Working Paper version of the Article presented in

The International Conference on “Technology Analysis, Fintech and Financial Services (TAFS), 2022

Price Discovery Efficiency and Resilience of Financial Futures - A Case

Study of Indian Banking Sector

Nivedita Mandal, Institute of Engineering and Management

Rituparna Das, Sharda University

DoI: 10.6084/m9.figshare.22216654

https://dx.doi.org/10.6084/m9.figshare.22216654

Abstract:

India has proven to be the second most attractive emerging market among other large
emerging economies in world. S&P has predicted India to be the one among fastest growing
emerging markets in FY’22. According to Morgan Stanley report, banking is found to be the
dominant sector in most of the emerging markets. Hence the banking sector in an emerging
economy like India has the potential to attract fresh investment, so as their financial derivatives
instruments. In this backdrop, the present work explores the price discovery mechanism
between futures and spot markets in particular to Indian banking industry to bring forth sector
specific insights. Alongside the paper tries to capture the impact of global slowdown due to
Covid-19 pandemic during 2020-21on the Indian banking industry to check for its ‘resilience’
which is always a major concern for any emerging markets. The authors have used Bi-variate
VEC-EGARCH framework to examine the price-discovery mechanism in the Bank-Nifty futures
and spot markets. The short run impact of Covid-19 shock is measured with help of the ‘market
model’ under ‘event study’ methodology.

Key Words: Market Model, E-GARCH, VECM, Granger Causality, Resilience, Abnormal
Return

JEL Code: G12

1. Introduction:

India attracts investors from various segments. High potential of growth in the banking and

financial sectors of the emerging markets (EMs) is reported by various studies. Growth

opportunities reflect on low credit penetration and a goof size of financial exclusion. The

banking industries are increasing becoming digitized, embracing fintech and mobile/cashless

Electronic copy available at: https://ssrn.com/abstract=4031224

payment modes in the EMs like China and India. Banking and financial services and the

derivatives markets are reported to be attracting huge investments.

Hence the present paper focuses to explore how the futures and spot markets of Indian banking

sector behave with respect to ‘informational efficiency’ and whether they are resilient towards

any global macroeconomic shock, like Covid-19 pandemic.

1.1 Informational Efficiency and Price Discovery:

Price discovery and risk transfer (i.e. hedging) are reported to be the pivot functions of the

futures market in all the economies. Price Discovery is process of convergence of the markets

towards the efficient price of the underlying asset containing its intrinsic value. At any point in

time any flow of new information into asset markets is incorporated the market prices for the

assets through readjustment of those prices. A news deemed by the market participants relevant

to asset pricing can be about the international or national macro-economic system, some

specific industry or corporate announcements or anything else relevant. Logically if multiple

markets of an asset get the same information arriving simultaneously, they should react at the

same time in a similar manner. In the case they do not react at the same time, one leads the

others. The former is viewed as contributing to price discovery mechanism for that asset. It has

been claimed that generally the futures market has a greater speed of assimilation of new

information compared to the spot market of the underlying asset because of their inherently

high leverage and low transaction costs. Sometimes the inflammation flows in the opposite

direction also. i.e. from the spot or cash market to the futures market or sometimes information

is reflected simultaneously in both the markets.. The microstructure of a market, the level of

transparency, the liquidity flow mechanism, the rules of orders, limitations of short sales and

settlement processes decide the contribution of a market to the price discovery process.

Electronic copy available at: https://ssrn.com/abstract=4031224

Since futures market has lesser trading costs, higher liquidity than spot market the information

is first expected to be reflected in the prices of futures and then it is expected to flow to cash

market. However, this may not be true in all circumstances. Sometimes it can happen that the

information is first discounted in the cash market and then moves on to futures market.

Alternatively, information is reflected simultaneously in both the markets.

There are mixed views regarding the price discovery efficiency of Indian equity futures market.

Thenmozhi (2002), Karmakar (2009), Pati, & Pradhan (2009), Wats, & Mishra (2009), Pati, &

Rajib (2011) reportedly agreed that price discovery happens in CNX-Nifty futures market and

it leads the spot market in information transmission. Whereas, Raju, & Karande (2003), Bhatia

(2007), Bose (2007), and Gupta, & Singh (2009) are reported to have found that although price

discovery happens in both futures and spot markets, as far as the information transmission is

considered the futures market leads the spot market of CNX-Nifty. Again, Srinivasan (2009),

Mallikarjunappa & E. M. (2010), Sakthivel, & Kamaiah (2010) reportedly concluded that there

is clear bi-directional causality between CNX-Nifty futures and spot markets and price

discovery happens in both the markets simultaneously. Only, Mukherjee, & Mishra (2003) is

reported to have found that there is bi-directional causality between futures and spot market,

and spot Nifty is more dominant in disseminating information. In the Indian context most of

the studies are reported to have been carried out on CNX Nifty index futures, and a few studies

on selected stock futures and that too are reported to have produced mixed results. These studies

are deemed failure in throwing any light to industry specific features.

