Demand Forecasting Using Time-Series Forecasting - Coursework Example

Quantitative Analysis and Decision Making: Assignment 2 Name Institution Date Executive Summary Quantitative techniques are used in decision making of logistics management to analyze data and uncover important information and hence make decisions on material acquisitions, production quantities, production capacity and the quantity of inventories and resources required. Demand forecasting is an important aspect in logistics management and one of the quantitative techniques that can be used in demand forecasting is time-series forecasting and index numbers where the historical data is analyzed and this provides the basis of decision making regarding the demand. Through forecasting time series, the seasonality, cyclic patterns and trends of demand can be identified because forecasting using time series, four key elements in the past demand are used and they include average, trend, seasonal, as well as cyclical element. Some time-series forecasting techniques include moving average, exponential smoothing, double exponential smoothing, and extended exponential smoothing. Some of the limitations with time series forecasting include that the quantitative technique can lead to erroneous forecasting because whereas the methods used in time-series forecasting can link historical patterns into future forecasts, the methods are not effective in integrating the input of projected future events. Therefore, to ensure accuracy during forecasting it is important to integrate the forecast techniques with suitable support and administrative systems. Another limitation is that some of the time-time series forecasting techniques can be significantly unresponsive to change and they also need updating and maintenance of huge historical data when calculating forecasts. This can be solved by combining the techniques during data analysis as this can allow detection of deviations and thus tracking of errors that may arise by using one technique. Assignment 2: Demand Forecasting using Time-series forecasting The focus of this assignment is on the relationship between data analysis and the decision making of logistics management. Tremendous amount of data is generated during logistics management and thus it is necessary to analyze the data sets to unearth important information thus data forms the first step in logistics management by providing the required information to provide basis for making decisions on components of logistics management such as acquisition of production materials, organization production capacity, quantity of inventories as well as making decisions regarding inventories (Krager, 2010). In particular, the paper will center on the use of time series forecasting in forecasting demand in order to predict the future demand and therefore make supply and logistics management decisions that can adequately meet the current demand. Demand forecasting serves as the foundation for planning decisions with regard to supply chain and logistics management. Basically, demand forecasting is very important for organizations that carry out pull or push strategies where organizations utilizing push strategy produce materials/products in anticipation prior to getting customer requests, while organizations that use pull strategy base their production on the real customer demand (Chan & Spedding, 2000). Consequently, demand forecast is required in pull strategy where a company basis its production on the real customer demand to establish the availability of production capacity as well as inventory. Similarly, in push strategy where the companies produce goods in anticipation before receiving customer requests, demand forecasting is done in order to determine the production quantities. Therefore, demand forecasting is required to ascertain the efficacy and profitability of the organization’s future operations (Donald, 2010). Time-series forecasting and index numbers can be used in forecasting the demand in the long-term in order for the organizational management to plan the organizational activities regarding material acquisitions or production quantities, production capacity as well as the amount of inventories and resources required. Normally, the basis of demand forecasting is the regularity in time series allied to the changes within economic indexes, for instance fluctuations in currency rate or adjustments in functioning conditions or environment modifications (Krager, 2010). As Donald (2010) explains, in demand forecasting a time series is stationery and the longer the observation period, the higher the likelihood that the determination of the regularities is effective. Nonetheless, since the tasks involved in demand forecasting have a relatively short cycle of time series, the time series forecasting will be based on a huge amount of relatively short time series where the considered historical demand data will be changed in each case (Krager, 2010). According to Jenkins (2012) through historical data, time series forecasting can be utilized in indentifying seasonality, cyclic patterns as well as trends. After identifying specific forecast elements, time series forecasting presumes that future will mirror the past. This presumption is normally practically correct within the short term and thus time series forecasting is most suitable in short-range forecasting. When there are significant changes in trends, the demand pattern will have a turning point. To carry out an analysis of the pattern and movement of historical data to determine the cyclical characteristics is done which is followed by developing time series forecasts depending on the particular characteristics (Hagan & Behr, 2008). Through time series forecasting, it is possible to identify demand patterns that have been there in the past and make an assumption that the same patterns will continue in future. In forecasting using time series, four key elements in the past demand are used and they include average, trend, seasonal as well as cyclical element. The average base in time series forecasting is the average level of demand which can or might not change with time (Jenkins, 2012). The trend is the general upward or downward flow in demand that happens with time while seasonal aspect is the repetitive pattern that occurs over time for instance where the demand maybe comparatively high in summer and drop in spring. The cyclical element is the demand patterns that are consistent over long time even years, though this is not common in time series forecasting because time series is most appropriate in short time changes (Donald, 2010). In forecasting, time series process uses past data in estimating parameters, the estimated parameters are used in determining how effective the time series model could have forecasted past demand and finally the estimated parameters are used in forecasting future demand. Time series forecasting is appropriate in demand forecasting because of its stability in that observations that are intermittent and do not fit with the other time series do not affect the forecasting. In addition, time series forecasting responds to changes fast in time series pattern and this indicates the responsiveness of the time series forecasting model (Hagan & Behr, 2008). The stability and responsiveness of time series in shown by the graph below: Figure 1: Stability vs. responsiveness. From the above figure, in the first eight periods, the demand was forecasted very well because the actual demand was forecasted. But in period nine the demand increased significantly and therefore this has a likelihood of causing operation problems in period 9. Therefore, there is need to establish if the demand upward increase is temporary or permanent through aspects such as support system and administration aspect (Krager, 2010). To forecast demand in supply chain and logistics management, four time series procedures are used as follows: Moving Average To forecast demand, moving average makes use of the latest duration’s sales where the average can utilize any number of preceding time periods. Some of the suitable average durations that can be used include 1-, 3-, 4-, and 12-period averages. A 1-duration moving average in demand forecasting results in successive duration’s forecast being forecasted using the last duration’s sales while a 12-duration moving average, for instance monthly utilize the average of the last 12 time durations. Every time a new time period of actual data is availed, the oldest time duration’s data is replaced by it and hence the number of time durations integrated within the average is constant. This method is suitable when it is not likely that there is trend within the current demand. The moving average method has one parameter k which indicates the number of durations utilized within the average (Molina, 2010). Therefore, the forecast for a specific duration is the average of the preceding k durations. For instance, for duration t+1 If the value of k is high, it indicates stability in demand while a lower value of k indicates responsiveness in demand. When the demand forecasts are necessary for numerous durations in future, then it is appropriate to use all succeeding periods in forecasting the first period (Molina, 2010). Exponential smoothing In order to forecast demand, this method bases the estimations of future sales on the weighted average of preceding demand and forecast levels where the new demand forecast is a function of the previous forecast augmented by fraction of the variance between the previous forecast and actual sale gotten (Molina, 2010). The calculation of exponential smoothing is as follows: Where α is the smoothing constant and its value is supposed to be between 0-1. Additionally, there should be higher values that lead to more responsiveness and lower values to more stability (Molina, 2010). This can be illustrated through expansion of the above calculation as follows: This expression indicates that higher weights are given to latest demand trends. In case the forecasts are needed for numerous time durations in future, the forecast of the initial duration is utilized for all successive durations (Patel, 2013). This can be illustrated by the following example where the table presents the actual demand data for the last 9 months of 201 together with forecasting that were obtained through simple moving average (6months) as well as exponential smoothing where α = 0.2. Month April 2012 May 2012 June 2012 July 2012 August 2012 September 2012 October 2012 November 2012 December 2012 Actual 115 111 120 99 132 120 141 116 141 Simple Moving Average 104.8 108.0 111.7 111.5 109.5 114.8 116.2 120.5 121.3 Exponential Smoothing 104.8 106.8 107.7 110.1 107.9 112.7 114.2 119.5 118.8 It is possible to forecast the demand for the first 3 months of 2013 using both exponential smoothing and simple moving average. Because the two procedures do not deal with trend or seasonality, the forecasts for the first 3 months of 2013 will be similar for every month and the forecasts can be obtained as follows: Simple moving average forecast = (99+132+120+141+116+141)/6 = 124.8 Exponential smoothing forecast = 0.2(141) + (1.0 - 0.2) (118.8) = 123.3 Double exponential smoothing This method forecasts the demand where there is trend. Here, the demand is forecasted by calculating an estimate of the expected demand