International Review of Economics and Finance 39 (2015) 224–238

Contents lists available at ScienceDirect

International Review of Economics and Finance

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

Frequency domain causality analysis of stock market and

economic activity in India
Aviral Kumar Tiwari a,⁎, Mihai Ioan Mutascu b,c,
Claudiu Tiberiu Albulescu d,e, Phouphet Kyophilavong f
a
Faculty of Management, IBS Hyderabad, IFHE University, Dontanpalli, Hyderabad, Pin-501203, India
b
Finance Department, West University of Timisoara, Bl. V. Parvan, 4, 300223 Timisoara, Romania
c
LEO, University of Orléans, Rue de Blois, 45067 Orléans, France
d
Management Department, Politehnica University of Timisoara, P-ta. Victoriei, 2, 300006 Timisoara, Romania
e
CRIEF, University of Poitiers, Rue Jean Carbonnier, 2, 86022 Poitiers, France
f
Faculty of Economics and Business Management, National University of Laos, Laos

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

Article history: In this study, we analyze the Granger-causality in frequency domain between stock prices and
Received 19 September 2013 economic growth in India, in order to identify the direction of the causality at different frequen-
Received in revised form 8 April 2015 cies. For this purpose we use in the first step different seasonal and structural breaks unit root
Accepted 9 April 2015
tests. In the second step we use a conditional VAR model as benchmark, and we focus on the con-
Available online 18 April 2015
ditional and non-conditional frequency domain causality tests. We find evidence of unidirectional
causal relationship between stock prices and industrial production in the long-run, running from
JEL classification: stock prices to industrial production. When using the non-conditional model, we find evidence of
C32
insignificant business cycle causality from both directions. Our study shows that stock prices are a
C53
leading indicator for growth in the industrial production in India. In this case, in order to adjust the
E4
industrial production in the long-term, the Indian economic policies should be focused with pre-
Keywords:
dilection on the stock market environment.
Stock market
© 2015 Elsevier Inc. All rights reserved.
Economic activity
Frequency domain Granger-causality

1. Introduction

One of the most sticking concerns for policy makers in economics is the relationship between financial development and economic
growth. Starting with Schumpeter (1912), and continuing with Gurley and Shaw (1955), Goldsmith (1969) and Schwarz (1978), a
fierce debate appears, both at theoretical and policy levels, regarding whether the economic activity increases the financial develop-
ment or vice-versa. A more difficult question is related to the forward-looking nature of stock prices, which can serve as evidence for
the causality between stock markets and the economic growth. At the same time, general savings, and an efficient allocation of capital
for productive investments, are facilitated by a well-developed stock market, which in turn will promote the economic growth. Therefore,
a bidirectional causal relationship between stock markets and the economic activity might exist.
There are several possible explanations for the strong association between stock prices and the real economic activity (see
Antonios, 2010; Fama, 1990; Schwert, 1990). First, information about the future economic growth may be reflected in stock prices.
In this case, stock prices are a leading indicator for the well-being of the economy. This idea is explored for example by the Economic
Tracking Portfolios (ETP) theory, which connects asset prices with news about economic variables (Lamont, 2001). Thus, investors
who wish to insure against economic growth fluctuations could take a position in the economic growth tracking portfolio. Second,

⁎ Corresponding author.
E-mail address: [email protected] (A.K. Tiwari).

http://dx.doi.org/10.1016/j.iref.2015.04.007
1059-0560/© 2015 Elsevier Inc. All rights reserved.
 A.K. Tiwari et al. / International Review of Economics and Finance 39 (2015) 224–238 225

changes in discount rates may affect stock prices and real investments in the same time, but the output generated by investments ap-
pears only after a certain period. Finally, changes in stock prices are associated with changes in wealth. In this case, the demand for
consumption and investment goods is affected, and influences the economic growth level.
A considerable number of studies investigate the stock prices — economic activity nexus, in both developed and emerging markets.
While for developed markets the researches document in general the causality between macroeconomic variables and stock markets
(Choi, Hauser, & Kopecky, 1999; Hassapis & Kalyvitis, 2002; Henry, Olekalns, & Shields, 2010; Panopoulou, 2009; Reboredo & Rivera-
Castro, 2014), for the emerging markets the findings are inconclusive (Naifar & Al Dohaiman, 2013; Tsouma, 2009; Zhu, Li, & Li, 2014).
This is also the case of the Indian context, where an appreciable number of studies report contradictory results. If the causality
between the stock market and the economic activity is not found by Pethe and Karnik (2000) or Bhattachary and Mukherjee
(2006), it is highlighted by Naka, Mukherjee, and Tufte (1998), Padhan (2007), Ahmed (2008), Singh (2010), Srivastava (2010),
Tripathy (2011), Pal and Mittal (2011), Ray (2012), Naik and Padhi (2012), Kalra (2012), Dasgupta (2013), Kumar (2013), Fang and
You (2014) or Ray and Sarkar (2014). Only few papers focus on the stock prices — economic growth relation, and none of these
researches employ the Granger causality in the frequency domain.
As far as we know, our paper is the first study addressing the link between stock prices and the economic growth in India, in a
frequency domain framework. Croux and Reusens (2013) investigate the predictive power of stock prices for the future domestic eco-
nomic activity in a frequency domain framework, but with a focus on G7 countries and within a bivariate framework only. Against this
background, we bring forward several contributions to the literature, including Croux and Reusens (2013), by relying on conditional
frequency domain framework.
First we test the short- and long-run Granger causality using the conditional and unconditional frequency domain approach of
Breitung and Candelon (2006). In the frequency domain, a stationary process can be expressed as a weighted sum of sinusoidal com-
ponents with a certain frequency. These frequency components can be analyzed separately and are often put in two heads, namely
quickly and slowly fluctuating components, associated with the short- and long-run respectively (Lemmens, Croux, & Dekimpe,
2008). Hence, the Granger causality analysis will help in finding out the concentration of the predictive power in the two fluctuating
components i.e., whether the predictive power is concentrated at the quickly fluctuating components, or at the slowly fluctuating
components. A priori, it is believed that the stock market can forecast the slowly fluctuating components of the future economic ac-
tivity more accurately than the quickly fluctuating components (Croux & Reusens, 2013; Rua, 2010). Therefore, we assume that the
Granger causality from the stock prices to the economic activity is significant for the slowly fluctuating components, but insignificant
for the quickly fluctuating components. However, the causality from the economic activity to the stock prices can manifest both in the
short- and long-terms, due to the theoretical reasons underlined above.
Second, we use a conditional vector autoregression (VAR) model as benchmark for comparing the results of the Granger causality
in the frequency domain analysis, and to reveal the advantages of the proposed methodology. Because our primary focus is on the eco-
nomic growth, only this variable is considered as endogenous in the conditional VAR model (similar to the conditional frequency do-
main model). Other macroeconomic variables frequently employed in the literature, as the exchange rate, trade balance, international
reserve or inflation, are treated as exogenous. This assumption is made in order to facilitate the comparison between the benchmark
and the frequency domain analysis.
Third, different from previous studies which use linear unit root tests for establishing the stationarity or non-stationarity of the
series, we use different seasonal and structural breaks unit root tests. More precisely, we start with the HEGY seasonal unit root
test of Hylleberg, Engle, Granger, and Yoo (1990). We continue with the seasonal unit root test proposed by Dickey and Zhang
(2010), and we then compare the reported results with the findings of the Zivot and Andrews (1992) unit root tests with a structural
break. We use these tests to see if our series are stationary and if it is recommended to proceed with the frequency domain analysis.
Finally, we focus on the Indian context which is particularly appealing because the Indian stock market is an emerging market,
prone to fluctuations and vulnerable to the international context, and because the previous results are rather contradictory. In this
context, our purpose is to add additional clarifications to the fact that the relation between stock prices and the economic growth
is non-linear and the causality changes at different frequencies. Moreover, we want to see if similarities appear between developed
and emerging countries when studying the stock prices — economic activity nexus in the frequency domain.
The remaining part of the study is organized as follows: the second section briefly presents the review of literature on the relation
between the stock market and the economic activity in India; the third section shortly discusses the data source and the methodology
employed; the fourth section presents the results of our analysis, while the fifth section offers conclusions and draws policy implications.

