Sunteți pe pagina 1din 10

TEACHING BRIEF

Spreadsheet Modeling of (Q, R) Inventory Policies


Barry R. Cobb
Department of Economics and Business, Virginia Military Institute, Lexington, VA 24450, email: cobbbr@vmi.edu, phone: (540)-464-7452.

Abstract This teaching brief describes a method for nding an approximately optimal combination of order quantity and reorder point in a continuous review inventory model using a discrete expected shortage calculation. The technique is an alternative to a model where expected shortage is calculated by integration, and can allow students who have not had a calculus-based statistics course to better understand how uncertainty aects inventory decisions.

Barry R. Cobb is a professor in the Department of Economics and Business at the Virginia Military Institute (VMI) in Lexington. He teaches operations management, managerial economics, and management science. His research in the areas of decision analysis, applied probability, and probabilistic graphical models has been published in such journals as Decision Analysis, Decision Support Systems, INFORMS Transactions on Education, and the Journal of the Operational Research Society.

Spreadsheet Modeling of (Q, R) Inventory Policies


Abstract This teaching brief describes a method for nding an approximately optimal combination of order quantity and reorder point in a continuous review inventory model using a discrete expected shortage calculation. The technique is an alternative to a model where expected shortage is calculated by integration, and can allow students who have not had a calculus-based statistics course to better understand how uncertainty aects inventory decisions. INTRODUCTION After a thorough explanation of the economic order quantity (EOQ) model, operations management and management science textbooks, including those by Heizer and Render (2006) and Stevenson (2012), typically treat the determination of an inventory reorder point (R) as a problem of ensuring a desired service level. Stevenson (2012) also briey discusses how to calculate an expected shortage level for a given R from a unit normal loss table. Ragsdale (2004) addresses optimization in the context of the EOQ model with certain demand using the Excel Solver provided by Frontline Systems (Solver), while Render et al. (2002) introduce a discrete expected value calculation for the optimal R that decouples this decision from the selection of optimal order quantity (Q). This teaching brief describes a method for jointly establishing optimal order quantity and reorder point policies (a (Q, R) model) using an Excel spreadsheet in conjunction with Solver. The next section denes the problem and the notation used in the paper. PROBLEM DEFINITION The problem addressed in this teaching brief is the purchase of a single inventory item at a single price under uncertain demand and lead time. Figure 1 gives a depiction of the inputs, process, and outputs of the problem.
Inputs Assumptions 1. Mean (expected) annual demand (AD) 2. Standard deviation of annual demand (AD) 3. Mean (expected) lead time (L) 4. Standard deviation of lead time (L) 5. Fixed ordering cost (per order) (K) 6. Holding cost per unit per year (hC) 7. Shortage cost per unit () Decision Variables 1. Order quantity (Q) 2. Reorder point ( R) Outputs 1. Optimal policies for Q and R. 2. Minimum total annual expected ordering, holding, and shortage costs . 3. Expected service level for optimal policies

Process Minimize Total of Annual Expected Ordering, Holding, and Shortage Costs

Figure 1: Inputs, process, and outputs for the inventory problem under uncertain demand. The company denes the parameters (mean and standard deviation) for annual demand (AD) and lead time in years (L). Under the simplifying assumption that lead time demand

(X ) is a sum of independent and identically distributed periodic demand values, the mean and standard deviation of X are determined as X = L AD and X =
2 2 L AD + 2 AD L .

(1)

2 ). The The probability density function fX for X is assumed to be normal, i.e. X N (X , X company also estimates its xed ordering (per order), unit holding (per unit per year), and unit shortage (per unit short) costs. The values under control of the manager are the order quantity and reorder point (where the reorder point is used to dene the safety stock (SS )). These are chosen to minimize the total expected annual inventory cost, calculated (Hadley and Whitin, 1963) as

Q AD AD K + + R X hc + S (R) . (2) Q 2 Q The rst term in (2) is the expected ordering cost, calculated as the expected number of annual order cycles (AD /Q) times the cost per order. The second term is the expected holding cost, where Q/2 is the average inventory level and R X is the average safety stock. The nal term is the expected shortage cost, which is determined by the expected number of order cycles, the shortage cost per unit, and where the expected shortage per cycle is T C (Q, R) =

S (R) =

(x R) fX (x) dx

or

S (k ) = X

(z k ) (z ) dz .

