Forecasting
1
What is forecasting?
Name two forecasts you utilize: ______, ______
What do we forecast in operations management?
______,______, ______, ______
Why is forecasting important in operations management?
Accuracy of forecasts:
2
1. not an exact science.
2. involves error
3. assumes the past predicts the future
4. can’t account for unplanned occurrences
5. forecast is NOT EQUAL to actual demand
6. accuracy decreases as the time period to forecast increases
Who has the advantage in forecasting and why?
3
1. Craft production? ______
2. Mass production? ______
3. Lean production? ______
How do you know if your forecast was accurate?
Week / Forecast (predicted) / Actual Demand1 / 200,000 / 205,000
2 / 198,000 / 200,000
3 / 197,000 / 196,000
4 / 196,500 / 196,000
5 / 194,000 / 195,000
6 / 193,500 / 192,000
7 / 192,000 / 191,000
8 / 190,000 / 191,000
4
1. What is the trend (pattern) in the actual demand data over 8 weeks?
2. How does actual demand compare to predicted or forecasted demand?
3. Plot the demand data and determine the trend.
4. Plot the forecast data on the same chart. How does the forecast compare to the actual demand data?
Approaches to Forecasting
7
1. Qualitative – using opinions, hunches, and subjective input to determine forecast. Where does the information come from?
2. Quantitative – using numbers. Data driven forecasting. Example: forecasting weekly demand for electricity. What are the best predictors of electricity for week 1 of January?
Determining Trends (patterns) in Historical Data
Open the file from the web page titled “ Trends in Historical Data”
Homework
Open the file from the web page titled “Production Trends Homework”
Forecasting – Averaging Techniques
9
1. Smooths fluctuations in time series data
2. Exhibits less variability than actual demand
3. Reflects recent values of a times series
4. Works best for slow, incremental changes
5. Types of averaging forecasts: Na ve, Moving Averages, Weighted Averages, Exponential Smoothing
Naïve Forecasts – any period’s forecast, equals the previous period’s actual demand.
Actual Demand / ForecastWeek 1 / 49
Week 2 / 52
Week 3
Week 4
What is the forecast for Week 3? ______
If actual demand in Week 3 was 55, what is the Naïve forecast for Week 4? ______
Advantages of Naive Forecasting
11
1. Cheap
2. Quick and Easy
3. No data analysis
4. Easy to understand
Is Naïve forecasting a good tool? ______
If you have seasonal demand, how would you use naïve forecasting?
Lawn Mower Demand
2000 Actual Demand / 2001 Naïve Forecast / 2001 Actual DemandJan. / 2 / 10
Feb / 4 / 15
March / 10 / 25
April / 30
12
1. Use Naive Forecasting to prepare the 2001 forecast for Lawn mowers.
2. In 2001, suppose your actual demand was MUCH higher than your forecast. In other words, your company was experiencing actual growth in sales. You noticed this growth trend in March of 2001. How would you adjust your Na ve forecasts for this new growth trend?
Moving Averages - averages a number of recent actual demand values, updated as new values become available.
Demand for Peanut Butter
Week / Actual Demand / Forecast1 / 50
2 / 52
3 / 54
4 / 49
5 / 51
6
7
14
1. Use a three-period moving average to generate a forecast for week 6. ______
2. Use a five-period moving average to generate a forecast for week 6.______
3. If actual demand for week 6= 53, what is the three-period moving average forecast for week 7? ______
4. What is the five-period moving average forecast for week 7? ______
Advantages of Moving Averages
16
1. Moving averages lag actual demand.
2. The number of data points in an average, determine the sensitivity to each new data point. Huh? Explain this. As the number of data points in an average increases (say we go from a 2-week moving average to a 7-week moving average), the sensitivity to a new data point increase or decrease? Explain.
3. Easy to compute
4. Easy to understand
Disadvantages
17
1. All values are weighted equally. The oldest value has the same weight as the newest value.
In-Class Assignment
Demand for Peanut butter.
Week / Actual Demand / 3-week Moving Average Forecast / 5-week Moving Average Forecast1 / 50
2 / 52
3 / 54
4 / 49
5 / 51
6 / 53
7 / 54
8 / 47
9 / 51
10 / 53
11 / 49
12 / 48
13 / 51
14 / 52
15 / 65
16 / 67
17 / 69
18 / 50
19 / 51
20 / 49
1. Calculate the three-week moving average forecasts for weeks 4-20. Why must you begin with week 4?
2. Calculate the five-week moving average forecasts for weeks 6-20. Why must you begin with week 6?
18
3. Plot actual demand, the three week forecasts, and the five week forecasts. What does the picture tell you about the accuracy of your two forecasting techniques?