2. Literature review

The literature regarding the relationship between the stock market and the economic activity in India is prolific. Using different
investigation tools, a first strand of researches does not identify any causality between stock prices and the economic activity,
while the second group reveals a significant causal relationship.
Pethe and Karnik (2000) and Bhattachary and Mukherjee (2006) are the main exponents of the first group. Pethe and Karnik
(2000) use monthly data and investigate the period 1992–1997. Their cointegration and error correction model show that there is
no long-run relationship between the stock price and the state of the economy. In the same line, Bhattachary and Mukherjee (2006)
employ both a VAR and Toda and Yamamoto non-Granger causality techniques (Toda & Yamamoto, 1995), using a monthly sample,
for the time-span 1992–2001. No causal connection is found between stock returns and macroeconomic variables (i.e. money supply,
index of industrial production, GNP, real effective exchange rate, foreign exchange reserve and trade balance). At the same time, they
found significant evidence of a bidirectional causality between stock returns and the rate of inflation.
226 A.K. Tiwari et al. / International Review of Economics and Finance 39 (2015) 224–238

The second group of researches relies on cointegration techniques and includes the representative works of Naka et al. (1998), Padhan
(2007), Ahmed (2008), Singh (2010), Srivastava (2010), Tripathy (2011), Pal and Mittal (2011), Ray (2012), Naik and Padhi (2012), Kalra
(2012), Dasgupta (2013), Kumar (2013) and Ray and Sarkar (2014).
Naka et al. (1998) explore the relationships between macroeconomic variables (output, inflation, money in circulation, interest
rate) and the Indian stock market, employing a vector error correction (VEC) model developed by Johansen (1991). The researchers
find three long-term equilibrium relationships among these variables.
Padhan (2007) brings to front a new methodology for investigating the relation between stock prices and the economic activity in
India. His study relies on Granger non-causality tests proposed by Toda-Yamamota, Dolado and Lutkephol (known as the TYDL
model). The TYDL model's estimations show a bidirectional causality between stock prices and the economic activity, especially dur-
ing the post-liberalization period (similar results are reported by Padhan, 2007). Ahmed (2008) follows in his study the Johansen's
approach of cointegration and the Toda-Yamamoto Granger causality test, but uses, as novelty, quarterly data for the period 1995–
2007. A long-run relationship is demonstrated between stock prices and foreign direct investments (FDI), money supply and the
index of industrial production.
Singh (2010) uses monthly data for the period 1995–2009 and selects the correlation analysis, unit root tests and the Granger causality,
as the main tools for his investigation. The Granger causality test reveals a bilateral causal relationship only between the index of industrial
production and stock prices, while no causality is reported for the inflation, interest rate or foreign exchange rate. With similar variables,
but applying the Johansen's (1991) cointegration methodology, Srivastava (2010) finds that macroeconomic determinants have a sig-
nificant impact on the stock market in the long-run (i.e. the industrial production, the Wholesale Price Index and the interest rate).
Tripathy (2011) focuses on the period 01:2005 to 02:2011 and uses Ljung–Box Q, Breusch–Godfrey LM, unit root and Granger cau-
sality tests. He concludes that there is a bidirectional relationship between the interest rate and stock prices on the one hand, and be-
tween the exchange rate and the stock market on the other hand. Pal and Mittal (2011) obtain similar results for the period 1995–2008,
using quarterly data. Their study provides evidence of bidirectional causal relationship between financial development and economic
growth, and a unidirectional causal relationship from economic growth to the stock market development.
Recent investigations are performed by Ray (2012), Naik and Padhi (2012), Kalra (2012), Dasgupta (2013) and Ray and Sarkar
(2014). All these papers use the classical Johansen (1991)'s cointegration technique and the VEC model. Quasi-all authors confirm
that the macroeconomic variables and the stock market index are cointegrated in India and, as a consequence, a long-run equilibrium
relationship exists between them. A different approach is used by Kumar (2013). Its factor analysis shows that the Indian stock market
is highly responsive to the macroenvironment.
We notice that till date, the time domain analyses cannot provide to the policy makers a clear cut answer to the question whether
stock market promotes the economic growth, whether the economic growth stimulates the stock market, or both. However, we must
be aware that the relationship between stock prices and economic growth is dynamic and non-linear in nature. Therefore, the present
study analyzes the Granger causality between the stock market and economic activity in the framework of frequency domain. Based
on the theoretical arguments advanced in the previous section, we formulate the following hypotheses:

• H1: A bidirectional relationship exists between stock prices and economic growth.
• H2: The Granger causality from stock prices to economic activity is significant for the slowly fluctuating components, but insignif-
icant for the quickly fluctuating components.
• H3: The causality from the economic activity to the stock prices can manifest both in the short- and long-term, because, on the one hand
the markets are sensitive to the macroeconomic news and, on the other hand, a long-run economic growth stimulates financial
activities.

3. Data source and methodology

3.1. Data

Our primary focus is the relation between stock prices and economic growth. While the stock prices are measured by share prices
(SP), the economic growth is measured by the Industrial Index of Production (IIP). However, because we use a conditional VAR model
as benchmark, in line with previous researches, we include in our VAR a series of macroeconomic variables which explain the eco-
nomic activity and which are treated as exogenous (exchange rate, inflation, foreign exchange reserves, trade balance). A description
of these variables is made in Table 1.
All measures are index numbers (2005 = 100), in natural logarithm form, while foreign exchange reserves and trade balance are
expressed as percentage of Gross Domestic Product (GDP). All variables are taken from the International Monetary Fund (IMF) CD-
ROM-2012, excepting Real Effective Exchange Rates (REER), which are offered by Reserve Bank of India (RBI) online database
2013. The monthly dataset covers the period 1993 M4–2011 M1.1

3.2. Unit root tests

We start our empirical analysis with a series of unit root tests. Only stationary data can be explored in the frequency domain frame-
work. As we use monthly data, seasonality might influence our results. Therefore, different from previous studies, we adopt a three

1
The study period is limited by availability of the data for variables under consideration.
 A.K. Tiwari et al. / International Review of Economics and Finance 39 (2015) 224–238 227

Table 1
Variables' description.

