Кwilinski Alex
25
www.virtual-economics.eu ISSN 2657-4047 (online)
Julia García Cabello
Virtual Economics, Vol. 3, No. 2, 2020
2020 Volume 3 Number 2 (April)
MONEY LEAKS IN BANKING ATM’S CASH-MANAGEMENT SYSTEMS
Julia García Cabello
Abstract. Some widely-accepted practices on banking ATM networks may negatively affect
efficient liquidity management. This paper analyses ATM cash management in light of
empirical evidence which suggests that banking ATMs tend to be overloaded beyond the
customer’s needs. This, in turn, results in high opportunity costs. While this is not perceived
by banks as particularly harmful, it might have a damaging impact on other business which
revolves exclusively around ATM networks, such as cashback sites. A dormant money case
may be solved by an appropriate tool matching the ATM’s cash to the user’s needs.
Supported by a large database of banking records, this paper also provides model validation
for a set of theorems previously developed by the author, resulting here in a cutting-edge,
reliable forecasting system, suitable for anticipating ATMs cash demand as well as coupling
with other supply chain planning processes.
Keywords: ATMs Cash Management, Stochastic Processes, Bank Data Processing, New
Methodology Tested, Cashback Sites
JEL Classification: C61, C63, G17, G21
Author:
Julia García Cabello
University of Granada, Campus Cartuja s/n, Granada, Spain, 18071
E-mail: cabello@ugr.es
https://orcid.org/0000-0003-0682-0678
Citation: García Cabello, J. (2020). Money Leaks in Banking ATM’s Cash-Management Systems. Virtual
Economics, 3(2), 25-42. https://doi.org/10.34021/ve.2020.03.02(2)
Received: January 24, 2020. Revised: March 12, 2020. Accepted: April 27, 2020.
© Author(s) 2020. Licensed under the Creative Commons License - Attribution 4.0 International (CC BY 4.0)
https://doi.org/10.34021/ve.2020.03.02(2)
https://creativecommons.org/licenses/by/4.0/
26
www.virtual-economics.eu ISSN 2657-4047 (online)
Julia García Cabello
Virtual Economics, Vol. 3, No. 2, 2020
1. Introduction
Cash management is one of the most important tasks performed by corporate firms as far as
it is the control key for handling crucial operations such as treasury administration and
working capital and mezzanine financing. Particularly in the banking sector, the financial
crisis which started in 2007, revealed that the liquidity was the most vulnerable aspect of the
banking system as long as the whole sector could be seriously affected by the crunch
whenever banks do not retain adequate safety liquidity levels. Actually, cash management is
of such paramount importance because a liquidity shortfall at a single institution can have
system-wide repercussions. By such crisis situations, the banking sector needs to incorporate
all technological knowledge which might involve a more efficient management of their liquid
resources: amongst other things, to meet the regulatory framework which fixes their safety
liquidity levels (Basel III rules).
As many authors claim, the branch efficiency study could significantly help improve the
global bank institution performance, (Camanho & Dyson, 1999), (Paradi & Zhu, 2013). Since
the concept of liquidity management covers a very broad spectrum of short/medium term
cash-based activities (cash management in ATMs is amongst these activities at the branch
level) this paper will focus on improvements in optimization of cash inventories at branch
ATM level under the premise that any improvement at the aggregate level has beneficial
repercussions on the global institution’s efficiency.
For what reasons is it recommendable to undertake a new revision of the costs of ATMs as
cash manipulation channels? It is mainly due to the spectacular increase in the number of
ATM machines. Indeed, forty years after the first ATM (called DAC, De La Rue Automatic
Cash System) a total of 3 million ATMs have been distributed across the length and breadth
of the whole world
1
. Then, although the introduction of ATMs along with other technological
innovations such as e-banking has reduced the management costs of bank liquid assets
(Valverde & Humphrey, 2009) the impressive usage of ATMs recommends upgrading this cash
supply channel. It should not be forgotten that the current situation of fierce competition
requires an effort on the part of the banking sector in order to oversee keeping costs. Bank
managers, however, may argue that current low interest rates mitigate the impact of these
potential losses. Even in that case, banks would still incur opportunity costs of not generating
profits if cash is invested in appropriate financial products.
But, apart from the banking case, there are other examples in which business revolves
exclusively around ATMs and, in consequence, such liquidity management is of capital
importance for keeping the company afloat. Exchange currency companies provide an
example of this. This is also the case of cashback sites. Cashback sites - physical and
websites- are currently a highly topical subject since they are being thought to be
1
Date research was conducted: March 29, 2019. Sources: ATMIA, National ATM Council, see
http://www.statisticbrain.com/atm-machine-statistics/
27
www.virtual-economics.eu ISSN 2657-4047 (online)
Julia García Cabello
Virtual Economics, Vol. 3, No. 2, 2020
implemented in Spain, although they have already been operating for a long time in some
countries like UK throughout supermarkets, post offices etc. Cashback sites offer services to
retail buyers under which a quantity of money (payout) is added to the total purchase price
of a transaction made by debit cards, in such a way that the customer receives the payout in
cash along with the purchase
2
. As cashback sites provide cash to the customers if required,
they act as ATMs: thus, they should anticipate uncertain demand without generating
dormant money to avoid opportunity costs. While opportunity costs are not perceived by
banks as particularly harmful, it might have a damaging impact on cashback sites.