In the present paper, the authors try to explore the price discovery mechanism and lead-lag

relationship, if any, between the spot and futures markets of banking sector in India. They try

also to capture the short run impact of Covid-19 shock on the Indian banking spot and futures

markets, so as to understand their ‘resilience’ towards any external shock. The present work

Electronic copy available at: https://ssrn.com/abstract=4031224

will surely help to bring forth some sector specific insights for potential investors of the Indian

market.

Rest of the paper is organised as follows: Section 2 followed by introductory section 1,

describes sample selection, period of the study, data collection, and results from preliminary

examinations. Section 3 explains model and methodology adopted for empirical analysis.

Results and findings are presented in section 4. Section 5 draws the conclusions of the study.

2. Sample Data and Preliminary Analysis:

2.1 Sample Selection: For selecting the sample of the study the ‘Judgmental Sampling’

technique is adopted. There are two major stock exchanges in India, viz. Bombay Stock

Exchange (BSE) and National Stock Exchange (NSE). Among these two, NSE is the lion in

the market as far as derivatives trading is concerned. Since inception in June 2000 till 2011,

NSE proudly bears 100-99% turnover in derivatives trading. In 2012, BSE launched new

incentive schemes and trading policies to revive its derivatives segment. Consequently, 2012

onwards NSE fetches 80-90% share in total turnover of F&O segment, while BSE is still

struggling to manage 10-20% share out of it. Therefore, NSE has been chosen purposefully

over BSE.

There are 6 indices for which NSE has 7 products in F&O segment. Among these, Bank Nifty

is in the second position after CNX Nifty, in terms of number of contracts traded and turnover

value (as on March’2018). As far as the sectoral indices are concerned, Bank Nifty is in the

top, leaving IT, Infra and PSE far behind.

2.2 Period of the Study:

a. For understanding the price discovery process, the period is taken from 13th June 2005, i.e.

from the inception of Bank Nifty Index, till 27th December 2019.

Electronic copy available at: https://ssrn.com/abstract=4031224

b. For capturing the short run impact of Covid-19 global pandemic on the banking sector the

impact year ‘2020’ is taken into consideration.

2.3 Data Collection: The daily closing prices data for Bank Nifty Index futures and the spot

are collected for the above mentioned periods of the study. For futures prices, only the near

month contract is accounted for its comparatively high trading volume than the other two

contracts (i.e. middle month and far month) available in the market. All the futures and spot

prices of Bank Nifty Index are collected from the official website of NSE.

2.4 Preliminary Examinations: The collected data on stock prices are clearly time series data

which span over a long period of time on daily basis. Before stepping into the main analysis,

some preliminary examinations have been carried out to get an idea about the type and nature

of the data, their distributional properties, time series properties, etc.

Most financial studies involve returns, instead of raw prices of the securities. Campbell, Lo, &

Mackinlay (1997) cited two main reasons for that: firstly, for average investors returns are

complete and scale free summary for investment opportunity, and secondly its attractive

statistical properties makes it more amenable for various analysis.

‘Returns’ are computed as continuously compounded return, i.e. natural logarithmic

differences of lagged price series: FRt = (ln Ft – ln Ft-1), and SRt = (ln St – ln St-1).

Here FRt and SRt are futures and spot returns respectively, at time ‘t’ and Ft and St are futures

and spot prices respectively, at time ‘t’.

The computed values of descriptive statistics reveal the fact that the average daily futures and

spot returns are almost equal over the sample period, albeit the volatility sometimes differs

from, sometimes matches with each other. The coefficients of skewness and kurtosis of return

series reveal that none of the distributions are alike to normal distribution. Moreover, the

Electronic copy available at: https://ssrn.com/abstract=4031224

Jarque-Bera test statistically proves that futures and spot returns of all the sample variables are

not normally distributed. There are evidences that all the series are suffering from the problem

of auto correlation. None of the series is an independent series. Moreover, clear ARCH effect

is present in all the returns series which implies dynamic conditional variance process. The

significant LB2-Q values and ARCH-LM values double signifies that the residuals of returns

have non constant time varying variance which results in to clustering of volatility in the series.

(See results in Table.1 in Appendix)

The futures and spot returns of the Bank Nifty Index have been tested for structural break points

during the study period following the method of Quandt-Andrews unknown breakpoint test

considering 15% trimmed data. The result shows no evidence of any notable structural change

during the study period.

Augmented Dickey Fuller (ADF) test (1979) and KPSS test (1992) have been conducted to

check for the ‘stationarity’ of the collected time series data. Results from the ADF test and the

KPSS test revealed that log normal futures and spot prices are first difference stationary and

have same order of integration, i.e. I (1). (See results in Table.2 in Appendix)

As all the futures and spot returns are having same order of integration, the next step is to check

for their cointegration property. Cointegration analysis provides important information about

the long term relationship among any group of time series data whose degree of integration is

same. The economic interpretation of cointegration is that if two or more variables are linked

to form an equilibrium relationship spanning the long-run, even though the series themselves

in the short run may deviate from the equilibrium, they will move close together in the long

run equilibrium. Thus, if futures and spot price series are found to be cointegrated, it ensures

that there exists a stable long-run relationship between futures and its underlying spot market.