Weighted Moving Average – assign greater weights to the most recent values in a time series.
Weights for Peanut Butter Demand
Recent = .50
Next = .30
Next = .20
Calculate the weighted three-week moving average for weeks 15, 16,17, 18, 19 for the Peanut Butter Demand data.
Week 15 forecast= week 14 demand (.50) + week 13 demand (.30) + week 12 demand (.20)
Week 15 forecast =52 (.50) + 51 (.30) + 48 (.20)
Week 15 forecast = 50.9
Week 16 forecast =
Week 17 forecast =
Week 18 forecast =
Week 19 forecast =
Plot these weighted average forecasts on your graph with the 3-week and 5-week moving average forecasts.
21
1. Which forecast appears to be a better predictor of demand? Why?
2. When demand increased in weeks 14, 15, and 16, which forecast appeared to be a better predictor? Why?
Exponential Smoothing – each new forecast is based on the previous forecast plus a percentage of the difference between the previous forecast and actual demand from the period.
Next forecast = Previous forecast + a (Actual demand from previous period – Previous forecast)
Where:
(Actual demand from previous period– Previous forecast) = Forecast Error
a = percentage of error
Where a=.10, use exponential smoothing to calculate forecasts for Weeks 5, 6, and 7 of Peanut Butter Demand, based on a three week moving forecast for week 4.
Week 5 forecast = week 4 forecast + .10 ( actual demand week 4 – week 4 forecast)
= 52 +.10 (49-52)
= 51.7
Week 6 forecast =
Week 7 forecast =
For practice, using a=.40, use exponential smoothing to forecast weeks 14, 15, and 16, using a week 13 forecast of 50.
Week 14 forecast =
Week 15 forecast =
Week 16 forecast =
a determines how quickly the forecast adjusts to changes in demand. The lower the value of a, the slower the forecast will adjust. The closer the value of a is to 1, the greater the responsiveness of the forecast.
Homework
Problems 3 & 4 at the end of the chapter.
Accuracy and Control of Forecasts
Will a forecast be accurate?
Why does the decision maker need a measure of accuracy?
Why should we monitor forecast error?
Forecast Error – difference between actual demand and the forecast.
Error= Actual demand – forecast
If Forecast Error is positive,
a. the forecast was too low
b. the forecast was too high
If forecast error is negative
c. the forecast was too low
d. the forecast was too high
Two techniques for measuring forecast accuracy:
Mean Absolute Deviation (MAD) and Mean Squared Error (MSE)
MAD = S ½ Actual demand – forecasted demand ½
n
MAD is the sum of the absolute value of error divided by n, the number of periods forecasted.
MSE = S (Actual demand – Forecasted demand)2 n-1
MSE is the sum of forecast error squared, divided by n-1, where n is the number of periods forecasted.
Both MAD and MSE summarize forecast error. When selecting a forecasting technique, do you want to pick a technique with a large or small value of MAD and/or MSE? Why?
MAD and MSE practice exercise
Examine the forecast accuracy of the 3- week moving average forecast and the 5-week moving average forecast for Peanut Butter Demand, weeks, 6-17. Which forecast is more accurate? Why?
Week / DEMAND / 3-wk Mov-ing avg. fore-cast / ERROR / Abso-lute value of error / ERROR2 / 5-wk Mov-ing avg. fore-cast / ERROR / Abso-lute value of error / ERROR21 / 50
2 / 52
3 / 54
4 / 49 / 52
5 / 51 / 51
6 / 53 / 51 / 51
7 / 54 / 51 / 51
8 / 47 / 52 / 52
9 / 51 / 51 / 50
10 / 53 / 50 / 51
11 / 49 / 50 / 51
12 / 48 / 51 / 50
13 / 51 / 50 / 49
14 / 52 / 49 / 51
15 / 65 / 50 / 50
16 / 67 / 56 / 53
17 / 69 / 61 / 56
MAD = S ½ Actual demand – forecasted demand ½
n
MSE = S (Actual demand – Forecasted demand)2
n-1
MAD for 3-wk:______
MSE for 3-wk:______
MAD for 5-wk:______
MSE for 5-wk:______
Homework:
Homework: page 126, problem 21, a,b,c,
page 127, problem 22, a
33