The primary objective of this paper is to show that the cash management of an ATM network
as significant room for improvement is particularly related to some practices which may be
generating losses and opportunity costs. We mainly refer to overloading ATMs beyond the
real cash necessities. Actually, this paper analyses branch ATM cash management in light of
empirical evidence (database formed by real ATM-level records) showing a mismatch
between quantities of cash placed in the ATMs and real cash needs of ATM’s consumers.
Along with this problem, this paper attempts to provide a potential solution to overcome
dormant money a new methodology as a handy decision making a support system for cash
managers. The theoretical fundamentals of the proposed methodology, developed earlier by
the author of this paper in (García Cabello, 2013a), were conceived only as a set of theorems
based on stochastic jump processes together with a dynamic mathematical setting in order
to model the ATMs cash flow. In this paper, by means of the corresponding model validation,
it will be proved that this set of theorems, pointed in the right direction, become an effective
forecasting system for ATMs. As a matter of fact, the aim of this paper is twofold. First, we
warn of the pressing need to improve the ATM cash management by specifically being aware
of some widely-accepted practices which may result in inefficiencies. The second aim of this
paper is to promote this new forecasting system as long as it shows an immediate practical
relevance for management practitioners.
The key decision for the bank as far as its ATMs are concerned is how much cash to maintain
in that account from an initial overall sum to be loaded. It should be noticed that, despite the
fact that IT technologies are present at branches (commonly used from centralized IT
planning centers), the procedure to compute the initial amount of cash to be loaded into the
ATM strongly relies on manager’s expertise, who further fine-tune the centralized
predictions by taking into account the specific branch features derived from the local
demographics. And the manager’s expertise relies on historical data handling as part of the
branch’s routines. That means that the branch registers the cash quantity on particular
weeks (workable, holidays …) and the resulting outcome (cash exceed or cash shortage) and
copies the successful amounts. However, in the decision-making process, branch managers
make a decision with only incomplete information. As a result, the branches tend to
overload the ATMs to avoid refilling the ATMs more than once a day. In truth, the banking
2
For instance, a customer purchasing 8.99\euro worth of goods at a supermarket might ask for thirty euros
cashback. He would pay a total of 38.99\euro (8.99 + 30.00) with their debit card and receive 30\euro in cash
along with their goods.
28
www.virtual-economics.eu ISSN 2657-4047 (online)
Julia García Cabello
Virtual Economics, Vol. 3, No. 2, 2020
information processed in this paper suggests that this could happen. This banking
information is database of real banking records on ATMs data transactions. As a matter of
fact, about 250,000 excel multicolumn cells have been compiled. As far as the author knows,
this is one of the few times that such dataset has been addressed in the literature, due to the
strict rules on using real data transactions at branch level. A large database of banking
records has been also employed in order to validate the new methodology proposed in this
paper, showing it as a matching solution which adjusts ATM’s cash to user’s needs. Actually,
the model validation process has been carried out in parallel with the attempt to find out
that ATMs tend to be overloaded, proceeding by comparing the following three items:
a) Real banking data on cash loaded into ATMs;
b) Consumer’s real cash necessities;
c) Forecasted cash amounts obtained from the method which predicts the right
quantities of cash which should be loaded into the ATM in order to meet an uncertain
demand.
It should be noticed that this new methodology has been mainly intended and designed to
forecast future ATM cash needs by analyzing past needs. Hence, as its forecasting
mechanism is based on past branch data, which implicitly include specific ATM features
inside, this methodology does take into account such specific branch characteristics for each
case. On the one hand, this forecasting system is very precise with minimum human
intervention. On the other hand, it is very simple to be implemented in the branch daily
practices
3
assuring costs reductions. These characteristics allow this methodology to co-exist
with other technologies as a complement which supports branch manager’s decisions and
helps notably ameliorate the ATM cash management. Moreover, it has the potential to be
applied to other contexts apart from the banking environment (or cashback sites) providing,
thus, sustainable competitive advantage since, in general terms, forecasting demand is an
important issue in any supply chain planning process (Hill et al., 2015; Xu et al., 2020).
Actually, the use of technology to anticipate demand establishes accurate management to
get to know up-to-date information for procuring both demand forecasting and supply
management, specifically by early warning of potential oversupply or stock-outs. As a matter
of fact, the generality of the methodology employed greatly extends the range of its possible
industrial uses
4
.
3
Following (Hill et al, 2015), implementing systems and procedures based on forecasting systems in real
organizations should not be taken lightly. However, the methodology exposed in this paper -since it may be
carried out through an Excel sheet or easily converted into an algorithm directly throughout the banking
institution’s own computer services- should be both inexpensive and easily implementable in the daily branch
routine.
4
In subsequent papers, applications for computer components suppliers and electric utility industry will be
studied.
29
www.virtual-economics.eu ISSN 2657-4047 (online)
Julia García Cabello
Virtual Economics, Vol. 3, No. 2, 2020
This paper is organized as follows: in section 2, a literature review is carried out. Section 3 is
devoted to outlining the general formulation of the theoretical methodology stating its main
features as well as running throughout a small sample intended for illustrative purposes.
Section 4 is devoted to the numerical experiments. Finally section 5 concludes the paper.
2. Literature Review
The body of the banking literature points at liquidity management as an essential function of
banks. In particular, the seminal references to the role of financial intermediaries (Diamond
& Rajan, 2011) already paid attention to deposits as the main input for banks, as well as to
the consequences of random deposit withdrawals and the role of deposit insurance to
reduce the risk of bank runs. These models have evolved over time, as in (Barth et al., 2004),
and have been largely revisited after the financial crunch by 2007, given that liquidity
tensions have been a main concern in the banking industry and for financial stability in
general, see, for example (Bolt, 2010).
