FORECASTING
What Are the Basic Rules of Forecasts
• Forecasts are almost always wrong.
o Important to measure forecast accuracy and take actions to improve when necessary
• Near-term forecasts tend to be more accurate.
• Forecasts for groups (product categories, multiple stores, etc.) tend to be more accurate.
• Forecasts are no substitute for calculated values.
Two Distinct Approaches to Forecasting
Qualitative Methods Quantitative Methods
• Used when situation is vague & little data exist • Used when situation is ‘stable’ & historical data
o New products exist
o New technology o Existing products
• Involves intuition, experience o Current technology
• ex., Forecasting sales to a new market • Heavy use of mathematical techniques
• ex., Forecasting sales of a mature products
What Are Time Series Models
• Quantitative forecasting models that use chronologically arranged data to develop forecasts.
• Assume that what happened in the past is a good starting point for predicting what will happen in the future.
• These models can be designed to account for:
o Randomness
o Trend
o Seasonality effects
• Advantages: Objective, good accuracy, predicts upturn / downturn, short to medium time, low to medium cost
• Disadvantages: technically complex, large historical data needed, software packages essential
Moving Average Models
• Based on last x periods
• Smoothes out random fluctuations
• Advantages: simple and easy to calculate, low cost, less time, good accuracy for short term and stable
conditions
• Disadvantages: can not predict downturn / upturn, not used for unstable market conditions and long-term
forecasts
What is Exponential Smoothing?
• A type of weighted moving averaging model
• Part of many forecasting packages; ideal for developing forecasts of lots of smaller items
• Needs only three numbers:
Ft-1 = Forecast for the period before current time period t
At-1 = Actual demand for the period before current time period t
a = Weight between 0 and 1
• Formula
• As “a” gets closer to 1, the more weight put on the most recent demand number
• Advantages: simple method, forecaster’s knowledge used, low cost, less time, good accuracy for short term
forecast
• Disadvantages: smoothing constant is arbitrary, not used for long-term and new product forecast
What is Single Regression?
• Develops a line equation y = a + b(x) that best fits a set of historical data points (x,y)
• Ideal for picking up trends in time series data
• Once the line is developed, x values can be plugged in to predict y (usually demand)
• For time series models, x is the time period for which we are forecasting
• For causal models (described later), x is some other variable that can be used to predict demand:
o Promotions
o Price changes
o Economic conditions
o Etc.
Naive / Ratio Method
Assumes: what happened in the immediate past will happen in immediate future
Simple formula used:
Advantages: simple to calculate, low cost, less time, accuracy good for short-term forecasting
Disadvantages: less accurate if past sales fluctuate
Common Qualitative Forecasting Methods
• Executive and outsider opinions
• Sales force composite
o This involves having product managers or sales reps developing individual forecasts, and then adding
them up
• Panel consensus & Delphi method
o Both methods have experts work together to develop forecasts
o The Delphi method has experts develop forecasts individually, then share their findings. The process
is repeated until a consensus emerges.
• Life cycle analogy
o Used when the product is new. The technique is based on the fact that many products have well-
defined life cycle stages (Growth, Maturity, and Decline)
Measures of Forecast Accuracy
Error = Actual demand – Forecast OR et = At - Ft
Mean Forecast Error (MFE)
For n time periods where we have actual demand and forecast values:
Ideal value = 0;
MFE > 0, model tends to under-forecast
MFE < 0, model tends to over-forecast
Mean Absolute Deviation (MAD)
For n time periods where we have actual demand and forecast values:
While MFE is a measure of forecast model bias, MAD indicates the absolute size of the errors
Example
Period Demand Forecast Error Absolute
Error
3 11 13.5 -2.5 2.5
4 9 13 -4.0 4.0
5 10 10 0 0.0
6 8 9.5 -1.5 1.5
7 14 9 5.0 5.0
8 12 11 1.0 1.0
Period Demand Forecast Error Absolute
Error
3 11 13.5 -2.5 2.5
4 9 13 -4.0 4.0
n=6 5 10 10 0 0.0
observations 6 8 9.5 -1.5 1.5
7 14 9 5.0 5.0
8 12 11 1.0 1.0
-2 14
MFE = -2/6 = -0.33
MAD = 14/6 = 2.33
Conclusion: Model tends to slightly over-forecast, with an average absolute error of 2.33 units.
Tracking Signal
Used to pinpoint forecasting models that need adjustment
Rule of Thumb:
As long as the tracking signal is between –4 and 4, assume the model is working correctly
Other Measures
