ea_2019_2 DOI: 10.28934/ea.19.52.2.pp12-27 ORIGINAL SCIENTIFIC PAPER Gauging the Effects of Modern Payment Technologies Adoption on the Demand for Money in Nigeria Tersoo Iorngurum1* 1 Veritas University Abuja, Faculty of Social Sciences, Department of Economics, Federal Capital Territory (FCT), Nigeria, West Africa ABSTRACT In contrast to the global intermediate goals of monetary policy, “financial exclusion” remains prevalent. Therefore, using the Nigerian economy as a point of reference, this paper attempts to shed more light on the role played by modern payment technologies in promoting financial inclusion, especially as it relates to the provision of currency in the hands of the Nigerian public for liquidity services during the period 2009:Q1 to 2017:Q4. In actualizing this objective, the Johansen cointegration method is employed to test for cointegration alongside vector error correction modeling (VECM) techniques, while the Gregory-Hansen cointegration method is employed to test for structural breaks and regime shifts. Subsequently, empirical results from the Johansen cointegration test and the normalized cointegrating coefficients of the estimated vector error correction model (VECM) reveal that real currency in the hands of the Nigerian public is positively cointegrated with real modern payment technologies transactions as well as real Gross Domestic Product (GDP), but negatively cointegrated with real savings interest rates, real quarterly time deposits interest rates, and inflation rate. On the other hand, empirical results from the Gregory- Hansen cointegration method indicate further that there are no structural breaks or regime shifts in the cointegrating coefficients during the period 2009:Q1 to 2017:Q4. In conclusion, the existence of a positive relationship between real modern payment technologies transactions and real currency in the hands of the Nigerian public implies that the former are partly responsible for the growth of the latter, thereby indicating that modern payment technologies are effective in promoting financial inclusion by providing access to liquidity services. Based on this finding the study recommends that the adoption of modern payment technologies should be promoted in order to further extend liquidity services to financially excluded Nigerians. Key words: modern payment technologies, money demand JEL Classification: E42, E41 INTRODUCTION Apart from the core mandate which primarily aims at achieving economic growth and curtailing inflation, monetary authorities all over the world aim at facilitating adequate access to financial services and financial products, otherwise known as “financial inclusion”, in their respective economies. In the Nigerian context, several policies and initiatives have been formulated and enacted in this regard, such as the Financial Stability Strategy (FSS2020), the Microfinance Policy, Non-Interest Banking, and until recently, the Cashless Policy Initiative * E-mai: tersoodavid@gmail.com Tersoo Iorngurum 13 which aims at extending digital financial services through modern electronic payment technologies (Kama & Adigun, 2013; CBN, 2014). In spite of these however, it is pertinent to note that inadequate access to financial services and financial products, otherwise known as “financial exclusion”, remains prevalent. With regards to the Nigerian economy, this problem prominently manifests itself with the majority of households (60.3%) lacking access to basic financial services and products such as liquidity services, savings accounts, deposit accounts, and credit services (World Bank, 2018). If this problem is allowed to persist, then the ultimate goal of achieving economic development may be unattainable in the short-term. This paper therefore attempts to shed more light on the possible solutions to this problem by examining the role played by modern payment technologies in promoting financial inclusion with reference to the Nigerian economy, especially as it relates to the provision of currency in the hands of the Nigerian public for liquidity services during the period 2009:Q1 to 2017:Q4. The empirical methods adopted in this study include the Johansen cointegration method, the Gregory-Hansen cointegration method, as well as theoretical considerations from Milton Friedman’s money demand theory. Further, in structuring the paper we adopt the following pattern: Section 2 deals with the literature review. Section 3 deals with the empirical methodology. Section 4 deals with the empirical results. And Section 5 deals with the conclusions. LITERATURE REVIEW In reviewing the empirical literature, a distinction is made between those early studies which focused on money demand and its conventional theoretical determinants such as income, inflation, interest rates, and exchange rates, and those contemporary studies which focused on money demand and modern payment technologies. Money demand, income, interest rates, and inInflation in Nigeria Pertaining to those early contemporary studies that laid empirical foundations for money demand investigations in Nigeria, Nwaobi (2002) examined the stability of money demand and the robustness of GDP as a scalar determinant of money demand in Nigeria during the period 1960 to 1995. With a relatively simple model-specifying vector valued autoregressive process, the hypothesis of the existence of cointegration vectors was tested and it was found that the demand for money was cointegrated with real income, interest rate and price level. Furthermore, adopting general to specific approach, an over parameterized dynamic money demand function was estimated. Thereafter, evidence gathered from the non-nested tests, suggested that income was the more appropriate scale variable in the estimation of the demand for money in Nigeria. These results