There has been also a strand of the literature dealing with the evolution of cash in the
economy, as well as on the impact that electronic payments and ATMs have on the demand
for currency. Humphrey et al. (2006) consider that transition from cash to electronic
payments could save around a 1% of GDP for a sample of 12 EU countries. Valverde &
Humphrey (2009) find similar gains for Spain. Attanasio et al. (2002) analyze the impact of
ATM transactions on the demand for currency. Additionally, other studies have shown the
importance of ATMs and cash in reducing the penetration of debit and credit cards in some
countries, as in (Valverde & Humphrey, 2009) for Spain.
From a bank management-level perspective there is, to the author’s knowledge, a more
limited number of studies dealing with efficiency improvements in liquidity management. In
particular, there is a paucity of research analyzing branch-level cash management. There are
only a few exceptions. The analysis of the early stages of ATM deployment in Greece and
empirical evidence found demonstrate that ATMs transactions could be improved only
through attracting new deposits, enlarging the ATM network and direct mail advertising
campaigns (Kouzelis, 1987). Other studies have directly focused on optimizing ATMs using
inventory models and, more recently, operational research techniques. Undertaken
simulations on how to optimize an ATM network found that up to 28% cost saving can be
achieved by improving the inventory policies and cash transportation decisions (Wagner,
2007).
As the problem stated and solved in this paper may be viewed as the optimal management
of an inventory of cash holdings within the bank’s ATM under uncertainty, models of supply
chain planning and inventory models should be mentioned. Some of them have relied on
supply management optimization techniques, as in (Alonso-Ayuso et al., 2003), where a
complete algorithmic approach for supply chain management under uncertainty is
developed. A good summary of these models can be found in (Osorio & Toro, 2012) who, in
turn, show that there are many similarities between cash supply chains and the typical
30
www.virtual-economics.eu ISSN 2657-4047 (online)
Julia García Cabello
Virtual Economics, Vol. 3, No. 2, 2020
chains for physical products. Castro (2009) follows an operational research perspective and
develops a solution to optimize the ATM cash management based on algorithms which
administer the cash in ATMs and banks. Finally, a more recent perspective has made use of
clustering and neural networks to forecast cash demand at ATMs. In particular, Venkatesh et
al. (2014) show that the cluster-wise cash demand forecast helps the bank’s top
management to design similar cash replenishment plans for all the ATMs in the same cluster.
This cluster-level replenishment plans could result in saving huge operational costs for ATMs
operating in a similar geographical region. Other papers of the existing literature focused on
ATMs are (Van der Heide et al., 2020), (Ekinci et al., 2015), (Jadwal et al., 2018) or (Teddy and
Ng, 2011) who optimize the ATM cash replenishment or develop different systems for
predicting the daily amounts withdrawn from ATM´s. Under the Inventory management
view, in (Naserabadi et al., 2014), an approach for an inventory system is developed. Other
approaches on ATM forecasting techniques are in (Darwish, 2013), where a brief summary of
the existing methods for cash forecasting are presented.
3. Methodology
This section, devoted to explaining the methodology used, is divided into two parts: the first
one provides an overview of those theoretical foundations which were conceived as a set of
theorems, developed in (García Cabello, 2013a). In next section, through the corresponding
model validation, it shall be proved that different ways of executing these theorems may
result in an accurate decision-making model for supporting ATM cash management. It should
be also mentioned that the theoretical foundations upon which the forecasting method
tested in this paper is based, do not specify how to determine the expected quantity of cash.
This leaves the door open to applying the method in several ways
5
. Hence, in the second part
of this section, one of the ways in which this method may work in practice beyond its
fundamentals shall be detailed. This will be carried out through a small sample (one week)
intended for illustrative purposes.
3.1. Fundamentals
In this section a brief summary of the theoretical setting developed in (García Cabello,
2013a), will be exposed. These are based on stochastic jump processes (compound Poisson
processes). With regard to the process of withdrawing cash from ATMs, there are two
stochastic unknowns: the number of ATM customers and the amounts of cash withdrawn
each time. The first one is described by means of the arrival process known as counting
process: if is the number of ATM users in the time interval (0, t), the main properties of
this arrival process considered as Poisson process with parameter λ are: the number of ATM
customers in a time lag (0, t) is a Poisson distribution with parameter λ·t: P* ] =
5
The best option to employ the method would depend on the (economic, social, demographic) circumstances
of each ATM bank branch. See the section 5 for complete explanations about possible methods of computing.
31
www.virtual-economics.eu ISSN 2657-4047 (online)
Julia García Cabello
Virtual Economics, Vol. 3, No. 2, 2020
provides the likelihood of having ATM customers throughout the time t. The
expected value and variance of are respectively E[ += λ·t, var[ += λ·t. Particularly, λ
represents the average of ATM withdrawals per day. As for the second stochastic unknown,
the amounts of cash withdrawn each time
, these are modelled by a compound Poisson
process:
∑
(1)
García Cabello (2013a) proved that the amount represents the quantity withdrawn by
the customers throughout the day. By the compound Poisson process properties, this is
equal to λ·E*
+ where λ is the average of the number of withdrawals per day whereas
E[
] stands for the average of quantity withdrawn from the ATM per day. Hence, if
represents the quantity to be loaded into the ATM at the beginning of the day, this may be
as follows:
[
]
(2)
This equation shall be at the heart of the subsequent model validation. Let it be noticed that
the proposed forecasting method Equation 2 does not predict ATM cash demand depending
on the total number of ATM arrivals but only on ATM arrivals in which money has been
withdrawn. That is, other arrivals without cash withdrawn at the ATM
6
, including an eventual
ATM failure, are not considered in Equation 2. The theoretical setting developed for ATMs in
(García Cabello, 2013a), on which the present model validation relies, was enlarged in
subsequent works for branches. Specifically, in (García Cabello, 2013b) a theoretical
programme of cash efficiency for bank’s branches is proposed thereby providing a significant
reduction of cash holdings at branches. In (García Cabello et al, 2017), an effective algorithm
to optimize branch cash holdings is designed as a cutting-edge methodology to enhance the
efficiency of bank branches regarding the liquidity management.
