Improving simulation model analysis and communication via design of experiment principles: an example from the simulation-based design of cost accounting systems

Abstract

Simulation offers management accounting research many benefits, such as the ability to model and to experiment with complex and large systems. At the same time, the acceptance of this method is hampered by a feeling of complexity often associated with simulation models and their behavior, as well as with challenges in communicating the models’ results. This study shows how these challenges can be addressed via the systematic use of design of experiment (DOE) principles. The DOE process framework is applied to a simulation model of a cost accounting system that is used to quantitatively evaluate two different methods for the allocation of service costs. As a result, we not only demonstrate the potential and benefits of simulation in the field of management accounting, but also show how DOE principles can help to improve understandings of simulation model behavior and the communication of simulation results.

Appendix

1.1 Numerical example

In the following, we present a numerical example to illustrate the process of the simulation experiment. Therefore, the simulation experiment is conducted once ¹⁰ with a given parameter setting, which is summarized in Table 10. Each step of the simulation process is shown and explained, as well as the calculations, to visualize how the output is generated.

Table 10 Input values for numerical experiment

Step 1: Initializing the cost vector and the consumption matrix

The variable Cost distribution on service departments (Dtr_Cost) sets the upper boundary for a uniformly distributed interval, which is 0.5. Thus, each number in this interval has the same probability being drawn. The number of service departments is defined by the variable Number of service departments (No_SD) and is set to 3. In the first step, we draw a number from the interval {0, 0.5} for each of the three service departments. Table 11 shows the resulting numbers. In the second step, we normalize the random numbers and multiply them with the total amount of costs [which is a control variable and set to 1,000,000 monetary units (MU)], to generate the direct costs at each service department (see Table 12).

Table 11 First step in generating the cost vector

Table 12 Cost vector

Next, we create the consumption matrix. The consumption matrix represents the amount of service flow from service departments to other departments. The first row represents the amount of service that the first service department provides to the department in the respective column. It is assumed that service departments do not consume their own service, but that all service departments provide services to production departments. The number of production departments is controlled by the variable Number of production departments (No_PD) and set to 5 in this example. Here again, random numbers are drawn from a uniformly distributed interval whose upper boundary is defined by the variable Variation of supplied services (Var_Serv). The variable is set to 0.5 and thus, for each matrix entry, we draw a number from the interval {0, 0.5}, except for the entries representing service flow between service departments. These entries are only filled to a certain degree, defined by the variable Degree of variety (Degr_Var). This variable is set to 0.4. This means that only 40 % of the entries are filled with numbers greater than zero. Table 13 shows the results of this step. One can see that between service departments 1 and 3, there is a reciprocal service exchange. Of the six possible service exchanges (the diagonal remains zero, as explained above), two are greater than zero. We normalize each row of the consumption matrix; in this way, for each service department, the sum of provided services adds up to 1. See Table 14 for the results of this step.

Table 13 First step in generating the consumption matrix

Table 14 The normalized consumption matrix

Step 2: Calculating the final costs at production departments for both allocation methods

Based on the cost vector and the consumption matrix, both allocation methods calculate their results. The result is a cost vector that represents the allocated cost at each production department.

• (a) Direct method

  The direct method ignores any services between service departments. Let $\hbox {x}_{\mathrm{SDi}\_\mathrm{SDn}}$ be the amount of service that is provided from service department i to service department n, and $\hbox {x}_{\mathrm{SDi}\_\mathrm{PDj}}$ the amount of service that is provided from service department i to production department j. Let $\hbox {cost}_{\mathrm{SDi}}$ be the total amount of costs at service department i, then $\hbox {cost}_{\mathrm{SDi}\_\mathrm{PDj}}$—the cost that is allocated from service department i to production department j—is calculated as follows, with n being the total number of service departments:

  $$\begin{aligned}&\hbox {cost}_{\mathrm{SDi}\_\mathrm{PDj}} =\frac{\hbox {x}_{\mathrm{SDiPDj}} }{1-\mathop \sum \nolimits _{j=1}^n x_{SD_i SD_j }}\times cost_{SD_i }\\&\hbox {cost}_{{\mathrm{SD}}1\_\mathrm{PD}1} =\frac{0.1423}{( {1-0.2130} )}\times 181,935\,\hbox {MU}=32,896\, \hbox {MU}. \end{aligned}$$

We ignored the amount of service provided to service department 3 (subtracted from the total amount). The respective share of service delivered to production department 1 is multiplied with the direct costs at service department 1. This is step repeated for each service flow. At the end, for each production department, we sum up the allocated costs.

(b) Reciprocal method

The reciprocal method acknowledges the service exchange among service departments. Therefore, a set of equations is solved simultaneously to generate the transfer prices for each service department that consider the service exchange. In this example, only two equations must be solved, since service department 2 neither receives nor delivers any services to other service departments. Let coco$_{\mathrm{SDi}}$ be the complete cost (including the reciprocal service) at service department i:

$$\begin{aligned}&coco_{SDi} =cost_{SDi} +\mathop \sum \limits _{i=1}^n x_{SD_i SD_n } \times coco_{SD_k }\\&coco_{SD1} =181,935\, MU+0.2645\times coco_{SD3}\\&coco_{SD3} =207,365\, MU+0.213\times coco_{SD1} \end{aligned}$$

Solving for cost$_{\mathrm{SD1}}$ and cost$_{\mathrm{SD3}}$:

$$\begin{aligned}&coco_{SD1} =250,920\, MU\\&coco_{SD3} =260,813\, MU \end{aligned}$$

The costs that are allocated to production department j are calculated as:

$$\begin{aligned} cost_{PDj}&=\mathop \sum \limits _{i=1}^n x_{SD_i PD_j } \times coco_{SD_k }\\ cost_{PD1}&=0.1423\times 250,920\, MU+0.2620\times 610,700\, MU\\&\quad +0.2836\times 260,818\, MU\\&=269,675\, MU \end{aligned}$$

We summarize the results from the simulation experiment in Table 15. For each production department, we show the allocated costs using the reciprocal method and using the direct method.

Table 15 Resulting allocation at production departments

Step 3: Evaluating the results

To evaluate the simulation results, we use the Euclidian distance measure (EUCD), taken from Balakrishnan et al. (2011). It is defined as $EUCD=\sqrt{\mathop \sum \nolimits _{i=1}^{Nr\_PD} ( {RM_i -DM_i } )^{2}}$, with RM being the results for the reciprocal method at the ith production department and DM the respective result for the direct method. This measure represents in monetary units the misallocation for the direct method compared to the reciprocal method. The calculation of the EUCD in this example is:

$$\begin{aligned}= & {} \sqrt{( {269,675-272,868} )^{2}+( {204,454-200,061} )^{2}+( {139,264-144,099} )^{2}\ldots }\\= & {} \sqrt{10,195,249+19,298,449+23,377,225+52,441+14,714,896}\\= & {} \sqrt{67,638,260}=8224 \end{aligned}$$

Thus, the output of the simulation experiment for this run is 8224 MU. It can be interpreted as 8224 monetary units are in total wrongly allocated to production departments. Considering a total amount of 1,000,000 MU in the system, the distortion is relatively low.