sharply contradict most findings based on developed countries but are in tune with the majority of studies that used income as the appropriate scale variable in demand for money functions estimated through techniques of cointegration and error correction mechanism. Akinlo (2006) examined the determinants of broad money demand in Nigeria and its functional stability during the period 1970Q1 to 2002Q4. In so doing, the study employed the bounds cointegration method and the CUSUM stability test. The bounds cointegration results showed that broad money was positively cointegrated with income, but negatively cointegrated with interest rate and exchange rate, while the CUSUM test reported a weakly stable money demand function. Omotor and Omotor (2011) studied the functional stability of the demand for money in Nigeria during the period 1960 to 2008. The method employed was the Gregory-Hansen cointegration method. The empirical results revealed an endogenous break in Nigeria’s money 14 Economic Analysis (2019, Vol. 52, No. 2, 12-27) demand function in 1994. However after estimation, money demand was found to be functionally stable, thereby confirming the findings of previous studies which found money demand to be functionally stable and also established the fact that the Central Bank of Nigeria can effectively use money supply as a monetary policy instrument. Aiyedogbon, Ibeh, Edafe, and Ohwofasa (2013) investigated the stability of money demand in Nigeria during the period 1986 to 2010. The study employed the Johansen cointegration test and the CUSUM/CUSUMSQ stability tests to test for cointegration and parametric stability respectively. The results showed that interest rate, inflation rate and openness had negative impacts on real money demand, while gross capital formation, exchange rate and government expenditure had positive impacts. Further, the results showed that the money demand was functionally stable during the study period. Doguwa, Olowofeso, Uyaebo, Adamu, and Bada (2014) studied Nigeria’s broad money demand function during the period 1991:Q1 to 2013:Q4. The procedures employed included the Gregory-Hansen cointegration method and the CUSUMSQ stability test. The Gregory-Hansen cointegration results revealed both intercept and regime shifts in 2007:Q1, thereby indicating that structural breaks existed in the long run relationship between real broad money demand, real income, real monetary policy rate, exchange rate spread and exchange rate movements in Nigeria. Further, the CUSUMSQ test provided evidence of a stable money demand function before and after the 2008/2009 global financial crisis. In conclusion, the study recommended that since the relationship among the aforementioned variables held over a fairly long period of time, the estimated money demand model can provide important foundations for monetary policy setting in Nigeria. Folarin and Asongu (2017) studied the long-run demand for money and its stability in Nigeria during the period 1992:Q1 to 2015:Q4. In this study, the bounds cointegration method was employed alongside the CUSUM/CUSUMSQ stability tests to determine the existence of a cointegrated and stable relationship between money demand and its determinants. On this note, the empirical results showed that a stable and cointegrated relationship existed between money demand, income, interest rate, and inflation. Also, inflation rate was found to be a better opportunity cost variable in explaining money demand when compared to interest rate. Tule, Okpanachi, Ogiji, and Usman (2018) investigated the determinants of broad money demand in Nigeria and its stability during the period 1985:Q1 to 2016:Q4. The procedures employed were the bounds cointegration method and the CUSUMSQ stability test. The empirical results of this study revealed that a stable long-run relationship existed between broad money (M2) and its determinants namely GDP, stock prices, foreign interest rates and real exchange rate. Particularly, stock prices showed a significant and positive effect on long-run broad money demand, which in some ways reflected increased “financialization” and integration of the Nigerian economy into the global economic system. Overall, the findings of this study gave credence to the continued relevance of broad money (M2) as a benchmark for monetary policy implementation in Nigeria. Nwude, Offor, and Udeh (2018) examined the determinants of broad money demand in Nigeria during the period 1991:Q1 to 2014Q4. The procedures employed included the bounds cointegration method and the CUSUM stability test. The empirical results showed that a cointegrated and stable relationship existed between real broad money, real income, domestic interest rate, inflation rate, exchange rate and foreign interest rate. To be precise, real income and exchange rate were found to be positively related to the demand for real broad money while domestic interest rate, inflation rate and foreign interest rate were found to be negatively related to the demand for real broad money. Tersoo Iorngurum 15 Money demand, modern payment technologies, and financial innovation in Nigeria Pertaining to those contemporary studies that attempted to investigate the impact of modern payment technologies and financial innovations on money demand in Nigeria, Odularu and Okunrinboye (2009) studied the impact of the financial innovations on the demand for real currency in Nigeria during the period 1970 to 2004. The procedure employed was the Engle and Granger Two Step Cointegration method. The empirical findings revealed that financial innovations did not significantly affect the demand for real currency in Nigeria. However, the empirical findings revealed that income was positively related to the demand for real currency whereas interest rate was inversely