3.2. Employing the Method through a Short Sample
As mentioned before, the set of theorems developed in (García Cabello, 2013a; 2017) may
be applied in several ways in order to produce adequate forecasted amounts of cash. This
subsection is devoted to detailing one of these: in few words, database processing will be
made in such a way that inputs for each step are the mean of cash withdrawn in all previous
stages. Other ways of employing the method would select only a group of inputs instead of
6
Like paying routine bills, fees, taxes, printing bank statements, updating passbooks, transferring money
between linked accounts, purchasing tickets -concert tickets, lottery tickets, movie tickets, train tickets etc. -
and many other functions.
32
www.virtual-economics.eu ISSN 2657-4047 (online)
Julia García Cabello
Virtual Economics, Vol. 3, No. 2, 2020
using all of them, simulating some widely-accepted manager’s practices of clustering the
time into groups (weeks/months/years) of similar features
7
.
For illustrative purposes, a small sample (one week) will be processed. The banking
information comes from partial extracts of daily ATM cash count sheets corresponding to a
representative office in demographic and sociological terms of an emblematic Spanish bank
firm. Due to confidentiality arrangements we provide some general descriptive statistics.
Throughout this section, the banking partial extracts of daily ATM cash count sheet’s specific
terminology has been kept: particularly, the term return means withdrawals while the label
Total Delivered coincides with the real needs of cash delivered by the ATM at the end of the
day. In order to explain how the method may be employed in practice, we will carry out the
contrast amongst:
a) Banking data on quantities of cash charged into ATMs;
b) Users’ real cash needs;
c) ATM forecasts.
The final result will be displayed in Table 3. Previous proceedings in order to get final
computations are shown in following Table 1.
Table 1. Previous Proceedings on the ATM Forecasting Method Employment
Source: Developed by the author.
Now, by applying previous Equation 2,
7
That refers to those periods of time where spending increases (pre-holidays such as beginning of July or
December) or decreases (periods of austerity such as the so-called ‘hard January’)
Total Delivered
TD
Total Returns
TR
Average of quantity withdrawn
from ATM=
Day 1 10,090 € 104 97.01 €
Day 2 3,160 € 17 185.8 €
Day 3 3,980 € 34 117.05 €
Day 4 3,090 € 24 128.75 €
Day 5 5,050 € 51 99.01 €
Day 6 6,540 € 79 82.78 €
Day 7 1,320 € 17 77.64 €
33
www.virtual-economics.eu ISSN 2657-4047 (online)
Julia García Cabello
Virtual Economics, Vol. 3, No. 2, 2020
(3)
This makes the following amount: 5,242.54 €.
That is to say, by processing the small sample of banking records, the ATM forecasting
methodology produces an output of 5,242.54 €, which should be enough to satisfy the ATM
users’ demand for cash every day. Before contrasting the aforementioned forecasted
amount with real necessities for cash, let us pay attention to the following data:
Table 2. A Mismatch between Real Needs and Cash Loaded
Total Delivered Total Intro (Loaded)
Day 1 10,090 € 25,770 €
Day 2 3,160 € 47,100 €
Day 3 3,980 € 43,940 €
Day 4 3,090 € 39,960 €
Day 5 5,050 € 36,870 €
Day 6 6,540 € 31,820 €
Day 7 1,320 € 23, 680 €
33,230 €
Total Delivered
249,140 €
Total Intro
Source: Developed by the author.
If compared with the previous Table 2, here the big difference between the real daily needs
for cash (labeled Total Delivered) and the amounts of cash loaded into the ATM (labeled
Total Intro) is prominently displayed. Later on, when processing a large database of real
banking records (more than 250,000 excel multicolumn cells with information about urban
and rural ATMs), the mentioned hypothesis of overloading ATMs will be reinforced. This
alone should make it reasonable enough to revisit the current ATM cash management
procedures. In order to finally draw an overall comparison amongst a) provision of funds for
the ATMs on the banking firms (i.e., average of the Total Intro by the branch staff), b) ATM
users real cash needs (i.e., average of the ATM Total Delivered) and c) predictions of cash on
the proposed methodology, as the two first quantities are both averages per week, the
corresponding quantities should be now
34
www.virtual-economics.eu ISSN 2657-4047 (online)
Julia García Cabello
Virtual Economics, Vol. 3, No. 2, 2020
€ and
€, i.e., 35,591.43€ and 4,741.14€, as it is shown as the mentioned
global comparison is following Table 3:
Table 3. Overall Comparison
Average Total Delivered Average Total Intro Forecasted amount
4,741.14€ 35,591.40€ 5,242.54€
Source: Developed by the author.
