International Journal of Accounting and Information Management

Emerald Article: Integrating activity-based costing with simulation and data mining H. Kostakis, C. Sarigiannidis, B. Boutsinas, K. Varvakis, V. Tampakas

Article information:
To cite this document: H. Kostakis, C. Sarigiannidis, B. Boutsinas, K. Varvakis, V. Tampakas, (2008),"Integrating activity-based costing with simulation and data mining", International Journal of Accounting and Information Management, Vol. 16 Iss: 1 pp. 25 35 Permanent link to this document: http://dx.doi.org/10.1108/18347640810887744 Downloaded on: 04-04-2012 References: This document contains references to 19 other documents To copy this document: permissions@emeraldinsight.com This document has been downloaded 1366 times.

Access to this document was granted through an Emerald subscription provided by UNIVERSIDADE FEDERAL DA PARAIBA For Authors: If you would like to write for this, or any other Emerald publication, then please use our Emerald for Authors service. Information about how to choose which publication to write for and submission guidelines are available for all. Additional help for authors is available for Emerald subscribers. Please visit www.emeraldinsight.com/authors for more information. About Emerald www.emeraldinsight.com With over forty years' experience, Emerald Group Publishing is a leading independent publisher of global research with impact in business, society, public policy and education. In total, Emerald publishes over 275 journals and more than 130 book series, as well as an extensive range of online products and services. Emerald is both COUNTER 3 and TRANSFER compliant. The organization is a partner of the Committee on Publication Ethics (COPE) and also works with Portico and the LOCKSS initiative for digital archive preservation.
*Related content and download information correct at time of download.

The current issue and full text archive of this journal is available at www.emeraldinsight.com/1834-7649.htm

Integrating activity-based costing with simulation and data mining

H. Kostakis
Department of Business Administration, University of Patras, Patras, Greece

Simulation and data mining

25

C. Sarigiannidis
Department of Accounting, Technological Educational Institute of Patras, Patras, Greece

B. Boutsinas
Department of Business Administration, University of Patras, Patras, Greece and Articial Intelligence Research Center, University of Patras, Patras, Greece

K. Varvakis
Department of Business Administration, University of the Aegean, Mytilene, Greece, and

V. Tampakas
Department of Accounting, Technological Educational Institute of Patras, Patras, Greece
Abstract
Purpose This paper aims to present a methodology for activity-based costing, which combines simulation modeling and association rule mining, one of the core data-mining techniques. The objective of the proposed methodology is to deal with the problem of dening cost drivers. Design/methodology/approach Activity-based costing uses the output produced by the simulation of cost drivers as inputs. As opposed to the integration of the ABC technique with simulation modeling, the possibility of estimating an empirical distribution of the simulated cost drivers does not exist in the proposed methodology. This is achieved with the use of data-mining techniques and is based on the proposition that, if an association is found between a cost driver, whose estimation or calculation is time-consuming, and another cost driver, which can easily be estimated or calculated, then the latter can lead to the estimation or calculation of the former. Findings The extracted association rules correspond to existing dependencies between the cost drivers. Originality/value The paper presents a combined methodology to deal with the problem of dening cost drivers in activity-based costing. An example of the proposed methodology in healthcare is also presented. Keywords Activity based costs, Simulation, Data handling, Information retrieval Paper type Research paper

Introduction ABC is a cost accounting methodology, aimed at allocating overhead costs properly, unlike traditional cost systems that misallocate them. Traditional cost systems were
The project is co-funded by the European Social Fund and National Resources EPEAEK II.

International Journal of Accounting and Information Management Vol. 16 No. 1, 2008 pp. 25-35 q Emerald Group Publishing Limited 1834-7649 DOI 10.1108/18347640810887744

