Predictions for Daily Cash Withdrawals at ATMs
Business Intelligence and Analytics Division
Introduction
Maintaining cash in ATMs (Automated Teller Machines) is a significant operating condition for commercial banks, inorder to fulfill customer demand and avoid penalties by central banks. According to Article 28 - Violations against regulations on bank card activities (Decree No. 96/2014/NĐ-CP) issued by State Bank of Vietnam (SBV):
The fines of VND 10,000,000 – 15,000,000 shall be applied to following violation: “Failing to supervise the remaining cash in an ATM, or fill the machine with cash to meet the withdrawal demands of the clients as demanded in related regulations”.
In addition, commercial banks’ motivations of forecasting daily cash demands are:
- Reducing total cost of cash (transport and replenishment, cash monitoring and ordering, opportunity costs).
- Increasing customer satisfaction and bank reputation.
Southeast Asia Joint Stock Commercial Bank (SeABank), established in 1994, is one of the earliest joint stock commercial banks and among Top 10 largest banks in Vietnam with nearly 160 offices at 30 big cities/provinces and 1,000 ATMs/POSs across the country. Targeting to become a model retail banker of Vietnam, SeABank provides a diversification of retail products and services tailored in accordance with demands and financial capability of each segment of individual as well as corporate customers nationwide. Therefore, it becomes really challenging and virtal for SeABank to maintain sufficient amount of cash in ATMs. To solve this problem, the bank need to create daily estimates for each ATM, which can result into “Out of Cash” or “Over Stock” situations. This requires a solution which can resonably predict daily cash demand needed by customers based on past transactional data.
Currently, SeABank has nearly 100 ATMs in Hanoi and most of them are located in central districts or business zones:
## user system elapsed
## 148.015 2.347 150.363
Daily cash withdrawals at ATMs can be shown as follows:
Motivations
Cash management for automated teller machines (ATMs) is a key service area for financial institutions such as banks. Cash-related costs constitute around 35%–60% of the overall costs of running an ATM. Studies using actual ATM investment data suggest that ATM usage has a positive impact on the cost efficiency of the banks. As the size and complexity of ATM networks increases, it becomes critical for financial institutions to optimize ATM cash flows to improve return on cash assets, reduce operation costs, and deliver high quality service to their customers. The factor that reduces the return on cash assets, referred to as the idle cash cost, is due to the more than necessary amount of cash residing in ATMs. Idle cash in ATM constitutes a cost to the financial institution since the institution cannot generate additional revenue by investments such as daily interest. In addition, Transfer of cash between cash center and ATM points is carried out by firms called “cash in transit (or CIT).” Banks pay the CIT a certain amount of money for each visit of an ATM and this payment constitutes the logistic costs, which are major com- ponents of operational costs. Optimal ATM cash management involves the analysis of idle cash cost and logistic cost. A vital, yet unexplored, issue in ATM cash management stems from the tradeoff between these costs: an ATM cash management system should minimize the overall idle cash and logistic cost while at the same time providing the customers with a quality of service by ensuring that ATMs do not run out of cash, i.e., by deciding on the optimum amount of money that should be placed in the ATMs to satisfy the customer demands.
For efficient cash management in an ATM network, a necessary amount of cash should be held in each ATM because having insufficient amount of cash leads to customer dissatis- faction. On the other hand, since the money held in an ATM is in cash, it is not possible for the banks to invest that money and generate additional income through daily interest rates. Therefore, having more than necessary amount of cash in the ATMs has a financial cost for the banks. Furthermore, the route of the CIT vehicles should be decided in an optimal way such that the cash collected from ATMs is delivered to the cash center (e.g., central bank) within working hours so that additional income can be generated through daily interest rates; otherwise, the cash is counted as idle.
CIT firms (or banks) carry out the delivery of cash to the ATMs; this action is referred to as the replenishment of the ATMs. Financial institutions such as banks pay the CIT firms a cer- tain amount of money for their service. We call this cost CIT cost. Daily replenishment of the ATMs decreases the customer dissatisfaction and the idle cash cost; however, it increases the CIT cost. On the other hand, replenishing the ATMs in long intervals decreases the CIT cost, but increases the idle cash cost. As a result, the frequency of ATM replenishment is an important decision.
