Financial Inclusion and Economic Growth in WAEMU: A Multiscale Heterogeneity Panel Causality Approach

This paper examines the causal relationship between Financial Inclusion and economic growth in the West African Economic and Monetary Union (WAEMU) from 2006 to 2015. We combined the heterogeneity panel causality test proposed by Dimitrescu and Hurlin (2012) with the Maximal Overlap Discrete Wavelet Transform (MODWT) to analyze the bi-directional causality at different time scales. We used two Financial Inclusion indicators: the overall rate of demographic penetration of financial services and the overall rate of use of financial services. Our results show that at scale 1 (2-4 years), there is no causality between economic growth and Financial Inclusion indicators. However, at scale 2 (4-8 years), we found a bi-directional causality between economic growth and Financial Inclusion. Policymakers should, therefore, while promoting Financial Inclusion reforms that are beneficial to Financial Inclusion, make more efficient the levers favoring macroeconomic growth, which also seems to be a decisive factor of Financial Inclusion.


Introduction
In recent years, the Central Bank of West African States (BCEAO 1 ) have implemented several reforms to promote Financial Inclusion (FI) in WAEMU 2 . These reforms focus on the establishment of a legal framework and financial infrastructures more adapted to the banking activity, in support of the decentralized financial sector and the implementation of action promoting access to financial services 3 (BCEAO, 2017).
1 These reforms had a positive effect on the use of financial services. Indeed, Demirguc-Kunt et al. (2015) has shown a recent expansion of Financial Inclusion through mobile money accounts in WAEMU countries, particularly in Ivory Coast and Mali. BCEAO (2017) found that about 21.9 million individuals now have a mobile phone account against 11 million in 2013 in WAEMU. These studies show the positive repercussions of the measures taken by the BCEAO on Financial Inclusion within WAEMU. However, the Financial Inclusion despite this recent expansion in WAEMU is still weak relative to other regions (see Mlachila et al., 2016).
The Financial Inclusion can be defined as "the pursuit of making financial services accessible at affordable costs to all individuals and businesses, irrespective of net worth and size respectively" 4 or according to the World Bank (2014), "typically defined as the proportion of individuals and firms that use financial services".
According to the World Bank and the AfDB 5 , the access to essential financial services would enable populations to have better-living conditions (health, investment in business, education . . . ).
Several authors have highlighted the beneficial effects of Financial Inclusion on economic growth. However, macroeconomic studies remain low. Some works such as those of Hariharan and Marktanner (2012) have shown that Financial Inclusion had the potential to enhance economic growth and development. Sahay et al. (2015) have demonstrated that Financial Inclusion indicators had a positive impact on growth but had to be coupled with financial development. Sharma (2016) has found that various dimensions of Financial Inclusion promoted economic growth.
In subsaharian Africa countries, several studies related to FI have been taken. Kpodar and Andrianaivo (2011) argued that the joint impact of Financial Inclusion and mobile phone development on growth was stronger. Oruo (2013) found a strong positive correlation between Financial Inclusion and economic growth in Kenya. Onaolapo (2015) and Babajide et al. (2015) showed that effects of Financial Inclusion on the economic growth of Nigeria are positives. The Outlook Regional Economic (2015) argued that Financial Inclusion by lowering constraints to access credit generally boosted growth in African emerging and developing countries.
The purpose of this paper is to determine whether the measures taken in recent years to promote Financial Inclusion have actually caused the economic growth in the WAEMU. Two variables will be used as proxies for Financial Inclusion: the overall rate of demographic penetration of financial services and the overall rate of use of financial services. 6 Moreover, we will try to see which of these variables has the most influence on economic growth. We favor a dynamic analysis of the causality between Financial Inclusion and growth. Indeed, Sahay et al. (2015) argued the possibility of a reverse causation in this relationship. We have combined the wavelet methods and the panel causality test proposed by Dimitrescu and Hurlin (2012). Firstly, we have implemented the MODWT 7 to get the data at different time scales and then, we have applied the panel causality test at each time scale. The wavelet analysis allows to accurately choose and to analyze the time scale where we want to study the causality between variables. In addition, this approach allows the analysis of non-stationary series dynamics (see Percival and Walden, 2000) and releases the hypothesis of co-integration of data of ECM 8 . time scales, to show precisely the causality period between the variables. Finally, this study is the first on the nexus analysis using the wavelet analysis.
The rest of the study is structured as follows. Section 2 presents a brief review of the literature on Financial Inclusion and growth relationship. Section 3 provides the econometric methodology used to analyze the Financial Inclusion and growth causality. Section 4 examines the data and empirical results and Section 5 concludes.

