Heterogeneous Causality Between Telecommunications and Economic Growth: An Application of Hurlin

Abstract

Keeping in mind the importance of telecommunications sector for India especially post the reform of 1991, the study seeks to study the causal relationship between telecommunication and economic growth for selected panel of 17 Indian states by implicitly taking into account panel heterogeneity. Using Hurlin–Venet causal mechanism, the study investigates homogeneous causality (HC) as against heterogeneous causality (HEC). Homogeneous non-causality and HC hypothesis both of which assume homogeneity are rejected in both the directions thereby implying that Indian panel is made up of heterogeneous cross sections. After this HEC tests are conducted namely heterogeneous non-causality (HENC) and HEC for each and every cross section. The results from HENC and HEC test show that six states show up unidirectional causality from economic growth to telecommunication, four states show unidirectional causality from telecommunications to economic growth, four states show up bidirectional relationship and three states show up no causality. The results are robust to different lags for both our dependent as well as independent variables. This implies that state level differences are very much relevant for India and thus policies suitable for each state would be guided majorly by direction of relationship that is true in a particular state economy.

JEL Classifications: C23, L96, O40

Keywords

Telecommunications economic growth India Hurlin–Venet

1. Introduction

From the age of letters, telegrams and typewriters, the world today has entered into the age of smartphones, computers, internet and hi-tech technology that has revolutionised the 21st century. The smartphones have established themselves as the ‘Swiss army knife’ of the tele world which acts as a hub for myriad activities. The changes in the technology have been so rampant that what seems to be new today loses its novelty in just a few months. This is spurting innovations in the field of ICT. The pace of transformation has been staggering. From 1G technology that dates back to 1970s, 2G technologies found its origin in 1990s. Thus the pace of transition was almost 20 years. But the subsequent upcoming generations have witnessed a quick period of progression. The world moved from 2G to 3G in just 14 years while the shift from 3G to 4G took place in just 12 years. The likely period for 5G to unfold is 2020. So what emerges out from this technology evolution is that subsequent transformation is taking lesser time to unveil. This marks off innovations that are characterising the Information and Communication Technology (ICT) industry world over. World Bank (2016) has identified three mechanisms that act as mediator between digital technology and economic growth. These are: inclusion, efficiency and innovation (Fig. 1).

To talk about the direction of causation between these variables is important to know beforehand so as to guide the policy planners in aiding the government at different levels. There are number of causality hypotheses pertinent to this relationship. The 1st one is a Supply leading hypothesis which postulates that availability of telecommunication infrastructure is a pre-requisite for economic growth. A proper infrastructure of telecommunication will engender economic growth by disseminating market information timely among the prospective buyers and also by improving the provision of public services like health, education, finance and most importantly governance. An uninterrupted telecommunications infrastructure will help in transmitting information swiftly thereby benefitting all. The 2nd one is Demand following hypothesis which asserts that the direction of causality is from economic growth to telecommunication infrastructure. According to this hypothesis, more economic growth in the economy will command more sophisticated and advanced telecommunications operation that will support the business activities which are entirely based upon the unceasing internet connectivity together with greater speeds. As an economy grows, it commands advanced tele infrastructure for which it can pay. Thus, increased ability to afford for trailblazing telecommunications technology is what that prompts economies especially the developed ones to go for innovative tele structures. Not only this, a growing economy is concomitant with a flourishing service sector whose need for telecommunication infrastructure tend to escalate. Thus, economic growth would entail telecommunication investment in the economy. The 3rd hypothesis contends that there is mutual Feedback relationship (Feedback hypothesis) between the two variables. This means that advancement in economic growth will make it mandatory for the economy to invest in telecommunications sector. This will cause the economy to further improve its growth position by facilitating commerce effectively in the economy. The last hypothesis is Neutrality hypothesis that asserts that there is literally no cause and effect relationship between the two variables. A brief summary of the studies that conform to each of these hypotheses have been presented in the Table 1.

Table 1.

Telecommunications and Growth Causality (International as well as Indian Evidence).¹

Hypothesis	International Evidence	Indian Evidence
Supply-led hypothesis (Telecom↓Growth)	Roller and Waverman (2019) for OECD Dutta (2019) for 30 countries	Vodafone Group Plc. (2009) Ghosh and Prasad (2012)
	Sridhar and Sridhar (2009) 63 for developing countries Kumar et al. (2015) for Small Pacific Island States Pradhan et al. (2018) for G-20 countries
Demand-led hypothesis (Growth↓Telecom)	Beil et al. (2005) for US economy Pradhan et al. (2013) for 34 OECD countries	Kateja and Jha (2008) Barman et al. (2018)
Feedback hypothesis (Growth↔Telecom)	Cronin et al. (1991) for US economy Madden and Savage (1998) for CEE countries Yoo and Kwak (2004) for Korea Pradhan et al. (2014) for G-20 David (2019) for African countries
Neutrality hypothesis	Brynjolfsson and Yang (1996)

Source: Authors’ compilation.

Note: CEE: Central and Eastern Europe; OECD: Organization for Economic Co-operation and Development.

