Steel consumption and economic growth: Evidence from India

ARTICLE IN PRESS

Resources Policy 31 (2006) 7–11 www.elsevier.com/locate/resourpol

Steel consumption and economic growth: Evidence from India Sajal Ghosh Management Development Institute (MDI), Gurgaon, India Received 6 June 2005; received in revised form 7 February 2006; accepted 28 March 2006

Abstract This paper examines cointegration and Granger causality between steel consumption and economic growth in India for the time span of 1951–1952 to 2003–2004 in a bivariate vector autoregression format. Augmented Dickey–Fuller tests reveal that both series, after logarithmic transformation, are non-stationary and individually integrated of order one. This study finds the absence of cointegration but the existence of unidirectional Granger causality running from economic growth to steel consumption. Thus, a growth in income is found to be responsible for a higher level of steel consumption. The impulse response function divulges that GDP growth does not seem to be responsive due to a shock in steel consumption growth. Finally, this study forecasts steel demand in India till 2011–2012 and highlights the preparedness of the Indian steel industry along with required policy prescriptions to meet this demand. r 2006 Elsevier Ltd. All rights reserved. JEL classification: C32; C53 Keywords: India; Steel consumption; Economic growth; Cointegration; Granger causality

Introduction Being a core sector, the Indian steel industry tracks the overall economic growth of the country. The sector was delicensed and de-regulated in 1991 and 1992, respectively. The steel industry has been removed from the list of industries reserved for the public sector. Price and distribution controls have been removed with a view to make the steel industry efficient and competitive. The role of the public sector, which accounted for more than 50% output few years back, has been reduced and today, private-sector production accounts for about two-third of the steel production in the country. Although India is the eighth largest producer of steel in the world, which accounts for 1.3% of India’s GDP and has a cumulative capital investment of over Rs. 1 trillion, per capita consumption of steel is only 30 kg as against 180 kg in China and an average of over 400 kg in the developed countries. This wide gap in relative steel consumption indicates that the potential ahead for India to raise its steel consumption is high. Experts opine that demand for steel in  Tel.: +91 124 2349831; fax: +91 124 2341189.

E-mail address: [email protected]. 0301-4207/$ - see front matter r 2006 Elsevier Ltd. All rights reserved. doi:10.1016/j.resourpol.2006.03.005

India is strong and likely to remain so, as the economy sustains higher growth path. The objective of this study was to examine cointegration and Granger causality between steel consumption and economic growth in India in bivariate vector autoregression (VAR) framework. The answer to these queries is expected to play a crucial role in policy formulation. In many cases, economic theory tells us that two or more variables should be cointegrated and a test for cointegration is the test of the theory. The presence of cointegration among the variables rules out the possibility of ‘‘spurious’’ correlation. Again, if, for example, there is unidirectional Granger causality running from steel consumption to economic growth, reducing domestic steel consumption could lead to a fall in national income. On the other hand, no causality in either direction (neutrality hypothesis) would indicate that steel consumption would not affect India’s economic growth and vice versa. Economic methodology and data description Engle and Granger (1987) showed that if the series X and Y (say) are individually I(1) (i.e. integrated of order one)

ARTICLE IN PRESS S. Ghosh / Resources Policy 31 (2006) 7–11

8

and cointegrated, then there would be a causal relationship at least in one direction. However, the direction of causality can be detected through the vector error correction model of long-run cointegrating vectors. Furthermore, Granger’s representation theorem demonstrates how to model a cointegrated I(1) series in a VAR format. VAR can be constructed either in terms of the level of the data or in terms of their first differences, i.e. I(0) variables, with the addition of an error correction term to capture the short-run dynamics. If the series are I(1) but not cointegrated, the causality test may give misleading results unless the data are transformed to induce stationarity. A three-stage procedure has been employed to test the existence of causality. The first step tests for the order of integration of the natural logarithm of the variables by using augmented Dickey and Fuller (ADF) (1981) statistics. Conditional on the outcome of the tests, the second stage involves investigating the cointegration relationship among the variables using the VAR approach of Johansen (1988, 1991) and Johansen and Juselius (1990). The third stage (or second if bivariate cointegration is rejected) involves constructing standard Granger-type causality tests, augmented where appropriate with a lagged error correction term.