Variable Description Database

Share prices (SP) The Share Price Index (Bombay Stock Exchange Index — BSE Sensex) offers information about the stock IMF
exchange market status and captures the current market price of certain group of shares.
Industrial Index of Production (IIP) Represents a particular economic Indian index, which measures the growth of mining, electricity and IMF
manufacturing sectors.
Exchange rate Real Effective Exchange Rates, based on trade weights (REER) and exports (REER1), respectively, illustrate the RBI
(REER and REER1) weighted average of nominal exchange rates, calculated by considering the purchasing power parity (PPP)
hypothesis, adjusted according to the relative price differential between domestic and foreign countries.
Whole Sale Price Index (WPI) It is a measure of the inflation rate. Whole Sale Price Index also captures the variation of prices of items, IMF
during the time, at the first main commercial transaction.
Reserves, less gold (R) Foreign exchange reserves quantify the special drawing rights, country reserves at IMF and holdings of IMF
foreign exchange which are under control of monetary authorities, excepting the gold holdings.
Trade balance (TB) Represents the difference between the exports and imports. IMF

step approach. In the first step we test for seasonal unit root in the data using the popular test proposed by Hylleberg et al. (1990),
called HEGY test. In the second step, in order to test the robustness of the results, we select the recently developed seasonal unit
root test proposed by Dickey and Zhang (2010). Finally, in the third step, we follow a structural break test proposed by Zivot and
Andrews (1992), to reconfirm our unit root findings regarding the retained variable.
Several methods are proposed in the econometric literature for testing unit roots for seasonal time series (see for example Dickey,
Hasza, & Fuller, 1984; Hylleberg et al., 1990; Canova & Hansen, 1995; Caner, 1998; Shin & So, 2000). However, the Hylleberg et al.'s
(1990) proposed test, HEGY, in comparison with other seasonal unit root tests has the advantage of the appropriate transformations
as it removes possible (seasonal) unit roots directly from the procedure itself and do not have to be implemented a priori. A descrip-
tion of this test can be found in Appendix A.

3.3. Causality analysis in frequency domain

In statistics, frequency domain is a term used to describe the domain for analysis of mathematical functions or signals with respect
to frequencies, rather than time. In the frequency domain, a stationary process can be expressed as a weighted sum of sinusoidal
components with a certain frequency (ω). However, in the frequency domain, a very similar definition holds for the Granger causality,
as in the time domain.2 Put it in a non-technical way, a time-domain graph shows how a signal changes over time, whereas a
frequency-domain graph shows how much of the signal lies within each given frequency band over a range of frequencies. In very
simple terms, “time” means the ability of indicating when a certain variation happens, whereas “frequency” is a component that mea-
sures the degree of a certain variation. Though there are other approaches, such as the Partial Directed Coherence (PDC) measures for
testing of Ganger-causality in frequency domain however, we focus on a slightly different approach of Granger causality, refined by
Geweke (1982) and later by Breitung and Candelon (2006). This approach provides an elegant interpretation of the frequency-
domain Granger causality as a decomposition of the total spectral interdependence between the two series (based on the bivariate
spectral density matrix, and directly related to the coherence) into a sum of “instantaneous”, “feedforward” and “feedback” causality
terms. Breitung and Candelon (2006) approach can be explained as follows:
Let zt = [xt, yt]′ be a two-dimensional vector of time series observed at t = 1, …, T and it has a finite-order VAR representation of the
form
ΘðLÞzt ¼ εt ð1Þ

where: Θ(L) = I − Θ1L − … − ΘpLp is a 2 × 2 lag polynomial with Lkzt = zt − k. We assume that the error vector εt is white noise with
E(εt) = 0 and E(εt εt′) = Σ,; where: Σ is positive definite. For ease of exposition we neglect any deterministic terms in Eq. (1).
Let G be the lower triangular matrix of the Cholesky decomposition G′G = Σ−1 such that E(ηtηt′) = I and ηt = Gεt. If the system is
assumed to be stationary, the MA representation of the system is:



Φ11 ðLÞ Φ12 ðLÞ ε 1t
zt ¼ ΦðLÞε t ¼ ð2Þ
Φ21 ðLÞ Φ22 ðLÞ ε2t



Ψ11 ðLÞ Ψ12 ðLÞ η1t
¼ ΨðLÞηt ¼ ð3Þ
Ψ21 ðLÞ Ψ22 ðLÞ η2t

where: Φ(L) = Θ(L)−1 and Ψ(L) = Φ(L)G−1. Using this representation the spectral density of xt can be expressed as:
1 n  
−iω 2


  o
−iω 2
f x ðωÞ ¼ Ψ11 e  þ Ψ12 e  : ð4Þ
2π

2
It is important to mention that the Granger (1969) approach to the question of whether X causes Y is to determine how much of the current Y can be explained by
past values of Y, and then to see whether adding lagged values of X can improve the explanation. Y is said to be Granger-caused by X if X helps in the prediction of Y, or if
the coefficients of the lagged Xs are statistically significant. It is important to note that the statement “X Granger causes Y” does not imply that Y is the effect or the result
of X. Granger causality measures precedence and information content but does not of itself indicate causality in the more common use of the term.
228 A.K. Tiwari et al. / International Review of Economics and Finance 39 (2015) 224–238

The measure of causality suggested by Geweke (1982) is defined as:

" #
2π f x ðωÞ
My→x ðωÞ ¼ log     ð5Þ
Ψ e−iω 2
11

   
  −iω 
 Ψ12 e 
¼ log1 þ   −iω   ð6Þ
  Ψ11 e  

If |ψ12(e−iω)|2 = 0, then the Geweke's measure will be zero, then y will not Granger cause x at frequency ω.
If the elements of zt are I(1) and co-integrated, in that case in the frequency domain the measure of causality can be defined by
using the orthogonalized MA representation:

e ðLÞε ¼ Ψ
Δzt ¼ Φ e ðLÞη ð7Þ
t t

where: Ψe ðLÞ ¼ Φ
e ðLÞG−1 ; η ¼ Gεt ; and G is a lower triangular matrix such that E(ηtηt′) = I. Note that in a bivariate co-integrated system
t
0e
β Ψð1Þ ¼ 0; where β is a co-integration vector such that β ′ zt is stationary (Engle & Granger, 1987). As in the stationary case the
resulting causality measure is:
   
 e −iω 
 Ψ12 e 

My→x ðωÞ ¼ log1 þ   −iω  : ð8Þ
 e
Ψ11 e 

To test the hypothesis that y does not cause x at frequency ω we consider the null hypothesis, within a bivariate framework:

My→x ðωÞ ¼ 0: ð9Þ

Breitung and Candelon (2006) present this test by reformulating the relationship between x and y in a VAR equation:

xt ¼ a1 xt−1 þ … þ ap xt−p þ β1 yt−1 þ … þ βp yt−p þ ε1t : ð10Þ

The null hypothesis tested by Geweke, My → x(ω) = 0, corresponds to the null hypothesis of:

H0 : RðωÞβ ¼ 0 ð11Þ

where: β is the vector of the coefficients of y and


cosðωÞ cosð2ωÞ ::::: cosðpωÞ
RðωÞ ¼ : ð12Þ
sinðωÞ sinð2ωÞ ::::::: sinðpωÞ

The ordinary F statistic for Eq. (12) is approximately distributed as F(2, T − 2p) for ω ∈ (0, π). It is interesting to consider the fre-
quency domain causality test within a co-integrating framework. To this end Breitung and Candelon (2006) suggested to replace xt in
regression (7) by Δxt, with the right-hand side of the equation remaining the same. Further, it is important to mention that in
cointegrated systems, the definition of causality at frequency zero is equivalent to the concept of “long-run causality”, and in station-
ary framework, there exists no long-run relationship between time series; a series may nevertheless explain future low frequency var-
iation of another time series. Hence, in a stationary system, causality at low frequencies implies that the additional variable is able to
forecast the low frequency component of the variable of interest on one period ahead.
In the frequency domain analysis the time dimension is lost but the frequency approach has different advantages, as in short series
as ours, a seasonal pattern may be important and the frequency domain allows eliminating these variations. Furthermore, the
methodology allows for observing non-linearities and causality cycles, that is the causality in high or low frequencies.

4. Data analysis and empirical findings

First of all, we present the descriptive statistics of variables in order to see the properties of the sample (Table 2).3 Greater volatility
is noticed in the trade balance. The negative values of the skewness statistic suggest a greater probability of large decreases for the
inflation and trade balance. The other variables are positively skewed. The kurtosis values indicate that changes in variables' values
are frequent. To test the normality of the variables we use a battery of normality tests and we find that each of the analyzed variables
follows a non-normal distribution as the null hypothesis of normality is rejected for each variable, by at least one normality test.

