Ima supply chain 2 sales forecasting and the myth of exponential smoothing
1. IMA SUPPLY CHAIN 2:
SALES FORECASTING AND
THE MYTH OF EXPONENTIAL
SMOOTHING
Tony Dear
IMA
add@invman.com
2. Mathematics and Managers
It’s easy to sell Mathematical Algorithms to Managers
Take Sales Forecasting
You are looking for a new Forecasting System
You are considering two vendors
System A uses a Moving Average to forecast
System B uses forecasts with Exponential Smoothing
Which would you choose?
3. Mathematics and Managers
• It’s easy to sell Mathematical Algorithms to Managers
Take Sales Forecasting
You are looking for a new Forecasting System
You are considering two vendors
System A uses a Moving Average to forecast
System B uses forecasts with Exponential Smoothing
Which would you choose?
4. Mathematics and Managers
• It’s easy to sell Mathematical Algorithms to Managers
• Take Sales Forecasting
You are looking for a new Forecasting System
You are considering two vendors
System A uses a Moving Average to forecast
System B uses forecasts with Exponential Smoothing
Which would you choose?
5. Mathematics and Managers
• It’s easy to sell Mathematical Algorithms to Managers
• Take Sales Forecasting
• You are looking for a new Forecasting System
You are considering two vendors
System A uses a Moving Average to forecast
System B uses forecasts with Exponential Smoothing
Which would you choose?
6. Mathematics and Managers
• It’s easy to sell Mathematical Algorithms to Managers
• Take Sales Forecasting
• You are looking for a new Forecasting System
• You are considering two vendors
System A uses a Moving Average to forecast
System B uses forecasts with Exponential Smoothing
Which would you choose?
7. Mathematics and Managers
• It’s easy to sell Mathematical Algorithms to Managers
• Take Sales Forecasting
• You are looking for a new Forecasting System
• You are considering two vendors
• System A uses a Moving Average to forecast
System B uses forecasts with Exponential Smoothing
Which would you choose?
8. Mathematics and Managers
• It’s easy to sell Mathematical Algorithms to Managers
• Take Sales Forecasting
• You are looking for a new Forecasting System
• You are considering two vendors
• System A uses a Moving Average to forecast
• System B uses forecasts with Exponential Smoothing
Which would you choose?
9. Mathematics and Managers
• It’s easy to sell Mathematical Algorithms to Managers
• Take Sales Forecasting
• You are looking for a new Forecasting System
• You are considering two vendors
• System A uses a Moving Average to forecast
• System B uses forecasts with Exponential Smoothing
• Which would you choose?
10. The Benefits of Exponential Smoothing
Now you may not have heard of Exponential Smoothing – but it sounds
impressive especially when its benefits are explained to you
Exponential Smoothing has two primary advantages.
1. It automatically adjusts for errors in forecasting. If the forecast for the current period is lower
than the actual demand, then the forecast for next period is automatically adjusted upwards and
vice versa. The larger the error the greater the adjustment made.
2. It gives progressively greater weights to more recent demands because these are more relevant
in forecasting the future than demands further back in history.
A simple moving average, on the other hand, makes no adjustment for
forecast error and gives equal weight to all the periods of demand history
included in the average.
Exponential Smoothing is therefore a superior method of forecasting when
compared with averaging as it is able to more readily adapt to a changing
pattern of demand.
11. The Benefits of Exponential Smoothing
• Now you may not have heard of Exponential Smoothing – but it sounds
impressive especially when its benefits are explained to you
Exponential Smoothing has two primary advantages.
1. It automatically adjusts for errors in forecasting. If the forecast for the current period is lower
than the actual demand, then the forecast for next period is automatically adjusted upwards and
vice versa. The larger the error the greater the adjustment made.
2. It gives progressively greater weights to more recent demands because these are more relevant
in forecasting the future than demands further back in history.
A simple moving average, on the other hand, makes no adjustment for
forecast error and gives equal weight to all the periods of demand history
included in the average.
Exponential Smoothing is therefore a superior method of forecasting when
compared with averaging as it is able to more readily adapt to a changing
pattern of demand.
12. The Benefits of Exponential Smoothing
• Now you may not have heard of Exponential Smoothing – but it sounds
impressive especially when its benefits are explained to you
• Exponential Smoothing has two primary advantages.