Table 3 displays a mismatch between provision of funds and banking ATMs’ real needs for
cash (see the two items underlined). This mismatch is shown as well in Figure 1, where the
area between green and blue/red lines (blue and red lines practically coincide) represents
the ATM surplus cash. Incidentally, both Table 3 and Figure 1 also show the high level of
precision and reliability of the proposed ATM forecasting methodology:
Figure 1. The Comparative Graph
Source: Developed by the author.
4. Numerical Experiments
This section is devoted to performing some numerical tests aimed at validating the
theoretical model. It could be considered as one of the most important contribution of this
paper as it clarifies how to employ the model in practice (see also the next section, where
€0
€5 000
€10 000
€15 000
€20 000
€25 000
€30 000
€35 000
€40 000
€45 000
€50 000
day 1 day 2 day 3 day 4 day 5 day 6 day 7
Provision of funds for the ATM on the banking firms
ATM users real needs of cash
Predictions of cash of the García Cabello's program
35
www.virtual-economics.eu ISSN 2657-4047 (online)
Julia García Cabello
Virtual Economics, Vol. 3, No. 2, 2020
different ways of applying the model are discussed). The data set is formed by more than
250,000 daily ATM branch transactions of two different branches of an emblematic Spanish
bank. These experiments have been carried out as a sensitivity test on ATMs withdrawals for
two kinds of branches: urban and rural. Despite our initial data set was originally written
using the entity’s specific code, significant external operations have been
extracted/separated from those internal organizational orders (accounting entries) as part of
the database processing. To comply with legislation, the name of the bank must be kept
confidential. For both kinds of ATMs, in urban/rural locations, two graphs have been
developed: the first one (a bar chart) draws a comparison between the quantities of cash
charged into ATMs and the real needs for cash in order to define trends in ATMs practices as
the possible overload. The second graph (a diagram of functions) displays jointly the three
functions corresponding to:
a) banking data on quantities of cash loaded into ATMs;
b) real needs for cash;
c) ATM forecasts, aimed at establishing the degree of accuracy of the proposed
forecasting methodology for ATMs.
For all these graphs, the x axis shows months and the y axis displays cash amounts (in Euros).
As mentioned before, there is more than one way to employ the methodology, which would
be more or less suitable depending on the context. In order to be consistent with former
sections, database processing will be still made in such a way that each inputs iteration is the
average of cash withdrawn in the sum of previous stages. Let it be noticed that the
distinction between city and rural branches is the usual categorization of branches amongst
branch managers although it does not correspond only to demographic parameters. On the
contrary, it includes other factors like branch size, for example.
At this point, let us make a few comments on the branch size. The branch size in a notion
that represents somehow the branch’s solvency. Practitioners use a wide range of
parameters to delimit the size of a branch, the volume of loans, the maximum volume of
cash allowed to be stored or the number of staff being amongst the most used. Moreover, it
is a notion closely related to local demographics in the sense that the size of a branch
strongly depends on the number and the volume of branch cash transactions which, in turn,
depend on the branch clients’ needs for cash (which have a high level of dependence on the
clients’ demographic area). For the model validation, we consider two main categories of
branches: city center branches versus rural ones. According to the branch managers’ view,
the corresponding data sets (one per each of these categories) include the corresponding
demographic features inside to record the branch managers’ normal practice of grouping
branches according to their solvency, not according to their geographical location. This
means that branches which are geographically placed in rural locations may be treated by
practitioners as urban if their cash benchmarks exceed the corresponding values for rural
ones.
4.1 A City Center Branch
36
www.virtual-economics.eu ISSN 2657-4047 (online)
Julia García Cabello
Virtual Economics, Vol. 3, No. 2, 2020
According to experts, the main feature of urban bank branches ATMs is a constant and high
client flow, with over 50% of them not being regular customers. Branch ATM consumers’
habits are not therefore fully known. In Figure 2, the blue bars display real needs for cash,
while the red ones show the quantities loaded into ATMs. A huge difference can be observed
between both items.
Figure 2. City Center ATMs’ Overload
Source: Devoloped by the author.
Figure 3. The Overall Comparison of City Center ATMs
Source: Devoloped by the author.
0
200 000
400 000
600 000
800 000
1 000 000
1 200 000
1 400 000
Real needs of cash Cash loaded
0
200 000
400 000
600 000
800 000
1 000 000
1 200 000
1 400 000
Real needs of cash Cash loaded Forecasted amounts
37
www.virtual-economics.eu ISSN 2657-4047 (online)
Julia García Cabello
Virtual Economics, Vol. 3, No. 2, 2020
The same conclusion is reached as far as Figure 3 is concerned, where, additionally, a high
degree of coincidence may be observed between the forecasted amounts of cash and the
real needs. Importantly, Figure 3 shows a small distortion of forecasts (January, beginning of
February) corresponding to the so-called ‘hard January’. This seasonal dissimilitude is
expected by the city-center branch managers.
4.2. A Rural Site Branch
The main features of ATM bank branches located in rural areas are a constant and
medium/low client flow, with less than 20% not being regular customers. Consequently,
branch consumer habits are well known by the branch staff. The withdrawals flow is
homogeneous with medium/low level of cash quantity.
Figure 4 and Figure 5 provide findings similar to those of the urban case: firstly, the
mismatch between quantities of cash placed in the ATMs and real needs for cash of ATMs’
consumers; and secondly, a high accuracy level of the proposed methodology in forecasting
the quantity of cash to be loaded into the ATM in response to an uncertain demand for cash.
One further conclusion may be drawn: in rural locations, the ATM consumers’ habits tend to
be more homogeneous than in urban areas. This is the probable reason as to why such
branch managers do not overload ATMs as disproportionately as those in the urban
locations.