3
Time series plots of the variables are presented in Fig. 1B (Appendix B), to show the nature of the variables over the period.
 A.K. Tiwari et al. / International Review of Economics and Finance 39 (2015) 224–238 229

Table 2
Descriptive statistics.

SP IIP REER REER1 WPI R TB

Statistics
Mean 4.3533 4.4114 4.5987 4.5998 4.4422 11.006 −124.05
Median 4.0626 4.3750 4.5956 4.6037 4.4271 10.819 −39.071
Maximum 5.6503 5.1061 4.7022 4.7012 4.9519 12.629 11.055
Minimum 3.3896 3.7233 4.5032 4.5003 3.9044 8.8361 −699.24
Std. dev. 0.6569 0.3521 0.0410 0.0406 0.2658 1.0770 160.83
Skewness 0.7023 0.0849 0.3649 −0.0127 −0.0085 0.0165 −1.5460
Kurtosis 1.9989 2.0756 2.8146 3.1909 2.0426 1.7365 4.4571
Normality tests
Jarque–Bera 26.528 7.8754 5.0564 0.3307 8.1748 14.243 104.18
(0.000) (0.019) (0.079) (0.847) (0.016) (0.000) (0.000)
Doornik–Hansen 108.889 11.070 6.8489 0.9507 11.358 25.653 257.88
(0.000) (0.003) (0.032) (0.621) (0.003) (0.000) (0.000)
Shapiro–Wilk 0.85943 0.9744 0.9819 0.9831 0.9754 0.9280 0.7523
(0.000) (0.000) (0.007) (0.011) (0.000) (0.000) (0.000)
Lilliefors 0.18322 0.0715 0.07373 0.0814 0.0569 0.1116 0.2511
(0.000) (0.010) (0.010) (0.000) (0.090) (0.000) (0.000)
Observations 214 214 214 214 214 214 214

Note: p-values in brackets (…).

Table 3
HEGY unit root analysis for constant and trend model.

Test statistics SP IIP REER REER1 WPI R TB

[0] [3] [1] [1] [0] [0] [10]

t (π1) 2.119 2.484 3.315 3.542 2.181 2.606 0.288

t (π2) 1.591 2.419 3.332 3.562 2.185 2.447 0.777
F(π3, π4) 15.86⁎⁎⁎ 6.777⁎⁎ 14.99⁎⁎⁎ 13.48⁎⁎⁎ 21.49⁎⁎⁎ 16.94⁎⁎⁎ 11.87⁎⁎⁎
F(π5, π6) 23.87⁎⁎⁎ 16.00⁎⁎⁎ 27.90⁎⁎⁎ 24.33⁎⁎⁎ 14.04a 18.97⁎⁎⁎ 8.435⁎⁎
F(π7, π8) 12.27⁎⁎⁎ 6.876⁎⁎ 34.43⁎⁎⁎ 31.53⁎⁎⁎ 16.35⁎⁎⁎ 23.74⁎⁎⁎ 5.609⁎
F(π9, π10) 21.97⁎⁎⁎ 12.57⁎⁎⁎ 30.49⁎⁎⁎ 28.52⁎⁎⁎ 19.17⁎⁎⁎ 18.61⁎⁎⁎ 2.167
F(π11, π12) 15.52⁎⁎⁎ 9.711⁎⁎⁎ 16.18⁎⁎⁎ 14.61⁎⁎⁎ 14.91⁎⁎⁎ 17.28⁎⁎⁎ 8.209⁎⁎
F(π1…… π12) 308.1⁎⁎⁎ 11.16⁎⁎⁎ 28.09⁎⁎⁎ 25.64⁎⁎⁎ 324.5⁎⁎⁎ 906.2⁎⁎⁎ 15.46⁎⁎⁎
F(π2…… π12) 335.9⁎⁎⁎ 11.01⁎⁎⁎ 24.24⁎⁎⁎ 22.08⁎⁎⁎ 336.4⁎⁎⁎ 987.9⁎⁎⁎ 16.74⁎⁎⁎

Notes: [k] shows the lag length used.

⁎⁎⁎ Denotes significance at 1%.
⁎⁎ Denotes significance at 5%.
⁎ Denotes significance at 10% level.

In the next step, we test for the presence of unit roots in our series. We start with the seasonal unit root HEGY test (Table 3 presents
the results).4 Table 3 shows that for all the variables, the null hypothesis of unit root at annual and semi-annual frequencies is accepted
at 1% level of significance. However, based on the F-value, the null hypothesis of unit root at quarterly and all other higher frequencies
is rejected at least at 10% level of significance (except for the trade balance — F9,10). This result suggests that all the variables are non-
stationary at annual and quarterly level, but not at the monthly or higher frequency levels. As our series have a monthly frequency, we
can therefore proceed with the VAR analysis.
Nevertheless, in order to check for the robustness of HEGY unit root test findings, we perform a test recently developed by Dickey
and Zhang (2010), known as the DST test statistic. Finally, to further validate the results obtained from the Dickey and Zhang (2010)
proposed unit root test, we use the Zivot and Andrews (1992) structural break test. The results of both unit root tests are presented in
Table 4 below.
The estimations indicate that both test statistics reject the null hypothesis of unit root at least with 10% level of significance, for all
the variables analyzed in our model. Hence, we conclude that all retained variables are stationary in level. The DST test and the Zivot
and Andrews (1992) test confirm thus the results of the HEGY test for the monthly frequency.
The third step of our empirical approach consists in the estimation of a classical conditional VAR. We use a one-shot causality mea-
sure in a VAR framework to test for the causality between share prices and the economic activity (Table 5). We consider all remaining
variables as exogenous, to facilitate the comparative analysis of the results obtained from the VAR model with the frequency domain
causality analysis. The VAR model diagnostics test of stability, serial correlation, lag-exclusion, heteroskedasticity, normality and the
conditional impulse response function are presented in Appendix D. (See Tables 1D–4D and Fig. 1D.)
It is evident from Table 5 that there is bidirectional causal relation between SP and IIP, after conditioning the VAR model. Although the
impulse response analysis shows that the response of IIP and SP, due to one standard deviation shock in the other variable, is positive, the

4
Tables 1C and 2C (Appendix C) present the tests of seasonal unit root in monthly data and critical values for the HEGY test.
230 A.K. Tiwari et al. / International Review of Economics and Finance 39 (2015) 224–238

Table 4
Unit root tests for constant and trend model.

Test SP IIP REER REER1 WPI R TB

DST statistic (corrected) −12.00 −8.803 −15.89 −15.14 −9.407 −10.46 −8.634
[k] [4] [3] [1] [1] [9] [4] [3]
(Probability) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000)
Zivot–Andrews test −3.960 −3.281 −5.623 −3.681 −4.510 −4.531 −5.279
[k] [1] [12] [8] [0] [3] [3] [1]
(Probability) (0.001) (0.012) (0.040) (0.000) (0.089) (0.020) (0.000)

Notes: (i) p-values in brackets (…). (ii) [k] denotes that final AR(p) order selected by BIC in case of DST test statistics and in case of Zivot–Andrews test it shows the lag
length used.

Table 5
VAR Granger causality/Block Exogeneity Wald Tests.

Dependent variable: IIP Dependent variable: SP

Excluded Chi-sq df Prob. Excluded Chi-sq df Prob.