While all the below presented studies exclusively talk about casual patterns at the overall level, none focus on subnational causal behaviour. In this vein we have selected works by Shiu and Lam (2008) that provide evidence for heterogeneous causality (HEC) behaviour for Chinese economy as well as for set of 105 countries of the world. For China, they found out that overall causality ran from gross domestic product (GDP) to teledensity while for high-income region the pattern was same but for other regions (Central and Western regions which are low income regions) there was no causality at all or it ran from GDP to teledensity. Similarly for 105 countries which were segregated on the basis of income, for high income countries the causality was bi-directional while for low- income countries it was from growth to teledensity.

The above review of literature highlights numerous gaps: First, despite extensive research on causality between telecom and economic growth, the research has still been inconclusive. Especially taking the Indian case, very few papers come to our knowledge that talk about causality. Despite dealing with the time series analysis the results are not unanimous. Second, the base of Indian studies is time series with a complete absence of any paper that has taken into account subnational causality aspect. In addition to that, dealing with panel studies offers added advantage in the form of verifying homogeneous causality (HC) against Heterogeneous Causality (HEC) which is the major novelty of the paper. The case for HEC is more relevant for panel studies because of higher degree of heterogeneity that is suspected among the cross sections. This HEC hypothesis is very much validated in other domains particularly for energy–economic growth relationship that too for the two most important economies of the world, the USA (Mahalingam and Orman 2018) and (Akkemik et al. 2012; Zhang and Xu 2012) China, and also for India. Having verified HEC between other economic variables particularly with respect to causality, the need for study is immense because the primary objective of the research is to study subnational causal relations. The case for differences in subnational causal relation is already validated by Shiu and Lam (2008) and the same can be expected in Indian case.

The research is undertaken with the sole objective of studying the causal behaviour between telecommunications and economic growth for 17 Indian states for the period 1991–2017 by addressing the issue of heterogeneity. Given the background of major surge in telecommunication facilities in India especially post reform of 1991, the results from the paper can have far reaching implications. As far as the contribution that the study makes to the overall subject, these can be found along three major lines: To the researcher’s knowledge, this study is the 1st one to discuss about causal process between telecommunications and economic growth for the panel of 17 Indian states so as to study the profound regional differences that are likely to exist within India. Secondly, this study makes a sharp departure from the existing causality studies by distinguishing between homogeneous and heterogeneous causality. Thirdly, the research is driven by the set of more recent econometric tools and methodologies comprising of Cross-sectionally Augmented Im–Pesaran–Shin (CIPS) unit root test and Westerlund cointegration test (both of which are 2nd generation testing tools which are more valid in the presence of cross section dependence (Latif et al. 2018) and Hurlin–Venet causal mechanism. Such an application of advanced econometric techniques imparts robustness and rigour to the analysis carried out.

Despite being a regional study catering only to the Indian states, the study offers its conclusions to the wider audience. Also it attempts to engage itself with the methodological debate centring on panel studies which are subjected to frequent changes. The application of Hurlin–Venet process and the subsequent validation of HEC in Indian case provide a strong case for verifying such patterns across other economies of the world. This can initiate a whole new set of literature discussing along the lines of heterogeneous causal relation instead of homogeneous relation which has been the major assumption of Granger causality process.

This study is divided into five sections including the present introductory one. The 2nd section provides the information on the database. The 3rd section details the methodology used in this study. The 4th section offers the empirical results and discussion of the analysis. Conclusions and relevant policy implications are given in the last section, that is, 5th section.

2. Data

To perform the causal analysis between the variables, we require two variables of telecommunications and economic growth. For telecommunications, we have used teledensity (number of telephones per 100 people) while for the variable of economic growth we have relied on three estimates: per capita net state domestic product (Y); gross domestic product (G) and net state domestic product (N). The data are collected for a set of 17 Indian states from 1991 to 2017. These states are Andhra Pradesh, Assam, Bihar, Gujarat, Assam, Haryana, Himachal Pradesh, Jammu and Kashmir, Karnataka, Kerala, Madhya Pradesh, Maharashtra, Odisha, Punjab, Rajasthan, Tamil Nadu, Uttar Pradesh and West Bengal. Few Indian states witnessed bifurcation over this time which includes Uttar Pradesh, Bihar and Madhya Pradesh in 2000 and Andhra Pradesh in 2014. In order to maintain coherence in the dataset, the states originally (before bifurcation) have been retained in the analysis.

3. Methodology

The estimation strategy adopted in this study is given in the Figure 2. To test the cross section dependence we employed Lagrange Multiplier (LM) test (Breusch and Pagan 1980; Pesaran 2007). The notation of LM test is given as:

L M = T Σ_{i = 0}^{N - 1} Σ_{j = i + 1}^{N} ρ_{i j}^{2}

Where $ρ_{i j}$ is the sample estimate of the pair-wise correlation of the residuals.

ρ_{i j} = \frac{Σ_{t = 1}^{T} u_{i t} u_{j t}}{{(Σ_{t = 1}^{T} u^{2})}^{1 / 2} {(Σ_{t = 1}^{T} u^{j t}_{2})}^{1 / 2}}

While the Pesaran test of cross section dependence is given as:

C D = \sqrt{\frac{2 T}{N (N - 1)}} (Σ_{i = 1}^{N - 1} Σ_{i = j + 1}^{N} ρ_{i j})

After checking the degree of dependence (or independence) as the case is, we conduct preliminary testing of stationarity and cointegration properties of our database. For this, we employed both CIPS (Im et al. 2003) and Cross-sectional Augmented Dickey–Fuller test statistic developed by Pesaran (2007), which can be expressed as follows:

CIPS (N, T) = 1 Σ t_{i} (N, T)

where ti (N, T) is nothing but the t value of our variable of interest.