1. It automatically adjusts for errors in forecasting. If the forecast for the current period is lower
than the actual demand, then the forecast for next period is automatically adjusted upwards and
vice versa. The larger the error the greater the adjustment made.
2. It gives progressively greater weights to more recent demands because these are more relevant
in forecasting the future than demands further back in history.
A simple moving average, on the other hand, makes no adjustment for
forecast error and gives equal weight to all the periods of demand history
included in the average.
Exponential Smoothing is therefore a superior method of forecasting when
compared with averaging as it is able to more readily adapt to a changing
pattern of demand.
13. The Benefits of Exponential Smoothing
• Now you may not have heard of Exponential Smoothing – but it sounds
impressive especially when its benefits are explained to you
• Exponential Smoothing has two primary advantages.
1. It automatically adjusts for errors in forecasting. If the forecast for the current period is lower
than the actual demand, then the forecast for next period is automatically adjusted upwards and
vice versa. The larger the error the greater the adjustment made.
2. It gives progressively greater weights to more recent demands because these are more relevant
in forecasting the future than demands further back in history.
A simple moving average, on the other hand, makes no adjustment for
forecast error and gives equal weight to all the periods of demand history
included in the average.
Exponential Smoothing is therefore a superior method of forecasting when
compared with averaging as it is able to more readily adapt to a changing
pattern of demand.
14. The Benefits of Exponential Smoothing
• Now you may not have heard of Exponential Smoothing – but it sounds
impressive especially when its benefits are explained to you
• Exponential Smoothing has two primary advantages.
1. It automatically adjusts for errors in forecasting. If the forecast for the current period is lower
than the actual demand, then the forecast for next period is automatically adjusted upwards and
vice versa. The larger the error the greater the adjustment made.
2. It gives progressively greater weights to more recent demands because these are more relevant
in forecasting the future than demands further back in history.
A simple moving average, on the other hand, makes no adjustment for
forecast error and gives equal weight to all the periods of demand history
included in the average.
Exponential Smoothing is therefore a superior method of forecasting when
compared with averaging as it is able to more readily adapt to a changing
pattern of demand.
15. The Benefits of Exponential Smoothing
• Now you may not have heard of Exponential Smoothing – but it sounds
impressive especially when its benefits are explained to you
• Exponential Smoothing has two primary advantages.
1. It automatically adjusts for errors in forecasting. If the forecast for the current period is lower
than the actual demand, then the forecast for next period is automatically adjusted upwards and
vice versa. The larger the error the greater the adjustment made.
2. It gives progressively greater weights to more recent demands because these are more relevant
in forecasting the future than demands further back in history.
• A simple moving average, on the other hand, makes no adjustment for
forecast error and gives equal weight to all the periods of demand history
included in the average.
Exponential Smoothing is therefore a superior method of forecasting when
compared with averaging as it is able to more readily adapt to a changing
pattern of demand.
16. The Benefits of Exponential Smoothing
• Now you may not have heard of Exponential Smoothing – but it sounds
impressive especially when its benefits are explained to you
• Exponential Smoothing has two primary advantages.
1. It automatically adjusts for errors in forecasting. If the forecast for the current period is lower
than the actual demand, then the forecast for next period is automatically adjusted upwards and
vice versa. The larger the error the greater the adjustment made.
2. It gives progressively greater weights to more recent demands because these are more relevant
in forecasting the future than demands further back in history.
• A simple moving average, on the other hand, makes no adjustment for
forecast error and gives equal weight to all the periods of demand history
included in the average.
• Exponential Smoothing is therefore a superior method of forecasting
when compared with averaging as it is able to more readily adapt to a
changing pattern of demand.
17. A Slight Problem
• There is a slight problem with this argument – viz
that ES adapts more quickly to changes in demand
than a simple moving average
It doesn’t
Let’s look at an example
18. A Slight Problem
• There is a slight problem with this argument – viz
that ES adapts more quickly to changes in demand
than a simple moving average
• It doesn’t
Let’s look at an example
19. A Slight Problem
• There is a slight problem with this argument – viz
that ES adapts more quickly to changes in demand
than a simple moving average
• It doesn’t
• Let’s look at an example
20. A Slight Problem
• There is a slight problem with this argument – viz
that ES adapts more quickly to changes in demand
than a simple moving average
• It doesn’t
• Let’s look at an example
22. Comparing Forecast Performance
Let’s see how Exponential Smoothing and Averaging
compare in forecasting this sales increase.