SP 20.72292 6 0.0021 IIP 11.94804 6 0.0631

overall results are not similar. Interestingly, the response of the industrial production index, due to one standard deviation shock in share
prices, is fairly stable, whereas the response of share prices, due to one standard deviation shock in the industrial production, shows
positive increasing trend (Fig. 2D). So, even if we admit the bidirectional causality as other papers do (Dasgupta, 2013; Kalra, 2012;
Naik & Padhi, 2012; Naka et al., 1998; Ray, 2012; Ray & Sarkar, 2014; Singh, 2010; Srivastava, 2010), we find that the SP is more sensitive
to the IIP fluctuations than conversely.
The final, and the main step in our empirical approach, is to present the results of the frequency domain analysis. There are two
series of results. In the first case, a bivariate unconditional model is estimated (Fig. 1). Fig. 1 provides evidence of business cycle
causality from both directions, but we are not able to reject the null hypothesis of non-causality for any frequency.
In the second case, a bivariate conditional model is estimated (Fig. 2). In this case, we have much more clear and interesting results,
which can be compared with the VAR results presented above. Our findings show that the null hypothesis, that share prices do not
Grange-cause the economic activity in frequency domain (SP ≠ IIP), is rejected at 5% level of significance at frequencies less than 0.4
that corresponds to the 15.7 months, and higher month-cycles.5 For other cases, though we find evidence of business cycle causality
from both directions, yet for none of the frequencies we are able to reject the null hypothesis of non-causality. At the same time, the
null hypothesis of IIP ≠ SP cannot be rejected. We find thus a unidirectional causality, from the share prices to the economic activity.6
When we use the bivariate unconditional frequency domain causality, our results fully confirm the first strand of literature
(e.g. Bhattachary & Mukherjee, 2006; Pethe & Karnik, 2000), which does not identify any causality between the stock market
and the economic activity. On the other hand, when the bivariate frequency domain causality is estimated by conditioning
the model, the findings partially mirror the second group of contributions, as the share price index is a good predictor for the econom-
ic activity in India, but only for low frequencies. In general, empirical studies investigating the bidirectional causality between stock
prices and the economic activity indicate a positive linkage running from stock returns to the economic activity (Tsouma, 2009).
While the recent literature usually documents bidirectional causality (this is also the case of our VAR model), using the frequency
domain analysis we discover only a unidirectional causality. This output might arise from new methodological used tools and period
of analysis (i.e. 1993M4–2011M1), which is applied for the first time in the exploration of the Indian economy.
However, our results are in line with those reported by Croux and Reusens (2013) in the case of G7 group of countries. Using
quarterly data, the authors find that the predictive power of the stock prices for the economic activity is predominantly present
at low frequencies. Because the stock market volatility affects the level of consumption and investment, and thus the economic
growth, it is expected that economic agents react to a smaller extent to financial signals which are not persistent in time. Con-
sequently, in order to foresee the evolution of the future economic activity, economic agents must consider the long-run trend of the
stock market.

5. Conclusions and policy implications

In the present study we analyze the Granger-causality between share prices and the economic activity for the Indian economy,
using monthly data covering the period 1993M4–2011M1. Resorting to different seasonal and structural breaks unit root tests, we

5
Since high frequencies are associated with short periods and low frequencies with the long-run, the figures of the Granger causality in the frequency domain stand
reversed (the short term fluctuations/cycles are presented at the right side of the figure).
6
For additional robustness purpose, we use REER1 instead of REER in the conditioned model, and we find similar results (Fig. 1E — Appendix E).
 A.K. Tiwari et al. / International Review of Economics and Finance 39 (2015) 224–238 231

Calculated values of F-statistic

5

4
BC critical value

2.97…
0.17…
0.33…
0.50…
0.66…
0.83…
0.99…
1.16…
1.32…
1.49…
1.65…
1.82…
1.98…
2.15…
2.31…
2.48…
2.64…
2.81…
0.01

Note: Frequency (ω )= 2π /cycle length (T)

Fig. 1. Breitung and Candelon (2006) frequency domain Granger-causality — unconditioned model. Note: Frequency (ω) = 2π/cycle length (T).

find that all the variables are stationary in level. Therefore we investigate the causality between share prices and the economic activity
using in the first stage a conditional VAR model. The main findings reveals that there is a bidirectional causal relationship between the
tested variables, after conditioning the VAR model.
In the second phase, we perform a frequency domain analysis. Our findings show that, for the Indian economy, the causal
and reverse causal relations between SP and IIP vary across frequencies. More concretely, the frequency domain approach
reveals that only the share price Grange-causes the economic activity for frequencies less than 0.4, which correspond to
the 15.7 months and higher month-cycles. For the rest of the situations, the results illustrate business cycle causality from
both directions, without any significant causality across the rest of the frequencies.
In conclusion, our findings provided evidence of a unidirectional frequency domain causal relationship between share
prices and the industrial production, only in the long-run (i.e., for low frequency components), which is running from
share prices to the industrial production. Our study shows that share prices are a leading indicator for growth in the indus-
trial production. In this case, in order to adjust the industrial production in the long-run, the Indian economic policies shall be
focused with predilection on the stock market environment. At the same time, a business cycle effect appears between the
stock market and the economic activity, even if the results are not significant at 5% significance level. However, in order to
implement counter-cyclical measures, the governmental policy shall be oriented on both the industrial activities and the
stock market.

35
Calculated values of F-statistic

30

25
BC critical value

20

15

10

0
0.17 …
0.33 …
0.50 …
0.66 …
0.83 …
0.99 …
1.16 …
1.32 …
1.49 …
1.65 …
1.82 …
1.98 …
2.15 …
2.31 …
2.48 …
2.64 …
2.81 …
2.97 …
0.01

Note: Frequency (ω )= 2π/cycle length (T)

Fig. 2. Breitung and Candelon (2006) frequency domain Granger-causality — conditional model. Note: Frequency (ω) = 2π/cycle length (T).
232 A.K. Tiwari et al. / International Review of Economics and Finance 39 (2015) 224–238

Acknowledgments

We would like to thank the two referees for their comments and suggestions, leading to a considerably improved manuscript.

Appendix A. HEGY seasonal unit root test proposed by Hylleberg et al. (1990)

Hylleberg et al. (1990) propose a method to test whether a time series contains seasonal unit roots in the presence of other unit
roots and seasonal processes. Applying a (1 − L4) to quarterly series, where L is the usual lag operator, implies that one assumes the
presence of four unit roots, as (l − L4) = (l − L)(l + L)(l − iL)(l + iL) = (l − L)(l + L)(l + L2), hence the unit roots are 1, −1, i and −i′.
Hylleberg et al. (1990) show that testing for seasonal unit roots amounts to testing the significance of the parameters of an auxiliary
regression, which may also contain deterministic elements, like a constant, trend and seasonal dummies. The auxiliary regression
they derive is

