Behavioral finance FAQ / Glossary (Model / System trading)
This is a separate page of the M and S sections of the Glossary
Dates of related message(s) in the
Behavioral-Finance group (*):
Year/month, d: developed / discussed,
(trading, pricing, investment...) Model, Modeling
02/7i - 08/11i - 09/3i + see quant,
system trading, decision-making,
computer trading, numeracy bias,
Robotic maids to dust the family portfolio.
Can they really be trusted?
Or might they break the crockery?
A financial market model is typically a mathematical software that processes
market data and tries to detect investment opportunities.
Such models are usually created, operated and adjusted, by "Quants" (see that
Modeling is quite popular in finance.
Plenty of mathematical algorithms and computer
models related to asset markets (stock markets among others) have
This might be because numbers are available, and because to process
numbers gives a feeling of certainty in a field that is actually a realm
of uncertainty (see numeracy bias) but where money flows to pay the
best mathematical minds.
A luck for quants, who, poor things, if not would earn a pitance as
rocket scientist or production engineers .
Those financial robots are used for
valuation, day trading, portfolio management, and
generally to performall types of market plays on
risks and returns.
They focus usually more on short term
operationsthan on long term ones (although some
are used for long term previsions and periodic asset
To cover the whole ground, economists also have been using for a long
time statistical models and the related mathematics.
Econometrics is based on those.
How are financial (and economic) models built?
Number grinding through mathematical mills.
Those models are software that use intricate mathematics, derived from hard
sciences like physics.
They thrive in historical data series and in crossing a multitude
of data (fundamental ratios, market fluctuations...).
=> They usually hunt for "statistical correlations" and phenomena
to which probability laws can be applied via
"stochastic calculations" (see that phrase)
This has been amplified by the surge of big data that explores a
massive quantity of seemingly unrelated data in which appears
* hidden relations (or accidents) seen on long periods
* or on the contrary nascent trends
Bayesian probabilities can be then applied to confirmed or
infirm those occurences.
We have here market robots guided by mathematicians.
Such man-man machine bionic combination gives them their strength ...and
weakness as seen below.
Main practical uses in finance
Appliances for every purpose in the quants' kitchen.
Here are the basic - let us say traditional - financial models (not citing
many more elaborate ones derived from them):
The CAPM (see that glossary article),
The various option pricing models,
The stochastic trend analysis models,
The risk analysis models (VaR...).
Not only some "quant funds", use intensively such tools, but more and more
institutionsuse them in some way and commercial software are available to
Here are the main uses:
Generally, for researchers as well as for players,
To understand the current market working and its price
trends,and adapt to it.
To detect short or long term trends that are liable to go on
In an opposite approach, to determine what are the "efficient"
financial prices, so as to spot market anomalies.
To draw from those findings automatic buy / sell
orientations to help in trading and investing decisions (see
In very short term trading and arbitrage, to emit direct
buy and sell orders, without human intermediation:
computer driven trading / system trading / algorithm
trading / high frequency trading.
This approach, that leaves robots decide and play the
market, tries to fill the need of a quasi instant reaction to
"opportunities" before they evaporate.
To build customized portfolios that fit any possible strategy
and management style, more or less risky, with either short
or long term horizons...
Little by little, the market gets in the hand of robots trained at
finding and unearthing truffles, and at fighting each other to
grab them. Are they good at it?
This is the question below.
Can those space ships fly?
All this, probabilistic mathematics, statistical data, fast computers, is impressive.
Rocket science, as some say. But can it work?
Before talking about financial models, let us appreciate predictive models
Models are supposed to be more rational than humans are.
Specifically, they are better at mimicking (what is known about) realities.
Progresses in artificial intelligence (see that word) are expected, by
which computers might draw lessons from their own experience.
Numbers do not fully reflect the pitfalls and opportunities linked to
the real world, above all in highly human / social areas such as
economicsand finance: see numeracy bias.
Also, as they are developed from historical data, it is OK when the
new situation has similitude with past ones, but they are rather unfit to
anticipate fully new situations, unless (wise) man-made scenario and
imaginative disruptive parameters are forced fed into them.
They might master what is seen as a measurable risk,
but what about uncertainty? (see that word).
To what extent financial modeling
can be reliable?
Caveat, financial model users!
Kick the tires and see
if bugs and rusty cogs
are hidden in the machine!
The degree of reliability / predictability of a pricing model is obviously an issue,
and here the Murphy law can be at play, as several obstacles intervene:
Various models are built on the "efficient market"
theory, and strict"ideal" or "rational" assumptions
such as random distribution
Those assumptions might be a weakness, as not fitting the
full reality in human / social fields such as economics (or even
in other fields showing unexpected event, even mutations).
Thus models might underestimate the real risks, by
ignoring some types of anomalies, notably the
rare but highly dangerous ones (black swans, see below).
They are "backtested" on past market statistics.
They might miss, if the statistical time span is too short, "black
swans", which means rare events (see that phrase) with crucial
Also, statistic-fed machines seem to have less capacity than the
human mind to imagine extreme scenarios
Robots' fuses blow when exceptional events strike!