level in specific time duration and the duration is denoted by Et. The estimate of the trend which represents the rise or drop per duration is denoted by Tt 9 (Patel, 2013). The forecast for duration t+n where n is time durations after the present time period is given by: And the base and trend values are updated as follows: The factor α (0 < α < 1) is utilized to smooth the base and the factor β (0 < β < 1) is utilized in smoothing the trend. Higher values of the smoothing parameters result to an increased responsiveness whereas lower values result to more stability. Extended exponential smoothing This method integrates the effect of both trend and seasonality where particular for the values can be identified. This method allows new forecast to be calculated fast with small data and the ability of the method to respond is dependent on the smoothing constant values. When the smoothing constant values are higher, there is fast responsiveness although this can result to overreaction and forecast precision problems (Patel, 2013). Therefore when managers use extended exponential smoothing to forecast demand, methodical and steady procedure updating of alpha replaces the managerial decisions. The method can be utilized for any number of future durations provided the managers interpret the last term to be the latest estimate of seasonality for the duration that is being forecasted (Krager, 2010). Limitations of Time-series Forecasting As per Patel (2013) explains the major limitation with time series forecasting is that it can lead to erroneous forecasting because whereas the methods used in time-series forecasting can link historical patterns into future forecasts, the methods are not effective in integrating the input of projected future events. Therefore, to ensure accuracy during forecasting it is important to integrate the forecast techniques with suitable support and administrative systems (Patel, 2013). The forecast support system should consist of the supply chain intelligence that should collect data and do the analysis, develop the forecast and inform the pertinent staff and planning systems regarding the forecast. This will enable all external aspects like the effect of advertisement, changes in price, product line changes, competition activities as well as economic situation to be taken into account when calculating the forecasts. It is therefore extremely vital for the time series forecast procedure to incorporate support system in order to aid in the maintenance, updating and manipulating of the historical database and the forecast as well (Borisov & Kornienko, 2008). Additionally, it is important to integrate the forecast with administration which consists of the organizational, technical, motivational, and workforce elements of forecasting and its incorporation into the other organizational functions. The aspect of organizational includes the specific roles and responsibilities and thus the roles and responsibilities should be defines clearly and in detail (Arnis, 2010). Other limitations with time series forecasting is that some techniques used can be considerably unresponsive to change and they also need updating and maintenance of huge historical data when calculating forecasts. Where the historical sales differences are big, it is impossible to rely on the average values to make accurate forecasts. In addition, even though techniques such as adaptive method can adjust errors automatically, the methods can at times react excessively through unsystematic interpretations of the errors to be trend or seasonality and these errors can result to increased errors in future (Chen, 2011). To overcome these limitations, it is recommended to combine several time-series forecasting techniques instead of relying on one method to do the forecasting. For instance, combining moving average with exponential smoothing allows a speedy calculation of a new forecast without having to use enormous historical data or updating since exponential smoothing is very adjustable to computerized forecasting and the method also allows monitoring and changing of technique sensitivity. To track possible errors in forecasts, for instance an automatic tracking signal can be included in the adaptive smoothing to allow monitoring of errors (Borisov & Kornienko, 2008). Bibliography Arnis, K, 2010, Demand Forecasting Based on the Set of Short Time Series, Scientific Journal of Riga Technical University, 2(2):80-105. Borisov, A & Kornienko, Y, 2008, A study of methods of classifier construction and updating, Automatic control and computer sciences, 42(6): 300-305. Chan, K & Spedding, T, 2000, Forecasting demand and inventory management using Bayesian time series, Integrated Manufacturing Systems, 11(5): 331 – 339. Chen, H, 2011, Forecasting Tourism Demand Using Time Series, Artificial Neural Networks and Multivariate Adaptive Regression Splines: Evidence from Taiwan, International Journal of Business Administration, 2(2) 12-24. Donald, J, 2010, Supply Chain Logistics Management, Sydney: McGraw-Hill. Hagan, M & Behr, S, 2008, The time series approach to short term load forecasting, IEEE Trans. Power System, 2(3). Jenkins, V, 2012, Time series analysis: forecasting and control, San Francisco: Holden-Day Inc. Krager, H, 2010, Demand Forecasting in a Supply Chain, Journal of International Money and Finance, 12 (3): 513-534. Molina, B, 2010, Day-ahead electricity price forecasting using the wavelet transform and ARIMA models, IEEE Trans. Power System, 20(4). Patel, M, 2013, An Analysis of Short Term Load Forecasting by Using Time Series Analysis, International Journal of Research in Computer and Communication Technology, 2(2): 2278-5841. Read More