X
p

Δ4 yt ¼ π1 z1;t−1 þ π 2 z2;t−1 þ π3 z3;t−1 þ π 4 z3;t−2 þ α j Δ4 yt− j þ εt ðA1:1Þ
j¼1

where: yt is the time being tested, z1t = (1 + L + L2 + L3)yt, z2t = (− 1 + L − L2 + L3)yt, z3t = (− 1 + L2)yt and z4t =
(1 − L4)yt, = Δ4yt = yt − yt − 4, L denoting the usual lag operator, and where {εt} is assumed to be a white noise process.
Applying an OLS regression to this auxiliary regression gives estimates of the πi's. Hylleberg et al. (1990) show that, when π1 = 0,
the series contains the (non-seasonal or zero frequency) root 1, when π2 = 0 the (semi-annual) root − 1 is present i.e., root −1
corresponds to unit roots ½ cycle per quarter or 2 cycle per year, the presence of the (annual) roots ±i (i = √−1) implying that
π3 = π4 = 0 (the stationary alternatives being π1 b 0, π2 b 0 and π3 b 0 and/or π4 = 0) i.e., ±i corresponds to unit roots at 1/4 cycle
per quarter or one cycle per year.
Thus, inference on the presence of seasonal unit roots may be carried out through the t-ratios associated to the last three πi
coefficients: tπ2, tπ3 and tπ4. On the other hand, evidence on the presence (absence) of a non-seasonal unit root is given by
tπ1. However, the analysis of stochastic seasonal non-stationarity becomes simpler if, instead of testing three separate hy-
potheses, we test some joint null hypotheses. To that end, one can use the F-statistics F 34 , which tests H 0 : π 3 = π 4 = 00,
and F234, associated to H0: π2 = π3 = π4 = 0. Finally, one can also test whether all the πi parameters are zero [i.e., whether the
Δ4 = (1 − L4) filter is appropriate] using F1234. The asymptotic distributions of the test statistics under the respective null hypotheses
depend on the deterministic terms in the model.
The number of lagged seasonal differences Δ4 yt − j has to be chosen before the HEGY tests can be performed. This again may be
done by using model selection criteria or parameter significance tests.
As for quarterly series, the test for monthly time series also amount to testing the significance in an auxiliary regression. In monthly
series the (1 − L12) filter has twelve unit roots. We then have:

pffiffiffi ! pffiffiffi ! pffiffiffi !

12 3þi 3−i 3þi
1−L ¼ ð1−LÞð1 þ LÞð1−iLÞð1 þ iLÞ 1 þ L 1þ 1− L
2 2 2
pffiffiffi ! pffiffiffi ! pffiffiffi ! pffiffiffi ! pffiffiffi ! ðA1:2Þ
3−i i 3þ1 i 3þ1 i 3þ1 i 3−1
1− L 1þ L 1− L 1− L 1þ L :
2 2 2 2 2

Collecting two terms at a time this equation can be written as

            
12 2 2 2 2 2 2 4 2 4 2 4
l−L ¼ l−L lþL 1 þ √3L þ L 1−√3L þ L 1þLþL l þ L þ L ¼ 1−L l−L þ L lþL þL :

For monthly series, the corresponding tests for seasonal unit roots are discussed by Franses (1990), based on the model proposed
by Hylleberg et al. (1990).

Δ12 yt ¼ π1 z1;t−1 þ π2 z2;t−1 þ π3 z3;t−1 þ π 4 z3;t−2 þ π5 z4;t−1 þ π6 z4;t−2 þ π7 z5;t−1 þ π8 z5;t−2 þ π9 z6;t−1 þ π10 z6;t−2
Xp

þ π 11 z7;t−1 þ π 12 z7;t−2 þ α j Δ12 yt− j þ εt ðA1:3Þ
j¼1
 A.K. Tiwari et al. / International Review of Economics and Finance 39 (2015) 224–238 233

where:

  
2 4 8
z1;t ¼ ð1 þ LÞ 1 þ L 1 þ L þ L yt
  
2 4 8
z2;t ¼ −ð1−LÞ 1 þ L 1 þ L þ L yt
  
2 4 8
z3;t ¼ − 1−L 1 þ L þ L yt
  pffiffiffi  
4 2 4 8
z4;t ¼ − 1−L 1− 3L þ L 1 þ L þ L yt
  pffiffiffi  
4 2 4 8
z5;t ¼ − 1−L 1 þ 3L þ L 1 þ L þ L yt
   
4 2 4 2
z6;t ¼ − 1−L 1−L þ L 1−L þ L yt
   
4 2 4 2
z7;t ¼ − 1−L 1−L þ L 1 þ L þ L yt
 
12
z8;t ¼ 1−L yt :

The process yt has a regular (zero frequency) unit root if π1 = 0 and it has seasonal unit roots if any one of the other πi (i = 2,…,12) is
zero. For the conjugate complex roots, πi = πi+1 = 0 (i = 3, 5, 7, 9, 11) is required. The corresponding statistical hypotheses can again be
checked by t- and F-statistics, critical values for which are given by Franses and Hobijn (1997). If all the πi (i = 2,…,12) are zero, then a
stationary model for the monthly seasonal differences of the series is suitable. As in the case of quarterly series, it is also possible to
include deterministic terms in the model (A1.3).

Appendix B. Variables' dynamics

IIP SHARE_PRICES REER

5.2 6.0 4.75

5.5
4.70
4.8
5.0
4.65
4.4 4.5
4.60
4.0
4.0
4.55
3.5

3.6 3.0 4.50

1994 1996 1998 2000 2002 2004 2006 2008 2010 1994 1996 1998 2000 2002 2004 2006 2008 2010 1994 1996 1998 2000 2002 2004 2006 2008 2010

RESERVES_LESS_GOLD TRADE_BALANCE WPI

13 200 5.0

4.8
12 0

4.6
11 -200
4.4
10 -400
4.2

9 -600
4.0

8 -800 3.8
1994 1996 1998 2000 2002 2004 2006 2008 2010 1994 1996 1998 2000 2002 2004 2006 2008 2010 1994 1996 1998 2000 2002 2004 2006 2008 2010

REER1
4.72

4.68

4.64

4.60

4.56

4.52

4.48
1994 1996 1998 2000 2002 2004 2006 2008 2010

Fig. 1B. Time series plots of the variables.

234 A.K. Tiwari et al. / International Review of Economics and Finance 39 (2015) 224–238

Appendix C. Tests of seasonal unit root in monthly data and critical values for the HEGY test

Table 1C
Tests of seasonal unit root in monthly data.

Null hypothesis Alternative hypothesis Test statistic

π1 = 0 π1 b 0 t (π1)
π2 = 0 π2 b 0 t (π2)
π3 ∩ π4 = 0 π3 ∪ π4 ≠ 0 F(π3, π4)
π5 ∩ π6 = 0 π5 ∪ π6 ≠0 F(π5, π6)
π7 ∩ π8 = 0 π7 ∪ π8 ≠ 0 F(π7, π8)
π9 ∩ π10 = 0 π9 ∪ π10 ≠ 0 F(π9, π10)
π11 ∩ π12 = 0 π11 ∪ π12 ≠ 0 F(π11, π12)
π2∩……..∩π12 = 0 π2∪….∪π4 ≠ 0 F(π2…… π12)
π1∩……..∩π12 = 0 π1∪….∪π4 ≠ 0 F(π1…… π12)

Table 2C
Critical values for the HEGY test.

Test statistic 1% 5% 10%

HEGY Test: Intercept, time trend and seasonal dummies.

t (π1) −3.91 −3.35 −3.08

t (π2) −3.34 −2.81 −2.51
F(π3, π4) 8.38 6.35 5.45
F(π5, π6) 8.55 6.48 5.46
F(π7, π8) 8.39 6.30 5.33
F(π9, π10) 8.50 6.40 5.47
F(π11, π12) 8.75 6.46 5.36
F(π2…π12) 5.34 4.58 4.26
F(π1… π12) 5.15 4.44 4.07

Note: Critical values are obtained from Franses and Hobijn (1997).

Appendix D. The VAR model diagnostics test and the impulse response function

Table 1D
VAR Lag Order Selection Criteria.

