Assessing the Impact of ATM and POS Transactions on Currency Circulation in Nigeria: A Comparison of Artificial Neural Network and Linear Regression Models

In the rapidly evolving landscape of financial technology, the assessment of the impact of ATM (Automated Teller Machine) and POS (Point of Sale) transactions on currency circulation in Nigeria has become a topic of paramount importance. POS (Point of Sales) is a payment method that allows individuals to make transactions like fund transfers and cash withdrawals without visiting banking halls or ATMs. However, charges apply depending on transaction volume. Nigeria, as one of the largest economies in Africa, has experienced a significant surge in the use of electronic payment methods, such as ATMs and POS terminals, in recent years [1, 2]. These technologies have transformed the way people conduct transactions, leading to a shift away from traditional cash-based transactions [3]. This transformation raises essential questions about its effects on the circulation of physical currency within the country.

ATMs and POS terminals have revolutionized the accessibility and convenience of financial transactions, making them a pivotal component of the modern banking and retail ecosystem [4]. ATM machines enable individuals to withdraw cash, check account balances, and make deposits at any time of day or night, reducing the need for physical bank visits [1]. On the other hand, POS terminals facilitate cashless transactions at various points of sale, further encouraging the use of digital payments [2]. These technological advancements not only enhance financial inclusion but also have the potential to influence the circulation of banknotes within the economy.

To decrease the amount of cash in circulation in Nigeria and promote the use of electronic-based systems or e-payment systems for transactions, the Central Bank of Nigeria (CBN) introduced the cashless policy initiative in 2011. The purpose was to help Nigeria reach its Vision 2020 objective of joining the top 20 economies on the planet. The program aimed to stimulate the use of electronic-based systems or e-payment systems for transactions and decrease the amount of currency in circulation [5, 6].

In many developed countries, cash remains the most common payment method at the point of sale (POS), but technological advancements have caused a decline in cash usage. Central banks are talking about the potential of using cash, but cashless payment is considered as a more viable substitute [7]. For the stability of the financial system to be ensured, to understand the demand for cash, and to assess the costs of payment systems, it is crucial to study the impact of POS payment on the use of cash [8]. Identifying potential trends in the demand for cash is a pertinent monetary policy issue.

While there is still a significant amount of cash in circulation, the current governor of the CBN, Godwin Emefiele, stated in an interview in February 2020 that the cashless policy would be reinstated in April 2020. This is demonstrated by the steadily rising volume of bills that the CBN prints each year [9, 10]. The situation surrounding the apex bank's potential reintroduction of the cashless policy was the impetus for this study.  Interestingly, a rise in the usage of electronic payment methods does not appear to have resulted in a decline in the use of cash. Nonetheless, as seen in Figure 1 below, the amount of currency in circulation has increased.

1.png

Figure 1. Nigerian currency in circulation

Source: CBN Annual Report for 2018

POS transactions have, in comparison to ATM terminals, contributed to the increased circulation of cash in Nigeria. This is primarily due to the greater accessibility of POS terminals for individuals across the country, even in remote rural areas where traditional banks and ATMs are scarce. In such areas, POS vendors play a crucial role by offering transaction services through these terminals. The driving force behind this study lies in the imperative to comprehend the impact of both POS and ATM transactions on the overall volume of currency in circulation within Nigeria. Remarkably, this research represents the pioneering endeavor of its kind in Nigeria since the stringent implementation of the cashless policy by the Central Bank of Nigeria. Furthermore, this study aims to expand the general public's understanding of how POS and ATM transactions can potentially curtail the circulation of physical cash. Thus, this study aims at studying the impact of ATM and POS transactions on currency in circulation in Nigeria using a multiple linear regression model and Artificial Neural Network. The Multilayer preceptor of ANN was adopted, with 70 percent of the data initially subjected to training and 30 percent for validity.

