Content uploaded by Kesten Green
Author content
All content in this area was uploaded by Kesten Green on Dec 12, 2018
Content may be subject to copyright.
Comments on
“
Data Models vs. Knowledge Models in F
orecasting:
The Way Forward”
J. Scott Armstrong
Kesten C. Green
!"
#
$%&'
()*+,-.#
$%!&)!!*/!+%!*01
*
Procedure for this talk
•")#)#"-
)))
•2)"3 #
))#
•"--4
#-
•- ))
#
/
Success of the &rst M-
competition
•*50+)
•)1)
•2$!&)-2 !
1)
1)
•")6)
))! )
•" ) ))
)) )
)
•$))7
8
An outlet for experimental forecasting research
•-*50/
1)
)
•$!&)1)
•#$!)9
,7-:*50/!08
)-;
)
) )
%
Data models vs. knowledge models
2
)-
#- 91)):
667
#- )
)
<
Can you imagine anything that might convince you
that data models may harm forecasting accuracy=
•)--
-
•") >
•" )
1)2
2
?
In conclusion
(not)
&)
)
=
@)-
;
Why? Reason 1: Data Models Violate
the
Golden Rule of Forecasting
A #-
)B
* CD/0!
/ -2*+<-*<+
1))E
8 " )
%+F
0
Knowledge models
)G#---A#-
)B
* )#- '
•)
•6
•)
/ ))
8 4- )
-
!
% H- )1)
)-1))#-
5
Combine Prior Knowledge with Extrapolation Methods
*1)
#- !)
19)213))
#- :
/$1)#- 1-
)-#
8D! )#- )
1/0"55
))*0
G1!
%/F4- )
Rule-Based Forecasting: Development and Validation of a
n Expert Systems Approach to Combining Time Series Extr
apolations *+
Using Prior Knowledge with Extrapolation
Methods: Examples
* ())!
"#$%9/+*<:*;<<!*;<0I
/ ()&G'G
&)1)) &J.#)
&'(9/++<:/<!8?I
8 &G' H- )!1
) J&)('9*558:
*+8!**<
**
Why? Reason 3: Data models violate Occam’s razor
* -8/5;)-)
)1)9
))1 ':
,)1)
>)1
/;F/<-4)
/ ) )
9$#):
8 &)1)) !)1
% ) ! -
*8
Why? Reason 5: Data mining enables advocacy leading to
unscienti&c (unethical?) practices
*)
L-
/D
2)92
) 2)7
:
8, 2
-)!)#
*<
Data models should not be used for forecasting
* ()
/ ()1-2
2
8 - 2
#-
-))
6)
<+F9G
M):
% -2)
A B
< 1-
)#,9)
-#.
*;
Rain Maker Theories
* A)# -
)) )#B
/ A."#-3N3#
=B
8 )#-
3-N
--
*0
To know the way forward, you must know where you started
#- )
9
Forecasting Methods and Principles: Evidence-Based Checklists :
)- 7)
))
G ) G )
9):*<)
-G)DC
*5