VAR Lag Order Selection Criteria

Endogenous variables: IIP, share prices

Exogenous variables: C REER, reserves (less Gold), trade-balance, WPI

Sample: 1993 M04 2011 M01

Included observations: 206

Lag LogL LR FPE AIC SC

0 342.4081 NA 0.000136 −3.227263 −3.065716

1 624.3807 544.7820 9.15e − 06 −5.926026 −5.699860
2 640.9350 31.66202 8.10e − 06 −6.047912 −5.757127⁎
3 641.9075 1.841246 8.34e − 06 −6.018520 −5.663115
4 652.6358 20.10248 7.82e − 06 −6.083843 −5.663819
5 657.9714 9.894192⁎ 7.72e − 06 −6.096810 −5.612168
6 662.0311 7.449319 7.71e − 06⁎ −6.097389⁎ −5.548128
7 662.6156 1.061244 7.98e − 06 −6.064229 −5.450349
8 664.5987 3.561763 8.14e − 06 −6.044647 −5.366148

Notes:(i) LR: sequential modified LR test statistic (each test at 5% level); FPE: Final prediction error; AIC: Akaike information criterion; SC: Schwarz information
criterion; HQ: Hannan–Quinn information criterion. (ii) According to the AIC criterion, the optimal lag is 6.
⁎ Indicates lag order selected by the criterion.
 A.K. Tiwari et al. / International Review of Economics and Finance 39 (2015) 224–238 235

Table 2D
VAR analysis.

Vector Autoregression Estimates

Sample (adjusted): 1993 M10 2011 M01

Included observations: 208 after adjustments

Standard errors in () & t-statistics in []

IIP Share prices

IIP(−1) 0.286286 0.250941

(0.07271) (0.11361)
[3.93757] [2.20877]
IIP(−2) 0.407541 0.066576
(0.07343) (0.11474)
[5.54999] [0.58022]
IIP(−3) 0.116697 −0.115727
(0.07699) (0.12030)
[1.51574] [−0.96195]
IIP(−4) −0.280002 0.003976
(0.07795) (0.12180)
[−3.59221] [0.03265]
IIP(−5) −0.214262 −0.005707
(0.07568) (0.11826)
[−2.83124] [−0.04826]
IIP(−6) 0.101197 0.303823
(0.07525) (0.11758)
[1.34485] [2.58390]
Share prices(−1) 0.124836 1.187320
(0.04669) (0.07296)
[2.67377] [16.2743]
Share prices (−2) −0.138439 −0.193286
(0.07145) (0.11166)
[−1.93745] [−1.73109]
Share prices (−3) 0.110503 −0.086027
(0.07112) (0.11113)
[1.55378] [−0.77410]
Share prices (−4) −0.076737 0.076861
(0.07122) (0.11129)
[−1.07747] [0.69064]
Share prices (−5) 0.020572 −0.066224
(0.07106) (0.11104)
[0.28951] [−0.59641]
Share prices (−6) 0.019607 0.011121
(0.04554) (0.07117)
[0.43050] [0.15626]
C −0.184793 0.641996
(0.48701) (0.76101)
[−0.37944] [0.84361]
REER −0.113291 0.034148
(0.08499) (0.13280)
[−1.33306] [0.25714]
Reserves (less Gold) −0.001253 0.051484
(0.01564) (0.02444)
[−0.08009] [2.10645]
Trade balance 8.84E − 05 −5.73E − 05
(4.1E − 05) (6.4E−05)
[2.15826] [−0.89500]
WPI 0.683882 −0.736841
(0.15999) (0.25000)
[4.27449] [−2.94731]
R-squared 0.986635 0.991108
Adj. R-squared 0.985515 0.990364
Sum sq. resids 0.316790 0.773521
S.E. equation 0.040726 0.063638
F-statistic 881.2519 1330.628
Log likelihood 379.5145 286.6722
Akaike AIC −3.485717 −2.593002
Schwarz SC −3.212937 −2.320222
Mean dependent 4.430602 4.378961
S.D. dependent 0.338388 0.648279
Determinant resid covariance (dof adj.) 6.54E − 06
(continued on next page)
236 A.K. Tiwari et al. / International Review of Economics and Finance 39 (2015) 224–238

Table 2D (continued)

Vector Autoregression Estimates

Sample (adjusted): 1993 M10 2011 M01

Included observations: 208 after adjustments

Standard errors in () & t-statistics in []

IIP Share prices

Determinant resid covariance .52E − 06

Log likelihood 668.8952
Akaike information criterion −6.104762
Schwarz criterion −5.559203

Inverse Roots of AR Characteristic Polynomial

1.5

1.0

0.5

0.0

-0.5

-1.0

-1.5
-1.5 -1.0 -0.5 0.0 0.5 1.0 1.5

Fig. 1D. VAR stability analysis — unit circle.

Table 3D
VAR stability analysis.

Roots of characteristic polynomial

Endogenous variables: IIP, share prices

Exogenous variables: C REER, reserves (less Gold) trade balance, WPI

Lag specification: 1 6

Root Modulus

0.979251 0.979251
0.724781 − 0.383950i 0.820199
0.724781 + 0.383950i 0.820199
−0.397935 − 0.662938i 0.773201
−0.397935 + 0.662938i 0.773201
0.611473 − 0.305894i 0.683718
0.611473 + 0.305894i 0.683718
−0.654923 − 0.151569i 0.672233
−0.654923 + 0.151569i 0.672233
0.061340 − 0.542102i 0.545562
0.061340 + 0.542102i 0.545562
−0.195121 0.195121

No root lies outside the unit circle.

VAR satisfies the stability condition.
 A.K. Tiwari et al. / International Review of Economics and Finance 39 (2015) 224–238 237

Table 4D
Diagnostic checks analysis.

VAR Residual Serial Correlation LM Tests Test statistics p-Values

6lag 1.616882 0.8058
VAR Lag Exclusion Wald Tests (Joint test)
6lag 7.795094*** [0.099379]
VAR Residual Normality Tests-Joint J–B test (Orthogonalization: Residual Covariance (Urzua)): Null Hypothesis: residuals are multivariate normal
12.27516 0.1982
VAR Residual Heteroskedasticity Tests: Includes Cross Terms
493.6044 0.1085
Multivariate ARCH–LM TEST with 2 lags
VARCH–LM test statistic: 198.7159 0.5124

Note: (1) *, ** and *** denotes significant at 1%, 5%, and 10% level respectively.
Source: Author's calculation.

Response to Generalized One S.D. Innovations ± 2 S.E.

Response of IIP to IIP Response of IIP to SHARE_PRICES
.05 .05

.04 .04

.03 .03

.02 .02

.01 .01

.00 .00

-.01 -.01

-.02 -.02
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10

Response of SHARE_PRICES to IIP Response of SHARE_PRICES to SHARE_PRICES

.12 .12

.10 .10

.08 .08

.06 .06

.04 .04

.02 .02

.00 .00
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10

Fig. 2D. Conditional impulse response functions analysis.

