INFLUENCE OF DEMAND FORECASTS ACCURACY ON SUPPLY CHAINS DISTRIBUTION SYSTEMS DEPENDABILITY. Natalia SZOZDA 1, Sylwia WERBIŃSKA-WOJCIECHOWSKA 2 1 Wroclaw University of Economics, Wroclaw, Poland, e-mail: natalia.szozda@ue.wroc.pl 2 Wroclaw University of Technology, Wroclaw, Poland, e-mail: sylwia.werbinska@pwr.wroc.pl Abstract The goal of this paper is to investigate how the demand forecasts inaccuracy influence the performance of distribution processes in supply chains. According to this, the main definitions of supply chain have been described. Later, authors focus on a simulation model development. The paper is ended by summary and directions for further research. Keywords: supply chain, forecast, dependability, forecasts accuracy. 1. INTRODUCTION Supply chain may be defined as an integrated process wherein a number of various business entities (like suppliers, manufacturers, distributors, and retailers) work together in an effort to: (1) acquire raw materials, (2) convert these raw materials into specified final, (3) deliver these final to retailers and final customers [2]. Such a logistic network is then characterized by a forward flow of materials and a backward flow of information. As a result, the reliability and efficiency of supply chain performance can be affected by many different factors. Supply chain networks are vulnerable to disruptions and failure at any point in the supply chain may cause the entire network to fail. A key factor in effective supply chain management is the ability to minimize the effects of such undesired events/disruptions occurrence. The factors which influence supply chain performance, as well as supply chain reliability/dependability definitions are investigated in e.g. [6, 7]. In these works, it is stated that random customer demand is the most challenge problem, as well as the distribution process parameters and the condition of distribution company. An inaccurate forecast of customer demand could lead to overstock or under stock, what could cause the increase of inventory costs of organization performance. Moreover, decisions made during supply chain designing process in the area of distribution process performance also have the influence on chain reliability. The variability of retailers/distribution centres performance, quality of transportation process, quality of demand planning process or information availability and accuracy may significantly disturb chain performance and lead to fail customer satisfaction achieving. Another problem is connected with demand uncertainty [5]. The errors in forecasting the customers demand could be e.g. changing customers preference, irregularity of customer orders in terms of time, quantity and quality. Thus, in this paper authors focus on demand planning process and its quality. The paper is a continuation of such research issues investigation given in [6, 7], where also a comprehensive literature overview is provided. Following this, in the next Section, there are presented obtained performance analysis results in comparison with the knowledge about the case company present condition The work ends up with summary and directions for further research.
2. CASE STUDY 2.1 Model building To fulfil the goal of the research a simulation model has been established. The model bases on really existing network (Fig. 1) distribute their by their points of sale (POS) all over the country (Poland). Manufacturer orders Central warehouse orders / forecasts orders / forecasts orders / forecasts Retailer 1... Retailer 32 Fig. 1. Analysed network The information about the network and the demand has been gathered through the internet, particular network s offers and in interviews with salespeople. It has allowed authors to gather the demand and delivery data needed to the simulation carrying out. 2.2 Distribution In the model the product Y is sold by 32 points of sale (POS) that operate within one network. The product Y is introduced to the market with the first weeks of January (Fig. 2). In June and September the marketing activities are planned. It resulted with an increase in analysed product demand. The company plans to withdraw the product Y from the market with the end of a year; it means the end of the product life cycle. In all POS the product Y is sold in one price determined by the manufacturer. Fig. 2. Product Y life cycle demand.
2.3 Model assumptions Each POS gathers following data each period: week number which is a time unit, customer demand which is an expressed willingness and readiness to purchase, initial inventory level which is a result of final inventory level from last period, sales which is a result of the demand met in a certain period and initial inventory level thus: if inventory level is bigger or equal the customer demand sales is equal the customer demand, if inventory level is smaller than the customer demand sales is equal the initial inventory level, inventory replenishment which is the order placed at a certain network one period ago, final inventory level which is a difference between an initial inventory level and sales, order put in a particular network which is a result of applied inventory system, lost benefits an amount of goods that could be potentially sold (because of customer demand) and weren t because of stock shortage. As we can see, one basic assumption was made in a model: the only possibility of sales being lesser than the demand in period t is when a POS hasn t got sufficient inventory level in period t. Inventory systems The authors applied one classic inventory system which is adequate for the demand of product Y [1]: the system of variable order cycle and variable order quantity (s,s), where s is a certain inventory level setting the time of order placement, and S is an order-up-to point measured in stock keeping units. The quantity of order is described as S inventory level. Parameters construction The parameters used for implementation particular inventory systems were defined as (compare [4]): s = d 1 ) (1) S = i ( V i d i where: (2) d i average forecast met in a certain POS i operating within a network V variability coefficient for a certain POS i operating within a network the coefficient for parameter s for a certain network the coefficient for parameter S for a certain network The coefficients and that were taken into the parameters s and S construction give a possibility to manipulate the values of these parameters among the network, but in the same time leave individually suited for each POS form of them. Cost ingredients construction In the simulation three components of cost were considered: Unit ordering cost: If a POS in a certain week puts an order in the upstream stage (network) has to bear unit ordering cost = C o, if not he doesn t bear any costs. Holding cost (compare [4]): C h = ( r u ) / O (3) i s