Post this, we carry out the cointegration using Westerlund’s (2007) testing procedure. Under this, we have four different statistics: Panel statistics (Pt and Pa) and group statistics (Ga and Gt). The tests are implemented by inferring whether the error-correction terms are zero as a test for the null of no cointegration. The error-correction-based panel cointegration tests allow for cross-sectional dependence and heterogeneity both in the long-run cointegrating relationship and in the short-run dynamics.

The process of assessing the causality is not simple especially in panel data and choosing the most appropriate technique is the major corner stone for research. This is because it addresses the empirical and theoretical issues in a more explicit way. The conventional Granger causality technique is ridden with most important econometric issue of homogeneity across cross sections which in the present times is hard to meet. To address this issue, Erdil and Yetkiner (2005) grouped the panel causality into two broad categories: one in which the auto regressive coefficients and slope coefficients are taken as variables in the panel Vector Autoregressive (VAR) model and on the other side Hurlin and Venet (2001) suggested a method wherein these coefficients are assumed to be constant. The main criterion to decide between the two is the time length of the study. If the time span is long, then the 1st method is adopted otherwise we go with the 2nd approach. Given the time frame of this study, 1991–2017 (27 years), we go with Hurlin–Venet approach. In this technique, we make use of fixed effects that are time invariant and the lags of dependent variable are included in the regression model. Thus, this study employs heterogeneous fixed effects dynamic panel data estimation procedure. The regression equations for estimation are given below:

I n T_{i t} = Σ_{l = 1}^{n} γ^{l} T_{i}_{, t - j} + Σ_{l = 1}^{n} β^{l}_{i \neq} I n Y_{i, t - l} + u_{i, t}, u_{i, t} = α_{i} + ε_{i, t}

(1)

I n Y_{i t} = Σ_{l = 1}^{n} γ^{l} Y_{i}_{, t - j} + Σ_{l = 1}^{n} β^{l}_{i \neq} I n T_{i, t - l} + u_{i, t}, u_{i, t} = α_{i} + ε_{i, t}

(2)

Here i refers to the individual cross sections (which in our case is Indian states), t denotes the time and l is the number of lags, α, β and γ are the parameters that are to be estimated. Here the slope coefficients β^′s are assumed to be constant over a period of time but vary across cross sections. Equation 1 tests for causality from economic growth to telecommunications while the Equation 2 tests for the other causality direction.

Hurlin and Venet (2001) gave four types of causal relationship for panel analysis: HC, homogeneous non-causality (HNC), HEC and heterogeneous non-causality (HENC). The test statistics are elaborated in Table 2.

Table 2.

Hurlin–Venet Causality Hypotheses.

Name	Hypothesis	F-statistics
Homogeneous non causality (HNC)	$\begin{array}{l} H_{0} : β_{i}^{l} = 0 for all i ε [1, N] and \\ For all 1 ε [1, n] (4) \\ H_{1} = β_{i}^{l} \neq 0 \end{array}$	$F_{HNC} = \frac{(R S S 2 - R S S 1) / N l}{R S S 1 / [N T - N (1 + l) - l]}$
Homogeneous causality (HC)	$\begin{array}{l} H_{0} : β_{i}^{l} = β_{}^{l} for all i ε [1, N] and \\ for all 1 ε [1, n] (7) \\ H_{1} = β_{i}^{l} \neq β_{k}^{l} \end{array}$	$F_{HC} = \frac{(R S S 3 - R S S 1) / (N - 1) n}{R S S 1 / [N T - N (1 + 2 n) + n]}$
Heterogeneous non causality (HENC).	$\begin{array}{l} H_{0} : β_{i}^{l} = 0 for all i ε [1, N] and \\ for all l ε [1, n] (10) \\ H_{1} = β_{i}^{l} \neq 0 for all i ε [1, n] and \\ for all l ε [1, n] \end{array}$	$\begin{array}{l} F_{H E N C}^{i} = \frac{(R S S 2, i - R S S 1) / n}{R S S 1 / [N T - N (1 + 2 n) + n]} \\ F_{HENC} = \frac{(R S S 4 - R S S 1) / N n c n}{R S S 1 / [N T - N (1 + n) - N c n]} \end{array}$

Source: Ça (2013)

4. Results and Discussion

To begin with, we performed cross section dependence test. The results from cross section dependence test are very much important because they aid in deciding the techniques of integration and cointegration that will be adopted at the subsequent stages. The results from cross section dependence test are presented in the Table 3. These results verify that there is a strong degree of dependence that exists across cross section of Indian states.

Table 3.

Cross Section Dependence Test Statistics.

Cross Section Dependence Test↓ Models↓	Breusch–Pagan Lagrange Multiplier	Pesaran Cross-sectional Dependence	Average Absolute Value of the Off-diagonal Elements
PCNSDP–Telecommunication	1248.241***	15.823***	0.513
GDP–Telecommunication	1274.637***	21.011***	0.518
NDP–Telecommunication	1253.198***	19.667***	0.513
Telecommunication–PCNSDP	1316.665***	14.229***	0.535
Telecommunication–GDP	1339.299***	20.254***	0.535
Telecommunication–NDP	1301.114***	18.830***	0.527

Source: The authors.