Each month we generate a forecast based only on past
demand.
A commonly used average is a 6 month average
Forecast=(Sum of last 6 months sales)/6
Exponential Smoothing (ES) uses a Smoothing Factor.
The most commonly used Smoothing Factor is 0.1.
Forecast = 0.9*(Last Forecast) + 0.1*(Latest Demand)
23. Comparing Forecast Performance
• Let’s see how Exponential Smoothing and Averaging
compare in forecasting this sales increase.
Each month we generate a forecast based only on past
demand.
A commonly used average is a 6 month average
Forecast=(Sum of last 6 months sales)/6
Exponential Smoothing (ES) uses a Smoothing Factor.
The most commonly used Smoothing Factor is 0.1.
Forecast = 0.9*(Last Forecast) + 0.1*(Latest Demand)
24. Comparing Forecast Performance
• Let’s see how Exponential Smoothing and Averaging
compare in forecasting this sales increase.
• Each month we generate a forecast based only on
past demand.
A commonly used average is a 6 month average
Forecast=(Sum of last 6 months sales)/6
Exponential Smoothing (ES) uses a Smoothing Factor.
The most commonly used Smoothing Factor is 0.1.
Forecast = 0.9*(Last Forecast) + 0.1*(Latest Demand)
25. Comparing Forecast Performance
• Let’s see how Exponential Smoothing and Averaging
compare in forecasting this sales increase.
• Each month we generate a forecast based only on
past demand.
• A commonly used average is a 6 month average
• Forecast=(Sum of last 6 months sales)/6
Exponential Smoothing (ES) uses a Smoothing Factor.
The most commonly used Smoothing Factor is 0.1.
Forecast = 0.9*(Last Forecast) + 0.1*(Latest Demand)
26. Comparing Forecast Performance
• Let’s see how Exponential Smoothing and Averaging
compare in forecasting this sales increase.
• Each month we generate a forecast based only on
past demand.
• A commonly used average is a 6 month average
• Forecast=(Sum of last 6 months sales)/6
• Exponential Smoothing (ES) uses a Smoothing Factor.
The most commonly used Smoothing Factor is 0.1.
• Forecast = 0.9*(Last Forecast) + 0.1*(Latest Demand)
29. And here’s the 6M Average
The Moving Average adapts more quickly than Exponential Smoothing
30. And here’s the 6M Average
The Moving Average adapts more quickly than Exponential Smoothing
31. Now Wait a Minute
(says the ES proponent)
If you change the Smoothing Factor to 0.2 then it will adapt
better than a 6 month average.
True. But you could also change the months used in the
average. Change them to 3 and they will perform better than ES
(0.2) (Take my word for it – but it’s easy to check)
We could go on like this. In fact the most adaptable method is
to use a smoothing factor of 1
– ie next month’s forecast = last month’s demand.
You can’t beat this for adaptability - but no uses it
Now here’s the key point: The adaptability of the forecast
depends less on the method and more on the parameter you
use in the method.
32. Now Wait a Minute
(says the ES proponent)
• If you change the Smoothing Factor to 0.2 then it will adapt
better than a 6 month average.
True. But you could also change the months used in the
average. Change them to 3 and they will perform better than ES
(0.2) (Take my word for it – but it’s easy to check)
We could go on like this. In fact the most adaptable method is
to use a smoothing factor of 1
– ie next month’s forecast = last month’s demand.
You can’t beat this for adaptability - but no uses it
Now here’s the key point: The adaptability of the forecast
depends less on the method and more on the parameter you
use in the method.
33. Now Wait a Minute
(says the ES proponent)
• If you change the Smoothing Factor to 0.2 then it will adapt
better than a 6 month average.
• True. But you could also change the months used in the
average. Change them to 3 and they will perform better
than ES (0.2) (Take my word for it – but it’s easy to check)
We could go on like this. In fact the most adaptable method is
to use a smoothing factor of 1
– ie next month’s forecast = last month’s demand.
You can’t beat this for adaptability - but no uses it
Now here’s the key point: The adaptability of the forecast
depends less on the method and more on the parameter you
use in the method.
