| | | | Chapter X - Maximum Performance Efficiency Approaches for Estimating Best Practice Costs | Data Mining: Opportunities and Challenges | by John Wang (ed) | | Idea Group Publishing 2003 | | | | **Brought to you by Team-Fly** | | | There are two uses of the word "model" in this chapter. By the basic cost model, we mean the assumed relationship between the cost driver of an activity and its contribution to the total cost pool. The simplest case is that of proportional variation. Let yr be the amount of the r-th cost driver. Under the basic cost model, the contribution to the total cost pool will be a_{r}y_{r}, where a_{r} may be called the average cost per unit of y_{r} or simply the cost rate. This chapter focuses on this basic cost model. Some other models such as for estimating marginal costs are discussed briefly. The other use of the word "model" refers to the method used to solve for estimates of the basic cost model parameters the a*_{r}. The estimation models used here are LP models. As noted above, this usage of LP does not appear to be similar to previous uses in the accounting literature, but is a consequence of the estimation principle used. Suppose there are r=1 R_{r} activities with associated cost driver quantities of y_{r}, respectively. Then, as in Cooper and Kaplan (1992), the cost of *resources consumed* is given by ∑a_{r}y_{r}. If x is the actual total cost pool associated with these activities, the cost of resources *supplied*, then there may be a difference, s ≥ 0, such that Cooper and Kaplan (1992) call s the *cost of unused capacity*. In their analysis, the x value was a budget figure and the y_{r} could vary due to decreased demands. We start from the same construct but regard variations of the y_{r} and x as due to more or less efficiency. By defining what may be called best practice a_{r} values, we therefore associate the s value with possible inefficiency of a business unit, a cost of inefficiency. Thus suppose we have j = 1 N *comparable* business units, achieving y_{rj} units of driver r, respectively, and with associated cost pools, x_{j}. In addition to having the same activities and cost drivers, we further require that comparable units be similar in the sense that the practices, policies, technologies, employee competence levels, and managerial actions of any one should be transferable, in principle, to any other. Define a_{r}* as the vector of cost rates associated with the most efficient unit or units under comparison. Then in the equation the cost of unused capacity, s_{j}, may be interpreted as an inefficiency suffered by a failure to achieve larger y_{rj} values, smaller x_{j} value, or some combination of these. The ratio v_{j} = ∑ a_{r}*y_{rj}/ xj, is an efficiency measure for the j-th unit when the a_{r}* are the true benchmark cost rates and ∑ a_{r}*y_{rj}≤x_{j} holds for all units. A technique for estimating parameters in efficiency ratios of the above form was proposed in Troutt (1995). In that paper, the primary data were specific decisions. Here a more general form of that approach called *maximum performance efficiency* (MPE) is proposed and applied to estimate the benchmark a_{r}* values. Assume that each unit j = 1 N, seeks to achieve maximum (1.0) efficiency. Then the whole set of units may be regarded as attempting to maximize the sum of these efficiency ratios, namely, ∑∑a_{r}*y_{rj}/ x_{j}. The maximum performance efficiency estimation principle proposes estimates of the a_{r}* as those which render the total, or equivalently the average, of these efficiencies a maximum. **Maximum Performance Efficiency (MPE) Estimation Principle:** *In a performance model depending on an unknown parameter vector, select as the estimate of the parameter vector that value for which the total performance is greatest.* Use of the word performance is stressed in the MPE name to emphasize the more general utility of the approach than was indicated by the earlier MDE term. Managerial performance, such as in the units under study here, involves many kinds, levels, and horizons of decisions. Define the data elements Y_{rj} by Y_{rj} = y_{rj}/x_{j}. Then the MPE estimation criterion for the benchmark a_{r}* values is given by Problem MPE is a linear programming problem. Its unknown variables are the best practice cost rates, the a_{r}* values. Solution of this model provides values for the a_{r}* as well as the unit efficiencies, v_{j}. A model adequacy or validation approach for this estimation procedure is proposed in a later section. As a linear programming problem, the MPE model is readily solved by a wide variety of commonly available software products. MPE may be seen here to be an estimation criterion analogous to the ordinary least squares error criterion in regression. The cost model, ∑ a_{r} Y_{rj} is analogous to the regression model. | | **Brought to you by Team-Fly** | | | |