Note: GDP: Gross domestic product; NDP: Net domestic product; PCNSDP: Per capita net state domestic product.

*** means significance level of 1 percent.

Having verified dependence, we now employ 2nd generation unit root testing and cointegration testing procedures to our data generating process. The results of stationarity tests are presented in the Table 4.

Table 4.

Unit Root Results.

Variables	Levels		First Differences
Variables	Individual Intercept	Individual Intercept + Trend	Individual Intercept	Individual Intercept + Trend
PCNSDP (Y)
IPS	12.808	1.415	–14.625***	–14.665***
CIPS	–2.435	–3.003	–5.292*	–5.581
CADF	–2.127*	–2.756**	–3.677***	–3.782***
Telecommunications (T)
IPS	1.799	–4.117***	–9.640***	–6.905***
CIPS	–1.580	–2.267	–3.397	–3.349***
CADF	–1.596	–2.741**	–3.091***	–3.151***
GDP
IPS	12.337	1.063	–16.735***	–14.849***
CIPS	–2.434	–3.218*	–5.378*	–5.716*
CADF	–2.164**	–2.971***	–3.783***	–3.912***
‘NDP
IPS	10.713	0.577	–17.600***	–14.314***
CIPS	–2.524	–3.126	–5.233*	–5.593*
CADF	–2.198	–2.976***	–3.499***	–3.607***

Source: The authors.

Notes: CADF: Cross-sectional Augmented Dickey–Fuller; CIPS: Cross-sectionally Augmented Im–Pesaran–Shin; GDP: Gross domestic product; IPS: Im–Pesaran–Shin; NDP: Net domestic product; PCNSDP: Per capita net state domestic product.

* means 10% significance level.** imply significance level of 5%.*** implies significance level of 1%.

The results from unit root tests clearly state that all our variables are stationary at first difference. Thus the order of integration of the variables is I(1). This non-stationary property of the data is revealed when the estimations are conducted using different models involving intercept and trend, following which long run equilibrium relationship between the two variables is investigated using cointegration analysis.

Given that the variables are non-stationary at level, and stationary at first difference, we try establishing a long term relation between the selected variables. The results from Westerlund’s cointegration test are given in the Table 5. From the cointegration test, we can say that there is no long run relation between any variable in any of the model. The null hypothesis which states that there is no cointegration is not rejected in any of the model.

Table 5.

Westerlund Cointegration Test.

Statistics	Values	Z-value	P-value	Robust P-value
Model 1: PCNSDP–Telecommunication
Gt	–1.375	1.848	.968	.630
Ga	–5.708	1.087	.861	.240
Pt	–6.106	–0.138	.445	.380
Pa	–6.045	–1.686	.046	.050
Model 2: GDP–Telecommunications
Gt	–1.614	0.752	.774	.460
Ga	–5.282	1.409	.921	.300
Pt	–6.275	–0.308	.379	.300
Pa	–4.872	–0.596	.276	.130
Model 3: NDP–Telecommunications
Gt	–1.616	0.743	.771	.440
Ga	–5.764	1.044	.852	.210
Pt	–6.536	–0.571	.284	.250
Pa	–5.380	–1.068	.143	.110
Model 4: Telecommunication–PCNSDP
Gt	–1.238	2.478	.993	.790
Ga	–5.233	1.446	.926	.500
Pt	–2.945	3.041	.999	.880
Pa	–2.738	1.386	.917	.620
Model 5: Telecommunication–GDP
Gt	–0.776	4.595	1.000	.970
Ga	–3.146	3.027	.999	.930
Pt	–1.714	4.279	1.000	.940
Pa	–1.464	2.569	.995	.850
Model 6: Telecommunication–NDP
Gt	–0.761	4.664	1.000	.980
Ga	–3.216	2.974	.999	.940
Pt	–1.648	4.346	1.000	.980
Pa	–1.434	2.597	.995	.950

Source: The authors.

Note: GDP: Gross domestic product; NDP: Net domestic product; PCNSDP: Per capita net state domestic product.

After testing the fundamentals of panel analysis, we now move onto assessing the causal relationship between telecommunications and economic growth. There are two causal relationships to be tested: one from telecommunications to economic growth and the 2nd one is from economic growth to telecommunications. Unlike the selection of lags which is guided by the lag length selection criteria, we carry out our analysis using three lags for both the causal relationships. This is done to check the sensitivity of the causal results. Also the causal regressions are carried out using fixed effects estimations. The 1st hypothesis that is tested is the hypothesis of HNC. The test statistics of this hypothesis is presented in the Table 6.

Table 6.

Test Results for Homogeneous Non-causality (HNC Hypothesis).

Causality from Economic Growth to Telecommunications
Lags	Y→T	G→T	N→T
1	3.07*	7.05***	6.11**
2	3.87**	8.30***	7.23***
3	4.90***	10.23***	8.79***
Causality from Telecommunications to Economic Growth
Lags	T→Y	T→G	T→N
1	31.77***	18.66***	18.62***
2	17.97***	13.69***	12.51***
3	8.13***	6.94***	6.93***

Source: The authors.