IJAIM 16,1

26

proper at times when direct labor costs represented 40-60 percent of the manufacturing cost of a product. However, now-a-days, the increase of overhead costs and to the decrease of direct labor and material costs justies the use of ABC (Cokins et al., 1993). Moreover, ABC can offer decision-making support to management. ABC traces costs by using resource and activity cost drivers that reveal activities and objects consumption patterns on the basis of a cause and effect relationship. The selection of a cost driver is one of the most difcult tasks in ABC. For instance, the survey conducted by Cobb et al. (1992) in 1991 on 21 companies revealed that the respondents doubted whether cost drivers, which can reect a causal relationship between cost objects and activities, could be found. As the survey pointed out, it seems that some companies do not succeed in translating theory into action because they cannot cope with problems such as: data collection, identication of activities, lack of resources and selection of cost drivers (Cobb et al., 1992). One of the most common methods the cost system designers use to identify activities and to select cost drivers is interviewing the heads of departments. The ABC designers also collect data needed to link resource costs to activities through labor reporting systems, work order systems, employee surveys, observation, timekeeping systems and storyboards (Kaplan and Cooper, 1998). The conclusion of Velcu (2002) is that some data, like cost driver information, can be more easily picked up from the organizational information system of the company, such as ERP systems. However, despite the companys technological equipment that facilitates data collection and the selection of cost drivers from the information system, the major problem dwells on collecting the appropriate information based on the employees interviews (Velcu, 2002). In order to overcome observed bottlenecks in the relevant ABC literature, this paper places an emphasis on the introduction of an effective methodology aimed to deal with the problem of dening cost drivers in ABC models where real data are not available. The new methodology serves as a tool to improve cost estimates in an activity-based costing context. The objectives of this paper are as follows: . to review existing literature covering the estimation of cost drivers in ABC models; . to present the drawbacks of current methodologies estimating cost drivers; and . to apply a new methodology for effectively addressing and optimizing the accuracy of estimating cost drivers in a medical domain. Basic techniques Activity-based costing Business resources, activities and cost objects are the three building blocks of ABC. Business resources can include people, machines, computers, equipment, technology, supplies, etc. Business activities are those performed in an organization to satisfy customers requirements. Business activities are best dened using an active verb and object convention (Cokins, 1996). Cost objects are products, customers or services. After the identication of resources, activities and cost objects, ABC establishes a transitional mapping between them. More specically, the design of an ABC system

involves two phases. During the rst phase, resource costs are mapped to activities through resource cost drivers, taking into consideration the resources consumed by each activity performed. During the second phase, activity costs are mapped to the cost objects that use the activities, through activity cost drivers. During the two stages of the aforementioned methodology, relevant data might not be available. These data include the proportions of each resource consumed by each activity and the level of usage of every activity in the production of the nal products/services. This means that, when real data are not available, the level (or proportion) of the cost of each activity consumed by each product/service must be estimated (Wiersema, 1995). This is done to achieve the greatest possible accuracy in the cost estimation of the nal products/services. Three levels of data accuracy can be used to estimate the aforementioned proportions and levels: educated guess, systematic appraisal and collection of real data. An educated guess is used when real data cannot be obtained or data collection efforts cannot be nancially justied. Data are collected from experts opinions and are combined to make estimates. The method requires less time but is also less accurate. It is suggested that when people estimate how much time they spent on a list of activities, they report percentages that add up to 100 percent and do not record idle or unused time. Therefore, almost all ABC systems calculate cost driver rates assuming that resources work at full capacity (Anderson and Kaplan, 2007). The systematic appraisal method is a more scientic way to obtain proportions. It uses scientic methods, such as analytic hierarchical processing, to pull subjective individual opinions into more representative information. This method is capable of allocating expenses from cost pools to each product/service, whereas the appropriate cost driver estimation is needed in order to achieve the desired level of accuracy. The method is more accurate than the previous; however, it requires personnel with analytical skills as well as specialized computer software. It also adds to computational time. The third method, which is most accurate, is the actual data collection. The method requires systematic data collection (sampling), followed by analysis of data using statistical methods. The main drawback of this method is the cost associated with its implementation. The paper introduces a new tool for estimating the values of cost drivers in activity-based costing. In particular, it is based on the proposition that if an association is found between a cost driver(s), whose estimation is either difcult or time-consuming, with another cost driver, which can easily be estimated, then the latter can lead us to the estimation of the former. In the relevant literature, distributions for all cost drivers are needed in order to make cost estimates (Beck and Nowak, 2000). The proposed methodology requires information regarding the estimation of only a few cost drivers and it can be used for the estimation of unknown distributions of cost drivers due to a lack of relevant data. In regards to of the above, the proposed methodology is complementary to the three methods of estimating cost drivers, namely educated guess, systematic appraisal and actual data collection. This means that, it cannot substitute for the three levels of data accuracy, as it is necessary to estimate at least one cost driver. The application of the methodology implies that considerable time is saved, as data collection for all cost drivers with the use of current methods can be extremely difcult and time-consuming.

