SCM 2160 Forecasting PDF

SCM 2160 Professor: Dan Shin For Today: Forecasting and Understanding Demand Forecasting Process Forecasting Methods Forecasting Errors NOT for distribution outside of SCM 2160 [DS] 1 Forecasting Forecasting: a method used to predict future...

SCM 2160 Professor: Dan Shin For Today: Forecasting and Understanding Demand Forecasting Process Forecasting Methods Forecasting Errors NOT for distribution outside of SCM 2160 [DS] 1 Forecasting Forecasting: a method used to predict future events using past and present data Forecasting in OM: Strategic: future products and markets Planning: demand of products and services Based on multiple types of data or analysis: mathematical models, historical data, simulations, and expert opinions Can be used for process management and supply chain management NOT for distribution outside of SCM 2160 [DS] 3 Forecasting Forecasting horizon: Long-range (longer than 2 years): mainly for strategy Mid-range (weekly/monthly up to 2 years): for planning Short-range (hourly/daily for up to several months): for scheduling Why is forecasting important for firms? NOT for distribution outside of SCM 2160 [DS] 4 Understanding Demands For this course, we will be mainly looking at time series forecasting Time series: a set of repeated observations of demand measured for a product/service over successive time periods Assumptions: 1. Past demand is a useful predictor of future demand 2. Record of past demand is available Time series data can reveal patterns of demand over time 5 basic patterns you can find from time series data: NOT for distribution outside of SCM 2160 [DS] 5 Demand Patterns 5 basic patterns you can find from time series data: Demand Average: stable demand over time near the mean Time NOT for distribution outside of SCM 2160 [DS] 6 Demand Patterns 5 basic patterns you can find from time series data: Random movement Demand Trend: Systematic increase or decrease in the mean over time Time NOT for distribution outside of SCM 2160 [DS] 7 Demand Patterns 5 basic patterns you can find from time series data: Demand Seasonal: Repeatable pattern of increase or decrease in demand, depending on day, week, month, season Time NOT for distribution outside of SCM 2160 [DS] 8 Demand Patterns 5 basic patterns you can find from time series data: Demand Cyclical: Longer term periodic behaviour of gradual increase/decrease that is less predictable (years or decades) Time NOT for distribution outside of SCM 2160 [DS] 9 Demand Patterns 5 basic patterns you can find from time series data: Demand Random variation: any remaining variability that cannot be explained and unforecastable Time NOT for distribution outside of SCM 2160 [DS] 10 Managing Demand Same resources Complementary Services Different demand cycles Increase demand Promotional Pricing Shift demand to new period Level demand based on capacity Prescheduled Appointment Adjust prices in real-time based demand Revenue Management Backlogs, Backorders, Accumulate orders for future delivery Stockouts Decrease service level and risk losing customers NOT for distribution outside of SCM 2160 [DS] 11 Forecasting Process 1. Identify the purpose of the forecast 2. Availability and collection of inputs required 3. Selection of techniques and models to use 4. Check for accuracy of the results and model (if accuracy is not acceptable, go back to step 3 and adjust the parameters or select new models) 5. Forecast over planning horizon and incorporate qualitative information NOT for distribution outside of SCM 2160 [DS] 12 Forecasting Process 1. Identify the purpose of the forecast What are you trying to predict? What are your inputs? Individual Product vs. Units vs. $ vs. Labour Product Families NOT for distribution outside of SCM 2160 [DS] 13 Forecasting Process 1. Identify