Notes: G: Gross domestic product; N: Net domestic product; T: Teledensity; Y: Per capita net state domestic product.

***, ** and * imply significance level of 1%, 5% and 10%, respectively.

For the causality running from economic growth to telecommunications, we see sufficient evidence against no causality hypothesis. This is true for all definitions of economic growth as well as for all three lags. Most of the test statistics are significant at 1% level. Same is true for causality running from telecommunications to economic growth where also we see that all the test statistics are highly significant (level of significance being 1%). Thus in all we can say that there does exist causality between both the variables in both the directions.

Having established the presence of causality, the next step is to check whether that causal behaviour is homogeneous or heterogeneous. First of all we try checking the HC relationship. The results from HC process are tabulated in the Table 7.

It is clear evidently, that in both the cases, the hypothesis stating that the causal relationship is homogeneous in all cross sections is rejected given the higher value of F-statistics over the critical F-values. For the causality direction from economic growth to telecommunications, we see that the null hypothesis is rejected either at 5% or 10%, but for the causal direction in the other way, all the statistics are significant at 1% level. Thus convincingly, we can reject our null hypothesis in each of these cases.

Table 7.

Test Results for Homogeneous Causality (HC Hypothesis).

Causality from Economic Growth to Telecommunications
Lags	Y→T	G→T	N→T
2	3.63*	4.17**	3.08*
3	3.43**	4.07**	2.74*
Causality from Telecommunications to Economic Growth
Lags	T→Y	T→G	T→N
2	12.53***	13.67***	13.03***
3	7.98***	9.06***	9.22***

Source: The authors.

Notes: G: Gross domestic product; N: Net domestic product; T: Teledensity; Y: Per capita net state domestic product.

***, ** and * imply significance level of 1%, 5% and 10%, respectively.

Because the hypothesis of HC is rejected vehemently in both the directions, we now seek to work out the heterogeneous causal relationships. The heterogeneous causal relationship is tested in two steps: 1st, the causal relationship is studied for each and every cross section of the panel and 2nd, the cross sections that depict causality in the 1st step are then jointly taken up as a group to test heterogeneous non-causal relationship.

In the 1st step we run our regression model for all 17 cross sections in both the ways subject to the restriction i of β^l = 0. Thus for the causality running from economic growth to telecommunication, we will run the regression equation 51 times (17 × 3) and similarly for the causality running from telecommunications to economic growth again we will run the model for 51 times. In all, 102 regressions equations will be calculated. The results from HEC running from growth to telecommunications are presented in the Tables 8 and 9

As far as causality from growth to telecommunications is concerned we see that 10 Indian states show up the causal relation in that direction. These Indian states are Himachal Pradesh, Madhya Pradesh, West Bengal, Kerala, Andhra Pradesh, Assam, Haryana, Maharashtra, Odisha and Punjab and for the causality direction from telecommunications to economic growth we see that eight out of 17 Indian states validate the causal behaviour. These states comprise of Bihar, Gujarat, Rajasthan, Tamil Nadu, Himachal Pradesh, Madhya Pradesh, West Bengal and Kerala.

Table 8.

Results from Heterogeneous Causality Between Growth and Telecommunication.

States↓ Lags→	Y→T			G→T			N→T
States↓ Lags→	1	2	3	1	2	3	1	2	3
Andhra Pradesh	0.71	0.67	2.57*	0.10	3.87**	4.93**	0.59	5.49**	4.70**
Assam	7.73**	2.53	1.94	5.89**	2.42	2.28	6.50**	2.89*	2.21
Bihar	4.73**	2.32	1.10	1.88	0.25	0.20	1.92	0.41	0.14
Gujarat	0.11	0.08	0.25	1.00	1.03	1.58	0.55	0.68	1.19
Haryana	7.80**	2.55	4.35**	5.68**	2.08	3.27**	5.86**	2.11	3.37**
Himachal Pradesh	3.51*	2.23	1.57	2.30	1.90	1.66	3.07*	2.19	1.54
Jammu and Kashmir	0.00	1.52	1.49	0.54	2.11	1.69	0.96	2.43	1.81
Karnataka	0.26	1.17	1.89	0.46	1.36	2.22	0.41	1.37	2.22
Kerala	9.59***	2.98*	5.61***	8.15***	2.86*	5.51***	8.07***	2.91*	7.73***
Madhya Pradesh	9.68***	0.50	0.46	3.45*	0.40	0.93	4.10*	0.42	0.90
Maharashtra	4.21*	4.84**	3.94**	2.69*	5.15**	4.91**	2.44	5.03**	5.28***
Odisha	11.51***	4.78**	5.39***	9.83***	4.20**	3.80**	8.05***	4.17**	4.34**
Punjab	5.59**	1.76	2.10	5.78**	2.02	2.37	5.44**	1.92	2.31
Rajasthan	0.08	0.04	0.27	0.09	0.37	0.94	0.06	0.26	0.86
Tamil Nadu	0.59	0.85	1.03	0.49	1.43	1.58	0.49	1.26	1.41
Uttar Pradesh	4.60**	2.14	3.61	0.13	0.48	2.25	0.59	0.15	2.42
West Bengal	1.91	3.70**	3.79**	1.31	4.74**	3.35**	1.27	4.39**	3.48**

Source: The authors.