Simulation and data mining

27

IJAIM 16,1

28

Simulation and association rule mining, ARM As already mentioned, simulation is also used in the proposed methodology. Simulation is mimicking a system over a period of time, usually with the aid of a computer. Computer simulation methods have developed since the early 1960s and may well be the most commonly used analytical tools of management science. The basic simulation principles are as follows (Pidd, 1998): . The analyst builds a model of the system under study. . The analyst writes a computer program embodying the model. . The analyst uses a computer to imitate the systems behavior when subjected to a variety of operating policies. Thus, the most desirable policy may be selected. Simulation is used as a tool for experimenting on a computer-based model of a system, usually in the form of trial and error. The aim is to present the effects of various policies. The policies that produce the best results in the model are implemented in the real system. Simulation models can be divided in various categories, depending on the nature of the system under simulation (stochastic or deterministic, discrete- or continuous-time), its evolvement over time (static or dynamic), time handling (time-slicing or next-event), etc. Finally, association rule mining (ARM), one of the core data mining techniques, is used in the proposed methodology. Data mining allows the user to discover the important relations in data using knowledge discovery methods, while the algorithms for ARM extract dependencies between factors in databases (Agrawal et al., 1993). The fundamental application of ARM is in market basket analysis, where the purchase of a list of items is used as a single transaction. The objective is to discover trends/dependencies/relations among a series of transactions, which in turn could be analyzed to reveal patterns of useful as information (Witten and Frank, 2001). An example of an association rule might be that 70 percent of customers that purchase bread also purchase butter, and that 60 percent of the transactions contain bread and butter. An association rule is an implication of the form X $ Y, where X and Y are a set of values. The rule X $ Y holds in the transaction set with condence c, if c% of transactions that contain X also contain Y. The rule X $ Y has support s in the transaction set if s% of transactions contain X and Y. The purpose of mining association rules is to generate all association rules that have support and condence greater than the user-specied minimum support and minimum condence. The proposed methodology integrates simulation and ARM in an activity-based costing context. Proposed methodology Activity-based costing has already been combined with simulation (Beck and Nowak, 2000), where the integration of the ABC technique with simulation modeling presupposes the possibility of estimating an empirical distribution of the simulated cost drivers. Discrete-event simulation generates values for the cost drivers, which are used in ABC to produce condence interval estimations. Thus, the system produces a series of estimations for the cost, while the conditions of the system vary. The proposed methodology was developed with the aim to overcome observed bottlenecks in dening cost drivers as it appears in the existing activity-based costing literature. In the proposed methodology, it is found necessary to have an empirical data

distribution of each cost driver to be simulated and applied to the interview and survey process. This is also true for any other cost driver calculation method, such as systematic appraisal and collection of real data only to a limited number of cost drivers. This is achieved with the use of ARM, a method which has not been applied in activity-based costing in such a context before. Association rule mining is used to identify relationships between cost drivers. It is assumed that one cost driver can easily be estimated, while the other (or others) cost driver is either difcult or time-consuming to estimate. Once the relationships between the cost drivers are established, it is possible to estimate those cost drivers, which was difcult to estimate before. A simulation model is built in order to mimic the systems behavior and produce values for the cost drivers. These values are in turn used in the ABC model to produce condence interval estimations for the cost, while the conditions of the system vary. The methodology is applied in a healthcare domain, where the accurate estimation of cost is of great signicance and new tools in costing and planning are increasing in importance. The methodology combines three methods, ARM, simulation, and ABC, for the purpose of making cost estimations, thus making it complicated to implement. However, this slight complexity leads to optimal accuracy of results. One implication of applying the methodology is time saving, as estimating all cost drivers can be particularly time-consuming (Figure 1). The black arrows show the ow of the traditional ABC approach, starting from the identication of resources to the determination of the cost of products/services (cost objects). The simple dotted arrows represent the ow of discrete-event simulation.