By deploying Johansen-Juselius (1990) maximum likelihood method of Cointegration test, it

Electronic copy available at: https://ssrn.com/abstract=4031224

is found that these two related price series are not only having same order of integration, but

they are also sharing same stochastic trend, i.e. futures and spot prices are cointegrated of order

one, CI(1). This implies co-movement of futures and spot prices and ensures existence of a

stable long run equilibrium relationship. For conducting J-J cointegration test, the optimal lag

length has been selected following Schwartz Information Criteria (SIC) and the best fitted

model has been considered following Pantula Principle1. (See Table.3 in Appendix)

Now depending on these findings, for further analysis model specifications has been set such

as to incorporate non-normal distribution, serial correlation, and ARCH effects in residual

process. And the fitted models have been tested for their adequacy along these lines.

3. Model Specification:

3.1 Model to identify the Price Discovery Mechanism -

Granger (1988) is reported to have pointed out that, if a pair of time series is cointegrated then

there must be some causality between the two series in at least one direction, and if possible in

both the directions. This causality is the reason behind the co-movements of two cointegrated

time series. The study tries to detect the direction of causal relationship between futures and

spot prices by applying standard Granger Causality test augmented with a lagged Error

Correction Term, i.e. Error Correction Model (hereafter ECM). Error correction model is

capable to capture the short run and long run components of Granger causality distinctly. The

effect of causality that flows from long run equilibrium relation between the two variables,

during temporary deviations from long run equilibrium path, which gets captured by the

coefficient of lagged Error Correction Term, i.e. the long run component, and the effect of

causality that arises from previous period’s spot price or futures price, i.e. the short run

1
Pantula Principle is the method of testing the joint hypothesis of both the rank order and deterministic
components as discussed in Johansen (1992).

Electronic copy available at: https://ssrn.com/abstract=4031224

component. Thus ECM is an appropriate statistical tool to examine the immediate impact of

news flows on asset prices and its transmission process from one market to another and the

speed of adjustment to the long run equilibrium between the futures and spot markets, under

static set up.

In the present study, the traditional VECM is extended to Vector Error Correction- Exponential

Generalized Auto Regressive Conditional Heteroscedasticity (hereafter VEC-EGARCH)

Model. VECM, being a restricted version of VAR set up can tackle the problem of serial

correlation by incorporating the auto regressive terms of dependent variable as regressors of

the model equations. But to address the problem of heteroscedasticity in residual process, i.e.

the ARCH effect, there is a need to extend the VECM to a GARCH set up. The VEC-EGARCH

framework helps to incorporate the time varying volatility effect in interpreting the dynamics

of Granger causality from ECM. In addition, the Exponential GARCH specification will help

to understand the asymmetric effect of volatility, i.e. how market responds to good and bad

news differently. Nelson (1991) showed that negative or bad news bear more impact on market

volatility, than any positive or good news to the market. This asymmetric response of the

market can be well captured by this exponential variant of GARCH family, as the EGARCH

specification avoids the non-negativity constraint on the conditional variance parameters, in

addition to incorporating asymmetry in return volatilities (Nelson, 1991). The proposed model

is superior to VECM, since the traditional approach is limiting in several ways. First, the model

does not leave scope for the possibility that volatility may be time varying in nature. Secondly,

the traditional VECM framework can only address linear price dynamics in the conditional

mean of price changes. Finally, VECM estimation that relies on ordinary least squares (OLS)

assumes that the distribution of price changes is characterized by a constant variance.

Past studies in Indian context have reportedly mostly applied VECM to identify the price

discovery efficiency of equity futures markets. Some studies have gone little far by deploying

Electronic copy available at: https://ssrn.com/abstract=4031224

Impulse Response Functions (IRFs) and Variance Decomposition (VD) to capture the

mechanism in dynamic set up. But the problem lies in the very basic if the time varying

volatility, i.e. the inherent nature of the financial series, is not accounted for and hence could

lead to spurious results.

