problems in measuring portfolio performance: An application to
contrarian investment strategies. Journal of Financial Economics, 38(1):79–107.

Basu, S. (1977). The investment performance of common stocks in relation to their price-earnings ratios: A test of
the efficient market hypothesis. Journal of Finance, 32:632–682.

Bondt, W. D. and Thaler, R. (1985). Does the stock market overreact? Journal of Finance, 40:793–808.

Chan, K. (1988). On the contrarian investment strategy. Journal of Business, 61:147–163.

Conrad, J. and Kaul, G. (1998). An anatomy of trading strategies. Review of Financial Studies, 11(3):489–519.

Glosten, L. R. and Harris, L. (1988). Estimating the components of the bid-ask spread. Journal of Financial
Economics, 14:123–142.

Glosten, L. R. and Milgrom, P. R. (1985). Bid, ask and transaction prices in a specialist market with heterogenously
informed traders. Journal of Financial Economics, 14:71–100.

18
Grundy, B. and Martin, J. S. (2001). Understanding the nature of the risks and the source of the rewards to
momentum investing. Review of Financial Studies, 14(1):29–78.

Hameed, A. and Kusnadi, Y. (2002). Momentum strategies: Evidence from pacific basin stock markets. Journal
of Financial Research, 25(3):383–397.

Harris, L. (2003). Trading and Exchanges: Market Microstructure for Practitioners. Financial Management Asso-
ciation: Survey and Synthesis Series. Oxford University Press, New York.

Jegadeesh, N. and Titman, S. (1995). Overreaction, delayed reaction and contrarian profits. Review of Financial
Studies, 8(4):973–993.

Jegadeesh, N. and Titman, S. (2002). Cross-sectional and time-series determinants of momentum returns. Review
of Financial Studies, 15(1):143–157.

Kang, J., Liu, M.-H., and Ni, S. X. (2002). Contrarian and momentum strategies in the china stock market:
1993-2000. Pacific Basin Finance Journal, 10:243–265.

Keim, D. B. (1983). Stock return seasonality and the size effect. Journal of Financial Economics, 12:33–56.

Korajczyk, R. and Sadka, R. (2002). Are momentum profits robust to trading costs? Technical report, Department
of Finance, Kellogg Graduate School of Management, Northwestern University.

Lee, C. M. C. and Ready, M. J. (1991). Inferring trade direction from intraday data. Journal of Finance, 46:733–746.

Lehmann, B. N. (1990). Fads, martingales and market efficiency. Quarterly Journal of Economics, 60:1–28.

Lo, A. W. and MacKinlay, A. C. (1990). When are contrarian profits due to stock market overreaction? Review of
Financial Studies, 3(2):175–205.

NSE (2002). Indian securities market - a review: 2002. Technical Report V, National Stock Exchange of India
Limited., Exchange Plaza, Bandra Kurla Complex, Mumbai 400 051.

O’Hara, M. (1995). Market Microstructure Theory. Basil Blackwell, Oxford.

Roll, R. (1984). A simple measure of the implicit bid-ask spread in an efficient market. Journal of Finance,
39:1127–1139.

19