(3)

In the second formulation above, is the standard normal density function and k = (R X )/X is often referred to as the safety factor, because safety stock is determined as SS = R X . MOTIVATION Winston and Albright (hereafter WA) (2007) introduce the (Q, R) continuous review inventory model and present an optimization model that uses Excel formulas to replicate the expected shortage function in (3). This function can be re-written as S (k ) = X
k

z (z ) dz k (1 (k ))
k

(4) z (z ) dz ,

where is the standard normal cumulative distribution function. Since (k ) = this further simplies to S (k ) = X ((k ) k (1 (k ))) .

Producing this result in Excel requires use of the NORM.S.DIST function to return values for the normal probability density function and the normal cumulative distribution function. The formula is X (NORM.S.DIST(k, 0) k (1 NORM.S.DIST(k, 1))) . Students without a background in calculus-based statistics are likely to have diculty understanding the concept of this expected shortage function. In fact, WA state the following in a footnote when introducing the Excel formula: 2

This is one of the few times in this book where you will have to take our word for it. The derivation of this formula is beyond the level of this book (p. 767). This quote motivated the method presented in this teaching brief. The goal is for students to gain an understanding of how uncertainty aects inventory decisions. Recognizing that students may not gain an appreciation of the expected shortage concept by simply memorizing a formula, this method exploits a concept that students recall well from statistics, calculating cumulative probabilities from a normal distribution. The model uses a discrete approximation to the lead time demand distribution. MODELING APPROACH The following example (WA, pp. 764771) will be used to demonstrate the model: Macheys department store sells, on average, 1200 cameras per year. The store pays a setup cost of $125 per order, and the holding cost is $8 per camera per year. Assume annual demand (AD) is normally distributed with a mean of AD = 1200 and a standard deviation of AD = 70. The store assumes there is a cost of $10 per unit short. Suppose the lead time (L) averages one week, or L = 1/52 of a year, with a standard deviation of one-half of a week, or L = 1/104 of a year. The (Q, R) inventory solution method is broken into three parts: 1) a lead time demand calculation; 2) an expected shortage calculation; and 3) an inventory cost model. To begin, students enter the problems inputs (see Figure 1) in an Excel worksheet. Figure 2 shows the Excel worksheet with the inputs in cells B2:B8. Lead Time Demand Calculation The rst step to solving the (Q, R) inventory problem is nding the mean (X ) and standard deviation (X ) of lead time demand (X ). This is accomplished by using the formulas in (1) as X = 1200
1 52

23 and X =

1 52

702 + 12002

1 2 104

15

These calculations are entered in a section of the worksheet labeled Lead Time Demand (cells B11:B12) and the ROUND formula is used to produce integer answers. This facilitates a discrete approximation to the expected shortage per cycle calculation. Cell B11 contains the formula ROUND(B2*B4,0), while cell B12 is calculated as ROUND(SQRT(B4*B3^2+B2^2*B5^2),0) . As stated previously, having students use the expected shortage formula without explaining its origin can be problematic, and an instructor may want to be equally cautious about having them use the X formula in (1). The formula can be introduced in stages. First, demonstrate the calculation of X with certain lead time (L = 0). Under this scenario, the formula has one term associated with one source of uncertainty and the answer in this example is X = 10. Next, re-calculate X as shown above to show that the variability in lead time demand increases to X = 15 when the second source of uncertainty is introduced. 3