Note: ***, ** and * imply significance level of 1%, 5% and 10%, respectively.

Table 9.

Results from Heterogeneous Causality Between Telecommunications and Growth.

States↓ Lags→	T→Y			T→G			T→N
States↓ Lags→	1	2	3	1	2	3	1	2	3
Andhra Pradesh	0.66	1.63	1.07	0.11	0.53	1.72	0.20	0.55	1.37
Assam	0.02	0.08	0.11	0.69	0.29	0.91	0.53	0.22	0.66
Bihar	19.48***	16.98***	7.34***	13.27***	7.95***	3.76**	16.37***	9.85***	4.05**
Gujarat	11.21***	10.51***	3.59**	5.21**	10.14***	4.17**	7.74**	10.44***	4.15**
Haryana	0.07	0.56	0.91	0.01	0.91	0.73	0.00	0.88	0.77
Himachal Pradesh	3.23*	0.55	0.48	5.44**	0.38	0.56	1.93	0.31	0.42
Jammu and Kashmir	8.40***	1.75	1.13	1.30	0.33	0.82	1.52	0.33	0.61
Karnataka	1.91	1.87	1.26	1.24	1.34	1.06	1.38	1.42	1.05
Kerala	1.46	2.53	1.42	0.83	4.08**	3.63**	0.96	5.96***	5.20***
Madhya Pradesh	7.76**	2.43	1.50	7.37**	3.94**	2.39	6.79**	3.14*	2.03
Maharashtra	0.20	0.45	0.88	0.03	0.73	1.46	0.01	0.39	1.36
Odisha	0.13	0.36	0.53	0.05	0.42	0.72	0.00	0.61	0.77
Punjab	1.90	1.55	1.28	1.24	1.01	0.84	1.73	1.22	1.16
Rajasthan	17.53***	5.56**	4.99**	8.25***	4.34**	3.67**	9.60***	3.87**	3.52**
Tamil Nadu	4.50**	1.30	1.30	3.21**	0.75	1.86	3.38*	0.88	1.61
Uttar Pradesh	4.62**	1.95	1.27	1.04	0.48	0.31	1.62	0.74	0.72
West Bengal	0.17	1.64	3.80**	0.06	2.65*	4.09**	0.04	2.16	3.62**

Source: The authors.

Note: ***, ** and * imply significance level of 1%, 5% and 10%, respectively.

After carrying out the 1st step of HEC, we now move onto carrying the 2nd step. For this we take up 10 Indian states as a single group to test HENC and the results are presented in the lower panel of Table 10. As is clear, in all the six cases, that is, for all definition of economic growth and at all lag values, the F-statistic is greater than the critical limit of F thereby rejecting the null hypothesis for all these subgroups of Indian state.

For other set of eight states that show causality from telecommunications to economic growth, when we bundle them into one group and test the hypothesis, we again find rejection of our null hypothesis. Thus, these eight Indian states show up causality in this direction.

Table 10.

Tests of Heterogeneous Non-causality.

Causality from Telecommunications to Economic Growth
Lags	T→Y	T→G	T→N
1	30.15***	23.64***	24.13***
2	19.10***	19.71***	16.94***
3	8.61***	9.87***	8.42***
Causality from Economic Growth to Telecommunications
Lags	Y→T	G→T	N→T
1	17.00***	17.32***	16.26***
2	8.36***	10.85***	9.86***
3	8.26***	11.29***	10.06***

Source: The authors.

Note: *** implies significance level of 1%.

5. Grouping of Indian States

Based on the evidence of causality direction in the 17 Indian states, these can be grouped under four heads:

Unidirectional causality from telecommunication to economic growth: These include states of Bihar, Gujarat, Rajasthan and Tamil Nadu. Here we see that the states are in fact scattered rather than being located in a particular geographic location. While Gujarat and Rajasthan are located on the West, Tamil Nadu in the South and Bihar in the North.

Unidirectional causality from economic growth to telecommunication: This group includes Andhra Pradesh, Assam, Haryana, Maharashtra, Odisha and Punjab. Here again a scatter plot is validated wherein some are coastal states (Maharashtra, Odisha and Andhra Pradesh) and others are landlocked states (Punjab and Assam).

Bidirectional causality between telecommunication and economic growth: Only four states of Indian origin show up causality in both the directions. These include Himachal Pradesh, Madhya Pradesh, West Bengal and Kerala.

No causality between telecommunication and economic growth: The three states of Jammu and Kashmir, Uttar Pradesh and Karnataka show causality in none of the direction.

The above set of states can be well elucidated in the form of a geographical map (Figure 3).

6. Conclusions

This study brings to the forefront the case of HEC between telecommunications and economic growth for 17 Indian states. After applying Hurlin–Venet process of Granger causality, we strongly refute the assumption of HC which has been upheld by the researchers in this domain. The results from the causal process are summarised as: HC is rejected in both the directions leading us to test for HEC; HEC holds for 10 states from economic growth to telecommunication (Supply led hypothesis) and eight states show causality from telecommunications to economic growth (Demand led hypothesis) and three states show up no causality (Neutrality hypothesis). The results are robust to different lags as well as definition of our economic growth variable. Thus, heterogeneous causal pattern between economic growth and telecommunications can mainly be attributed to differences among states on account of development, population, natural resources, consumption patterns and many others.