the purpose of the forecast What are you trying to predict? What are your inputs? History of past Consumer Past forecasts demands research Data from partners Future promotions and collaborators NOT for distribution outside of SCM 2160 [DS] 14 CPFR Collaborative Planning, Forecasting, and Replenishment Requires collaboration and sharing of information with suppliers and customers Independent forecasts generated, and then compared, which is then adjusted continuously until forecasts approach consensus Why would this be of value to the buyers and suppliers? NOT for distribution outside of SCM 2160 [DS] 15 Forecasting Process 2. Availability and collection of inputs required Do you have enough inputs? Is there a way to collect the inputs? Sample size: large vs. small Cost: feasible? NOT for distribution outside of SCM 2160 [DS] 16  Forecasting Process 3. Selection of techniques and models to use Does the technique match the inputs? Correct technique for your purpose? Judgement methods Causal methods Time-series analysis NOT for distribution outside of SCM 2160 [DS] 17 Forecasting Process 4. Check for accuracy of the results and model Is the accuracy acceptable? (if accuracy is not acceptable, go back to step 3 and adjust the parameters or select new models) 5. Forecast over planning horizon and incorporate qualitative information Are you looking to forecast short-term, mid-term, or long-term? Plan your production and capacity according to the planning horizon NOT for distribution outside of SCM 2160 [DS] 18 Qualitative vs. Quantitative Qualitative Quantitative Subjective Objective Can incorporate a variety of Can incorporate large a information volume of information Do not require numerical Do not have to rely on few data individuals Results may be biased Numerical data may not be Results may be conflicting available Mathematical models may be too simplistic NOT for distribution outside of SCM 2160 [DS] 19 Time-series Methods Makes predictions based only on historical data about the dependent variable Assumptions (previously discussed) 1. Past demand is a useful predictor of future demand 2. Information on past demand is available Popular methods: 1. Naïve 2. Simple moving average 3. Weighted moving average 4. Exponential smoothing NOT for distribution outside of SCM 2160 [DS] 20 Naïve Method Naïve method: forecast for next period is the demand for most recently observed period Only applicable for short-term forecasts Sensitive to random variation 𝐹𝑡+1 = 𝐷𝑡 𝐹𝑡+1 = 𝑓𝑜𝑟𝑒𝑐𝑎𝑠𝑡 𝑓𝑜𝑟 𝑝𝑒𝑟𝑖𝑜𝑑 𝑡 + 1 𝑜𝑟 𝑎𝑛𝑦 𝑓𝑢𝑡𝑢𝑟𝑒 𝑝𝑒𝑟𝑖𝑜𝑑 𝐷𝑡 = 𝑎𝑐𝑡𝑢𝑎𝑙 𝑑𝑒𝑚𝑎𝑑𝑛 𝑖𝑛 𝑡ℎ𝑒 𝑚𝑜𝑠𝑡 𝑟𝑒𝑐𝑒𝑛𝑡 𝑝𝑒𝑟𝑖𝑜𝑑 𝑡 When would it be a good idea to use the naïve model? NOT for distribution outside of SCM 2160 [DS] 21 Naïve Method Week Demand Forecast 1 63 2 70 3 78 4 51 5 56 6 67 7 80 NOT for distribution outside of SCM 2160 [DS] 22 Simple Moving Average Simple moving average: forecast for next period is the average demand for n most recent periods Smooths out random variations Useful for stable demand 𝐹𝑡+1 = (𝐷𝑡 + 𝐷𝑡−1 + … + 𝐷𝑡− 𝑁−1 /𝑁) 𝑁 = 𝑡𝑜𝑡𝑎𝑙 𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑝𝑒𝑟𝑖𝑜𝑑𝑠 𝑢𝑠𝑒𝑑 𝑖𝑛 𝑡ℎ𝑒 𝑐𝑎𝑙𝑐𝑢𝑙𝑎𝑡𝑖𝑜𝑛 − 𝑠𝑒𝑛𝑠𝑖𝑡𝑖𝑣𝑖𝑡𝑦 𝑡𝑜 𝑟𝑎𝑛𝑑𝑜𝑚 𝑣𝑎𝑟𝑖𝑎𝑡𝑖𝑜𝑛 𝑑𝑒𝑝𝑒𝑛𝑑𝑠 𝑜𝑛 𝑁 − 𝑁 𝑖𝑠 𝑠𝑢𝑏𝑗𝑒𝑐𝑡𝑖𝑣𝑒 NOT for distribution outside of SCM 2160 [DS] 24 Simple Moving Average Week Demand Forecast 1 63 2 70 3 78 4 51 5 56 6 67 7 80 Calculate 2-period moving average NOT for distribution outside of SCM 2160 [DS] 25 Weighted Moving Average Weighted moving average: forecast for next period equals average demand for n most recent periods, and each observation of demand can have its own weight; the sum of weights are always equal to 1 Weights represent the varying amounts of influence of past demand on forecast 𝐹𝑡+1 = 𝑊𝑡 𝐷𝑡 + 𝑊𝑡−1 𝐷𝑡−1 + … + 𝑊𝑡− 𝑁−1 𝐷𝑡− 𝑁−1 𝑊𝑡 = 𝑤𝑒𝑖𝑔ℎ𝑡 𝑎𝑠𝑠𝑖𝑔𝑛𝑒𝑑 𝑡𝑜 𝑡 ′ 𝑠 𝑑𝑒𝑚𝑎𝑛𝑑 − 𝑀𝑜𝑠𝑡 𝑐𝑜𝑚𝑚𝑜𝑛: 𝑚𝑜𝑠𝑡 𝑐𝑢𝑟𝑟𝑒𝑛𝑡 𝑑𝑒𝑚𝑎𝑛𝑑 ℎ𝑎𝑠 