Appendix E. Robustness check for the frequency domain conditional model

35
30
Calculated values of F

25
20 BC critical value
statistic

15
10
5
0
0.17…
0.33…
0.50…
0.66…
0.83…
0.99…
1.16…
1.32…
1.49…
1.65…
1.82…
1.98…
2.15…
2.31…
2.48…
2.64…
2.81…
2.97…
0.01

Note: Frequency (ω ) = 2π/cycle length (T)

Fig. 1E. Breitung and Candelon (2006) frequency domain Granger causality — conditional model (robustness check with REER1). Note: Frequency (ω) = 2π/cycle length (T).
238 A.K. Tiwari et al. / International Review of Economics and Finance 39 (2015) 224–238

References

Ahmed, S. (2008). Aggregate economic variables and stock market in India. International Research Journal of Finance and Economics, 14, 14–64.
Antonios, A. (2010). Stock market and economic growth: An empirical analysis for Germany. Business and Economics Journal, BEJ-1, 1–12.
Bhattachary, B. B., & Mukherjee, J. (2006). Indian stock price movement and the macroeconomic context — A time series analysis. Journal of International Business and
Economics, 5, 167–181.
Breitung, J., & Candelon, B. (2006). Testing for short and long-run causality: A frequency domain approach. Journal of Econometrics, 132, 363–378.
Caner, M. (1998). A local optimal seasonal unit root. Journal of Business and Economics Statistics, 16, 349–359.
Canova, F., & Hansen, B. F. (1995). Are seasonal patterns constant over time: A test for seasonal stability. Journal of Business and Economics Statistics, 13, 237–252.
Choi, J., Hauser, S., & Kopecky, K. (1999). Does the stock market predict real activity? Time series evidence from the G-7 countries. Journal of Banking and Finance, 23,
1771–1792.
Croux, C., & Reusens, P. (2013). Do stock prices contain predictive power for the future economic activity? A granger causality analysis in the frequency domain. Journal
of Macroeconomics, 35, 93–103.
Dasgupta, R. (2013). The influence of macroeconomic variables on Indian stock market — An empirical study. Asian Journal of Research in Banking and Finance, 3, 1–25.
Dickey, D. A., Hasza, D. P., & Fuller, W. A. (1984). Testing for unit roots in seasonal time series. Journal of the American Statistical Association, 79, 355–367.
Dickey, A., & Zhang, Y. (2010). Seasonal unit root tests in long periodicity cases. Journal of the Korean Statistical Society, 39, 271–279.
Engle, R. F., & Granger, C. W. J. (1987). Co-integration and error correction: Representation, estimation, and testing. Econometrica, 55, 251–276.
Fama, E. (1990). Stock returns, expected returns, and real activity. Journal of Finance, 45, 1089–1108.
Fang, C. -R., & You, S. -Y. (2014). The impact of oil price shocks on the large emerging countries' stock prices: Evidence from China, India and Russia. International
Review of Economics and Finance, 29, 330–338.
Franses, P. H. (1990). Testing for seasonal unit roots in monthly data. Econometric Institute report no. 9032A. Rotterdam: Erasmus University.
Franses, P. H., & Hobijn, B. (1997). Critical values for unit root tests in seasonal time series. Journal of Applied Statistics, 24, 25–46.
Geweke, J. (1982). Measurement of linear dependence and feedback between multiple time series. Journal of American Statistical Association, 77, 304–324.
Goldsmith, R. W. (1969). Financial structure and development. New Haven: Yale University Press.
Granger, C. W. J. (1969). Investigation causal relations by econometric models and cross-spectral methods. Econometrica, 37, 424–438.
Gurley, J., & Shaw, E. (1955). Financial aspects of economic development. American Economic Review, 45, 515–537.
Hassapis, C., & Kalyvitis, S. (2002). Investigating the links between growth and real stock price changes with empirical evidence from the G-7 economies. The Quarterly
Review of Economics and Finance, 42, 543–575.
Henry, O., Olekalns, N., & Shields, K. (2010). Sign and phase asymmetry: News, economic activity and the stock market. Journal of Macroeconomics, 32, 1083–1100.
Hylleberg, S., Engle, R. F., Granger, C. W. J., & Yoo, B. S. (1990). Seasonal integration and cointegration. Journal of Econometrics, 44, 215–238.
Johansen, S. (1991). Estimation and hypothesis testing of cointegrating vectors in Gaussian vector autoregressive models. Econometrica, 59, 1551–1580.
Kalra, R. (2012). Impact of macroeconomic variables on Indian stock market. IUP Journal of Financial Risk Management, 9, 43–54.
Kumar, R. (2013). The Effect of Macroeconomic Factors on Indian Stock Market Performance: A Factor Analysis Approach. IOSR Journal of Economics and Finance (IOSR-
JEF), 1(3), 14–21.
Lamont, O. A. (2001). Economic tracking portfolios. Journal of Econometrics, 105, 161–184.
Lemmens, A., Croux, C., & Dekimpe, M. G. (2008). Measuring and testing Granger-causality over the spectrum: An application to European production expectation
surveys. International Journal of Forecasting, 24, 414–431.
Naifar, N., & Al Dohaiman, M. S. (2013). Nonlinear analysis among crude oil prices, stock markets' return and macroeconomic variables. International Review of
Economics and Finance, 27, 416–431.
Naik, P. K., & Padhi, P. (2012). The impact of macroeconomic fundamentals on stock prices revisited: Evidence from Indian data. Eurasian Journal of Business and
Economics, 5, 25–44.
Naka, A., Mukherjee, T. K., & Tufte, D. R. (1998). Macroeconomic variables and the performance of the Indian stock market. Working paper 1998–06. University of New
Orleans, Department of Economics and Finance.
Padhan, P. C. (2007). The nexus between stock market and economic activity: An empirical analysis for India. International Journal of Social Economics, 34, 741–753.
Pal, K., & Mittal, R. (2011). Impact of macroeconomic indicators on Indian capital markets. Journal of Risk Finance, 12, 84–97.
Panopoulou, E. (2009). Financial variables and euro area growth: A non-parametric causality analysis. Economic Modelling, 26, 1414–1419.
Pethe, A., & Karnik, A. (2000). Do Indian stock market maters? Stock market indices and macroeconomic variables. Economic and Political Weekly, 35, 349–356.
Ray, P. (2012). Testing Granger causal relationship between macroeconomic variables and stock price behaviour: Evidence from India. Advances in Applied Economics
and Finance, 3, 470–481.
Ray, P., & Sarkar, J. (2014). Macroeconomic link to Indian capital market: A post-liberalization evidence. Modern Economy, 5, 272–288.
Reboredo, J. C., & Rivera-Castro, M. A. (2014). Wavelet-based evidence of the impact of oil prices on stock returns. International Review of Economics and Finance, 29,
145–176.
Rua, A. (2010). Measuring comovement in the time–frequency space. Journal of Macroeconomics, 32, 685–691.
Schumpeter, J. (1912). The theory of economic development. Cambridge, M.A.: Harvard University Press.
Schwarz, W. G. (1978). Estimating the dimension of a model. Annals of Statistics, 6, 461–464.
Schwert, W. G. (1990). Stock returns and real activity: A century of evidence. Journal of Finance, 45, 1237–1257.
Shin, D. W., & So, B. S. (2000). Gaussian tests for seasonal unit roots based on Cauchy estimation and recursive mean adjustments. Journal of Econometrics, 99, 107–137.
Singh, D. (2010). Causal relationship between macro-economic variables and stock market: A case study for India. Pakistan Journal of Social Sciences, 30, 263–274.
Srivastava, A. (2010). Relevance of macro economic factors for the Indian stock market. Decision, 37, 69–89.
Toda, H. Y., & Yamamoto, T. (1995). Statistical inference in vector autoregressions with possibly integrated processes. Journal of Econometrics, 66, 225–250.
Tripathy, N. (2011). Causal relationship between macro-economic indicators and stock market in India. Asian Journal of Finance & Accounting, 3, 208–226.
Tsouma, E. (2009). Stock returns and economic activity in mature and emerging markets. The Quarterly Review of Economics and Finance, 49, 668–685.
Zhu, H. -M., Li, R., & Li, S. (2014). Modelling dynamic dependence between crude oil prices and Asia-Pacific stock market returns. International Review of Economics and
Finance, 29, 208–223.
Zivot, E., & Andrews, K. (1992). Further evidence on the great crash, the oil price shock, and the unit root hypothesis. Journal of Business and Economic Statistics, 10,
251–270.