Figure 4. Rural ATMs Overload
Source: Developed by the author.
0
10 000
20 000
30 000
40 000
50 000
60 000
70 000
80 000
90 000
Real needs of cash
Cash loaded
38
www.virtual-economics.eu ISSN 2657-4047 (online)
Julia García Cabello
Virtual Economics, Vol. 3, No. 2, 2020
Similar to Figure 3, in Figure 5 there is a small deviation between forecasts and real needs for
cash. It is caused by the population increase in rural areas due to their travelling from city
centers to rural districts (holidays / vacations, July, August). Seasonal varieties are expected
by the rural branch managers.
Figure 5. The Overall Comparison for Rural ATMs
Source: Developed by the author.
5. Exploring the Forecasting Method Opportunities
The precise way of computing used in the model validation is detailed when employing the
method throughout a small sample (see section 3.2). However, the existence of alternative
ways of computing is suggested. This is because, as mentioned earlier, the set of seminal
theorems at the heart of the proposed methodology does not specify how to compute the
forecasted amounts for ATMs leaving the door open to several opportunities, whose
effectiveness may be tested in different scenarios.
Let us remind that the forecasting model (summarized in the central Equation 2) provides
the explicit formula for determining the expected amount of case by means of two
unknowns: i) withdrawals and ii) quantities withdrawn per day. Of course, the days from
which these data are extracted (i.e., the reference days) must be prior to the forecasting day.
The fact that these reference days may be chosen following several methods, open many
opportunities of computing according to the needs. Some ways of computation are listed
below as well as some of their intrinsic characteristic (such as learning capabilities, i.e.,
progressively improving performance) which would help to identify those contexts where
such way of computing would fit better:
0
10 000
20 000
30 000
40 000
50 000
60 000
70 000
80 000
90 000
Real needs of cash Cash loaded Forecasted amounts
39
www.virtual-economics.eu ISSN 2657-4047 (online)
Julia García Cabello
Virtual Economics, Vol. 3, No. 2, 2020
- The last day: to use for the current day the data extracted from the last one. This
method assumes that all days are similar. This way of computation would be suitable for
branches with not too many peaks and falls.
- The last similar day: to use for the current day the data extracted from the last similar
one where similar means with similar specific features. In such case, days are grouped
(clustering) depending on specific features such as work days, holidays, etc. in order to take
samples from the corresponding cluster. This way of computation would be suitable for
those branches with more extreme swings.
- An accumulated average: to use the historical average of ATM cash needs. This
method assumes that all days are equal and will never account for extreme values. It will
however, slowly adapt to rising or decreasing needs since the corresponding temporal
sequence has steps which become broader. Here, the outputs starting at that give rise to
more information (on both withdrawals and quantities withdrawn) are used as inputs for the
next steps. Thus, these outputs meet a temporal sequence which becomes larger with each
new iteration in such a way that the cumulative error becomes smaller. That is, the proposed
methodology has learning capabilities if performed this way.
- An accumulated average with the initial learning period: A modification of the former
because during the initial period, the average is very sensible to extreme variations.
Both an accumulated-average and an accumulated-average-with-initial-learning-period
methods would be suitable versions for branches with large volumes of ATMs transactions.
These are general guidelines while exploring potential fine-tuning of the proposed method
should be carried out by testing the procedure with the real data of each kind of a branch
(see Conclusions section for further details). Thus, in addition to the versatility in employing
the method, other further fine-tunings could be implemented in order to fit best the
characteristic of each scenario. As a matter of fact, each ATM location represents itself a
particular scenario whose set of features ranges from the market conditions to the special
conditions of the site where the ATM is located.
6. Conclusions
The employment of ATMs network as an additional alternative to cash window has spread
enormously amongst the bank entities’ users now reaching massive proportions. This alone
should be a reason significant enough to revisit the current ATM cash management
procedures in order to detect money leaks. Moreover, in Spain within a foreseeable period
of time, the new companies’ establishing (cashback sites) may have a high chance of
occurrence.
Cashback sites would act as ATMs by offering services to retail customers while providing
cash added to the total purchase price of the debit card transactions, as shown in Figure 6. A
similar provision exists, with regard to other companies which use ATMs machines to expend
40
www.virtual-economics.eu ISSN 2657-4047 (online)
Julia García Cabello
Virtual Economics, Vol. 3, No. 2, 2020
money, such as exchange currency companies. All these settings should anticipate uncertain
demand without generating dormant money to avoid opportunity costs. While opportunity
costs are not perceived by banks as particularly harmful, it might have a damaging impact on
cashback sites or similar companies.
Figure 6. Cashback Сompanies
Source: Developed by the author.
This paper puts on the table the pressing necessity of enhancing ATMs performance as well
as learning from possible inefficient ATM branching practices such as overloading the ATMs
beyond the users’ needs. As a matter of fact, the large dataset formed by real banking
information used in this paper suggest that it is probably the common practice for banks to
overload ATMs with cash, which, in turn, can generate large losses and opportunity costs.
