Forecasting for
Operations
21/10/19
Forecasting
Definition
Forecasting is the process to estimate the future value of a variable of interest.
Why to forecast
• Scheduling
Efficient use of resources requires the scheduling of production, transportation,
cash, personnel and so on. Forecast of the level of demand for product, material,
labor, financing or service are essential input to such scheduling.
• Acquiring resources
The lead time for acquiring raw materials, hiring personnel, or buying machinery
and equipment vary from a few days to several years. Forecasting is required to
determine future resource requirements.
• Determining resource requirement
Application of forecasting
• Economic forecasting
• Earthquake prediction
• Player and team performance in sports
• Political forecasting
• Product forecasting
• Sales forecasting
• Telecommunications forecasting
• Weather forecasting
• Flood forecasting
• Supply chain management
• Forecasting can be used in supply chain management to ensure that the right product is at the right
place at the right time. Accurate forecasting will help a retailers to
• Reduce excess inventory
• Increase service level
Strategic importance of forecasting
Supply chain management
Human resources
Capacity
Contd…
• Comparison between forecasting existing products versus new product
forecasting
Forecasting existing New product forecasting
products
Data History Assumption
Analytics Statistical Judgmental
Forecast Point Range
Plan Certainties Contingencies
Measurement Accuracy Meaningfulness
Type of forecasts
• Duration
• Short-range forecast: less than 3months and span of up to 1 year
• Medium-range forecast: 3month – 3years
• Long-range forecast: more than 3years
• Nature of output
• Point forecast
• Interval forecast
• Density forecast
• Procedure
• Qualitative
• Quantitative
Methods of Demand Forecasting
Demand Forecasting
Qualitative Analysis Quantitative Analysis
Customer Survey Sales Force Composite
Time Series Analysis Causal Analysis
Executive Delphi
Opinion Method Simple Moving Simple
Average Exponential
Smoothing
Past Analogy
Holt’s Double Winters’s Triple
Exponential Exponential
Smoothing Smoothing
Forecast by Linear
Regression Analysis
Seven steps in the forecasting
1. Determine the use of forecasting
2. Select the items to be forecast
3. Determine the time horizon of the forecast
4. Select the forecasting model(s)
5. Gather the data needed to make forecast
6. Make the forecast
7. Validate and implement the results
Time series data pattern
Identification of data pattern is an important step in selecting appropriate model
for forecasting.
Different data patterns are
• Horizontal: fluctuate around the constant mean also called stationary data
series.
• Seasonal: series is influence by seasonal factor
• Cyclical: repeat the pattern with varies time period, rise and falls are not of a
fixed period.
• Trend: pattern exist when there is long term increase/decrease in the data.
Actual Actual
Demand Demand
Time
Time
No growth or decline trend; no seasonal variation – simple (or No growth or decline trend; seasonal variation present –
weighted) moving average; simple exponential smoothing simple moving average
Actual
Actual
Demand
Demand
Time Time
Linear growth (or decline) trend; no seasonal variation – Linear growth (or decline) trend; seasonal variation present –
Holt’s double exponential smoothing Winters’s triple exponential smoothing; linear regression
analysis
Naïve approach
•
Simple moving average
t n
1
Ft
n
D
t t 1
t
Disadvantage Moving average:
1. used selected data points
2. assign equals points to all data point
3. applicable only for stationary data series
• Exponential smoothing
𝑡 +1= ∝ 𝐷 𝑡 + ( 1 −∝ ) 𝐹𝑡
𝐹
Forecasting by linear regression
Best fit line is
extrapolated to find
the forecast for the
future
Forecast
Actual Best fit line with slope b
Demand / y = a + b. x
Forecast (Least squares method) using
the past demand data
y intercept = a
0
Time
Scatter Diagram and Best Fit Line (Forecasting by Linear Regression
Analysis)
Forecast control limit
Forecast Error
0 + 3. s Upper Control Limit (UCL)
Targeted or Aimed-at Mean Forecast Central Line (CL)
Error = 0
Time
0 - 3. s Lower Control Limit (LCL)
Forecast Control Limits
Holt’s model
•
• is the level which represents the smoothed value up to and including the
last data slop of the line given by
• the forecasted value for next period i.e. t+1
• The value of and can be updated using
• To initialize the value
• i.e. first period actual demand
•/
• Capture the idea of exponential smoothing
• Level and trend at all points
Holt’s-Winter model
• Multiplicative seasonality
Yt
Level : L t 1 L t 1 b t 1
S t s
Treand : b t L t L t 1 1 b t 1
Yt
seasonal : S t 1 S t s
Lt
Forecast : Ft m L t b t m S t s m
s : lenghth of seasonaliy, L t : level of series; b t : trend; S t : seasonal component, Ft m : is the forecast for m period ahead
• Additive seasonality
Level : L t Yt St s 1 L t 1 b t 1
Treand : b t L t L t 1 1 b t 1
seasonal : St Yt L t 1 St s
Forecast : Ft m L t b t m St s m
Measurement of forecast error
1 n
Mean Error(ME) e t
n t 1
1 n
Mean Absolute Error(MAE) e t
n t 1
1 n 2
Mean square Error(MSE) e t
n t 1
1 n
Mean Absolute Deviation(MAD) Yt Y
n t 1
2
1 n
Mean Standard Deviation (MSD) Yt Y
n t 1
Y Ft
Percentage Error (PE t ) t 100
Yt
1 n
Mean Percentage Error(MPE) PE t
n t 1
1 n
Mean Absolute Percentage Error(MAPE) PE t
n t 1