Literature review
The literature on the nexus between Financial Inclusion and economic growth from a macroeconomic point of view is recent and not very extensive. Hariharan and Marktanner (2012) have shown that Financial Inclusion could stimulate economic growth. They also argued that Financial Inclusion could create capital because of this strong positive correlation with the total factor productivity. They concluded that Financial Inclusion could increase the savings portfolio, the efficiency of intermediation of financial sector, foster entrepreneurship and thus economic growth. Sahay et al. (2015) have used macroeconometrics and microeconometrics methodologies to study the link between Financial Inclusion and GDP growth. The results showed that Financial Inclusion have a positive impact on GDP growth but must be combined with financial development. However, as more inclusion and financial development increases, the positive effect of inclusion on growth decreases. Sharma (2016) using the Vector autoregression (VAR) and the Granger causality, have shown that various dimensions of Financial Inclusion (banking penetration, availability, and usage of banking services) have positively impacted the economic growth. Author found a bi-directional causality between the geographical penetration of banking services and the economic development and a unidirectional causality between the number of deposits and the GDP.
In sub-Saharan African countries, Kpodar and Andrianaivo (2011) have addressed the question of whether Financial Inclusion was one of the channels through which the development of mobile telephony improved the economic growth. They have shown that mobile penetration had a positive impact on the economic growth by facilitating Financial Inclusion, but it has also consolidated the impact of Financial Inclusion on the economic growth. Greater penetration of mobile telephony increases access to deposits and loans. She have concluded that the joint impact of Financial Inclusion and mobile phone development on the growth was stronger. Oruo (2013) has investigated the relationship between Financial Inclusion and economic growth in Kenya. She has found that the economic growth had a strong positive correlation with Financial Inclusion, especially the branch networks of the banking sector, mobile money accounts and the users. Onaolapo (2015) studied the effects of Financial Inclusion on the economic growth of Nigeria. He found a significant positive relationship between financial inclusion and economic growth. The author also showed that Financial Inclusion greatly influenced poverty reduction and financial intermediation through positively impacted Bank Branch Networks, Loans to Rural Areas and small enterprises. Babajide et al. (2015), were interested in the impact of Financial Inclusion on growth. They found that Financial Inclusion positively impacted the total factor of production and the capital per worker, which impact positively the final output of the economy. The Outlook Regional Economic (2015) by using a micro-founded general equilibrium model, analyzed the impact of Financial Inclusion on growth in Africa. They showed that lowering credit access constraints and lowering participation costs to market for firms and companies could stimulate growth and productivity and reduce inequality.

Methodology
In this section, we present the econometric methodology used to study the causality between Financial Inclusion and economic growth. First, we provide an overview of Maximum Overlap Discrete Wavelet Transform and then, we present the panel causality test proposed by Dimitrescu and Hurlin (2012).

Maximum Overlap Discrete Wavelet Transform (MODWT)
We use the MODWT to implement the data at different time scales (see Percival and Walden, 2000). The MODWT localizes variations in the signal or time series in time and frequency simultaneously. The variability and the evolution over time can be captured by decomposing the time series at many timescales.
Let X t , the data. The time series can be decomposed by a sequence of projections onto wavelet basis: where j = 1, 2 . . . J, the level of multiresolution and J = log2(T ); Φ,the father wavelet and Ψ, the mother wavelet.
s J,k , the smooth wavelet coefficient (long run movements) provides a smooth or overall pattern of the original signal and d j,k , the wavelet detail coefficient (short run movements) capture local fluctuations in each scale over the entire period of a time series. Φ J,k and ψ j,k are scaling and translation obtained from Φ and Ψ and are defined as follow For the decomposition, we use Daubechies least asymmetric (LA) wavelet filter of length 8 because it is one of the best and most used in wavelets theory.
The decomposition of the series by the MODWT is usually implemented by the Pyramidal Algorithm (see Mallat, 1999). The multiresolution analysis of the X t using the MODWT can be written as follows