Secondly, Granger causality tests with cointegrated variables may utilize the I(0) data with an error correction term i.e.,

ADF test for stationarity

H0 :

ð1Þ

where Xt is the underlying variable at time t, et is the error term and a0, b, k and gj are the parameters to be estimated. The lag terms are introduced in order to justify that errors are uncorrelated with lag terms. For the abovespecified model the hypothesis, which would be of our interest, is

i¼1

DY t ¼ a þ

Yt ¼ a þ

q X i¼1

n X

gj Y t
j þ ut ,

(2)

cj X t
j þ vt ,

(3)

j¼1

bi Y t
i þ

bi DY t
i þ

i¼1

r X

cj DX t
j þ d ECMt
1 þ vt .

j¼1

(5) Thirdly, if the data are I(1) but not cointegrated, valid Granger-type tests require transformation to make them I(0). So, in this case the equations become DX t ¼ a þ

m X

bi DX t
i þ

i¼1

DY t ¼ a þ

q X

n X

gj DY t
j þ ut ,

(6)

cj DX t
j þ vt .

(7)

j¼1

bi DY t
i þ

r X j¼1

i¼1

The optimum lag lengths m, n, q and r are determined on the basis of Akaike’s (AIC) and/or Schwarz Bayesian (SBC) and/or log-likelihood ratio (LR) test criterion. Now, for Eqs. (2) and (3), Y Granger causes (GC) X if g1 ¼ g2 ¼ . . . ¼ gn ¼ 0 is rejected; ¼ at least one gj a0; j ¼ 1; . . . ; n

H0 :

c1 ¼ c2 ¼ . . . ¼ cn ¼ 0 is rejected;

Against HA :

¼ at least one cj a0; j ¼ 1 . . . r

For Eqs. (4) and (5), DY, GC DX if H0 :

g1 ¼ g2 ¼ . . . ¼ gn ¼ 0 is rejected;

Against HA :

¼ at least one cj a0; j ¼ 1; . . . ; n or da0

c1 ¼ c2 ¼ . . . ¼ cn ¼ 0 is rejected; ¼ at least one gj a0; j ¼ 1; . . . ; r or da0.

For Eqs. (6) and (7), DY, GC DX if

If the series, X and Y are individually I(1) and cointegrated then Granger causality tests may use I(1) data because of the superconsistency properties of estimation:

i¼1

q X

Against HA :

Granger causality test

bi X t
i þ

gj DY t
j þ dECMt
1 þ ut ,

j¼1

(4)

H0 :

m X

n X

and DX GC DY if

ð1
kÞ ¼ 0.

Xt ¼ a þ

bi LX t
i þ

and X GC Y if

DX t ¼ a0 þ ð1
kÞbt
ð1
kÞX t
1 þ Sgj DX t
j þ et ,

H0 :

m X

Against HA :

ADF test is conducted with the following model:

ðj : 1; 2; . . . ; pÞ,

DX t ¼ a þ

r X

H0 :

g1 ¼ g2 ¼ . . . ¼ gn ¼ 0 is rejected;

Against HA :

and DX GC DY if H0 :

c1 ¼ c2 ¼ . . . ¼ cn ¼ 0 is rejected;

Against HA :

j¼1

where ut and vt are zero-mean, serially uncorrelated, random disturbances.

¼ at least one gj a0; j ¼ 1; . . . ; n

¼ at least one cj a0; j ¼ 1; . . . ; r:

The tests are conducted on annual data for India covering the period of 1951–1952 to 2003–2004. Data on gross domestic product (GDP) at 1993–1994 prices, as a proxy to economic growth, have been collected from

ARTICLE IN PRESS S. Ghosh / Resources Policy 31 (2006) 7–11 35000 30000

Apparent consumption of finished steel (Mild) in 000 tones

25000

GDP at Factor Cost by Industry Origin at 1993-94 Price in billion Rs.