related to the demand for real currency. Oyelami and Yinusa (2013) examined the impact of modern payment technologies on money demand during the period 2008:M01 to 2010:M12. The procedures employed included Vector Error Correction Modeling (VECM) and impulse response function analysis. The findings revealed that money demand was cointegrated ATMs usage, PoS usage, web usage, and mobile money usage. Further, impulse response functions showed that money demand responded positively to innovations from Automated Teller Machines (ATM) and Point-of-Sales (PoS), but responded negatively to innovations from internet payment and mobile money. Sowunmi, Amoo, Olaleye, and Salako (2014) investigated the effects of Automated Teller Machine (ATM) on demand for money in Isolo Local Government Area of Lagos State. The procedure employed included probit analyses. The results revealed that Automated Teller Machines (ATMs) has significantly increased the frequency of demand for money when compared with non-users of ATM. However, the average volume of money withdrawn through ATM was found to be significantly less than the amount withdrawn though cheque. The study also found that the ability of customers to meet their precautionary cash needs was enhanced by the use of ATMs because customers have access to cash during weekends and national strikes. Further, ATMs did not only reduce long queues in the banking halls but also reduced the average time spent in withdrawing cash. Among the problems encountered, most of the ATM users (45.5%) complained of inadequate service due to technical fault or power outages. Therefore, it was recommended that the quality of ATMs should be improved through adequate investment. Egbetunde, Ayinde, and Adeyemo (2015) investigated the impact of modern payment technologies on broad money demand in order to determine the impact of Nigeria’s cash policy on money demand during the period 2010:M1 to 2013:M8. The procedures employed included the Johansen cointegration method and Vector Error Correction Modeling (VECM). The empirical results revealed that a negative cointegrated relationship existed between money demand, ATMs usage, PoS usage, monetary policy rate, and exchange rate, while a positive cointegrated relationship existed between money demand, inflation, government expenditure, real GDP, and web transfer. Apere (2017) studied the implications of financial innovations in modern payment technologies on money demand in Nigeria during the period 1981 to 2016. The procedures employed included impulse-response analysis and VAR Granger causality testing. The empirical results revealed that innovation in modern payment technologies is an important variable that affects money demand negatively. The results of the study also revealed that the long-run demand for money balances in Nigeria positively depends on income level but negatively depends on interest rate. THEORETICAL FRAMEWORK This study adopts Friedman’s (1956) money demand theory. According to this theory, the demand for real (money) balances is determined by the real yields of other assets (bonds, equities, and physical assets), the rate of inflation, real wealth, the ratio of human to non-human 16 Economic Analysis (2019, Vol. 52, No. 2, 12-27) wealth, and individuals’ tastes and preferences. This is captured in the following money demand function: �� = ��� = �� �, … … , #, $, %, &, '( )'(� (1) Here, md denotes demand for money balances in real terms, Md denotes demand for money balances in nominal terms, P denotes price level, ri denotes returns in real terms of the ith asset, π denotes rate of inflation, w denotes wealth in real terms, u denotes individuals’ tastes and preferences, and HW/NHW denotes the ratio of human to non-human wealth. Given that data on the ratio of human to non human wealth was not available at the time of Friedman’s (1956) postulation, the empirical version of the money demand function was and/or is often expressed as: �� = ��� = �� �, … … , #, $, %� (2) Based on this equation, Friedman (1956) empirically argued that real wealth w, is positively related to the demand for real balances md, whereas the returns on other assets ri and inflation π are inversely related to the demand for real balances md. Therefore, on the basis of this theory, it is expected that: *+� *, > 0 (3) *+� */0 < 0 (4) *+� *2 < 0 (5) In order to introduce modern payment technologies, we assume that the liquidity services of money or currency in the hands of the public can be substituted (or complemented) by utilizing modern payment technologies like Automated Teller Machines (ATMs), Point of Sales (PoS) terminals, web payment systems, and mobile payment systems for transactionary and precautionary purposes. Under this assumption, individuals’ decisions to use modern payment technologies in lieu of currency in the hands of the public for liquidity services at any given point in time reflects their tastes and preferences at that point in time. Therefore, assuming that the degree of individual’s tastes and preferences for money (u) in Friedman’s (1956) theory is proxied by the volume of modern payment technologies transactions (MPT) in the economy, it can be assumed that: *+� *3 ≈ *+� *��5 (6) where MPT denotes modern payment technologies transactions, and all other variables remain as previously denoted. Further, it is intuitively inferable that: *+� *��5 = *+� *65� + *+� *�78 + *+� *,9: + *+� *+;:�<9 (7) and: *+� *65� > 0 (8) *+� *�78 > 0 (9) Tersoo Iorngurum 17 *+� *,9: < 0 (10) *+� *+;: < 0 (11) such that the direction and magnitude of (6) depends on the predominance of (8) and (9) over (10) and (11). In other words, if ATMs and other cash-dispensing payment technologies are found to be predominant in the economy, it is expected that: *+� *��5 > 0 (12) But if mobile