Bi-variate VEC-EGARCH (1, 1, 1) Framework:

FRt = κf + αf (FRt-1) + βs (SRt-1) + δf (ECf, t-1) + ɛf, t ……………..…………...……… (1)

where, ɛf, t ǀ It-1 ~ t distribution with conditional variance hf,t

ln(hf,t) = ω0,f + θ1,f (zf,t-1) + γ1,f [ ǀ zf,t-1ǀ - E(ǀ zf,t-1ǀ)] + φ1,f ln(hf, t-1) …………...… (1.a)

ɛ𝑓,𝑡
where, 𝑧𝑓,𝑡 = is the standardized residual of FRt
√ℎ𝑓,𝑡

SRt = κs + αs (SRt-1) + βf (FRt-1) + δs (ECs, t-1) + ɛs, t ………………………………... (2)

where, ɛs, t ǀ It-1 ~ t distribution with conditional variance hs,t

ln(hs,t) = ω0,s + θ1,s (zs,t-1) + γ1,s [ ǀ zs,t-1ǀ - E(ǀ zs,t-1ǀ)] + φ1,s ln(hs, t-1) …………... (2.a)

ɛ𝑠,𝑡
where, 𝑧𝑠,𝑡 = is the standardized residual of SRt
√ℎ𝑠,𝑡

Here for sake of simplicity, the VEC-EGARCH framework is reported to have been presented

in order (1, 1, 1), i.e. the conditional variance equation has the ARCH term of order one,

asymmetry of order one, and GARCH term of order one. This is deemed to be the simplest

form of the model and it could take higher order also depending on the sample data. In addition,

only one period lagged difference variable, (i.e. FRt-1, SRt-1) in the mean equations was

considered. This lag order depends on the VAR optimal lag selection, where SIC has been

followed. It-1 is the set of all information regarding spot and futures markets in first as well as

Electronic copy available at: https://ssrn.com/abstract=4031224

in second moments, available at period ‘t-1’. Since preliminary examinations reportedly reveal

that none of the futures and spot returns follow normal distribution, and they have fat tailed

high kurtosis distributions, here the residuals were reportedly considered to follow t-

distribution. Enders (2004) is reported to have shown how t-distribution places a greater

likelihood on large realizations in any financial series than does the normal distribution.

Model Interpretation:

Mean Equations: The mean equations were prescribed for estimating the mean returns

(futures and spot) of sample data. In the Equations (1) and (2), the ‘β’ coefficients are deemed

to capture the effect of short run causality from one market to another. The statistically

significant non-zero value of βs (βf) are found to imply that spot (futures) return Granger causes

futures (spot) return in short run. That is, previous period’s value of spot (futures) return are

deemed to help predicting the current futures (spot) return, in a better way than only the past

values of futures (spot) returns do. If both the βs and βf are statistically significant, these may

indicate ‘feedback effect’ from one market to another, i.e. there is bi-directional causality in

between futures and spot markets. If any one of the β’s is significant, then there may be a flow

of unidirectional Granger causality either from spot to futures market or from futures to spot

market. The ‘α’ coefficients may be taken as the ‘own price effect’, i.e. how the past price

changes of a market can affect its current price changes. Statistically significant α value

indicates that a change in the past market price has either positive or negative impact on its

today’s market price moves. The Error Correction Term ‘EC’ indicates temporary deviations

from the long run equilibrium path of the futures and spot prices. The adjustments in the first

moments of the futures and spot returns to this temporary deviation should be captured by the

mean equation of the VEC-EGARCH model. The magnitudes of these EC terms are generally

derived from the cointegrating equations between the futures and spot prices. Thus the lagged

error correction terms, ECf, t-1 = (lnFt-1 – c2 lnSt-1 – c1), and ECs, t-1 = (lnSt-1 – c4 lnFt-1 – c3), are

10

Electronic copy available at: https://ssrn.com/abstract=4031224

taken as the long run equilibrium deviations in the previous period which can be corrected in

the next current period. The presence of this lagged EC term indicates the dynamics of long

run relation linking the two series. The loading δf (δs) is interpreted as the ‘speed of

adjustment’ of futures (spot) return towards the equilibrium path. Hence the values of these δ

coefficients indicate how speedy one market is to rectify the previous period’s deviation from

long run equilibrium, through the causality effect of another market. Here is the effect of long

run Granger causality found. If δf (δs) appears to highly different from zero, the spot (futures)

prices impact futures (spot) price changes through the long run price equilibrium channel. Thus,

a statistically significant non zero value of δf (δs) indicates that the spot (futures) market

Granger causes the futures (spot) market in long run.

Conditional Variance Equations: In the Equations (1.a) and (2.a), ‘ω’ is the intercept term

in conditional variance equations. ‘θ’ being the coefficient of asymmetry, captures how the

positive (good news) and negative (bad news) innovations in past affect the current volatility

of the market. Hence θ measures the ‘sign effect’ of past innovations. ‘γ’ is the ARCH

coefficient which measures the impact of past innovations on the current volatility of the

market, i.e. the ‘size effect’. Thus θ and γ together capture the effect of past innovations on

current volatility.

Now let us consider, Fi (Zi, t-1) = θi Zi, t-1 + γi [ ǀ Zi, t-1ǀ - E (ǀ Zi, t-1ǀ)], where i = f, s;

Here Fi is the function of lagged standardized innovation of ith market,

ɛi,t−1
i.e. zi,t−1 = ,
√hi,t−1

where hi,t−1 is the conditional variance of innovation at time ‘t-1’, ɛi,t−1 .