Along with this problem, this paper aims to provide a potential solution to ameliorate
banking cash management by optimizing the ATM replenishments through a cutting-edge
methodology matching the ATMs’ cash with the users’ needs. The tests performed in this
paper (through large ATM database in order to reduce noise as much as possible) show this
methodology as sound and reliable. Actually, our findings demonstrate that the proposed
method may significantly reduce the mismatch between provision of funds for the ATMs and
the ATM users’ real needs for cash. Furthermore, our approach is a non-expense-based
methodology aimed at co-existing with other IT technologies as an extra decision support
system for practitioners. It has also the potential to be applied to other contexts apart from
the banking environment, thus, providing a sustainable competitive advantage. As
mentioned along the paper, the set of seminal theorems at the proposed methodology’s
heart does not specify how to compute the forecasted amounts for ATMs, leaving the door
open to several opportunities, whose effectiveness may be tested in different scenarios.
41
www.virtual-economics.eu ISSN 2657-4047 (online)
Julia García Cabello
Virtual Economics, Vol. 3, No. 2, 2020
7. Acknowledgements
The financial support from the Spanish Ministry of Science and Innovation ``Regulación
Financiera y Sector Bancario en Tiempos de Inestabilidad: Mecanismos de Prevención y
Resolución de la Crisis'' (ECO2014-59584-P), Junta de Andalucía ``Excellence Groups''
(P12.SEJ.2463), and Junta de Andalucía (SEJ340) is gratefully acknowledged. Declarations of
interest: none.
References
Alonso-Ayuso, A., Escudero, L.F., Garín, A., Ortuño, M., & Pérez, T. (2003). An Approach for Strategic
Supply Chain Planning under Uncertainty Based on Stochastic 0-1 Programming. Journal of
Global Optimization, 26, 97-124. https://doi.org/10.1023/A:1023071216923
Attanasio, O.P., Guiso, L.M., & Japelli,T. (2002). The Demand for Money, Financial Innovation, and the
Welfare Cost of Inflation: An Analysis with Household Data. Journal of Political Economy, 110(2),
317-351. http://dx.doi.org/10.1086/338743
Barth, J.R., Caprio, G., & Levine, R. (2004). Bank Regulation and Supervision: What Works Best?
Journal of Financial Intermediation, 13, 205-248. https://doi.org/10.1016/j.jfi.2003.06.002
Bolt, W., & Humphrey, D.B. (2010). Bank Competition Efficiency in Europe: A Frontier Approach.
Journal of Banking & Finance, 34(8), 1808-1817. https://doi.org/10.1016/j.jbankfin.2009.09.019
Camanho, A.S., & Dyson, R.G. (1999, September). Efficiency, Size, Benchmarks and Targets for Bank
Branches: An Application of Data Envelopment Analysis. The Journal of the Operational Research
Society, 50(9), 903-915. https://doi.org/10.2307/3010188
Castro, J. (2009). A Stochastic Programming Approach to Cash Management in Banking. European
Journal of Operational Research, 192(3), 963-974. https://doi.org/10.1016/j.ejor.2007.10.015
Darwish, S.M. (2013). A Methodology to Improve Cash Demand Forecasting for ATM Network.
International Journal of Computer and Electrical Engineering, 5(4), 405-409.
https://doi.org/10.7763/IJCEE.2013.V5.741
Diamond, D.W., & Rajan, R.G. (2011, May). Fear of Fire Sales, Illiquidity Seeking, and Credit Freezes.
The Quarterly Journal of Economics, 126(2), 557-591. https://doi.org/10.1093/qje/qjr012
Ekinci, Y., Lu, J.Ch., & Duman, E. (2015, May 1). Optimization of ATM Cash Replenishment with
Group-Demand Forecasts. Expert Systems with Applications, 42(7), 3480-3490.
https://doi.org/10.1016/j.eswa.2014.12.011
García Cabello, J. (2013a). An Efficient Liquidity Management for ATMs. Aestimatio, The IEB
International Journal of Finance, 6, 50-75.
García Cabello, J. (2013b). Cash Efficiency for Bank Branches. Springerplus, 2, 334.
https://doi.org/10.1186/2193-1801-2-334
42
www.virtual-economics.eu ISSN 2657-4047 (online)
Julia García Cabello
Virtual Economics, Vol. 3, No. 2, 2020
García Cabello, J., & Lobillo, F. (2017). Sound Branch Cash Management for Less: A Low-Cost
Forecasting Algorithm under Uncertain Demand. Omega, 70, 118-134.
https://doi.org/10.1016/j.omega.2016.09.005
Hill, A.V., Zhang, W., & Burch, G.F. (2015). Forecasting the Forecastability Quotient for Inventory
Management. International Journal of Forecasting, 31(3), 651-663.
https://doi.org/10.1016/j.ijforecast.2014.10.006
Humphrey, D.B., Willesson, M., Bergendahl, G., & Lindblom, T. (2006). Benefits from a Changing
Payment Technology in European Banking. Journal of Banking & Finance, 30(6), 1631-1652.
https://doi.org/10.1016/j.jbankfin.2005.09.009
Jadwal, P.K., Jain, S., Gupta, U., & Khanna, P. (2018). K-Means Clustering with Neural Networks for
ATM Cash Repository Prediction. In S. Satapathy & A. Joshi (Eds.), Information and
Communication Technology for Intelligent Systems (ICTIS 2017) - Volume 1. ICTIS 2017. Smart
Innovation, Systems and Technologies, vol 83 (pp. 588-596). Cham, Switzerland: Springer
International Publishing. https://doi.org/10.1007/978-3-319-63673-3_71
Kouzelis, A. (1987, June). On the Determinants of ATM Performance. European Journal of Operational
Research, 30(1), 89-94. https://doi.org/10.1016/0377-2217(87)90015-4
Naserabadi, B., & Mirzazadeh, A., & Nodoust, S. (2014). A New Mathematical Inventory Model with
Stochastic and Fuzzy Deterioration Rate under Inflation. Chinese Journal of Engineering, Article ID
347857, 1-10. http://dx.doi.org/10.1155/2014/347857
Osorio, A.F., & Toro, H.H. (2012). A MIP model to optimize a Columbian cash supply chain.