This study is organized into several sections to systematically explore and analyze the impact of ATM and POS transactions on currency circulation in Nigeria. The research will begin with an in-depth review of the literature to provide a comprehensive background on the subject. Subsequently, the methodology section will outline the data collection process, model selection, and analytical techniques employed. The results of the analysis will be presented and discussed, followed by conclusions, policy recommendations, and avenues for future research. By comparing Artificial Neural Network and Linear Regression models, this research aims to contribute valuable insights into the evolving financial landscape of Nigeria and inform decisions aimed at optimizing currency circulation strategies in an increasingly digitized economy.

3.1 Pearson product moment correlation

A correlation coefficient measures the degree of relationship that exists between two variables and it takes the range of value -1≤r≤+1.

The Pearson product moment correlation coefficient is given by:

$r=\frac{n \sum_{i=1}^n x y-\left(\sum_{i=1}^n X\right)\left(\sum_{i=1}^n Y\right)}{\sqrt{\left(n \sum_{i=1}^n X^2-\sum_{i=1}^n(X)^2\right)\left(n \sum_{i=1}^n Y^2-\sum_{i=1}^n(Y)^2\right)}}$      (1)

3.2 Multiple linear regression model

An improved predictive model can be obtained for a dependent variable by using more than one independent variable. Models that contain extra independent variables are called multiple regression models.

A multiple regression equation is given as:

$\hat{Y}=\beta_0+\beta_1 X_1+\beta_2 X_2+\beta_3 X_3+\beta_k X_k+e_{i j}$      (2)

where, X_k, X_k, X_k, …, X_k are the independent variables, β₀ the intercept, β₁…β_k are the parameters to be estimated and Y the dependent variable.

The above model is a system of n equations that can be expressed as:

$Y=X \beta+e_i$      (3)

$\mathrm{Y}=\left[\begin{array}{c}Y_1 \\ Y_2 \\ \cdot \\ \cdot \\ Y_k\end{array}\right]$      (4)

$X=\left[\begin{array}{ccccccc}1 & X_{11} & X_{12} & X_{13} & . & . & X_{1 k} \\ 1 & X_{21} & X_{22} & X_{23} & . & . & X_{2 k} \\ . & . & \cdot & \cdot & . & . & \cdot \\ . & . & \cdot & . & . & . & \cdot \\ 1 & X_{n 1} & X_{n 2} & X_{n 3} & . & . & X_{n k}\end{array}\right]$      (5)

$\beta=\left[\begin{array}{c}\beta_0 \\ \beta_1 \\ \beta_2 \\ \cdot \\ \cdot \\ \beta_k\end{array}\right]$     (6) 

$e=\left[\begin{array}{c}e_1 \\ e_2 \\ e_3 \\ \cdot \\ \cdot \\ e_k\end{array}\right]$      (7)

In matrix notation, the normal equations written as:

$X^1 X \hat{\beta}=X^1 Y$      (8)

$\hat{\beta}=\left(X^1 X\right)^{-1} X^1 Y$      (9)

which gives the least square estimate of β.

The general form of X¹X is:

$\left[\begin{array}{cccccc}n & \sum X_{11} & \sum X_2 & \sum X_3 & \cdots & \sum X_k \\ \sum X_1 & \sum X_1^2 & \sum X_1 X_2 & \sum X_1 X_3 & \cdots & \sum X_1 X_k \\ \sum X_2 & \sum X_1 X_2 & \sum X_2^2 & \sum X_2 X_3 & \cdots & \sum X_2 X_k \\ \cdot & \cdot & \cdot & \cdot & \cdots & \cdot \\ \sum X_k & \sum X_1 X_k & \sum X_2 X_k & \sum X_3 X_k & \cdots & \sum X_k{ }^2\end{array}\right]$      (10)

$X^1 Y=\left[\begin{array}{c}\sum Y_i \\ \sum X_1 Y \\ \sum X_2 Y \\ \sum X_k Y\end{array}\right]$      (11)

3.3 Artificial Neural Network (ANN)

This is a machine learning algorithm and a semi parametric method used for predictive modeling and studying the trend in a data set. Artificial Neural Network is a form of semi parametric method where the output is executed by a programmed nonlinear function on multiple inputs. Neurons in the neural network model re grouped into the input layer, hidden layer and output layer. A neural network processes information from one layer to another through an activation function. Feed-forward neural network architecture with one hidden layer is given in Figure 2. Figure 2 is a feed-forward neural network with one input layer having three nodes, one hidden layer having 4 nodes and one output layer having two nodes.