Thus, we can say that instead of universal causality direction, we find out that multiple hypotheses are being validated at subnational level. Given these differences across Indian states a word of advice that is most suitable here is that regional differences need to be accounted for, at the time of policy recommendations because it is more of a state affair now. What is true at the national level might not be true for all its constituent units and that is exactly what is replicated in this study. The states wherein the two variables are intertwined are at an advantageous position because each of them serves as a causal factor for the other. For them, both growth and Information and Communication Technology (ICT) are equally important and thus emphasis on either of them will prove to be fructuous because of feedback effect that exists in such state economies. Shock to either of them will ravage the growing process. The states which do not show any causality from telecommunications to economic growth could be attributed to lower penetration rates than the ‘critical mass’ level which is needed for telecommunication to show up their impact on growth (Shiu and Lam 2008).

Given the above heterogeneity in causal relation which is one of its kind to work along these lines, the study is not bereft of limitations. The most important of this is the case of bivariate analysis. Because this study takes into account a bivariate relationship, this offers the problem of omitted variable bias. So the future line of research needs to take into account additional variables like capital formation, labour force among others to study the direction of causality.

Footnotes

Data Availability Statement

The data needed for the present empirical research is a part of doctoral thesis hence cannot be made available although, the same can be made available on request.

Declaration of Conflicting Interests

Funding

The authors received no financial support for the research, authorship and/or publication of this article.

Notes

References

Akkemik

K. Ali

, Göksal

Koray

, & Li

Jia

. 2012. ‘Energy Consumption and Income in Chinese Provinces: Heterogeneous Panel Causality Analysis’. Applied Energy 99: 445–454. https://doi.org/10.1016/j.apenergy.2012.05.025

Barman

Hemanta

, Dutta

Mrinal Kanti

, & Nath

Hiranya K.

. 2018. ‘The Telecommunications Divide Among Indian States’. Telecommunications Policy 42 (7): 530–551. https://doi.org/10.1016/j.telpol.2018.05.003

Beil

Richard O

, Ford

George S.

, & Jackson

John D.

. 2005. ‘On the Relationship Between Telecommunications Investment and Economic Growth in the United States’. International Economic Journal 19 (1): 3–9. https://doi.org/10.1080/1351161042000320399

Breusch

T. S.

, & Pagan

Adrian

. 1980. ‘The Lagrange Multiplier Test and its Applications to Model Specification in Econometrics’. The Review of Economic Studies 47 (1): 239. https://doi.org/10.2307/2297111

Brynjolfsson

Erik

, & Yang

Shinkyu

. 1996. ‘Information Technology and Productivity: A Review of the Literature’. Advances in Computers 43 (C): 179–214. https://doi.org/10.1016/S0065-2458(08)60644-0

Ça

2013. ‘Identifying Tourism and Economic Growth Nexus: Hurlin–Venet Approach’. The Empirical Economics Letters 12 (4).

Cronin

Francis J.

, Parker

Edwin B.

, Colleran

Elisabeth K.

, & Gold

Mark A.

. 1991. ‘Telecommunications Infrastructure and Economic Growth. An Analysis of Causality’. Telecommunications Policy 15 (6): 529–535. https://doi.org/10.1016/0308-5961(91)90007-X

Olalekan

David Oladipo

. 2019. ‘Nexus Between Telecommunication Infrastructures, Economic Growth and Development in Africa: Panel Vector Autoregression (P-VAR) Analysis’. Telecommunications Policy 43 (8): 101816. https://doi.org/10.1016/j.telpol.2019.03.005

Dutta

Amitava

. 2019. ‘Telecommunications and Economic Activity: An Analysis of Granger Causality’. Journal of Management Information Systems 17 (4): 71–95.

10.

Erdil

Erkan

, & Yetkiner

I. Hakan

. 2005. ‘A Panel Data Approach for Income-Health Causality’. In The Economics of Health Reforms, edited by J. N. Yfantopoulos, 701–724. ATINER. https://EconPapers.repec.org/RePEc:sgc:wpaper:47

11.

Ghosh

Sajal

, & Prasad

Rohit

. 2012. ‘Telephone Penetrations and Economic Growth: Evidence from India‘. NETNOMICS: Economic Research and Electronic Networking 13 (1): 25–43. https://doi.org/10.1007/s11066-012-9067-z

12.

Hurlin

Christophe

, & Venet

Baptiste

. 2001. ‘Granger Causality Tests in Panel Data Models with Fixed Coefficients’. Cahier de Recherche EURISCO.

13.

Kyung So

, Hashem Pesaran

, & Shin

Yongcheol

. 2003. ‘Testing for Unit Roots in Heterogeneous Panels’. Journal of Econometrics 115 (1): 53–74. https://doi.org/10.1016/S0304-4076(03)00092-7

14.

Kateja

Alpana

, & Jha

Debajit

. 2008. ‘Exploring the Casual Relationship Between Telecommunications and GDP Growth in India’. Artha Vijnana 3: 195–208.

15.

Kumar

Ronald Ravinesh

, Kumar

Radika Devi

, & Patel

Arvind

. 2015. ‘Accounting for Telecommunications Contribution to Economic Growth: A Study of Small Pacific Island States’. Telecommunications Policy 39 (3–4): 284–295. https://doi.org/10.1016/j.telpol.2014.08.005

16.