At this point, introduce the assumption of normality for lead time demand and review the NORM.DIST function in the context of demonstrating the reorder point concept. Remind students that 99.73% of the area under the normal curve is within 3 standard deviations of the mean, and dene X + 3X 68 as the eective maximum for lead time demand. This calculation is added to the Lead Time Demand section of the worksheet in cell B13. The decision variablesorder quantity (Q) and reorder point (R)are included in the worksheet in cells B16:B17, then safety stock is calculated in cell B18 as SS = R X (or B17-B11). WA use the safety factor, k , a multiple of X , as a decision variable, then calculate R and SS . Reorder point (R) is used here a decision variable because it is directly referenced in the expected shortage calculation, and is the value implemented in the decision policy used by a manager. Expected Shortage Calculation The discrete approximation to expected shortage per cycle is prepared in a separate section of the worksheet labeled Expected Shortage, as shown in Figure 2. Column D lists the possible lead time demand values between the mean and eective maximum (some rows in Figure 2 are hidden). If reorder point is greater than lead time demand, no shortage occurs, so Column E calculates the shortage for each lead time demand value using the MAX function. Cell E3 contains MAX(0,D3-$B$17), where cell B17 contains the reorder point. This is copied to the remaining rows in Column E, and once lead time demand exceeds the reorder point a positive value results. The probability column contains the probability in one unit intervals of the normal distribution with mean in cell B11 and standard deviation in cell B12, where the value in the lead time column is the mid-point. Thus, the formula in cell F3 is NORM.DIST(D3+0.5,$B$11,$B$12,1)-NORM.DIST(D3-0.5,$B$11,$B$12,1). This formula can be interpreted as the probability that lead time demand equals 23. When copied down, the formula can also be interpreted as the probability that the shortage equals the amount in Column D for each row. The sum of all these cells equals about 51% in this case. The probability under the normal density function from [22.5, 23) actually exceeds the probability that lead time demand is above 68.5. In theory, the total probability should be normalized to 50%, but skipping this step has almost no eect on the nal solution. Using the discrete approximation can facilitate a straightforward graphical representation of the shortage values and the associated probabilities for a given reorder point. For example, these values are shown for a (not necessarily optimal) reorder point of R = 30 in Figure 3. The chart reveals the relative likelihood of the lead time demand values where a shortage occurs and does not occur for a given reorder point selection. The discrete approximation makes construction of this chart possible because the shortage and probability values are explicitly listed in Columns E and F of the model (see Figure 2). In cell F1, the expected shortage is calculated as SUMPRODUCT(E3:E48,F3:F48). This result will be referenced in the total inventory cost calculation. Prior to teaching students this formula, reviewing the concept of expected value using a simpler example is helpful. Students can be 4

Figure 2: Excel spreadsheet for the (Q, R) inventory problem.

Figure 3: Graph of the shortage function and associated probabilities for R = 30. assigned a homework problem that requires them to construct the expected shortage model prior to building the inventory cost model described in the next section. Inventory Cost Model Introducting the expected total cost formula in (2) by rst reviewing an alternate expression that has a verbal description of each component allows students to gain intuition behind the calculation prior to having to absorb the mathematical notation. The cost function is entered into Excel using several lines for each cost category. Expected orders per year are calculated in cell B22 as AD /Q (or B2/B16), then cell B23 multiplies the result times K to obtain total ordering costs as B22*B6. Average inventory in cell B26 is calculated as Q/2 (or B16/2), then safety stock is referenced from cell B18. This reinforces that there is an added holding cost for safety stock. The sum of average inventory and safety stock is multiplied in cell B28 times unit holding cost ((B26+B27)*B7). Expected shortage in cell B31 is referenced from cell F1 in the Expected Shortage model, then multiplied times orders (cycles) per year (linked to B22) and shortage cost per unit. Thus, cell B33 contains the formula B31*B32*B8. All costs are added to obtain total annual inventory costs in cell B35, which equals B23+B28+B33. Solver Solver is used to nd the best Q and R combination. The total annual inventory cost calculation in cell B35 is referenced as the Target Cell and Solvers objective is to minimize this value. The Solver window in Figure 4 shows three constraints. The constraint $B$16>=1 is included to avoid errors that might occur if a value too close to zero is entered during the nonlinear optimization routine. The constraints $B$17<=$B$13 and $B$17>=$C$11 limit the reorder point to values between X and the eective maximum lead time demand. The optimal solutions are Q = 200.67, R = 40, SS = 17, and T C = 1741.38. Using the Excel formula suggested by WA to calculate the expected shortage results in solutions of 6