payment systems and other cashless payment technologies such as internet payment systems are found to be predominant in the economy, it is expected that: =>? =@AB < C (13) EMPIRICAL METHODS AND RESULTS Data The data used in carrying out this study is quarterly time series data covering the period 2009Q1 to 2017Q4. The data is obtainable from the statistical bulletin of the Central Bank of Nigeria (2019). Model specification In accordance with Friedman’s (1956) money demand theory, the demand for real balances is specified as a function real wealth proxied by real Gross Domestic Product (y), real savings interest rate (r1), real quarterly time deposits interest rates (r2), inflation rate (π), and individuals’ tastes and preferences proxied by real modern payment technologies transactions (MPT) as seen in (14): �� = ��D� , �� , �� , $� , �E�� � (14) In VAR form we have: ��� = F + ∑ F�� D�H�I�J − ∑ F�� ��H�I�J − ∑ F�� ��H�I�J − ∑ F�� $�H�I�J + ∑ F�� �E��H�I�J + ∑ FL� ��H��I�J� + ��� (15) D� = M + ∑ M�� D�H�I�J� − ∑ M�� ��H�I�J − ∑ M�� ��H�I�J − ∑ M�� $�H�I�J + ∑ M�� �E��H�I�J + ∑ ML� ��H��I�J + ��� (16) �� = N + ∑ N�� D�H�I�J − ∑ N�� ��H�I�J� − ∑ N�� ��H�I�J − ∑ N�� $�H�I�J + ∑ N�� �E��H�I�J + ∑ NL� ��H��I�J + ��� (17) �� = O + ∑ O�� D�H�I�J − ∑ O�� ��H�I�J − ∑ O�� ��H�I�J� − ∑ O�� $�H�I�J + ∑ O�� �E��H�I�J + ∑ OL� ��H��I�J + ��� (18) $� = P + ∑ P�� D�H�I�J − ∑ P�� ��H�I�J − ∑ P�� ��H�I�J − ∑ P�� $�H�I�J� + ∑ P�� �E��H�I�J + ∑ PL� ��H��I�J + ��� (19) 18 Economic Analysis (2019, Vol. 52, No. 2, 12-27) �E�� = Q + ∑ Q�� D�H�I�J − ∑ Q�� ��H�I�J − ∑ Q�� ��H�I�J − ∑ Q�� $�H�I�J + ∑ Q�� �E��H�I�J� + ∑ QL� ��H��I�J + �L� (20) Estimation procedures Unit root testing with break points The order of integration of the given time series, yt, may be examined by a unit root test which accounts for structural breaks. For this purpose we adhere to Peron (1989) by utilizing a parametric approach to evaluate a null of non-stationarity (Φ = 0) against an alternative of broken trend-stationarity (Φ < 0) with the following equations: D� = � + RST� �U:� + OS� �U: � + FD�H� + ∑ V� WD�H�X�J� + &� (21) D� = � + M� + RST� �U: � + OS� �U: � + FD�H� + ∑ V� WD�H�X�J� + &� (22) D� = � + M� + YSU� �U: � + FD�H� + ∑ V� WD�H�X�J� + &� (23) D� = � + M� + RST� �U: � + YSU� �U:� + OS� �U: � + FD�H� + ∑ V� WD�H�X�J� + &� (24) Here, (21) represents a non-trending level-break model which allows for a change in level; (22) represents a trending level-break model which also allows for a change in level with a trend specification; (23) represents a trending trend-break model in which allows for a change in trend; and (24) represents a trending level/trend-break model which allows for a change in both level and trend. Further, in (21) to (24) the three dummy variables characterize the break points. The first is a level-break dummy defined by: ST� �U: � = Z0, [� � ≤ U:1, [� � > U: (25) The second is a trend-break dummy defined by: SU� �U: � = Z 0, [� � ≤ U:1�� − U: + 1�, [� � > U: (26) And the third is a one-time break dummy defined by: S� �U: � = Z0, [� � ≠ U:1, [� � = U: (27) Lag length selection In determining the optimal lag length of the VAR, it is pertinent to adopt appropriate information criteria. On this note, we adopted the Akaike Information Criterion (AIC) and the Schwarz Information Criterion (SIC) which involves the following computations: �_` = U ∗ abc│e│ + 2g (28) h_` = U ∗ abc│e│ + g ∗ abc�U� (29) Here, AIC denotes Akaike information criterion statistic, SIC denotes Schwarz information criterion statistic, T denotes number of usable observations, log│Σ│ denotes natural log of Tersoo Iorngurum 19 covariance matrix, and N denotes total number of parameters to be estimated in the VAR (Lutkepohl, 2005). Cointegration Testing In testing for cointegration, Johansen’s (1995) method utilizes two tests: the Maximum Eigenvalue test and the Trace test. The Maximum Eigenvalue test evaluates a null of r cointegrating relations against an alternative of r+1 cointegrating relations. The test statistic is computed as: ij+kl � | + 1� = −U abc�1 − n/o�� = ij�/ � |p� − ij�/ � + 1|p� (30) for r = 0,1, …, k-1. On the other hand, the Trace test evaluates a null of r cointegrating relations against an alternative of k cointegrating relations, where k is the number of endogenous variables, for r = 0, 1, …, k-1. The alternative of k cointegrating relations corresponds to the case where none of the series has a unit root such that a stationary VAR may be specified in terms of the levels of all the series. The trace test statistic for the null of r cointegrating relations is computed as: ij�/ � |p� = −U ∑ abc�1 − n� �X�J/o� (31) where λi denotes the i-th largest eigenvalue of the Π matrix (Johansen, 1995). Vector Error Correction Modeling (VECM) When evidence of cointegration abounds, Granger’s representation theorem allows one to estimate a vector error correction model which takes the following form: WD� = ∑ q�XH��J� WD�H� + r� D�HX + � + ɛ� (32) Here, yt is a px1 vector of endogenous I(1) variables; μ is a px1 vector of intercepts; ɛt is a px1 vector of stationary random processes with zero mean and constant variance; Γ is a pxp matrix of short-run coefficients; and Π is a matrix decomposed into M’F, where M’ is an pxr matrix of cointegrating vectors and F is a pxr matrix of error correction coefficients (Johansen, 1995). Statistical and econometric evaluation The t-test and the f-test may be utilized to impose restrictions on the estimated coefficients of the vector error correction model (VECM) and the cointegrating vectors in order to test for statistical significance. Further, a plethora of econometric tests may be employed to ensure that the estimated models possess the necessary second order econometric properties. To be precise, reference will be made to the Breusch-Godfrey