Simulation and data mining

29

Raw Materials

Resources

Resource Drivers

Data

Data Mining

Activities Events

Activity Drivers

Cost Objects

Figure 1. Proposed methodology

IJAIM 16,1

30

The values generated by simulation are used for the estimation of cost drivers (bold dotted arrows). Every simulation run leads to a single run in the ABC model. Finally, after the system has run several times, condence interval estimations for the cost are produced. Association rules are used to assist simulation modeling. The association is based on a small sample of real data. This small amount of sample data ensures its fast and inexpensive collection in real world conditions. Application of the proposed methodology The proposed methodology was applied in a healthcare domain, specically in a health clinic, where we consider four different types of tests to be performed: X-rays, blood test, urine blood test and ultra sound. In each test, we consider the following resources: direct labour, rent and consumables. Moreover, each test involves three activities, the administrative organization of the tests, changes in the adjustments of the machines and conduction of the tests. In Figure 2, we consider three resource drivers, the man-hours, the square meters of each production centre (tests) and the cost in euros (e) of the consumables of the machines. Also, we consider three activity drivers, the number of tests performed, the adjustment time for the preparation of the tests and the operation time of the machines. The objective is to apply the proposed methodology in three activity drivers number of tests, adjustment time and operation time. The empirical distribution of number of tests results from the fact that the 30 percent of the tests conducted in the clinic are X-rays, 30 percent are blood tests, 25 percent are urine blood tests and
DIRECT LABOUR

RENT

CONSUMABLES

Manhours

M2

ADMIN/TIVE ORGANIZATION NO. OF TESTS.

ADJUSTMENT CHANGES ADJUSTMENT TIME

TEST CONDUCTION OPERATION TIME

Figure 2. The proposed ABC model

X-RAYS

BLOOD TESTS

URINE BLOOD TESTS

ULTRA SOUND

15 percent are ultra sounds. The nature of activity cost drivers make them appropriate to use in a simulation model (Beck and Nowak, 2000). We assume that, the activity driver operation time is 1/5 of the activity driver number of tests (a trivial dependency). Similar assumptions are made in Beck and Nowak (2000). The innovation of the proposed methodology concerns the calculation of adjustment time of the machines after each change is made for the conduction of the four different types of tests. We assume that, the adjustment time is a function of the number of tests (X). This relation between number of tests and adjustment time is arbitrary. It is dened so as to represent a relation, which is difcult to dene with the common methods of dening cost drivers (e.g. questionnaires). We assume that, this relation is unknown to the designers of the ABC model. A sample of real measurements of number of tests and adjustment time will contain this relation, depending on the sample size. A sample of 100 pairs of number of tests and adjustment time was taken, which has been articially produced to satisfy the aforementioned relation. This relation was dened due to the lack of real measurements. After categorizing the continuous numerical values, the sample was used as the training set of the association rules algorithm Apriori (Agrawal and Srikant, 1994), which we implemented along with other data mining algorithms in Borland C Builder, during our former contributions (Boutsinas and Vrahatis, 2001; Vrahatis et al., 2002). The extracted association rules (Figure 3) depicted the aforementioned dependencies. For example, if we consider the association rules if number of tests of urine blood tests falls into the interval (95.5,118], then adjustment time falls into

Simulation and data mining

31

Figure 3. Extracted association rules

IJAIM 16,1

32

the interval (31, 539}, then it is obvious that this rule conrms the relation 0.3X-0.4X, as shown in Table I. Based on the extracted association rules, it is possible to dene the empirical distribution of adjustment time. Extendw software was used for the simulation (Figure 4). The model was run ten times. After each run, the values of number of tests and adjustment time were transferred to Microsoft Excelw software (operation time was calculated as 1/5 of number of tests). In each particular model, the cost of the resources, activities and services was calculated. Finally, based on ten simulation runs, condence interval estimations are produced for each type of tests. Discussion In this paper, we present a methodology for activity-based costing, which combines simulation and association rule mining for effectively addressing the problem of estimating cost drivers. The objectives of this work have been to highlight the drawbacks of existing methodologies apply the proposed methodology in a healthcare domain. In business, services and manufacturing processes involve a great number of interconnections, particularly in more complex systems. This means that cost drivers frequently interact with each other and have an effect on the production cost. These interactions should be accounted for when building an ABC model. ABC models are widely used in manufacturing, where costs are easily broken down into smaller divisions where they are assigned to activities; in practice, however, it is not easy to apply traditional ABC techniques in the services industry, as a great percentage of