Latif

Zahid

, mengke

Yang

, Danish

Shahid Latif

, Ximei

Liu

, Pathan

Zulfiqar Hussain

, Salam

Shafaq

, & Jianqiua

Zeng

. 2018. ‘The Dynamics of ICT, Foreign Direct Investment, Globalization and Economic Growth: Panel Estimation Robust to Heterogeneity and Cross-sectional Dependence’. Telematics and Informatics 35 (2): 318–328. https://doi.org/10.1016/j.tele.2017.12.006

17.

Gary

Madden

, and Savage

Scott

. 1998. ‘CEE Telecommunications Investment and Economic Growth’. Information Economics and Policy 10 (2): 173–195. https://doi.org/10.1016/S0167-6245(97)00020-6

18.

Mahalingam

Brinda

, & Orman

Wafa Hakim

. 2018. ‘GDP and Energy Consumption: A Panel Analysis of the US’. Applied Energy 213: 208–218. https://doi.org/10.1016/j.apenergy.2018.01.036

19.

Pesaran

M. Hashem

. 2007. ‘A Simple Panel Unit Root Test in the Presence of Cross-section Dependence’. Journal of Applied Econometrics 22 (2): 265–312. https://doi.org/10.1002/jae

20.

Pradhan

Rudra P.

, Mallik

Girijasankar

, & Bagchi

Tapan P.

. 2018. ‘Information Communication Technology (ICT) Infrastructure and Economic Growth: A Causality Evinced by Cross- country Panel Data’. IIMB Management Review 30 (1): 91–103. https://doi.org/10.1016/j.iimb.2018.01.001

21.

Pradhan

Rudra P.

, Arvin

Mak B.

, Norman

Neville R.

, & Bele

Samadhan K.

. 2014. ‘Economic Growth and the Development of Telecommunications Infrastructure in the G-20 Countries: A Panel-VAR Approach’. Telecommunications Policy 38 (7): 634–649. https://doi.org/10.1016/j.telpol.2014.03.001

22.

Pradhan

Rudra Prakash

, Bele

Samadhan

, & Pandey

Shashikant

. 2013. ‘The Link Between Telecommunication Infrastructure and Economic Growth in 34 OECD Countries’. International Journal of Technology, Policy and Management 13 (3): 278–293. https://doi.org/10.1504/IJTPM.2013.054915

23.

Lars-Hendrik

Roller

, and Waverman

Leonard

. 2019. ‘Telecommunications Infrastructure and Economic Development: A Simultaneous Approach’. The American Economic Review 91 (4): 909–923.

24.

Shiu

, Lam

(2008) Causal relationship between Telecommunications and Economic Growth: A study of 105 countries. Journal of Regional Studies 42(5): 705–718.

25.

Sridhar

Kala Seetharam

, & Sridhar

Varadharajan

. 2009. ‘Telecommunications Infrastructure and Economic Growth Evidence from Developing Countries’. Applied Econometrics and International Development 2: 91–116.

26.

Vodafone Group Plc. 2009. ‘India: The Impact of Mobile Phones (Issue 9)’. https://www.vodafone.com/content/dam/vodcom/sustainability/pdfs/public_policy_series_9.pdf

27.

Khuong M.

2011. ‘ICT as a Source of Economic Growth in the Information Age: Empirical Evidence from the 1996–2005 Period’. Telecommunications Policy 35 (4): 357–372. https://doi.org/10.1016/j.telpol.2011.02.008

28.

Westerlund

Joakim

. 2007. ‘Testing for Error Correction in Panel Data’. Oxford Bulletin of Economics and Statistics 69 (6): 709–748. https://doi.org/10.1111/j.1468-0084.2007.00477.x

29.

World Bank. 2016. ‘World Development Report: Digital Dividends’. World Bank Publications.

30.

Yoo

Seung Hoon

, & Kwak

Seung Jun

. 2004. ‘Information Technology and Economic Development in Korea: A Causality Study’. International Journal of Technology Management 27 (1): 57–67. https://doi.org/10.1504/IJTM.2004.003881

31.

Zhang

Chuanguo

, & Xu

Jiao

. 2012. ‘Retesting the Causality Between Energy Consumption and GDP in China: Evidence from Sectoral and Regional Analyses Using Dynamic Panel Data’. Energy Economics 34 (6): 1782–1789. https://doi.org/10.1016/j.eneco.2012.07.012

Heterogeneous Causality Between Telecommunications and Economic Growth: An Application of Hurlin–Venet Process to the Indian States

Abstract

Keywords

1. Introduction

Telecommunications and Growth Causality (International as well as Indian Evidence). 1

3. Methodology

Hurlin–Venet Causality Hypotheses.

Table 3.

Cross Section Dependence Test Statistics.

Unit Root Results.

Westerlund Cointegration Test.

Test Results for Homogeneous Non-causality (HNC Hypothesis).

Test Results for Homogeneous Causality (HC Hypothesis).

Results from Heterogeneous Causality Between Growth and Telecommunication.

Results from Heterogeneous Causality Between Telecommunications and Growth.

Tests of Heterogeneous Non-causality.

Footnotes

Data Availability Statement

Declaration of Conflicting Interests

Funding

Notes

References

Telecommunications and Growth Causality (International as well as Indian Evidence).¹