Some banks typically maintain as much as 40% more cash than it is needed, even though many experts consider cash excess of 15% to 20% to be sufficient. Through cash management optimization, banks can avoid falling into the trap of maintaining too much or too little cash. Therefore, it is very important to apply proper forecasting methods of cash demand. Each bank, in order to ensure the continuity of the customers’ service, must determine the level of the daily limit of cash, understood as the minimal level of cash, in the bank, ensuring the continuity of the customers’ service. This limit should be based on estimation of cash flows in the bank. The part of this market is determined and realized on the basis of contracts between banks. However, for the majority of banks, a significant part of their cash turnover is the customers’ deposits and withdrawals. Determining the amount of cash flows can be estimated on the basis of the knowledge of the past customers’ behavior and historical transaction data.
Our Approach for Forecasting Daily Cash Demand
In the literature, techniques used for cash demand forecasting can be broadly classified into four groups:
- Time series methods that predict future cash need based on the past values of variable and/or past errors.
- Factor analysis method, which is based on the determination of various factors that influencethe cash demand pattern and calculating their correlation with actual cash withdrawals. In this group, the econometric models can be included.
- Fuzzy expert system approaches that try to imitate the reasoning of a human operator. The idea is to reduce the analogical thinking behind the intuitive forecasting to formal steps of logic.
- Machine learning approaches that map the relationships between various factors affecting thecash withdrawals and the actual cash withdrawals.
With purpose of forecasting daily cash demand at SeABank ATMs, BI Team use methods of time series analysis in conjunction with deep learning approach belonging to machine learning methods. Experiments were performed on time series representing bank customers’ past daily cash withdrawals stored in SeABank’s Database Management System from Jan 2018 to Aug 2018.
Results and Assessment
## [1] 67
## Connection successful!
##
## R is connected to the H2O cluster:
## H2O cluster uptime: 31 minutes 1 seconds
## H2O cluster timezone: Asia/Ho_Chi_Minh
## H2O data parsing timezone: UTC
## H2O cluster version: 3.20.0.8
## H2O cluster version age: 1 month and 9 days
## H2O cluster name: H2O_started_from_R_chidung_rwi275
## H2O cluster total nodes: 1
## H2O cluster total memory: 7.84 GB
## H2O cluster total cores: 32
## H2O cluster allowed cores: 32
## H2O cluster healthy: TRUE
## H2O Connection ip: localhost
## H2O Connection port: 54321
## H2O Connection proxy: NA
## H2O Internal Security: FALSE
## H2O API Extensions: XGBoost, Algos, AutoML, Core V3, Core V4
## R Version: R version 3.5.1 (2018-07-02)
Numeric actuals and predictions for daily cash withdrawals at this ATM can be shown as follows:
| 2018-04-28 |
85.13 |
67.18 |
| 2018-04-29 |
62.13 |
62.29 |
| 2018-04-30 |
27.38 |
27.72 |
| 2018-05-01 |
21.24 |
22.49 |
| 2018-05-02 |
65.50 |
65.31 |
| 2018-05-03 |
67.31 |
67.53 |
| 2018-05-04 |
17.34 |
18.28 |
| 2018-05-05 |
19.39 |
18.80 |
| 2018-05-06 |
16.55 |
18.75 |
| 2018-05-07 |
27.35 |
26.23 |
| 2018-05-08 |
12.87 |
10.83 |
| 2018-05-09 |
341.75 |
341.02 |
| 2018-05-10 |
406.48 |
403.37 |
| 2018-05-11 |
181.22 |
181.52 |
| 2018-05-12 |
181.20 |
181.64 |
Total actual cash demand vs predicted for 15 consecutive days:
| Cash Actual |
1532.846 |
| Cash Prediction |
1512.941 |
Conclusion and Future Work
Forecasting daily cash demand in ATMs is difficult due to unpredictability of withdrawals. Therefore, finding the best match between cash stock and demand becomes crucial to improve. This BI Team presented a deep learning appproach aimed at searching optimal strategies to refill ATM cash stocks to meet non-stationary of cash demand. Such a system will help the bank for proper and efficient cash management and can be scaled for all branches of a bank by incorporating historical data from these branches.
The predicted daily cash demand derived from our approach showed good results and proved that this approach is feasible in suggesting reliable rules able to improve historical cash management. However, there are other factors like festival period and market activities that unpredictively influence cash demand and if these are quantified and included as influencing variables, the result will improve with lesser error. The future work will be directed on implementing coordinated route planning technique for reducing the ATM network’s management costs and build an adaptive ATM cash management and support system.