International Transactions in Operational Research, 19(5), 659-673.
https://doi.org/10.1111/j.1475-3995.2011.00850.x
Paradi, J.C., & Zhu, H. (2013). A Survey on Bank Branch Efficiency and Performance Research with
Data Envelopment Analysis. Omega, 41(1), 61-79. https://doi.org/10.1016/j.omega.2011.08.010
Teddy, S.D., & Ng, S.K. (2011, July-September). Forecasting ATM Cash Demands Using a Local
Learning Model of Cerebellar Associative Memory Network. International Journal of Forecasting,
27(3), 760-776. https://doi.org/10.1016/j.ijforecast.2010.02.013
Valverde , S.C., & Humphrey, D.B. (2009). Technological Innovation in Banking: The Shift to ATM’s and
Implicit Pricing of Network Convenience. In L. Anderloni, D. T. Llewellyn, & R. H. Schmidt (Eds.),
Financial Innovation in Retail and Corporate Banking (pp. 89-110). Cheltenham, United Kingdom:
Edward Elgar Publishing. https://doi.org/10.4337/9781848447189.00010
Van der Heide, L.M., Coelho, L.C., Vis, I.F.A., & Van Anholt, R.G. (2020, February). Replenishment and
Denomination Mix of Automated Teller Machines with Dynamic Forecast Demands. Computers
and Operations Research, 114, 104828. https://doi.org/10.1016/j.cor.2019.104828
Venkatesh, K., Ravi, V., Prinzie, A., & Van den Poel, D. (2014, January 16). Cash Demand Forecasting in
Atms by Clustering and Neural Networks. European Journal of Operational Research, 232(2), 383-
392. https://doi.org/10.1016/j.ejor.2013.07.027
Wagner, M. (2007). The Optimal Cash Deployment Strategy-Modelling a Network of Atms. M.Sc.
thesis. Finland: Swedish School of Economics and Business Administration.
Xu, Z., Elomri, A., Zhang, Q., Liu, C., & Shi, L. (2020). Status Review and Research Strategies on
Product-Service Supply Chain. Proceedings of the Institution of Mechanical Engineers, Part B:
Journal of Engineering Manufacture, 234(8), 1075-1086.
https://doi.org/10.1177/0954405420905199
javascript:void(0)
javascript:void(0)
https://www.scopus.com/authid/detail.uri?origin=resultslist&authorId=57195488930&zone=
https://www.scopus.com/authid/detail.uri?origin=resultslist&authorId=55437035500&zone=
https://www.scopus.com/authid/detail.uri?origin=resultslist&authorId=55605779560&zone=
https://www.scopus.com/authid/detail.uri?origin=resultslist&authorId=56530226200&zone=
https://doi.org/10.1016/0377-2217(87)90015-4
https://www.scopus.com/authid/detail.uri?origin=resultslist&authorId=57211441479&zone=
https://www.scopus.com/authid/detail.uri?origin=resultslist&authorId=54890623600&zone=
https://www.scopus.com/record/display.uri?eid=2-s2.0-85073982702&origin=resultslist&sort=plf-f&src=s&st1=forecasting+cash+ATM+&st2=&sid=b75e39ef3f078d01401202486b85ed7c&sot=b&sdt=b&sl=36&s=TITLE-ABS-KEY%28forecasting+cash+ATM+%29&relpos=0&citeCnt=0&searchTerm=
https://www.scopus.com/record/display.uri?eid=2-s2.0-85073982702&origin=resultslist&sort=plf-f&src=s&st1=forecasting+cash+ATM+&st2=&sid=b75e39ef3f078d01401202486b85ed7c&sot=b&sdt=b&sl=36&s=TITLE-ABS-KEY%28forecasting+cash+ATM+%29&relpos=0&citeCnt=0&searchTerm=
https://www.scopus.com/sourceid/24355?origin=resultslist
https://www.scopus.com/sourceid/24355?origin=resultslist
https://www.scopus.com/authid/detail.uri?origin=resultslist&authorId=54791755100&zone=
https://www.scopus.com/authid/detail.uri?origin=resultslist&authorId=56915869700&zone=
https://www.scopus.com/authid/detail.uri?origin=resultslist&authorId=55371383400&zone=
https://www.scopus.com/record/display.uri?eid=2-s2.0-85081976769&origin=resultslist&sort=plf-f&src=s&st1=demand+forecasting+supply+chain&nlo=&nlr=&nls=&sid=3d71fa07bf21877888c13145fa059614&sot=q&sdt=b&sl=51&s=TITLE-ABS-KEY-AUTH%28demand+forecasting+supply+chain%29&relpos=3&citeCnt=0&searchTerm=
https://www.scopus.com/record/display.uri?eid=2-s2.0-85081976769&origin=resultslist&sort=plf-f&src=s&st1=demand+forecasting+supply+chain&nlo=&nlr=&nls=&sid=3d71fa07bf21877888c13145fa059614&sot=q&sdt=b&sl=51&s=TITLE-ABS-KEY-AUTH%28demand+forecasting+supply+chain%29&relpos=3&citeCnt=0&searchTerm=
https://www.scopus.com/sourceid/20406?origin=resultslist
https://www.scopus.com/sourceid/20406?origin=resultslist