34. Now Wait a Minute
(says the ES proponent)
• If you change the Smoothing Factor to 0.2 then it will adapt better
than a 6 month average.
• True. But you could also change the months used in the average.
Change them to 3 and they will perform better than ES (0.2) (Take
my word for it – but it’s easy to check)
• We could go on like this.
In fact the most adaptable method is to use a smoothing factor of 1
ie next month’s forecast = last month’s demand.
You can’t beat this for adaptability - but no uses it
Now here’s the key point: The adaptability of the forecast depends
less on the method and more on the parameter you use in the
method.
35. Now Wait a Minute
(says the ES proponent)
• If you change the Smoothing Factor to 0.2 then it will adapt better
than a 6 month average.
• True. But you could also change the months used in the average.
Change them to 3 and they will perform better than ES (0.2) (Take
my word for it – but it’s easy to check)
• We could go on like this.
• In fact the most adaptable method is to use a smoothing factor of 1
ie next month’s forecast = last month’s demand.
You can’t beat this for adaptability - but no uses it
Now here’s the key point: The adaptability of the forecast depends
less on the method and more on the parameter you use in the
method.
36. Now Wait a Minute
(says the ES proponent)
• If you change the Smoothing Factor to 0.2 then it will adapt better
than a 6 month average.
• True. But you could also change the months used in the average.
Change them to 3 and they will perform better than ES (0.2) (Take
my word for it – but it’s easy to check)
• We could go on like this.
• In fact the most adaptable method is to use a smoothing factor of 1
ie next month’s forecast = last month’s demand.
You can’t beat this for adaptability - but no uses it
• Now here’s the key point: The adaptability of the forecast depends
less on the method and more on the parameter you use in the
method.
37. ES – History and Why It Survives
History: Exponential Smoothing was introduced not because it was a superior method of forecasting
but rather because it required less data to be held on computer. To calculate a six months average
we need to hold at least six buckets of data. The calculation for ES requires only two buckets – last
month’s demand and a forecast bucket. ES was introduced primarily because it required less disk
space in the 1950s when disk space was at a premium.
Survival: A primary reason for it being still so widely used in an era of vast gigabyte data retention
availability is that managers think it seems ‘scientific’ when compared with a simple average. If our
forecasting is lousy then we are likely to be less criticised if we say we use exponential smoothing
rather than a moving average. Or if we sell systems it sounds quite impressive when we tell you how
we use Exponential Smoothing to forecast. (This is why ES is found in so many software packages.)
38. ES – History and Why It Survives
• History: Exponential Smoothing was introduced not because it was a superior method of
forecasting but rather because it required less data to be held on computer. To calculate a six
months average we need to hold at least six buckets of data. The calculation for ES requires
only two buckets – last month’s demand and a forecast bucket. ES was introduced primarily
because it required less disk space in the 1950s when disk space was at a premium.
Survival: A primary reason for it being still so widely used in an era of vast gigabyte data retention
availability is that managers think it seems ‘scientific’ when compared with a simple average. If our
forecasting is lousy then we are likely to be less criticised if we say we use exponential smoothing
rather than a moving average. Or if we sell systems it sounds quite impressive when we tell you how
we use Exponential Smoothing to forecast. (This is why ES is found in so many software packages.)
39. ES – History and Why It Survives
• History: Exponential Smoothing was introduced not because it was a superior method of
forecasting but rather because it required less data to be held on computer. To calculate a six
months average we need to hold at least six buckets of data. The calculation for ES requires
only two buckets – last month’s demand and a forecast bucket. ES was introduced primarily
because it required less disk space in the 1950s when disk space was at a premium.
• Survival: A primary reason for it being still so widely used in an era of vast gigabyte data
retention availability is that managers think it seems ‘scientific’ when compared with a simple
average. If our forecasting is lousy then we are likely to be less criticised if we say we use
exponential smoothing rather than a moving average. Or if we sell systems it sounds quite
impressive when we tell you how we use Exponential Smoothing to forecast. (This is why ES is
found in so many software packages.)
40. In Conclusion
There is a moral to this story.
This issue goes well beyond Exponential Smoothing.
Many – certainly not all – people in business are impressed by mathematics which they
do not understand, especially when the claimed advantages of the algorithm can be
presented in an appealing manner by a competent proponent of the approach.