Number of tests Table I. Relation between tests and machine adjustment time X-rays Blood test Urine blood test Ultra sound

Adjustment time 0.1X-0.3X 0.1X-0.2X 0.3X-0.4X 0.4X-0.5X

Figure 4. Simulation model of Extendw software

total cost is labor cost and processes are highly variant. In fact, it is rather complicated to calculate the amount of time an employee has spent in a particular activity, since this can change vary regularly (Munoz and Oksan, 2006). Therefore, traditional ABC models have difculty coping with process variation in services, particularly in healthcare (Glick et al., 2000). In view of the above, simulation modeling is a particular useful tool since it models process variability, and the interactions between cost drivers in an activity-based costing context. Thus, it reveals the dynamic behavior of the cost in the production process (Beck and Nowak, 2000). Simulation of activities is inuenced by the extent of complication in transactions and thus by different types of rms. Simulation of complicated activities is harder. In the service industry, the results produced could be equally volatile and production can be inuenced by such variation. However, it is important to keep the model at low-complexity levels and to verify extrapolated results with the physical system, where this is possible. In order to simulate cost drivers one needs to gather performance data and develop statistical distributions accordingly. This task can be accomplished without interviewing and surveying employees, as a subtle and serious problem arises from the interview and survey process itself (Anderson and Kaplan, 2007). To facilitate this process, in the proposed methodology uses ARM to extract dependencies between cost drivers, based on the values of cost drivers generated by simulation. More specically, we are interested in denite dependencies between cost drivers that are hard to estimate. Various methodologies could be used for this purpose, such as regression analysis, which extracts models for dependent variables. However, ARM is a very effective technique, since it uses algorithms with low complexity. Also, ARM is based on a user-friendly paradigm and is not based on any working hypothesis that must be made by users. Finally, ARM extracts all dependencies in one run. The complexity of transactions or the type of rm does not affect the application of ARM methodology, since ARM is applied after simulation is performed. The proposed methodology offers no advantage when denite cost drivers, such as the resource driver oor space or the transaction activity drivers are being used. However, other cost drivers such as the percentage of an employees time spent on an activity, can be more accurately estimated from a simulation run (Beck and Nowak, 2000). Usually, ABC implementations take advantage of such trivial dependencies. For instance, the cost drivers items produced, machine hours and setup time could be computed by simply monitoring the cost driver number of orders. Their computation is then based on a constant multiplication with number of orders. The proposed methodology can take advantages of less obvious dependencies, which are extracted by using ARM. Such extracted dependencies are extremely useful in cases where the costs of activities are insensitive to their outputs and are widely spread to certain cost objects (resource drivers). They are also useful in cases where the amount of activities required for different cost objects differ signicantly from one cost object to another (activity drivers). For example, the case where the amount of maintenance (thus maintenance cost and time) needed differs signicantly from product A to products B and C, etc. The maintenance activity is performed after a certain time period, while during this same time period all products have been produced. In Kaplan and Atkinson (1998) and Cooper (1989), duration or intensity drivers are suggested for such a case. The proposed