test, the Breusch-Pagan-Godfrey test, and the Jarque-Bera test in order to test for serial correlation, heteroskedasticity, and normality respectively. The alternative of cointegration testing with structural breaks In Johansen’s (1995) method, the cointegrating coefficients may be biased if structural breaks actually exist in the cointegrating coefficients. A more suitable approach which allows for cointegration testing with structural breaks is Gregory and Hansen’s (1992b) method. In this method, the null of hypothesis of “no cointegration” is evaluated against four distinct alternative 20 Economic Analysis (2019, Vol. 52, No. 2, 12-27) hypotheses of cointegration with structural breaks. The first alternative hypothesis (GH 1) assumes that there is a level-break in the cointegrating relationship: D�� = �� + ��P�/ + F5 D�� + t� (33) The second alternative hypothesis (GH 2) assumes that there is a level break with a trend specification: D�� = �� + ��P�/ + M� + F5 D�� + t� (34) The third alternative hypothesis (GH 3) assumes that there is a regime-shift: D�� = �� + ��P�/ + F�5 D�� + F�5 D�� P�/ + t� (35) The fourth alternative hypothesis (GH 4) assumes that there is a regime shift and a trend break: D�� = �� + ��P�/ + M�� + M��P�/ + F�5 D�� + F�5 D�� P�/ + t� (36) In order to evaluate the null of “no cointegration” against the aforementioned alternative hypotheses, Gregory and Hansen’s (1992b) method employs the Philips tests (Zt and Zα) and the augmented Dickey-Fuller (ADF) test. However, the non-parametric Phillips tests are preferable because they are robust against misspecification and structural breaks which might encountered in parametric estimations. Empirical results Unit root test results The results of the unit root tests are presented in Table 1. Table 1. Unit Root Test Results Variable s Lags Included Trend Specification Break Date ADF Test Statistic 5% Critical Value Remarks mtd 1 Trend 2011Q4 -5.1573 -5.1757 I(1) Δmtd 2 No Trend 2011Q4 -5.9083 -4.4437 I(0) mptt 1 Trend 2015Q3 -3.9637 -5.1757 I(1) Δmptt 0 No Trend 2011Q1 -6.1528 -4.4436 I(0) yt 4 Trend 2011Q1 -3.6790 -5.1757 I(1) Δyt 0 No Trend 2017Q2 -6.8611 -4.4437 I(0) r1t 0 Trend 2013Q2 -4.0692 -5.1757 I(1) Δr1t 0 No Trend 2016Q1 -7.8116 -4.4436 I(0) r2t 1 Trend 2015Q3 -3.2078 -5.1757 I(1) Δr2t 0 No Trend 2011Q1 -5.0090 -4.4436 I(0) πt 3 Trend 2016Q1 -3.7669 -5.1757 I(1) Δπt 2 No Trend 2015Q4 -8.3801 -4.4436 I(0) Lag selection based on Schwarz Information Criterion (SIC) Source: Result processed using Eviews 9. For each variable, the null of non-stationarity is evaluated against the alternative of stationarity with a break point. In the level form, the null of non-stationarity is accepted at the 5% level of significance. But in the first difference, the null hypothesis of non-stationarity is Tersoo Iorngurum 21 rejected at the 5% level of significance. Therefore, in compliance with the Johansen cointegration method, the unit root test results shows that each variable is difference-stationary. Lag Length Selection Results The results of the AIC and the SIC are presented in Table 2. The AIC and the SIC both suggest that 4 lags should be utilized. Therefore in testing for cointegration with the Johansen method, 4 lags will be incorporated in estimating the VAR model. Table 2. Lag Length Selection Results Lag Length AIC SIC 0 25.5078 25.7827 1 19.8126 21.7364 2 19.9453 23.5180 3 19.0485 24.2701 4 13.8920* 20.7626* Note: * indicates lag order selected by criterion Source: Result processed using Eviews 9. Cointegration Test Results Table 3 presents the Trace test results while Table 4 presents the Maximum Eigenvalue test results. Table 3. Unrestricted Cointegration Test Results Trace Test Null Hypothesis Eigen-value Trace Statistic 5% critical Value P-value None* 0.8731 124.5164 95.7537 0.0001 At most 1 0.5804 60.5175 69.8189 0.2198 At most 2 0.4707 33.5922 47.8561 0.5242 At most 3 0.2462 13.8672 29.7971 0.8480 At most 4 0.1314 5.1039 15.4947 0.7977 At most 5 0.0235 0.7371 3.8415 0.3906 Notes: Trace test indicates 1 cointegrating equation * denotes rejection at 5% level Source: Result processed using Eviews 9. Table 4. Unrestricted Cointegration Test Results Maximum Eigenvalue Test Null Hypothesis Eigen-value Trace Statistic 5% critical Value P-value None* 0.8731 63.9989 40.0776 0.0000 At most 1 0.5804 26.9253 33.8769 0.2674 At most 2 0.4707 19.7250 27.5843 0.3605 At most 3 0.2462 8.7634 21.1316 0.8510 At most 4 0.1314 4.3668 14.2646 0.8187 At most 5 0.0235 0.7371 3.8415 0.3906 Notes: Maximum eigenvalue test indicates 1 cointegrating equation * denotes rejection at 5% level Source: Result processed using Eviews 9. 22 Economic Analysis (2019, Vol. 52, No. 2, 12-27) The decision rule is to reject the null of “no cointegrating vectors” if the Trace and Maximum eigenvalue statistics are greater than their corresponding 5% critical values. Therefore, from Tables 3 and 4, the null hypotheses of no cointegrating vectors are rejected at the 5% level of significance, thereby indicating that 1 cointegrating vector exists for the relationship between money demand and the other endogenous variables of the VAR. Vector Error Correction Modeling (VECM) Results The evidence of cointegration in Table 3 and Table 4 allows us to estimate a vector error correction model based on Granger’s representation theorem. The estimates of the vector error correction model are presented in Table 5, while the corresponding normalized cointegrating coefficients are presented in Table 7. In Tables 5 and 7, and the subsequent equations, md still denotes real currency in the hands of the public, y still denotes gross domestic product which serves as a proxy for wealth, r1 still denotes real savings interest rate, r2 still denotes real quarterly time deposits interest rate, π still denotes inflation