Traps and limitations of
economic prevision models

Hard models and soft anticipations
in social areas, such as economics


Can linear and binary economic models, that work with historical data and
clear-cut probability laws, really represent soft and uncertain phenomena
and prospects?

Can they be made flexible to fit foggy and changing realities?

More precisely can they mirror human and social behavior that can go
against all predictions?

fog Reality is often foggy.
Not easy to trap it in the black box!

Why do we need previsions?

Can simulation models help?

A fundamental Human game: anticipation

A fundamental trait of the Human being is its effort to anticipate not only
what will come next, but also what might occur in the long term, so as to
try to make optimal decisions.

However derided "speculation" is, it is what makes Humans differ
fromlettuces.
To speculate - in the general sense - is to decide moves based on the
estimation of future prospects and risks
.

This use of anticipations is crucial for
sous economic  decision decisions, from borrowing to buy a car or
house to investing one's nest egg.

What is true for individuals is also true for  a countries and businesses :
their economic choices must rest on (preferably as wise as possible)
anticipations

This explains why the "prediction game" led to a lot of
tool tools,
ranging from the highly fanciful and superstitious to the rather rational.
The issue is:
How effective are those tools,
however rational they are?

What are the traps and limitations of the game ?

Economics, as a land of prediction models

In the sous economic area, individuals, businesses, institutions
obviously need to make or obtain economic predictions so as to:

* Prepare and adapt their medium term / long term projects.
* Even make some current choices, decisions and actions

For that purpose, they use - or they rely on experts to use - what are
considered at the moment the best available data and the most objective
predictive models,
In our "big data" age, this brought a flurry of simulation tools
that churn a mass of statistics, with the help of highly
sophisticated mathematics,
among which
probability
probability is king and statistical correlations
between the evolutions of various factors are princesses.
Well, to use the most advanced tools to simulate reality is a laudable
practice.
But there is a risk that those analyzes and forecasts suffer from an
overconfidence
* in numbers (numeracy bias)
* and in underlying apparently rational assumptions
.


Actually, several flaws and obstacles, described below, might affect
as well the data as the models and methods that are used to make
projections.

Economic evolutions traits
that models cannot fully match

Biased human and social reactions

Consumers, producers, investors, borrowers, lenders, businesses, public
institutions might react in unforeseen ways and be affected by
distortasymm behavioral biases.

Those human and social factors and their effects hardly fit mathematical
equations and the use of past statistical data.

Non binary situations

Binary logic (yes or no, true or false, 0 or 1, black or white) might not be
appropriate when the reasoning is about:
  • Soft phenomena,
They are those that involve people and society,
unmeasurable or irrelevant probabilities (uncertainty).

Probabilities are highly important to replace
fanciful impressions
(see base rate fallacy) with
a clear picture of realities. But in their
turn they
become unadapted (see below: overreliance...) in rare
 (not well covered by statistics) or fully new situations.

When facing such situations, fuzzy logic, cursor theory or Bayesian
probabilities
(starting with scenarios and adapted any time a new
event confirm or infirm them) might do the job better than binary
reasoning.
More generally, the fundamental fog uncertainty in economic life,
as well as in ordinary life, can make illusory the "addiction to clear-cut
predictions".
=> Better be prepared to éventail many scenarios, including
        extreme ones
, than to find a false comfort in one clear-cut
        prediction
(pseudo-certainty).
But this wisdom goes often against the human tunnel vision and
"range estimate aversion".

Broken lines (non linear evolutions)

Economic evolutions might be disrupted by percolations, bifurcations
and other sudden
jump jumps and disruptions proper to
dynamical systems.
Those "perturbation" events baffle the linear extrapolation of past
trends.

Overreliance in historical probabilities
and mathematical laws

Numbers and equations have the appearance of rationality, but might have
illusive traits
.

  • Historical statistical series might be too short .
(or samples too small, or data too scarce)
They would therefore miss dramatic surprise "rare events"
such as the "100 year storm".
The law of large numbers does not apply to small ones
.