11

Electronic copy available at: https://ssrn.com/abstract=4031224

Then the asymmetric effect of standardised past innovations on current volatility may be

measured by,

𝜕𝐹𝑖 𝜃𝑖 + 𝛾𝑖, 𝑖𝑓 𝑍𝑖,𝑡−1 > 0

={
𝜕𝑍𝑖,𝑡−1 𝜃𝑖 − 𝛾𝑖 , 𝑖𝑓 𝑍𝑖,𝑡−1 < 0

|𝜃𝑖 −𝛾𝑖 |
The ‘Relative Asymmetry’ may be defined as, 𝜉𝑖 =
𝜃𝑖 +𝛾𝑖

The magnitude of ξi is greater than, equal to, or less than one for negative asymmetry,

symmetry, and positive asymmetry, respectively. If negative asymmetry is supposed to

characterize the market volatility, it may mean that the impact of any negative shock to the

market on its return volatility is more in impact, as compared to the same amount of positive

shock.

The ‘φ’ parameter is considered to represent the volatility persistence level of a market, which

means how the previous period’s conditional volatility continues to affect the return volatility

in current period. In other words, φ coefficient detects the phenomena of ‘volatility-clustering’

in the return series, i.e. the GARCH effect. For the conditional volatility process to be

stationary, it should be ǀ φ ǀ < 1.

3.2 Model to identify the short run impact of Covid-19 pandemic –

To understand the short run impact of the global pandemic on the Indian banking sector, a

separate methodology is adopted and the period of this pandemic is treated separately to avoid

the structural break in the data set. For this the ‘market model’ of ‘event study’ is applied as

described in Benninga (2014).

Methodology of Event Study -

12

Electronic copy available at: https://ssrn.com/abstract=4031224

The initial step in an event analysis is to define the event of interest and the event window. The

Government of India (GoI) imposed a nationwide lockdown on the evening of 24th March 2020.

Since the impact of that announcement on the stock market is expected to get realised on the

next day, i.e. 25th March 2020, is considered as the ‘event day’. The event window is the time

period during which the security prices get affected due to a particular event. The event window

consists of two components—the anticipation window and the adjustment window. The day of

impact or the event day is set as day ‘0’.

The authors have tried to capture the immediate or ‘very short run’ impact of the event, i.e. the

announcement of nationwide lockdown, on the Indian banking stocks.

In this study ‘estimation period’ of total 252 trading days have been considered to get the

anticipated return. In between ±3 days surrounding the event day (T=0) are considered as ‘event

window’. Thus we get,

T=-252 T= -3 T=0 T=+3

Estimation Window Event Window

Here a very short event window is considered, as a long event window (i) decreases the power

of the test statistics, (ii) leads to confounding effects, and (iii) results in false conclusions

(McWilliams and Siegel, 1997).

Market Model –

13

Electronic copy available at: https://ssrn.com/abstract=4031224

‘Market model’, is an OLS regression model, which regresses security returns against stock

market returns to estimate expected returns on each security/index. The model is as follows,

Rit = αi + βiRmt + εit

where, Rit = Return on ith security in period t

Rmt = Return on market index in period t
εit = Error or Residual term

The estimated regression equation is as follows:

𝑅̂ ̂𝑖 +𝛽̂i Rmt
𝑖𝑡 = 𝛼

In this model the Abnormal Return (AR) for any security/index ‘i’ in any period t, is defined

as the difference between the actual and estimated return of that security/index in the same

period t. Abnormal Returns are calculated for each individual security/index for their event

̂it ) = [Rit –(𝛼̂𝑖 +𝛽̂i Rmt )]

period as, ARit = εit = (Rit - R

Then to capture the cumulative effect, Cumulative Abnormal Return (CAR) has also been

calculated for the ‘event window’ where CARt = CARt-1 + ARt

Hypotheses:

H0: Abnormal returns surrounding the event day are zero, i.e. market is resilient to the impact

of sudden shock

H1: Abnormal returns surrounding the event day are not zero, i.e. market is not resilient to the

impact of sudden shock

14

Electronic copy available at: https://ssrn.com/abstract=4031224

By assumption, abnormal returns being the residuals of the market model follow normal

distribution. This justifies conducting t-test (Armitage, 1995) to check whether these ARs are

significantly different from zero or not.

4. Results & Findings:

4.1 Findings related to Price Discovery Efficiency of Bank Nifty Futures:

In this section the results of VEC-EGARCH estimations have been presented, and the findings

have been discussed in detail.