Table 1 Augmented Dickey–Fuller (ADF) unit root tests

20000 15000 10000 5000 0 1940

1950

1960

1970

1980

9

1990

2000

2010

Year

Variables

Constant, no trend

Constant, trend

Levels Lgdp Lstl

1.8884
0.71420


0.94368
2.5039

First difference DLgdp DLstl


7.1388a
5.9914a

Note: 95% critical values for ADF statistic (without trend) is
2.9287; 95% critical values for ADF statistic (with trend) is
3.5136. a Represents rejection of null hypothesis at 5% level of significance.

Fig. 1. Graphical representation of data.

Economic Survey 2005, published by Ministry of Finance and Company Affairs, Government of India (GOI). Annual consumption of finished steel (  103 tonnes) for the same time span is taken from the Steel Scenario Statistical Yearbook, published by Spark Steel & Economy Research Centre Private Limited, India. The graphical representation of data is shown in Fig. 1. Lgdp and Lstl represent GDP and steel consumption, respectively, after their logarithmic transformation.

Table 2 Johansen–Juselius likelihood cointegration tests Null

Alternative

Statistic (Lgdp and Lstl)

Critical value (95%)

Maximal eigenvalue test r¼0 r¼1 rp1 r¼2

40.1453 14.1658

15.8700 9.1600

Trace test r ¼ 00 rp1

54.3110 14.1658

20.1800 9.1600

rX1 r¼2

Empirical results In the first stage, the order of integration of the variables is investigated. Table 1 presents the results of unit root tests on the natural logarithms of the levels and the first differences of the variables. On the basis of ADF statistics, the null hypothesis of a unit root cannot be rejected at 5% level of significance. Stationarity is obtained by running the similar test on the first difference of the variables, indicating both the series are I(1) in nature. In the second stage, the Johansen maximum likelihood procedure is used to detect cointegration. This provides a unified framework for estimation and testing of cointegrating relations in the context of a VAR error correction model. The cointegration rank, r, of the time series is tested using two test statistics. Denoting the number of cointegrating vectors by r0, the maximum eigenvalue (lmax) test is calculated under the null hypothesis that r0 ¼ r, against the alternative of r04r. The trace test is calculated under the null hypothesis that r0pr, against r04r. Referring to Table 2, both maximal eigenvalue test and trace test reveal that the null hypothesis r ¼ 0 and rp1 between Lgdp and Lstl cannot be accepted against the alternative rX1 and r ¼ 2 at 5% level of significance. These imply the absence of cointegration between Lgdp and Lstl. Consequently, the bivariate system DLgdp and DLstl, where ‘D’ is the first difference operator and hence defines the growth of the respective variable, can be modeled as an unrestricted VAR. On the basis of SBC and adjusted LR test criteria, the optimal lag order of the VAR is chosen as 1. The absence

Table 3 Granger causality tests Null hypothesis

w2

DOFa

P-valueb

Non-causality DLstl XDLgdp Non-causality DLgdpXDLstl

0.4130E
3 6.6376

1 1

0.984 0.01

a

Degrees of freedom. Acceptance probability.

b

of residual serial correlation of the individual equations has also confirmed the correct order of VAR selection. Finally, Table 3 represents the results of the Granger causality tests. The null hypothesis of non-causality from DLgdp to DLstl, which is asymptotically distributed as a w2 variate with one degree of freedom, cannot be accepted at 5% and 10% levels of significance. While testing the noncausality from DLstl to DLgdp, the null hypothesis cannot be rejected at 5% and 10% level of significance. These imply the presence of unidirectional causality running from GDP growth to steel consumption without any feedback effect. Figs. 2 and 3 represent the impulse response paths due to the various shocks to the system. The response of steel consumption to one standard error (SE) shock in the equation of GDP growth is presented in Fig. 2. The steel consumption responds positively at the initial year after the shock but then responds negatively, eventually returning to its pre-shock level after a period of 4 years. As shown in

ARTICLE IN PRESS S. Ghosh / Resources Policy 31 (2006) 7–11

2008-09

2009-10

2010-11

2011-12

34.9

40.95

44.09

39.41

42.51

37.89

2007-08

33.52

36.42

2006-07

40

2005-06

50

2004-05

Fig. 3, GDP growth does not seem to be responsive to one SE shock in the equation for the growth in steel consumption. Fig. 4 represents in-sample forecasting for the demand of finished steel against the actual numbers for the period of 2000–2001 to 2003–2004. From the graphs it is clear that the model works pretty well. Finally, Fig. 5 projects the demand of finished steel in India (in Mt) till 2011–2012.