Simulation and data mining

33

IJAIM 16,1

34

methodology is capable of extracting dependencies between maintenance time and the number of items produced for each different product. Cooper and Kaplan (1988) introduced the concept of the optimal cost system as being the system that minimizes the sum of the cost measurement, i.e. those costs associated with the measurements required by the cost system (Cooper and Kaplan, 1988). The design of an ABC model is judged on the cost/benet basis. This means that, the cost of doing a very detailed and accurate cost tracing should not exceed its benets. Therefore, an optimal ABC model should aim at balancing the cost of errors made from inaccurate estimates with the cost of measurement (Kaplan and Atkinson, 1998). In the previous example, the use of duration and intensity drivers is followed by a high cost of measurement. The advantage of the proposed methodology is that it reduces the interview and survey process thus minimizing the sum of the cost measurement. This is achieved by applying the interview and survey process, or any other cost driver calculation method, such as systematic appraisal and collection of real data only to a limited number of cost drivers. The remaining cost drivers are calculated indirectly based on their dependencies (denite cost drivers can be directly calculated). Users need only to identify easily estimated or calculated cost drivers and then use ARM to extract dependencies between them and those which cannot be easily estimated. Conclusion The paper introduces a new and effective methodology which combines simulation modeling and association rule mining in an activity-based costing context. It proposes a way to deal with the problem of dening cost drivers. As opposed to the integration of the ABC technique with simulation modeling, the possibility of estimating an empirical distribution of the simulated cost drivers does not exist in the present methodology. This is achieved with association rule mining used to establish dependencies between cost drivers whose estimation or calculation is either difcult or time-consuming with those which can easily be estimated or calculated. It is shown that, the extracted association rules reasonably depict the existing dependencies between the cost drivers. The application presented in this paper uses simulated data, but the randomness with which they have been created guarantees the same results when real data are used. The application of the proposed methodology reduces the probability of making inaccurate estimations of cost drivers. This in turn leads to better costing and planning and improves decision making. More informed and consequently better decisions are directly associated with reducing the total cost in a rm. Although, the proposed methodology slightly adds to model complexity due to the further application of simulation and association rule mining, considerable time is saved since it reduces the interview and survey process for cost driver estimation.
References Agrawal, R. and Srikant, R. (1994), Fast algorithms for mining association rules, Proc. of the 20th VLDB Conference, pp. 487-99. Agrawal, R., Imielinski, T. and Swami, A. (1993), Mining association rules between sets of items in large database, Proc. of the ACM SIGMOD International Conference on the Management of Data, pp. 207-16.

Anderson, S.R. and Kaplan, R.S. (2007), Time-Driven Activity-Based Costing: A Simpler and More Powerful Path to Higher Prots, Harvard Business School Press, Boston, MA. Beck, V.U. and Nowak, J.W. (2000), The merger of discrete event simulation with activity based costing for cost estimation in manufacturing environments, Proc. of the 2000 Winter Simulation Conference, pp. 2048-54. Boutsinas, B. and Vrahatis, M.N. (2001), Articial nonmonotonic neural networks, Articial Intelligence, Vol. 132 No. 1, pp. 1-38. Cobb, I., Innes, J. and Mitchell, F. (1992), Activity Based Costing-problems in Practice, The Chartered Institute of Management Accounting, London. Cokins, G. (1996), Activity-based Cost Management, Irwin, Chicago, IL. Cokins, G., Straton, A. and Helbling, J. (1993), An ABC Managers Primer, Institute of Management Accountants, London. Cooper, R. (1989), The rise of activity-based costing part three: how many cost drivers do you need, and how you select them?, Journal of Cost Management, Vol. 3, pp. 34-46. Cooper, R. and Kaplan, R.S. (1988), Measure costs right: make the right decision, Harvard Business Review, September/October, pp. 96-103. Glick, N.D., Blackmore, C.C. and Zelman, W.N. (2000), Extending simulation modeling to activity-based costing for clinical procedures, Journal of Medical Systems, Vol. 24 No. 2, pp. 7-89. Kaplan, R.S. and Atkinson, A. (1998), Advanced Management Accounting, Prentice-Hall International, Englewood Cliffs, NJ. Kaplan, R.S. and Cooper, R. (1998), Cost and Effect, Harvard Business School Press, Boston, MA. Munoz, M. and Oksan, E. (2006), Method of determining task costs for activity-based costing models, US patent. Pidd, M. (1998), Computer Simulation in Management Science, Wiley, New York, NY. Velcu, O. (2002), Practical aspects in the implementation of an ABC model, M.Sc thesis in Accounting, Swedish School of Economics and Business Administration, Helsinki. Vrahatis, M.N., Boutsinas, B., Alevizos, P. and Pavlides, G. (2002), The new k-windows algorithm for improving the k-means clustering algorithm, Journal of Computing, Vol. 18, pp. 375-91. Wiersema, W.H. (1995), Activity Based Management-todays Powerful New Tool for Controlling Costs and Creating Prots, AMACOM, New York, NY. Witten, I.H. and Frank, E. (2001), Data Mining: Practical Machine Learning Tools and Techniques, The Morgan Kaufmann Series in Data Management Systems, Hanser Fachbuch, Munchen. Corresponding author H. Kostakis can be contacted at: kostaki@upatras.gr

Simulation and data mining

35

To purchase reprints of this article please e-mail: reprints@emeraldinsight.com Or visit our web site for further details: www.emeraldinsight.com/reprints