Figure 4: Solver window for the Inventory Cost Model. Q = 201.41, k = 1.11, SS = 16.68, R = 39.76, and T C = 1744.72. There is little dierence in the solutions. If we insert the (Q, R) = (200.67, 40) solutions in the WA model, the expected total cost is 1744.77. We can conclude there is a $0.05 opportunity cost to applying the suboptimal solutions in the exact model. Although the WA model incorporates a technically correct expected shortage calculation, the approach presented here produces results in terms of total cost that dier negligibly. While the expected shortage calculation is an approximation, other aspects of the model are also imprecise. The assumption of normally distributed demand during lead time, in particular, may be made for mathematical convenience and is likely an approximation to the actual demand distribution. Additionally, the shortage cost per unit may be dicult to ascertain with certainty in practice. IMPLEMENTATION This method was taught to 51 undergraduates divided into three sections of operations management. There was ample opportunity for hands-on class time to work in Excel. At this point in the semester, the students were well-versed in Excel and Solver. There was no control group that learned the (Q, R) problem using the WA expected shortage formula. Given the mathematical background of the students, the expected value method was more suitable in this setting than the WA approach. Therefore, three questions will be considered to evaluate eectiveness: 1) With ample practice, were students able to develop the model and solve the problem? 2) Did students understand the role of safety stock in inventory decisions? 3) Did students understand why selecting an optimal (Q, R) combination is important? Problems requiring students to use the method were given on two assignments, then a similar 7

problem was included on an exam. Answering the exam question correctly was an all-or-nothing proposition. The students had to build each part of the model correctly and use Solver to get the exact answers. Twenty-six of 51 students were successful. The average grade on the two assignments was 75% for students who answered the exam question correctly, and 54% for those who did not. Those who practiced did a better job of implementing the model. A regression of a binary dependent variable (1 = student answered correctly) on average assignment grade (0 to 100 scale) yielded this model: Prob. of answering exam question correctly = 0.2723 + 0.0037 (avg. assignment grade) . The coecient on average assignment grade has a p-value of 0.047. With no practice (0% average) a student only had a 27.23% chance of answering the exam question correctly. If one of two assignments were completed correctly (50% average), a student had a 45% chance of answering the exam question correctly. If both assignments were completed correctly, a student had a 65% chance of answering the exam question correctly. Exam grades in this course average about 70%, whereas the overall average course grade is about 75%. The results are as expected for a problem on a dicult topic. The students were also given two essay questions: 1) Explain the purpose of holding safety stock when demand is uncertain; 2) Explain how choosing the optimal combination of order quantity and safety stock helps a business reduce its costs when demand is uncertain. The responses were evaluated based on similarity to the following statements (WA, p. 764): 1) Safety stock is extra inventory used as a cushion in case demand during lead time is greater than expected; 2) When future demand is uncertain...a company has to determine the tradeo between ordering too much (and having excess inventory costs) and ordering too little (and having shortage costs and a low service level). Scores were 0 (not similar), 0.5 (somewhat similar), or 1 (very similar). The evaluation is admittedly subjective, but results from the essay questions were encouraging. The average score for the rst question on safety stock was 0.76, while the average score on the second question on balancing overall costs was 0.70. Interestingly, the students were not assigned reading on this topic and there was no lecture that specically covered the inventory decision and safety stock. The discussion of the issues addressed in the questions primarily occurred during the course of solving the problems as described in this teaching brief. ACKNOWLEDGEMENTS The author is grateful for valuable comments from two anonymous reviewers and the associate editor which substantially improved the teaching brief.

REFERENCES Hadley, G., T. Whitin. 1963. Analysis of Inventory Systems. Prentice Hall, Englewood Clis, NJ. Heizer, B., J. Render. 2006. Operations Management, 8th edition. Prentice Hall, Upper Saddle River, NJ. Ragsdale, C.T. 2004. Spreadsheet Modeling and Decision Analysis, 5th edition. Cengage, Mason, OH. Render, B., R.M. Stair, N. Balakrishnan. 2002. Managerial Decision Modeling with Spreadsheets. Prentice Hall, Upper Saddle River, NJ. Stevenson, W.J. 2012. Operations Management, 11th edition. McGraw-Hill, New York. Winston, W.L., S.C. Albright. 2007. Practical Management Science, 3rd edition. Thomson South-Western, Mason, OH.

S-ar putea să vă placă și