rate, and mpt still denotes real modern payment technologies transactions. Table 5.: Estimates of the Vector Error Correction Model (VECM) Eqn. Δmdt Δmptt Δyt Δr1t Δr2t Δπt Δ mdt-3 0.0814 (0.1563) [ 0.5209] 0.1422* (0.0673) [ 2.1115] 2.8012 (1.3838) [ 2.0241] 0.0779 (0.0872) [ 0.8929] 0.1554 (0.1584) [ 0.9811] -0.6301* (0.2628) [-2.3970] Δmptt-3 -1.1109* (0.4511) [-2.4625] -0.2500 (0.1943) [-1.2877] 1.0590 (3.9936) [ 0.2651] 0.2452 (0.2518) [ 0.9739] 0.8453 (0.4571) [ 1.8491] 0.2581 (0.7586) [ 0.3403] Δyt-3 -0.04007 (0.0207) [-1.9318] -0.0174* (0.0089) [-1.9517] 0.3530 (0.1836) [ 1.9225] 0.0144 (0.0115) [ 1.2450] 0.0145 (0.0210) [ 0.6930] -0.0176 (0.0348) [-0.5059] Δr1t-3 0.1448 (0.6912) [ 0.2095] 0.0826 (0.2977) [ 0.2774] -15.4736* (6.1187) [-2.5288] -0.2020 (0.3858) [-0.5236] -0.2419 (0.7004) [-0.3453] -0.4603* (1.1623) [-0.3960] Δr2t-3 -0.4504 (0.2424) [-1.8579] -0.2233* (0.1044) [-2.1392] 4.4526* (2.1459) [ 2.0749] -0.0229 (0.1353) [-0.1692] -0.0399 (0.2456) [-0.1625] 0.4736 (0.4076) [ 1.1620] Δπt-3 0.17808 (0.1725) [ 1.0327] 0.0197 (0.0742) [ 0.2664] -2.2950 (1.5266) [-1.5033] 0.0629 (0.0962) [ 0.6543] 0.0965 (0.1747) [ 0.5524] -0.4812 (0.2900) [-1.6596] C 0.01431 (0.2584) [ 0.0554] 0.3345* (0.1113) [ 3.0053] 3.3837 (2.2875) [ 1.4791] -0.0184 (0.1442) [-0.1278] -0.3041 (0.2618) [-1.1616] 0.1274 (0.4345) [ 0.2933] ECTt-1 -0.0202* (0.0089) [-2.2592] -0.0046 (0.0038) [-1.2021] 0.3932* (0.0789) [ 4.9796] 0.0007 (0.0049) [ 0.1530] -0.0306* (0.0090) [-3.3884] -0.0150 (0.0150) [-1.0045] Notes: * indicates significance at 5% level of significance Standard errors in parenthesis ( ), t-statistics in square brackets [ ] Source: Result processed using Eviews 9. Tersoo Iorngurum 23 Table 6. Summary Statistics of the Vector Error Correction Model (VECM) Δmdt Δmptt Δyt Δr1t Δr2t Δπt R-squared 0.5814 0.4640 0.6479 0.2056 0.4624 0.3451 Adj. R-SQ 0.4593 0.3077 0.5452 -0.0259 0.3057 0.1541 F-statistic 4.7628 2.9688 6.3101 0.8877 2.9501 1.8070 BG(4) 36.2591 {0.4565} BPG(294) 309.5488 {0.2553} JB(12) 3.8144 {0.9865} Notes: P-values in brackets { } F-test 5% critical value equals 2.49, at v1=6 and v2=26 degrees of freedom BG(4) denotes Breusch-Godfrey LM test for 4th order serial correlation BPG(294) denotes Breusch-Pagan-Godfrey heteroskedasticity test at 294 degrees of freedom JB(12) denotes Jarque-Bera joint normality test at 12 degrees of freedom, 2 for each of 6 components Source: Result processed using Eviews 9. Table 7. Normalized Cointegrating Coefficients mdt C mptt yt r1t r2t πt 1 -9.6496 -10.0091* (1.8327) [-5.4614] -0.5762* (0.2046) [-2.8163] 11.8729* (4.9138) [ 2.4163] 7.7256* (1.9776) [ 3.9066] 5.3337* (1.3068) [ 4.0814] Notes: * indicates significance at 5% level of significance Standard errors in parenthesis ( ), t-statistics in square brackets [ ] Source: Result processed using Eviews 9. For ease of interpretation, the estimates in Table 5 are also presented linearly in (37) to (42), while the coefficients in Table 7 are also presented linearly in (43). W��� = 0.0143 + 0.0814W��H�� − 1.1109W�E��H� − 0.0401WD�H� + 0.1448W ��H� − 0.4504W ��H� + 0.1781W$�H� − 0.0202�`U�H� + &�� (37) W�E�� = 0.3345 + 0.1422W��H�� − 0.250W�E��H� − 0.0174WD�H� + 0.0826W ��H� − 0.2233W ��H� + 0.0197W$�H� − 0.0046�`U�H� + &�� (38) WD� = 3.3837 + 2.8012W��H�� + 1.0590W�E��H� + 0.3530WD�H� − 15.4736W ��H� + 4.4526W ��H� − 2.2950W$�H� + 0.3932�`U�H� + &�� (39) W �� = −0.0184 + 0.0779W��H�� + 0.2452W�E��H� + 0.0144WD�H� − 0.2020W ��H� − 0.0229W ��H� + 0.0629W$�H� + 0.0007�`U�H� + &�� (40) W �� = −0.3041 + 1554W��H�� + 0.8453W�E��H� + 0.0145WD�H� − 0.2419W ��H� − 0.0399W ��H� + 0.0965W$�H� − 0.0306�`U�H� + &�� (41) W$� = 0.1274 − 0.6301W��H�� + 0.2581W�E��H� − 0.0176WD�H� − 0.4603W ��H� + 0.4736W ��H� − 0.4812W$�H� − 0.0150�`U�H� + &L� (42) Among the numerous coefficients, interest lies in the error correction coefficient (ECTt-1) which shows the rate of adjustment to the long-run cointegrated relationship between mdt and the other endogenous variables. In (37), the error correction coefficient is expectedly negative and statistically significant, thereby signifying equilibrium and indicating that 2.02 percent of 24 Economic Analysis (2019, Vol. 52, No. 2, 12-27) any discrepancies in long-run mdt will be corrected in each period. In (38), the error correction coefficient is negative but statistically insignificant, thereby implying that (38) is out of equilibrium. In (39), the error correction coefficient is positive and statistically significant, thereby implying that (39) is too high to be in equilibrium. In (40), the error correction coefficient is positive, but statistically insignificant, thereby implying that (40) is out of equilibrium. In (41), the error correction coefficient is negative and statistically significant, thereby implying that (41) reverts to equilibrium with respect to discrepancies in long-run mdt. In (42), the error correction coefficient is negative but statistically insignificant, thereby implying that (42) is out of equilibrium. Further, based on the adjusted R2s in Table 6, (39) seems to be the only component with a good fit (54.52%). On the other hand, based on the f-statistics, (37), (38), (39), and (41) seem to be statistically significant at the 5% level, while (40) and (42) seem to be statistically insignificant at the 5% level. Pertaining to the second order econometric criteria, the p-values (0.4564 and 0.2553) for the Breusch-Godfrey (BG) test and the Breusch-Pagan-Godfrey (BPG) test lead to the rejection of the null hypothesis of 4th order serial correlation and the null hypothesis of heteroskedasticity respectively, while the p-value (0.9865) for the Jarque-Bera (JB) test leads to the rejection of the null hypothesis of abnormally distributed residuals. Therefore, the VECM does not violate any of the second order