Panel Heterogeneity Causality Test
We apply to the data at different time scales, the heterogeneity panel causality test introduced by Dimitrescu and Hurlin (2012) 9 . This test is a extension to panel data version of the Granger (1969) causality test for time series. The underlying regression writes as follows 9 DH 4 where x i,t and y i,t are the observations of two stationary variables for individual i in period tand α i are the fixed effects. Coefficients are allowed to differ across individuals but are assumed time-invariant. The maximal lag order K is assumed to be identical for all individuals and the panel must be balanced.
As in Granger (1969), the procedure to determine the existence of causality is to test for significant effects of past values of x on the present value of y. The null hypothesis is therefore defined as which corresponds to the absence of causality for all individuals in the panel. The test assumes there can be causality for some individuals but not necessarily for all. The alternative hypothesis thus writes H1 : γi1 = γi2 = . . . = γik = 0, ∀i = 1, . . . , N 1 γi1 = 0 or γi2 = 0 or . . . or γik = 0, ∀i = 1, . . . , N 1 + 1, . . . , N where N 1 ∈ [0; N − 1] is unknown. If N 1 = 0, there is causality for all individuals in the panel. N 1 is strictly smaller than N , otherwise there is no causality for all individuals and H1 reduces to H0. Against this backdrop, DH propose the following procedure: run the N individual regressions implicitly enclosed in (6), perform F-tests of the K linear hypotheses γi1 = γi2 = . . . = γik = 0 to retrieve W i , and finally we compute W as the average of the N individual Wald statistics where W i is the standard adjusted Wald statistic for individual i observed during T periods. We emphasize that the test is designed to detect causality at the panel-level, and rejecting H0 does not exclude that there is no causality for some individuals. Using Monte Carlo simulations, DH show that W is asymptotically well-behaved and can genuinely be used to investigate panel causality. Under the assumption that Wald statistics W i are independently and identically distributed across individuals, it can be showed that the standardized statisticZ when T → ∞ and then N → ∞ (sometimes interpreted as T should be large relative to N ) follows a standard normal distribution Also, for a fixed T dimension with T > 5+3K, the approximated standardized statisticZ follows a standard normal distributionZ The testing procedure of the null hypothesis in (7)     Firstly, we compute the wavelet coefficients using the MODWT to obtain the data at different time scales. For the decomposition, we use Daubechies Least Asymmetric (LA) wavelet filter of length 8 10 (see Daubechies, 1992).
10 One of the best wavelets filters used in the theory  The maximum number of scales or decompositions allowed is log2(T) 11 where N is the number of observations. However, the wavelet coefficients become too small at large scales, then we have decided to stop to 2 decompositions or scales with 2 wavelet details and 1 smooth wavelet coefficient (long run dynamic or trend). Secondly, we apply the panel causality test of DH at each scales. The analysis of Finally, we apply the DH panel causality test at D1 and D2 scales. Given the number of data, the optimal number of lags allowed by the DH panel causality test is 1 15 . The test results in Table 3 indicates that at scale (2-4 years), there is no causality between the economic growth and the two indicators of Financial Inclusion. At scale 2 (4-8 years), the causality is present and is even bi-directional. The overall rate of demographic penetration of financial services and the overall rate of use of financial services cause GDP growth and vice versa. The analysis of the statistic tests (ZBAR) at scale 2 (4-8 years) provides further informations. Firstly we have found that economic growth causes more Financial Inclusion than this one causes economic growth. Then, the use of financial services causes more economic growth than their demographic penetration. Finally, economic growth causes more geographic penetration of financial services than their use. The results obtained seem to be logical. Indeed, with regard to the use of financial services, we have noticed an increase from 2.6 to 7.8 million bank accounts from 2006 to 2014 and from 366,000 in 2010 to 16 millions in 2016 of electronic money coin. This rise in savings certainly had a positive effect on the investment and therefore on the economic growth. In addition, demographic access and the supply of financial services has increased thanks to various measures that have been implemented and the economic growth of recent years. We have noticed an increase of to 18.4 points of services for 10,000 adults in 2014 against 0.9 points in 2006 and from 0 points of services of electronic currency issuer in 2009 to 24,300 in 2014 (see BCEAO, 2016).

Conclusion
In this study, the causal relationship between Financial Inclusion and economic growth is investigated using panel data from WAEMU from 2006 to 2015. We used the GDP per capita growth as the proxy of economic growth and two indicators as proxies of the Financial Inclusion: the overall rate of demographic penetration of financial services 11 log2(10) = 3.3219 12 Maddala and Wu 13 Im, Pesaran and Shin 14 Cross-sectionally augmented IPS 15 T > 5 + 3K, where K is the lag number and the overall rate of use of financial services. We combined MODWT and panel causality test from Dimitrescu and Hurlin (2012) to analyze this relationship.
The findings reveal that the causal relationship between the economic growth and the Financial Inclusion depends on the time scale. At scale 1 (2-4 years), there is no causality but at scale 2 (4-8 years), there is a bidirectional causality between the economic growth and the Financial Inclusion. We have also found that the use of financial services causes more the economic growth than the demographic penetration of financial services. But in the other sense, the economic growth causes more the demographic penetration of financial services than the use of financial services. The results are almost similar to those of Sharma (2016). There is no causality between the Financial Inclusion and the economic growth at short run but at medium or long run, there is a bi-directional causality. We can conclude that Financial Inclusion measures that have been implemented have actually simulated growth in WAEMU in long run.
The results from this study are relevant for policymakers. They could improve the Financial Inclusion and macroeconomic growth simultaneously to reach an inclusive and sustainable growth. Firstly, the policymakers should continue to encourage and even intensify policies and reforms promoting the demand for financial services 16 . This would stimulate the economic growth by increasing savings and therefore investments. At the same time, they should strengthen and liberalize the investment regulatory framework and create an environment conducive to exports by facilitating administrative procedures and fighting against corruption. All of this will promote the economic growth which in turn will increase the supply of available financial services.
Policymakers and financial authorities should, while promoting financial inclusion, simulate macroeconomic growth, given its strong positive impact on financial inclusion.