Table No. 4.1: VEC-EGARCH Results for Bank Nifty Index Futures and Spot Returns

Futures Return Spot Return

(1, 1, 1) (1, 1, 1)
Coefficients Estimates (P-values) Estimates (P-values)
Mean Intercept (κ) 0.0006*** (0.0674) 0.0006 (0.1066)
Equation ECT (δ) -0.1947** (0.0489) 0.0203 (0.8316)
Own (α) -0.3652* (0.0082) 0.2099 (0.1339)
Cross (β) 0.4807* (0.0006) -0.0788 (0.5659)
Variance Intercept (ω) -0.2412* (0.0000) -0.2218* (0.0000)
Equation ARCH (γ) 0.1487* (0.0000) 0.1382* (0.0000)
Asymmetry (θ) -0.0758* (0.0000) -0.0726* (0.0000)
GARCH (φ) 0.9839* (0.0000) 0.9855* (0.0000)
Others T distribution 8.1335* (0.0000) 8.3976* (0.0000)
D.O.F
Log likelihood 5418.933 5460.477
Relative 3.08 3.21
Asymmetry (ξ)
Residual LB2-Q (8) 6.513 (0.5900) 6.554 (0.4340)
Diagnostics ARCH-LM (8) 6.833 (0.5548) 6.739 (0.5650)

15

Electronic copy available at: https://ssrn.com/abstract=4031224

*
significant at 1% level, ** significant at 5% level, *** significant at 10% level

The VEC-EGARCH models for estimating Bank Nifty Index futures and spot returns both are

having order (1, 1, 1). For VECM the optimal lag orders for differenced endogenous variables

are considered according to VAR lag order selection following SIC. In case of Bank Nifty

Index futures and spot returns the optimal lag order is (1, 1). For cointegrating equation, the

deterministic trend is specified with that of model.32 that assumes a linear trend in the data and

an intercept in cointegrating equation.

The coefficient of the ECT is significant at 5% level with the value of -0.19 in the FR mean

equation, and for the SR mean equation the ECT is positive, but insignificant. This implies that

there is long run causality running from spot to futures market which enables the Index futures

market to adjust to the short run deviations from equilibrium path with 19% speed of

adjustment. Moreover, the insignificant ECT in the SR mean equation indicates exogenity of

the variable. The coefficient of lagged SR term in the FR mean equation is significant at 1%

level with the value of 0.48, which indicates short run causality from spot to futures market

with 48% leading effect. In the SR mean equation, the coefficient of Lagged FR is insignificant

implying no trace of causality from futures to index spot market. This clearly proves that there

is ‘unidirectional’ short run causality from Bank Nifty spot to futures market. The own market

effect of price change is significant, although negative for futures return. This implies that

Index futures market gets negatively affected by its past price changes; however, past price

2
The statistical package (EViews) offers five options in applying the J-J method of cointegration. The options
correspond to different specification of intercept and trend variable in the underlying VAR model. The options
are five: Model 1 assumes that there are no deterministic trends in the variables and the underlying data
generating process does not contain a deterministic trend. Model 2 is appropriate when the jointly determined
variables, i.e. cointegrated variables do not contain a deterministic trend, only restricted intercept. Model 3
assumes unrestricted intercept, but no trends in the VAR model. Model 4 is appropriate when the jointly
determined variables in the VAR have a linear deterministic trend as well as unrestricted intercept. Model 5
considers unrestricted intercept, and unrestricted trends in the VAR model. In general, model 1 and 5 are
irrelevant. So the present analysis is limited to three options: model 2, 3, and 4.

16

Electronic copy available at: https://ssrn.com/abstract=4031224

changes have insignificant effect on relatively more efficient spot Index market. The

coefficients of asymmetry in the variance equations of both FR and SR series are negative

which implies that both the market reacts to bad news more adversely than the good news. The

computed values of relative asymmetry are 3.08 and 3.21 respectively for FR and SR, which

states that the spot market reacts to negative shocks 3.21 times more than the same amount of

positive shocks; while the reaction is slight lower, i.e. 3.08 times in futures market.

In the above models, the ARCH and GARCH coefficients are significant in both conditional

variance equations, indicating that both the futures and spot markets are characterized by time

varying volatility and the clustering of volatility is present in their time plots. The significant

t-distribution degrees of freedom show that the error distributions for the estimated equations

are more alike to follow a t-distribution. Moreover, the models have been checked with

Generalized Error Distribution, but with t- distribution the models produce higher Log-

likelihood values and lower SIC and AIC values. Model adequacy has been checked for

remaining serial correlation in residuals and ARCH effect in residual process. The LB2-Q (8)

test statistics with high P-values indicate that the null of no serial correlation in the squared

residual series cannot be rejected, i.e. the models have no problem of serial correlation in

residuals. In addition, the computed values for ARCH-LM (8) test statistic are insignificant

implying non rejection of the null of no ARCH effect in squared residual series. Hence the

estimated models are free from ARCH effects in their residuals. Thus the residual diagnostics

prove that the estimated models are good fit for the sample data.

4.2 Findings related to short run impact of Covid-19 on Bank Nifty Spot market:

Following the methodology adopted by Benninga (2014), the estimated model has come as

follows –

RBank Nifty = - 0.000658 + 0.459247 RNifty50.

17

Electronic copy available at: https://ssrn.com/abstract=4031224

where R2 is 0.13 with Steyx value 0.016. The Steyx function measures the standard error of the

regression-predicted y –values. Here Nifty50 returns are considered as the proxy for market

return.