2.png

Figure 2. Feed-forward neural network architecture

Considering a feed-forward neural network with one hidden layer, the kth node in the hidden layer is defined as:

$t_k=f_k\left(\beta_{o k}+\sum_{i \rightarrow k} w_{i k} x_i\right)$      (12)

where, x_i the value of the ith input node is, f_k( ) is an activation function which is a logistic function.

$f_k(z)=\frac{\exp (z)}{1+\exp (z)}$      (13)

where, β_ok is the bias, the summation i→k means summing over all input nodes feeding to k, w_ik are the weights.

The node for the output layer is defined as:

$o=f_0\left(\alpha o_o+\sum_{j \rightarrow 0} w_{j o} h_j\right)$      (14)

The Artificial Neural Network used in this study is a multilayer preceptor consisting of one input layer, one hidden layer and one output layer. The MLP adopts a supervised learning approach called iteration or epoch for training. 70% of the data were initially used for training while 30% of the data were used for validity test along with hyperbolic tangent as the activation function. In order to improve the R-square, the amount of training data was increased to 61.4% while the testing data was increased to 38.6%.

This study is aimed at studying the impact of ATM and POS transactions on currency in circulation in Nigeria. The Pearson product moment correlation coefficient was used to assess the degree of relationship between currency in circulation and ATM transactions, currency in circulation and POS transactions, ATM transactions and POS transactions. In addition to this, a multiple linear regression model was also fitted on the data to in order to assess the impact of currency in circulation on ATM transactions and POS transactions. From the results obtained, it was observed that POS transactions have a significant impact on the amount of currency in circulation compared to ATM transactions. This is a clear indication that most Nigerians now patronize POS agents for their transactions such as cash withdrawal, cash deposits and so on. It can also be seen that a lot of Nigerians prefer the POS as a mode of transaction especially when it comes to cash withdrawal compared to the ATM terminals.

This was further studied using the Artificial Neural Network (ANN). The Multilayer preceptor of ANN was adopted, with 70 percent of the data initially subjected to training and 30 percent for validity. Thereafter, the amount of training data was increased to 61.4% and testing data was increased to 38.6% and it was observed that there was corresponding increase in the R-square. This shows that the R-square of the ANN can be improved by increasing the amount of training data and testing data. Ki et al (2017) showed that the R-square can be improved by increasing the amount of training data and reducing the number of hidden layers. However, it was observed in this study that the R-squared can be improved by increasing the training data and testing data. The ANN model which comprises the input layer, hidden layer and output layer shows that POS transactions account for 100% importance in the model.

In view of the above, the following recommendations have been made.

(1) There is need for more POS agents especially in rural areas considering the implementation of the cashless policy. The government can make loans more accessible for the small and medium scale enterprises so that people can access capital in order to have more POS operators across the country. Also, security measures must be put in place for the POS operators.

(2) The government through the central bank of Nigeria should come up with an empowerment scheme that will encourage Nigerians venture into the POS business as this will reduce poverty to some extent.

(3) Nigerian banks should improve the network services on their cards to avoid recurrent declined transactions on POS terminals.

(4) R-squares can be improved by increasing the amount of training data and testing data.

(5) Banks should improve the security architecture of their online services to avoid fraud on customers’ account. If the security of online transactions can be improved, individuals can seamlessly carryout transactions using their mobile devices without fear of their account being hacked.

If the above recommendations can be implemented, there will be an increase in the number of POS operators in the country which by extension will be creating jobs for the unemployed, there will be a significant improvement in the implementation of the cashless policy because individuals will be more confident carrying out seamless transactions online and by extension, it will greatly improve the country’s economy.