econometric criteria. For the long-run analysis, the normalized cointegrating coefficients in (41) will be interpreted accordingly. ��� = 9.6496 + 10.0091�E�� + 0.5762D� − 11.8729 �� − 7.7255 �� − 5.3337$� + t� (43) The coefficient of mptt is positive and statistically significant at the 5% level, thereby implying that a unit increase (decrease) in real modern payment technologies transactions causes real currency in the hands of the public to increase (decrease) by 10.0091 units. The coefficient of yt is expectedly positive and statistically significant at the 5% level, thereby implying that a unit increase (decrease) in the level of real GDP causes real currency in the hands of the public to increase (decrease) by 0.5762 units. The coefficients of r1t and r2t are expectedly negative and statistically significant at the % level, thereby implying that a unit increase (decrease) in real savings interest rates and real quarterly time deposits interest rates cause real currency in the hands of the public to decrease (increase) by 11.8729 units and 7.7255 units respectively. The coefficient of inflation rate is expectedly negative and statistically significant, thereby implying that a unit increase (decrease) in inflation rate causes real currency in the hands of the public to decrease (increase) by 5.3337 units. Finally, the intercept of the cointegrating equation is positive and it implies that the real currency demand function has a positive autonomous magnitude of 9.6496 units. Alternative evidence from cointegration testing with structural breaks Unlike the Johansen cointegration method, the Gregory-Hansen cointegration method is employed as an alternative cointegration method which accounts for structural breaks. The Gregory-Hansen test results are presented in Table 8. Table 8. Gregory-Hansen Test Results Specification Break Date Zt Statistic 5% Critical Value Accept H0 GH 1 2014Q4 -5.27 -5.56 Yes GH 2 2010Q1 -5.38 -5.83 Yes GH 3 2015Q2 -5.61 -6.41 Yes Specification Break Date Zα Statistic 5% Critical Value Accept H0 GH 1 2014Q4 -31.30 -59.40 Yes Tersoo Iorngurum 25 GH 2 2010Q1 -32.99 -65.44 Yes GH 3 2015Q2 -33.19 -78.52 Yes Specification Break Date ADF Statistic 5% Critical Value Accept H0 GH 1 2014Q3 -5.29 -5.56 Yes GH 2 2010Q1 -6.90 -5.83 No GH 3 2014Q4 -5.61 -6.41 Yes Note: r2 was excluded in order to obtain 4 regressors GH 4 was not computed due to limited observations Source: Result processed using Stata 13. In interpreting results, we adopt the non-parametric Zt and Zα statistics instead of the parametric ADF statistic because the former are robust against misspecification and structural breaks which might be encountered in parametric estimation. On this note, Table 8 shows that the Zt statistics are less than their 5% critical values, thereby leading to the acceptance of the null hypotheses of “no cointegration” in lieu of the alternative hypotheses of cointegration with level-break, level-break with trend, and regime shift. Similarly, Table 8 also shows that the Zα statistics are less than their 5% critical values, thereby leading to the acceptance of the null hypotheses of “no cointegration” in lieu of the alternative hypotheses of cointegration with level-break, level-break with trend, and regime shift. Therefore, based on the Zt and Zα statistics we accept the null of “no cointegration” but reject the alternative hypotheses of cointegration with structural breaks. CONCLUSION In order to evaluate the role of modern payment technologies in promoting financial inclusion in the Nigerian economy, this study mainly attempted to examine the effects of modern payment technologies adoption on the availability of currency in the hands of the Nigerian public during the period 2009:Q1 to 2017:Q4. On this note, the Johansen cointegration method was employed to test for cointegration alongside vector error correction modeling (VECM) techniques, while the Gregory-Hansen cointegration method was employed to test for structural breaks and regime shifts. Thereafter, empirical results from the Johansen cointegration test and the normalized cointegrating coefficients of the vector error correction model (VECM) revealed that real currency in the hands of the Nigerian public was positively cointegrated with real modern payment technologies transactions as well as real Gross Domestic Product (GDP), but negatively cointegrated with real savings interest rates, real quarterly time deposits interest rates, and inflation rate. On the other, empirical results from the Gregory-Hansen cointegration method indicated further that there were no structural breaks or regime shifts in the cointegrating coefficients during the period 2009:Q1 to 2017:Q4. In conclusion, the existence of a positive relationship between real modern payment technologies transactions and real currency in the hands of the Nigerian public indicated that the former were partly responsible for the growth of the latter during the period under investigation, thereby implying that modern payment technologies were effective in promoting financial inclusion by providing more access to liquidity services. Therefore, it was recommended that wide-spread adoption of modern payment technologies should be promoted in order to further extend liquidity services to financially excluded Nigerians. ACKNOWLEDGEMENTS I humbly wish to acknowledge my dissertation supervisor who also doubles as Chairman of Economics Department and visiting professor of the African Union, Distinguished Professor G.C. Nwaobi for his indispensible guidance during the preparation of this manuscript. 