Applying the above market model, the Expected Return is calculated and accordingly

Abnormal Return and Cumulative Abnormal Return are calculated as follows -

Table No. 4.2: Market Model Result for Bank Nifty Spot Market

Cumulative Abnormal Return

Date Expected Return Abnormal Return (AR) (CAR)
20-03-2020 0.025377384 -0.208507391* -0.208507391
23-03-2020 -0.064510553 0.075658393* -0.132848998
24-03-2020 0.010714042 0.066526564* -0.066322434
25-03-2020 0.028800638 0.030691956 -0.035630478
26-03-2020 0.016869955 0.001072597 -0.034557881
27-03-2020 0.000340057 -0.061600847* -0.096158728
(* values are significant at 5% level of significance.)

Fig. 4.1: Graph of Abnormal Returns on Bank Nifty Spot during the ‘event window’

Abnormal Return (AR)

0.1

0.05

-0.05

-0.1

-0.15

-0.2

-0.25

From the above table, it is evident that the impact of the shock was already expected much

ahead of the ‘announcement day’, and that is why the first day of the event window, i.e. 20 th

March 2020 shows a significant and the highest ‘negative’ return. The event day and its next

18

Electronic copy available at: https://ssrn.com/abstract=4031224

day didn’t have any significant impact, rather the market already started its correction process.

Hence it is apparent that the Bank Nifty had shown quick and efficient response towards the

market corrections which is a sign of ‘resilience’ for its futures counterpart as well.

5. Conclusions:

The above findings from VEC-EGARCH estimations reveal a different story about futures

market functioning in Indian banking industry: it is the spot market where price discovery

happens actually. For Bank Nifty Index there are both long run and short run Granger causality,

which is flowing unidirectional from spot to futures market. Hence it can be concluded that so

far as Bank Nifty is concerned, its’ futures market fails to function as a price discovery vehicle.

In short run, price discovery happens mainly in the spot markets of Indian banking sector.

Both the markets – futures and spot are characterized by asymmetric nature of market response

towards good and bad news, with different thrusts. They react more to any negative shock than

to any positive news, and the plunge is more or less same in futures and spot markets of the

underlying asset. CNX-Bank Index being composite of individual stocks is the most vulnerable

and sensitive towards any shock with comparatively high value of relative asymmetry.

The first part of the work establishes that the ‘price discovery’ mainly happens in the spot

market for Bank Nifty which flows eventually to affect its futures counterpart in short as well

as in long run, and thereby they maintain a long run equilibrium with continuous correction

processes. Hence in the second part, the authors have examined how the spot Bank Nifty

reacted towards the announcement of the news (event) – nationwide lockdown due to Covid-

19 pandemic and how long it took to adjust for the same vis-a-vis the national stock market.

The event study following the market model have shown that the impact was visible much

ahead of the announcement date which gradually dampened over a period of 4-5 days. Hence

19

Electronic copy available at: https://ssrn.com/abstract=4031224

it can be concluded that the banking sector in India has ‘efficiency’ to readjust with the market

shock though not fully ‘resilient’.

The present work leaves scope for future research by conducting the study with intra-day data

instead of daily price data, which might capture the speed of information transmission across

the markets in a better way. Moreover the study has considered only the concept of linear

Granger causality, while accounting for the non-linear causality could enlighten new findings.

References:

Benninga, S. (2014). Financial modeling. MIT press.

Bhatia, S. (2007). Do the S&P CNX Nifty index and Nifty futures really lead/lag? Error
correction model: A cointegration approach. NSE Research Initiative Paper, November.
Retrieved from http://www.nseindia.com/research/content/research_initiative.htm
Bose, S. (2007). Contribution of Indian index futures to price formation in the stock market.
Money & Finance, February, 39-56.
Campbell, J. Y., Lo, A. W., & Mackinlay, A. C. (1997). The econometrics of financial markets.
Princeton, New Jersey: Princeton University Press.
Dickey, D. A., & Fuller, W. A. (1979). Distribution of the estimators for autoregressive time
series with a unit root. Journal of The American Statistical Association, 74, 427-431.
Enders, W. (2004). Applied econometric time series (2nd ed.). Hoboken, New Jersey. Indian
Reprint, Noida, India: John Wiley & Sons Inc.
Granger, C. W. (1988). Some recent development in a concept of causality. Journal of
Econometrics, 39(1), 199-211.
Gupta, K., & Singh, B. (2009). Price discovery and arbitrage efficiency of Indian equity futures
and cash markets. NSE Research Initiative Paper, March. Retrieved from
http://www.nseindia.com/research/content/research_initiative.htm
Johansen, S. (1992). Testing weak exogeneity and the order of cointegration in UK money
demand data. Journal of Policy modeling, 14(3), 313-334.
Johansen, S., & Juselius, K. (1990). Maximum likelihood estimation and inference on
cointegration—with applications to the demand for money. Oxford Bulletin of Economics and
Statistics, 52(2), 169-210.