It is often assumed that more complex algorithms work better than simple algorithms,
which is not necessarily the case. This is rather like assuming that a medicine that tastes
horrible must be better than one that is easy to swallow.
Finally we constantly see many examples of business operations misusing forecasting
and planning algorithms they do not understand to achieve dismal performance – often
coupled with a lot of wasted time.
Commonsense is more important than algorithms.
41. In Conclusion
• There is a moral to this story.
This issue goes well beyond Exponential Smoothing.
Many – certainly not all – people in business are impressed by mathematics which they
do not understand, especially when the claimed advantages of the algorithm can be
presented in an appealing manner by a competent proponent of the approach.
It is often assumed that more complex algorithms work better than simple algorithms,
which is not necessarily the case. This is rather like assuming that a medicine that tastes
horrible must be better than one that is easy to swallow.
Finally we constantly see many examples of business operations misusing forecasting
and planning algorithms they do not understand to achieve dismal performance – often
coupled with a lot of wasted time.
Commonsense is more important than algorithms.
42. In Conclusion
• There is a moral to this story.
• This issue goes well beyond Exponential Smoothing.
Many – certainly not all – people in business are impressed by mathematics which they
do not understand, especially when the claimed advantages of the algorithm can be
presented in an appealing manner by a competent proponent of the approach.
It is often assumed that more complex algorithms work better than simple algorithms,
which is not necessarily the case. This is rather like assuming that a medicine that tastes
horrible must be better than one that is easy to swallow.
Finally we constantly see many examples of business operations misusing forecasting
and planning algorithms they do not understand to achieve dismal performance – often
coupled with a lot of wasted time.
Commonsense is more important than algorithms.
43. In Conclusion
• There is a moral to this story.
• This issue goes well beyond Exponential Smoothing.
• Many – certainly not all – people in business are impressed by mathematics which
they do not understand, especially when the claimed advantages of the algorithm
can be presented in an appealing manner by a competent proponent of the
approach.
It is often assumed that more complex algorithms work better than simple algorithms,
which is not necessarily the case. This is rather like assuming that a medicine that tastes
horrible must be better than one that is easy to swallow.
Finally we constantly see many examples of business operations misusing forecasting
and planning algorithms they do not understand to achieve dismal performance – often
coupled with a lot of wasted time.
Commonsense is more important than algorithms.
44. In Conclusion
• There is a moral to this story.
• This issue goes well beyond Exponential Smoothing.
• Many – certainly not all – people in business are impressed by mathematics which
they do not understand, especially when the claimed advantages of the algorithm
can be presented in an appealing manner by a competent proponent of the
approach.
• It is often assumed that more complex algorithms work better than simple
algorithms, which is not necessarily the case. This is rather like assuming that a
medicine that tastes horrible must be better than one that is easy to swallow.
Finally we constantly see many examples of business operations misusing forecasting
and planning algorithms they do not understand to achieve dismal performance – often
coupled with a lot of wasted time.
Commonsense is more important than algorithms.
45. In Conclusion
• There is a moral to this story.
• This issue goes well beyond Exponential Smoothing.
• Many – certainly not all – people in business are impressed by mathematics which
they do not understand, especially when the claimed advantages of the algorithm
can be presented in an appealing manner by a competent proponent of the
approach.
• It is often assumed that more complex algorithms work better than simple
algorithms, which is not necessarily the case. This is rather like assuming that a
medicine that tastes horrible must be better than one that is easy to swallow.
• Finally we constantly see many examples of business operations misusing forecasting
and planning algorithms they do not understand to achieve dismal performance –
often coupled with a lot of wasted time.
Commonsense is more important than algorithms.
46. In Conclusion
• There is a moral to this story.
• This issue goes well beyond Exponential Smoothing.
• Many – certainly not all – people in business are impressed by mathematics which
they do not understand, especially when the claimed advantages of the algorithm
can be presented in an appealing manner by a competent proponent of the
approach.
• It is often assumed that more complex algorithms work better than simple
algorithms, which is not necessarily the case. This is rather like assuming that a
medicine that tastes horrible must be better than one that is easy to swallow.
• Finally we constantly see many examples of business operations misusing forecasting
and planning algorithms they do not understand to achieve dismal performance –
often coupled with a lot of wasted time.
• Commonsense is more important than algorithms.