Inventory cost (compare [4]): C i = ( p where: I h ) pi I 2 ki (4) p product price for a certain network h holding rate for a certain network r i weekly rent cost for a certain POS i u s trade surface assigned for inventory (in percentage terms) I pi initial inventory level for a certain POS i I ki final inventory level for a certain POS i O quantity of that a certain network sells through their POS All cost components were counted periodically (weekly). Overall cost was determined as a simple sum of all components. 2.4 Demand forecast in the analysed network In order to verify how the demand forecasts accuracy influences the performance of distribution processes in supply chains three forecast are conducted (Fig. 3): Forecast No.1 is based on assumptions made in practice - the average sales of the product determined by experts, assumed to be constant throughout the analysed period; Forecast No.2 is set out as a weighted average demand of the last three periods; Forecast No.3 is set out as a forecast by analogy. Fig. 3. Demand forecast for a product Y. To compare the obtained forecasts results the forecast error is used (Table 1). Forecast error is calculated by using the mean absolute error [3] (MAE) which is a quality quantity used to measure how close forecasts or predictions are to the eventual outcomes. The mean absolute error is given by: 1 MAE n n t1 f t y t (5)
where: f t forecast for the time period t y t sales for the time period t Table 1. Forecast errors (MAE) Average Demand [pcs] 345 Forecast No.1 [pcs] 369 Forecast No.2 [pcs] 387 Forecast No.3 [pcs] 348 MAE No.1 3,34 MAE No.2 2,10 MAE No.3 1,63 The lowest forecast error was calculated for the third forecast and MAE equals 1,63. In the next section will be determined how the forecast accuracy affects the operation of the supply chain. 2.5 Simulation results In the simulation (s,s) inventory system was applied, what means that the order cycle is not fixed but it depends on parameter s. The values of both parameters were described as (compare [4]): s = d 1 ) ; (coefficient = 1) (6) S = i ( V i d i ; (coefficient = 2,5) (7) Table 2 shows the simulation results, and Fig. 4 shows the differences between the demand and sales curves for different forecasts. Table 2. Simulation results realistic constraints Forecast Lost benefits [pcs] Inventory cost [zl] Holding cost [zl] weekly per year weekly per year weekly per year 1-56 -2 902 591,07 29 535,70 2 757,40 137 785,00 2-44 -2 290 658,52 32 961,80 3 097,88 155 100,00 3-27 -1 372 592,90 29 557,94 2 697,55 134 997,50 Fig. 4. Demand and sales of product Y within the network.
The analysis showed that the applied models do not provide sufficient inventory level that allows for fulfilling demand in 100%. The differences between the demand and sales curves are particularly visible in the phase of active demand. (S,s) model in the best scenario (Forecast No. 3) gives 13% demand fulfilment per week which means that weekly 27 Y are not sold because of stock shortage. Lack of proper forecasting system (Forecast No. 1) increases the risk of lost benefits more than double (to the level of 56 pcs). It is surprising that in this case the costs are not lower but they are comparable. 3. SUMMARY The presented research results are the preliminary step in developing the supply chain performance simulation model for inventory analysis. In the next step, there are planned more sophisticated methods to be used [8]. However, the focus of this paper was to examine if demand forecasts accuracy influences a distribution network performance level. This goal was achieved by authors. In the presented paper one of the most commonly used in practice inventory system was applied with the relation of 3 methods of demand forecast implementation. Based on the obtained analysis results, the use of Forecast no. 3 gives the better (but not optimal) economic results than implementation of other demand forecasts methods. It occurs, that Forecast no. 3 use results in shorter time length of order cycle (more frequently orders placing). This situation let maintain lover average level of inventories and, as a result, lower holding and inventory costs bearing. The worst economic results give implementation of Forecast no. 2. However, it should be noted, that the inventory system implementation problem should be reconsidered with taking into account the issues of the best forecast method selection. In the presented article, authors decided to implement three most commonly used forecast methods because of their simplicity. This does not means that they are the best when taking into account demand profile for product Y. This problem will be investigated in the next step of authors research analyses in this area. LITERATURE [1] Bashyam S., Fu M. Optimization of (s, S) Inventory Systems with Random Lead Times and a Service Level Constraint. Management Science, 1998, Vol.44, No. 12 pp. 243-256. [2] Beamon B. M. Supply Chain Design and Analysis: Models and Methods. International Journal of Production Economic, 1998, Vol. 55, No. 3, pp. 281-294. [3] Hyndman R., Koehler A. Another look at measures of forecast accuracy. International Journal of Forecasting, Elsevier, 2006, vol. 22(4), pp. 679 688. [4] Krawczyk S. (ed.). Logistics. Theory and Practice (in Polish), Vol. 2. Warszawa, Difin Publishing House, 2011. [5] Li L., Schulze L. Uncertainty in Logistics Network Design: A Review. Proceedings of the International MultiConference of Engineers and Computer Scientist, 2011, Vol. II, Hong Kong. [6] Szozda N., Werbińska-Wociechowska S. Influence of the demand information quality on planning process accuracy in supply chain. Case studies. Article prep. for Carpathian Logistics Congress CLC 2011. [7] Szozda N., Werbińska-Wociechowska S. Influence of the demand planning process on logistic system reliability. Case study. Logistics and Transport, 2011, Vol.2, pp. 79-92. [8] Vila-Parrish A.R. Dynamic Inventory Policies for Short Life Cycle and Perishable Products with Demand Uncertainty. North Carolina State University, 2010.