20

Electronic copy available at: https://ssrn.com/abstract=4031224

Karmakar, M. (2009). Price discoveries and volatility spillovers in S&P CNX Nifty future and
its underlying index CNX Nifty. Vikalpa, 34(2), April-June, 41-56.
Kwiatkowski, D., Phillips, P. C., Schmidt, P., & Shin, Y. (1992). Testing the null hypothesis
of stationarity against the alternative of a unit root: How sure are we that economic time series
have a unit root?. Journal of Econometrics, 54(1), 159-178.
Mallikarjunappa, T., & E.M., Afsal. (2010). Price discovery process and volatility spillover in
spot and futures markets: Evidences of individual stocks. Vikalpa, 35(2), April-June, 49-62.
McWilliams, A., & Siegel, D. (1997). Event studies in management research: Theoretical and
empirical issues. Academy of management journal, 40(3), 626-657.
Mukherjee, K., & Mishra, R. K. (2006). Lead-lag relationship between equities and stock index
futures market and its variation around information release: Empirical evidence from India.
NSE Research Initiative Paper, July. Retrieved from
http://www.nseindia.com/research/content/research_initiative.htm
Nelson, D. (1991). Conditional heteroscedasticity in asset returns: A new approach.
Econometrica, 59 (2), 347-370.
Pati, P. C., & Padhan, P. C. (2009). Information, price discovery and causality in the Indian
stock index futures market. The IUP Journal of Financial Risk Management. VI (3&4), 07-21.
Pati, P. C., & Rajib, P. (2011). Intraday return dynamics and volatility spillovers between NSE
S&P CNX Nifty stock index and stock index futures. Applied Economics Letters, 18, 567-574.
Raju, M. T. & Karande, K. (2003). Price discovery and volatility on NSE futures market. SEBI
Working Paper Series, No.7, March. Retrieved from
http://www.sebi.gov.in/workingpaper/discovery.pdf
Sakthivel, P., & Kamaiah, B. (2010). Price discovery and volatility spillover between spot and
futures markets: Evidence from India. The IUP Journal of Applied Economics, IX(2), 81-97.
Srinivasan, P. (2009). An empirical analysis of price discovery in the NSE spot and futures
markets of India. The IUP Journal of Applied Finance, 15(11), 24-36.
Telser, G. L. (1981). Why there are organized futures markets?. Journal of Law and
Economics, 24(1), 1-22.
Thenmozhi, M. (2002). Futures trading, information and spot price volatility of NSE-50 index
futures contract. NSE Research Initiative Paper, December. Retrieved from
http://www.nseindia.com/research/content/research_initiative.htm
Tse, Y. (1999). Price discovery and volatility spillovers in the DJIA index and futures
markets. Journal of Futures markets, 19(8), 911-930.
Wats, S., & Misra, K. K. (2009). Price discovery efficiency of the Indian futures market: The
case study of National Stock Exchange. International Journal of Business Insights &
Transformation, October’08-March’09, 39-50.
http://www.bseindia.com
http://www.nseindia.com

21

Electronic copy available at: https://ssrn.com/abstract=4031224

http://www.sebi.gov.in

Declaration of Interest Statement: Authors prepared this paper with purely academic interest.

It is not funded. It uses secondary data collected from public domain. It is not funded.

Appendix

Table.1: Descriptive Statistics of Sample Variables

Bank Nifty Index

FR SR
Mean 0.0005 0.0005
Median 0.0009 0.0009
Standard 0.0220 0.0215
Deviation
Skewness 0.0489 0.0857
Kurtosis 7.2488 7.1557
Jarque-Bera 1597.72* 1530.24*
LB-Q(8) 40.13* 53.88*
LB2-Q(8) 295.97* 311.50*
ARCH-LM(8) 155.41* 161.67*
*
significant at 1% level.

Table.2: Results for Unit Root Tests

Securities Log Prices in Futures Price Spot Price

22

Electronic copy available at: https://ssrn.com/abstract=4031224

 ADF KPSS ADF KPSS
Bank Nifty Level -1.953 4.474* -1.939 4.479*
Index First Difference -41.383* 0.064 -40.418* 0.064
Critical Values at 1% level are, ADF = -3.433, and KPSS = 0.739. * significant at 1% level.

Table.3: Results for Cointegration Tests

Null Hypothesis Trace Statistic λTrace Max-Eigenvalue Statistic λMax

Bank Nifty r=0 189.602* (0.0001) 184.868* (0.0001)
Index r≤1 4.734 (0.3139) 4.734 (0.3139)
Cointegrating lnFP = -1.0026 lnSP + 0.0223
Equation: [0.0007] [0.0059]
r≤1 7.513 (0.2940) 7.513 (0.2940)
Cointegrating lnFP = -0.9974 lnSP – 3.03E-06 T
Equation: [0.0012] [9.3E-07]
P-values: ( ), Standard Errors: [ ]; *significant at 1% level.

23

Electronic copy available at: https://ssrn.com/abstract=4031224