𝑚𝑜𝑠𝑒 𝑖𝑛𝑓𝑙𝑢𝑒𝑛𝑐𝑒 − 𝑊𝑒𝑖𝑔ℎ𝑡𝑠 𝑎𝑛𝑑 𝑁 𝑎𝑟𝑒 𝑠𝑢𝑏𝑗𝑒𝑐𝑡𝑖𝑣𝑒 NOT for distribution outside of SCM 2160 [DS] 27 Weighted Moving Average Week Demand Forecast 1 63 2 70 3 78 4 51 5 56 6 67 7 80 Calculate 3-period weighted moving average Weights: 0.8 to most recent period, 0.15 to second most recent period, and 0.05 to most distant period NOT for distribution outside of SCM 2160 [DS] 28 Exponential Smoothing Exponential smoothing: a weighted moving average assigning differing levels of weight to recent demand compared to older historical data Requires only three data points: last period’s forecast, last period’s demand, smoothing parameter (𝛼) 𝐹𝑡+1 = 𝛼𝐷𝑡 + 1 − 𝛼 𝐹𝑡 𝛼 = 𝑠𝑚𝑜𝑜𝑡ℎ𝑖𝑛𝑔 𝑝𝑎𝑟𝑎𝑚𝑒𝑡𝑒𝑟 0 < 𝛼 ≤ 1 ; 𝑐𝑜𝑚𝑚𝑜𝑛𝑙𝑦 0.01 𝑡𝑜 0.5 − 𝑙𝑜𝑤 𝛼 → more smoothing; high 𝛼 → more responsive to variability − 𝛼 𝑖𝑠 𝑠𝑢𝑏𝑗𝑒𝑐𝑡𝑖𝑣𝑒 NOT for distribution outside of SCM 2160 [DS] 30 Exponential Smoothing Week Demand Forecast 1 63 2 70 3 78 4 51 5 56 6 67 7 80 Calculate exponential smoothing forecast with 𝛼 = 0.15 And an initial forecast of 60 for week 1 NOT for distribution outside of SCM 2160 [DS] 31  Forecast Error Forecast error: difference between predicted demand and actual demand 𝐸𝑡 = 𝐷𝑡 − 𝐹𝑡 Can be random or the result of an inappropriate model Measures of forecast error: 1. Cumulative sum of forecast errors (CFE) 2. Mean squared error (MSE) 3. Mean absolute deviation (MAD) 4. Mean absolute percentage deviation NOT for distribution outside of SCM 2160 [DS] 33 Addressing the Bias Cumulative sum of forecast errors (CFE): assesses total error in forecasts over time 𝐶𝐹𝐸 = ෍ 𝐸𝑡 Used to evaluate presence, amount, and direction of bias If forecast is consistently lower than demand, CFE will be highly positive If forecast is consistently higher than demand, CFE will be highly negative NOT for distribution outside of SCM 2160 [DS] 34 Assessing the Spread Mean square error (MSE) 2 Indicates on average, how close forecast is to demand σ 𝐸𝑡 Magnifies large errors Standard deviation accomplishes the same function 𝑛 Mean absolute deviation (MAD) σ |𝐸𝑡 | Simple measure of magnitude of error Does not reveal direction of the bias 𝑛 Mean absolute percentage error (MAPE) σ |𝐸𝑡 |/𝐷𝑡 Contextualizes magnitude of error relative to demand ∗ 100 𝑛 NOT for distribution outside of SCM 2160 [DS] 35 Seasonality Adjustment Multiplicative Seasonal Method: a seasonal factor is calculated, which is then multiplied by an estimate of average demand to adjust for seasonality 1. Take demand for a period, divide by the number of seasons in that period to get average demand per season (e.g., annual demand divided by 12 to calculate average monthly demand) 2. Divide actual demand for that season by predicted average demand per season calculated above to calculate seasonal index 3. Repeat for each period (e.g., year) of data you have, then calculate the average seasonal index across all periods 4. Forecast demand for the next period any way you choose. Divide forecast for period by number of seasons, multiply each period by its seasonal index NOT for distribution outside of SCM 2160 [DS] 36  Seasonality Adjustment Observed Average Demand (If Seasonal Index Demand No Seasonality) (Observed ÷ Expected) Winter 100 Spring 270 Summer 800 Fall 130 TOTAL 1300 NOT for distribution outside of SCM 2160 [DS] 37 Activity: Breakout into Learning Groups Please download the spreadsheet on UM Learn A company is trying to select a forecasting model. Using the attached spreadsheet and data provided, compute the naïve forecast, 3-period simple moving average, and exponential smoothing forecasts. (for exponential smoothing, use the 3-period moving forecast as your initial input for the first week) Which forecast method should management use if the performance criterion it chooses is: CFE? MSE? MAD? MAPE? NOT for distribution outside of SCM 2160 [DS] 40