Another trap in
short term series is that probabilities and/or
correlations might be just apparent, as only temporary
coincidences
, which leads to availability or representativeness
heuristic or even ...superstitions
(Gaussian or Laplace distribution...),
Although widely used as assumptions in many models, they
might not fully apply as plagued by
anomalcluster distortions:
asymmetries, fat tails, clusters, leptokurtosis...
One reason is that players are not always independent
from one another beliefs in their decisions
, contrarily
to what probability laws  supposes, as victims of mimicry
  • Even more important, past data can become irrelevant
This is the case in fully new events, circumstances and
situations.
Then,
uncertainty rather than measurable
risk
is involved.

As figures, however precious, do not tell everything (numeracy bias), it
might be good to avoid an overreliance on them and to take precautions
such as:
  • Whenever possible to look not only at numbers,
but to look also directly how things work on the field and
what tends to happen around.
  • And above all, to anticipate what will be the next move,
and the next move after the next move, instead of extrapolating
the
image past!
Models might be improved with "big data" (see below).
And also, here comes creative thinking (which mechanical
data lack)
, by fancying
a range of scenarios and entering
them into the parameters so that, if situations show signs
that
those occurrences emerge, they get soon detected.

Herd instinct among experts

Asking experts 'opinions are needed to complete information in fields
where extensive experience and knowledge is needed. But experts tend
to use similar data, equations, assumptions  in their models and also to
imitate each other
by fear of being wrong alone.
=> This can bring a superficial consensus based on what is the
        most salient, that does not help to detect hidden forces that
        might overturn the situation.
Another caveat is that deciders tend to ask experts to give the most
probable single projection / precise number (i.e. about inflation
rate, GDP growth rate, earnings per share, currency rate, stock index...)
=> This prevision that gives only one result, a single projected
        number, ignores all other scenarios that are crucial to
       
make a decision and be prepared for all contingencies (plan B...).
Technical lessons to draw 
for economic models

In the rescue of economists


Economic models, and as a consequence, economics (the "dismal science")
and economists, are currently highly criticized if not derided.

That does not mean that economic models should be scraped definitively.
But many of them, widely used for previsions with poor results, should be
made flexible on the basis of scenarios that take into account true
surprise  "surprises" (not to confound with run of the mill fluctuations
and "noise") such as:
  • Not only quasi mechanical extreme events / disruptions
  • But also the "soft" elements
    which are the possible changes of attitudes by economic players.
Moreover the surge of "big data" (processing huge masses of data that
might seem unrelated) allows to

* go much farther in the search for correlations as well back
   in the past as within the diversity of fields that might interfere
,

* also detect faster some new trends from a surge of signals
   and apply Bayesian probabilities to the suspected evolutions.


But big data is only oriented towards the present and the past, thus
creative
thinking, as already mentionned is a key tool when dealing
with future evolutions.

And what about management models?

An overreliance in quantitative models for financial market
trading
, that neglect "rare events" (see "
probability") can end up
in dramas.

As for Management Information Systems (MIS), they should not be
the only source for a business manager or controller.
It does not do its job, if it:
* does not go look at the reality in which things concretely take place
* also does not anticipate what might evolve, inside and outside the
    business.
There are limits also in applying Management by objectives, it can easily
become bureaucratic, putting a straight jacket against initiatives and
adaptations to new situations, and / or a temptation to cheat (
moral hazard),
and also a source of demotivating stress.


A related topic is an
overreliance on regulations that monitor only some
data that might be irrelevant as a substitute to smart watchdogs.

Sources and further readings

Details about those notions are found
in the
Behavioral finance glossary

Back to collection: Economic articles migrated from Knol
Back to collection: Management-marketing articles migrated from Knols

Pageviews for this article before migration from Knol: 4.2 k

 
M.a.j. / updated : 06 Aug. 2015
All my ex-knols / Tous mes ex knols
Disclaimer / Avertissement légal

This site tracked by OneStat.com. Get your own free site tracker.