Demand in Mt

10

30 20 10 0

0.04

Year DLgdp

0.03

Fig. 5. Projected demand of finished steel till 2011–2012.

DLstl

0.02 0.01

Conclusion

0 0

5

10

15

20

-0.01 -0.02

Horizon Fig. 2. Generalized impulse responses to one SE shock in the equation for DLgdp.

0.12 DLgdp

0.1

DLstl

0.08 0.06 0.04 0.02 0 0

5

10

15

20

25

30

-0.02

Horizon Fig. 3. Generalized impulse responses to one SE shock in the equation for DLstl.

32000

(ooo Tonnes)

31000

Steel Consumption (Actual) Steel Consumption (Predicted)

30000 29000 28000 27000 26000 2000

2001

2002 Year

Fig. 4. In-sample forecasts.

2003

2004

The study finds the absence of a long-run equilibrium relationship between steel consumption and economic growth in India but establishes the presence of unidirectional Granger causality running from GDP growth to steel consumption. Impulse response function also reveals that GDP growth does not seem to be responsive to a shock in steel consumption growth. Existence of Granger causality from economic growth to steel consumption without any feedback effect reveals that a growth in income is responsible for fostering steel consumption. This is quite obvious as with economic growth, the demands for consumer durables, automobiles, construction and infrastructure have been increased, where steel is used as one of the basic infrastructural inputs. According to experts, world’s steel demand should continue to grow by around 5%, driven by the continuing strong demand in China, India, Latin America and other Asian countries. Ballooning global demand coupled with increasing input costs for steel production, due to limited availability of iron ore and coking coal, should push forward the steel price in northeast direction. The Indian steel sector enjoys advantages of domestic availability of raw materials and cheap labor. The Indian steel-makers already have one foot in the global market and are therefore familiar with international conditions. Most of the major steel manufacturers in India have already envisaged huge expansion plans in the short and medium terms so that, in addition to meeting the strong domestic demand, they can compete in the international market to supply to regions such as Middle East, Southeast Asia and China. The companies that have already announced their capacity addition program include Steel Authority of India Limited (8 Mt), Tata Iron & Steel Company (9.5 Mt), ESSAR (1.5 Mt), Rashtriya Ispat Nigam Limited (7 Mt), Jindal Iron And Steel Company (3 Mt), ISPAT (1.7 Mt) and Nilachal Ispat (0.5 Mt). However, in order to make the domestic steel industry a leader in the global arena, providing adequate infrastructure

ARTICLE IN PRESS S. Ghosh / Resources Policy 31 (2006) 7–11

such as power, roads, ports and rail transport is an essential prerequisite. The other major deciding factor is the availability of raw materials. For the integrated steel plant, iron ore and coking coal supplies need to be tied up. The government has to look at the mineral policy and coal policy in the light of current reality and come out with suitable policies for mining leases, development of new mines, environmental clearance, trading and export of iron ore and removal of trade barriers imposed by developed countries like the USA and European Union. Indian companies, going global, should also emphasize on energy conservation measures and acquire coalfields abroad for captive purposes.

11

References Dickey, D.A., Fuller, W.A., 1981. Likelihood ratio statistics for autoregressive time series with a unit root. Econometrica 49 (4), 1057–1072. Engle, R.F., Granger, C.W.J., 1987. Co-integration and error-correction: representation, estimation and testing. Econometrica 55 (2), 251–276. Johansen, S., 1988. Statistical analysis of cointegration vectors. J. Econom. Dyn. Control 12 (2/3), 231–254. Johansen, S., 1991. Estimation and hypothesis testing of cointegrating vectors in Gaussian vector autoregressive models. Econometrics 59 (6), 1551–1580. Johansen, S., Juselius, K., 1990. Maximum likelihood estimation and inference on cointegration with application to money demand. Oxford Bulletin of Economics and Statistics 52 (2), 169–210.