Forecasting: The Judgement Dimension

1
Forecasting The Judgement Dimension
We have looked at RBF. It is the state of the
ART. We also looked at the push-button models.
Everyone had a intelligent way to consider the
Series Modification, the Feature Identification
Phase and the Rule Application. The bottom line
is if you are serious about doing forecastsuse
RBF you now have the intel to apply this model.
Let us now look at three other NON-Times series
modelling ways to do forecasting Statistical
Judgement, Expert Panels, and Delphi Models.
2
The Bi-Nominal Model

• The idea here is simple. We assume that the
  chance of the event that we are examining will
  occur is chance 50/50. We then collect as much
  information as possible about the event from
  where ever we can. We recorded the number of time
  these sources support one of the binary
  alternatives. Then we compute the probability as
  the Binomial Probability here summing over the
  first given Y to Ni.e. the End
• ? P(N,Y) NCY x Py (1 P)N-y
• Let us look at a timely example due to my friend
  Lisa Chen.

3
The EURO Where will it go?

• When the Euro first came on line, everyone in the
  currency trading world was collecting forecasting
  information. Lisas firm collected 14 opinions.
  Three said the Euro wouldafter the
  honeymoonslide to a level where it would trade
  below the 11 Dollar Peg. Two Rated it as a
  no-call, and nine felt it would trade-up from the
  dollar. This makes the probability test of the
  null
• P(gt) ?12C9 x (0,5)12
• 0,054 0,016
  0,003 0,0002
• This is lt 0,08. Reasonable to odds reject the
  50/50 null odds. Guess what actually happened?
  The EURO tanked for about a year!

4
Statistical Judgement
You have all had a basic statistics course where
you have studied the tests of proportion in both
the single and two sample cases. These methods
provide a good way to develop statically relevant
information. For example, Here I will just give
the results of a study assuming that you have
studied these methods. There are found in EVERY
statistics book ever written. I worked on the
forecast for the Democratic Mayoral Planning
CommitteeDMPC for the Mayoral Race in the city
of Philadelphia. The democratic candidate ED
Rendell was running in 1991. At the time he
announced his candidacyabout 13 months before
the election, we collected a sample of 550
individuals. We wound up with a usable sample of
487 individuals. The question was If you were
going to vote for the mayor of Philadelphia today
who would you select?
5
Ed Rendell Wins Decisively in the Philadelphia
Mayoral Race.
The person interview was presented with a list of
four possible contenders at that time. The
results were of the 487 individuals who gave
their opinions 173 indicated Ed Rendell as their
choice. The 95 confidence interval for this
population proportion was
0,355 1,96 x ((0,355) x (0,645))/487½ or
0,355 0,0425 In whole percentages 31,25
39,75 This suggested to the DMPC that Ed was
a viable candidate since he was probably
preferable to a random selectioni.e., 25.
Later, he was actually running against a
Republican candidate and we tested his voter
preference two months before the election and
found a CI of 56,2 to 63,6 in the actual
election his winning percentage was 58,1.
6
Expert Panels
Here the idea is to select a number of
individuals who have expertise in the area in
which you are interested. You collect their
opinions and/or their assessment on the actual
final question of interest. For example you are
trying to decide if the German economy is
recovery from the long slow down that has been in
evidence for about 19 months. You could get a
number a economists and ask them for their
opinions on the interpretation of data that they
use in making assessments of economic trend data.
Now based upon their opinions about their
variable of interest you make YOUR conclusion. Or
you could simply ask them their opinion and based
upon some decision rulelike the majority opinion
record their overall assessment. The are lots of
problems with expert panels Difficult to get
experts to take it seriously, Who are really
experts anyway, Very expensive to get quality
information, The results to date suggest that
experts do not outperform chance.
7
Delphi Panels Feedback Groups
These are really expert panel for individuals who
are part of the processi.e. they have a vested
interest in or some direct control regarding some
aspect of the forecasting problem. Now with
conferencing in Chat Rooms and Virtual board
rooms this technique is making a strong come
back. It works as follows You get a group,
usually about 3 0R 5 (never more than five) of
decision makers who have a common problem where
they need to act in consorti.e., together. When
they need to forecast or project the results of
their decisions they are asked to--in the group
setting indicate what the think they will do.
The expected results are fed back to the group
and they revise their decisions after, of course,
discussing it with the others. This continues
until until all are satisfiedi.e. the process
converges to acceptance. It can work.
8
Delphi Bottom Line
It seems that when you have a limited number of
individuals who are decision makersi.e. they can
control recourse and they share a common problem,
like management of the supply chain in a
vertically integrated organisation or vertically
ventured company, the Delphi Panel seems to
improve the forecasting in particular material
requirements planning. Delphi only can work if
you have a recognised problem, accepted by all
the decision makers as real, a Chat/Virtual
space, simulation sort of data links, feedback
loops, and end-stage criteria. It also helps to
believe in Post-auditing.
9
Some Fables True Ones
The Dancing Waters in Berlin Random?! Not for
me! Tibors
Game Whose got the Real
Model? Pie
Anyone! What we learn from these When things are
random, we think we see pattern and believe we
can explain it, when a model or a process exists,
we cannot find it, when someone tells us what it
is, we don't believe them and we rarely do see
what is going on in front of our noses. What does
this mean? Be carefulrarely do we know what we
are doing! Believe in Models, Statistical
Analysis and Judgement. Careful and sensible use
of ALL three is essentialif you rely on only one
or two aspects You are skiing downhill in a
forest blindfolded. Please do not be surprised if
a tree takes you out.