26 Economic Analysis (2019, Vol. 52, No. 2, 12-27) REFERENCES Aiyedogbon, J.O., Ibeh, S.E., Edafe, M., & Ohwofasa, B.O. (2013). “Empirical Analyses of Money Demand Function in Nigeria: 1986 – 2010.” International Journal of Humanities and Social Sciences, 3(8): 132–147. Akinlo, A. E. (2006). “The Stability of Money Demand in Nigeria: An Autoregressive Distributed Lag Approach.” Journal of Policy Modeling, 28(1): 445-452. Apere, T.O. (2017). “The Implications of Financial Innovation on Money Demand in Nigeria.” Paper presented at the ISER 94th International Conference, Zurich. Central Bank of Nigeria (2014). Payment System Transformation: Cashless Nigeria Implementation. Garki, Abuja: Central Bank of Nigeria. Central Bank of Nigeria (2019). Central Bank of Nigeria (CBN) Statistical Bulletin. Garki, Abuja: Central Bank of Nigeria. Doguwa, S.I., Olowofeso, O.E., Uyaebo, S.O.U., Adamu, I., & Bada, A.S. (2014). “Structural Breaks, Cointegration and Demand for Money in Nigeria.” CBN Journal of Applied Statistics, 5(1): 15–33. Egbetunde, T., Ayinde, T.O., & Adeyemo, A.A. (2015). “Demand for Money and Cashless Policy in Nigeria.” Indian Journal of Management Science, V(1): 83 – 90. Folarin, O. E., & Asongu, S. A. (2017). “Financial Liberalization and Long-Run Stability of Money Demand in Nigeria.” African Governance and Development Institute Working Paper WP/17/18. Friedman, M. (1956). Studies in the Quantity Theory of Money. Chicago, USA: University of Chicago Press. Gregory, A.W., & Hansen, B.E. (1992b). “Residual-Based Tests for Cointegration in Models with Regime-Shifts.” Queens University Working Paper 862. Johansen, S. (1995). Likelihood Based Inference in Cointegrated Vector Autoregressive Models. Oxford: Oxford University Press. Kama, U., & Adigun, M. (2013). Financial Inclusion in Nigeria: Issues and Challenges. Garki, Abuja: Central Bank of Nigeria. Lütkepohl, H. (2005). New Introduction to Multiple Time Series Analysis. Berlin, Germany: Springer. Nwaobi, G.C. (2002). “A Vector Error Correction and Non Nested Modeling of Money Demand Function in Nigeria.” Economics Bulletin, 3(4): 1–8. Nwude, E.C., Offor, K.O., & Udeh, S.N. (2018). “Determinants and Stability of Money Demand in Nigeria.” International Journal of Economics and Financial Issues, 8(3): 340-353. Odularu, G.O., & Okunrinboye, O.A. (2009). “Modeling the Impact of Financial Innovation on the Demand for Money in Nigeria.” Journal of Business Management, 3(2): 039–051. Omotor, D. G., & Omotor, P.E. (2011). “Structural Breaks, Demand for Money and Monetary Policy in Nigeria.” Ekonomski Pregled, 62(9): 559-582. Oyelami, L.O., & Yinusa, D.O. (2013). “Alternative Payment Systems Implication for Currency Demand and Monetary Policy in Developing Economies: A Case Study of Nigeria.” International Journal of Humanities and Social Science, 3(20): 253–260. Peron, P. (1989). “The Great Crash, the Oil Price Shock, and the Unit Root Hypothesis.” Econometrica, 1(1): 1361-1401. Sowunmi, F.A., Amoo, Z.O., Olaleye, S.O. & Salako, M.A. (2014). “Effect of Automated Teller Machine (ATM) on Demand for Money in Isolo Local Government Area of Lagos State, Nigeria.” Journal of Applied Business and Economics, 16(3): 171–180. Tule, M.K., Okpanachi, U.M., Ogiji, P., & Usman, N. (2018). “A Reassessment of Money Demand in Nigeria.” CBN Journal of Applied Statistics, 9(1): 47-75. World Bank (2018). The Little Data Book on Financial Inclusion. Washington DC: World Bank Group. Tersoo Iorngurum 27 APPENDIX Table A.1: Data Obs. Quarter md y r1 r2 π mpt 1 2009Q1 27.0800 60.1500 0.8629 11.3629 1.5500 1.6000 2 2009Q2 24.9400 62.7400 -0.4864 9.8736 2.8200 1.6400 3 2009Q3 23.4800 67.3000 -2.3838 7.6329 4.5800 1.8500 4 2009Q4 25.4700 68.1800 0.6174 10.8641 2.3100 1.6400 5 2010Q1 23.6300 120.5800 -0.6108 6.7925 3.8600 0.6700 6 2010Q2 22.9000 121.2000 0.3386 3.9153 2.3600 0.8300 7 2010Q3 22.5200 128.4100 -2.8753 0.2114 4.6800 1.1600 8 2010Q4 25.1600 130.6100 -0.1617 3.0250 1.8400 1.3800 9 2011Q1 27.1300 115.1000 -1.7568 1.8032 3.6300 3.1400 10 2011Q2 27.0500 115.8400 -0.2459 3.7408 1.9000 3.3400 11 2011Q3 25.4800 121.2900 -1.4598 2.4969 3.4200 3.3400 12 2011Q4 26.8000 123.7900 -0.9654 4.6613 2.8900 4.0000 13 2012Q1 25.3000 106.1300 -3.3073 3.1027 6.0500 3.5400 14 2012Q2 24.7900 106.9000 -0.4012 5.9955 2.8600 3.7600 15 2012Q3 23.6100 114.4300 -0.2625 6.5142 2.7500 3.8700 16 2012Q4 25.6700 114.5500 -0.7601 6.4866 3.3500 4.2700 17 2013Q1 24.9100 101.6600 -0.3474 6.3093 2.9100 4.7000 18 2013Q2 23.8300 103.5800 0.1098 5.7931 2.7700 5.1100 19 2013Q3 23.3900 111.1400 0.8520 6.1720 2.3000 5.5300 20 2013Q4 26.4300 113.3600 0.3996 5.8396 3.0800 6.2600 21 2014Q1 25.1400 100.1400 1.2885 7.4018 3.0400 6.0600 22 2014Q2 23.8200 102.1900 1.3249 7.2716 3.2300 6.4200 23 2014Q3 23.9100 108.9300 1.4159 7.1226 3.0600 7.5500 24 2014Q4 24.6300 111.2100 1.7262 7.7895 2.7500 7.6900 25 2015Q1 24.9600 96.1000 1.2356 7.0656 3.8100 6.8800 26 2015Q2 21.5100 96.0100 0.9297 6.6130 4.4600 6.9100 27 2015Q3 20.2400 102.4700 1.3624 8.0991 3.9400 7.1600 28 2015Q4 21.9500 103.7900 1.7043 5.7176 3.1500 7.6400 29 2016Q1 22.5800 85.7900 -0.7968 2.7332 7.2700 7.4300 30 2016Q2 21.3300 82.0500 -2.7846 0.5487 11.8200 7.5500 31 2016Q3 20.4700 85.1500 -0.3403 3.4064 8.5000 8.2000 32 2016Q4 23.3000 86.1100 1.5801 5.9601 5.3600 9.7900 33 2017Q1 22.3800 72.1000 0.6222 5.4422 7.6000 9.5600 34 2017Q2 19.9900 70.8000 -0.9766 4.1867 11.2000 9.5500 35 2017Q3 18.4000 123.1400 0.2332 6.1899 8.8500 9.2300 36 2017Q4 19.6100 127.5000 1.7387 7.5054 5.6000 10.7400 Notes: md denotes real currency in the hands of the Nigerian public, y denotes real Gross Domestic Product, r1 denotes real savings interest rate, r2 denotes real quarterly time deposits rate, π denotes inflation rate, and mpt denotes real modern payment technologies transactions. md, y, and mpt are in billions of Naira. π is calculated as change in Consumer Price Indices; r1 and r2 are expressed in percentages. Source: Central Bank of Nigeria (2019). Article history: Received: October 9, 2019 Accepted: November 3, 2019