For example a sudden market illiquidity
(see liquidity squeeze), in other words a sudden lack of
counterpart for buyers and sellers, might strike because
of an external shock or a reversal of investor trust
In the absence or counterpart, a trader would be
unable to get rid of a derivative contract that
can cause him total ruin.
Also a stop loss order would not work.
This illiquidity factor intervenes only exceptionally in large markets,
but when it does it has horrific effects on investors and in some
cases on the whole financial system (systemic crisis if it takes the
form of a contagious crash).
As this is quite rare, model makers can have problems to
integrate liquidity as a parameter.
They might even neglect it totally because of overconfidence.
Market models are poorly adapted to unforeseen and fully new
Such sudden and important changes in the general picture are proper
to evolutional or chaotic dynamical systems.
Standard "probability laws", with their pretty and clean graphs
and equations might misrepresent them, because of disruptions,
mutations and emergences (see percolation, bifurcation ,
Therefore most models work (or worked) well in a specific period,
market or state of affairs and might be out of step in another one.
If they are only trained to swim, how can they behave
when the lake dries up?
If they did not envision unprecedented scenarios (see
that word), how can they spot them if they emerge?
Therefore, they become easily obsolete.
Some even have a very short-lived usefulness.
Consider financial models as food with a
"best before" date.
After it you will have to scrape them ...or at
least give them a heavy 15,000 miles
maintenance and upgrade.
Better disconnect the autopilot (but not all
safety functions) when entering an
Use the manual mode, as rules might differ
from those implanted into the electronic brain.
Market models risk to be either over-simple (heuristics)
or over-elaborate (systems), either over-rigid or over-
All in all, they are slaves of what is already known, and also of the
current interpretations (paradigms).
They might tend to look for the lost keys under the lamppost,
not where they might have fallen.
Thus they might not be too able to understand the
unknown and to adapt to it.
Modeling might not take fully into account the
human factor , which might distort prices in
an unexpected way.
Instead of correcting those distortions, the human behavior might:
Either make those inefficiencies persistent, and then unprofitable
Or overcompensate them, bringing other anomalies, maybe
OK, but how to monitor human attitudes?
Polls are not fully reliable (see "sentiment").
An alternative, but only slightly better, would be to sit in a bus
or at a cafe terrace and hear people talk.
Most models that investor apply
use similar algorithms.
Here human herding (see that word) gets multiplied by massive and
instant "computer herding" (see herding).
This conjunction creates a lack of counterparties and illiquid markets.
This is usually found in market crashes (see that word) or in a milder
way in periods of excessive volatilities.
Also those operations are now done at the speed of light between
computers by big banks and other financial institutions, which brings
a "speed asymmetry" between traders
Also they usually have a short time horizon
Financial market models are usually not well adapted to long term
Models are built mostly for short term trading
(and, for some of them, for medium term allocation arbitrages).
The drawback is that unexpected long term / rare effects can
suddenly strike in the short term (here again, crashes are the most
typical of those effects).
As for agent-based models (see that phrase),
they work only if the agent categories are relevant
and their behaviors reliably recurrent.
to spot rightly the types of agents,
to identify their relative market strengths,
last but not least, to be sure enough that their behaviors
are fully predictable.
Anyway agent based models might be better predictors of dramatic
"non linear" disruptions and unbalances than models working on
standard mathematics equations
Therefore, however useful models might be, better look closely
at their assumptions, and at their construction.
The market effects of the recent "subprime crisis" have again
shown the limitations and some dangers of those trading systems.
Handle with care
Humans should respect the laws of robotics
when living with robots.
Financial models are more and more available as generic computer
Whatever their critical limitations seen above, they might bring
Robots, for those who remember Asimov novels, are supposed
to have less emotional and cognitive biases than us poor
They are more disciplined we might say. Greed and fear are
unknown to robots.
Of course we are not talking here of toy robots programmed
to mimic our basic behavior.
They might help to automate some safety portfolio management
They can spot, at the speed of light, small
n short-lived discrepancies,
Those are anomalies which humans would not see or have no time
to take advantage of.
Another thing is that they cannot be accused to be the main
cause of crashes (see that word).
Such financial meltdowns could have happened as well if human
beings were in control.
They are often the consequence of prior exuberant behaviors.
Or in some cases of unforeseeable external shocks.
Anyway, those market androids have to be used with some judgment.
This is also what can be said about technical analysis - see that phrase -
as a crude ancestor of quantitative analysis and system trading.
Fully automated trading can bring very bad surprises.
Not in normal time, but when something new and big happens suddenly
while traders trusted too much computers and advanced stochastic calculations
to show the way.
Some traders might suffer the comfortable illusion /
dependence /obedience / addiction that
those automatons would do their job or
offset their own "exuberant" lack of foresight,
self-discipline and common sense.
(*) To find those messages: reach that BF group and, once there,
1) click "messages", 2) enter your query in "search archives".
Members of the BF Group, please
vote on the glossary quality at BF polls