Rik Blok
A lecture for Physics 510: Stochastic Processes in Physics, by Birger Bergersen.
We have considered many models with autonomous individuals or
agents (eg. birth/death, citizens of cities, ants).
Agents have been modelled by assuming simple probability
distribution for possible actions and treating as unmotivated, ``dumb''
particles (eg. birth/death rate, diffusion).
Advantage: allowed analytical solutions via Master or
Fokker-Planck equations.
Disadvantage: often not very realistic.
Introducing intentional motivated agents can complicate
analysis but can be very rewarding
[1,2,3].
Simplest to make agents perfectly rational with complete
knowledge and unlimited computational ability.
Tends to produce simple, static equilibria (eg. ants make a
bee-line straight for food).
Aside: selfish motives can produce sub-optimal behaviour (eg.
Braes' paradox [4], Diner's Dilemma [5])
Limited knowledge and/or computational ability
Selfish motives
Related to St. Petersburg paradox [6]
Gambler playing a ``double or nothing'' type game repeatedly
against an infinitely rich adversary (multiplicative stochastic
process).
Chance of winning each time is p (known by gambler).
In each iteration gambles a fraction r of wealth.
Wealth after t iterations is Wt.
Aside: can also be interpreted as a simple market model with one
risky asset (price fluctuations as described above) and one riskless
asset paying no interest. r describes portfolio.
Expected wealth is
Naive goal is to maximize
w.r.t. r:
If
\frac{1}{2}$ */ ?>
then ``rational'' gambler will wager
everything.
But must eventually lose (if p<1):
Problem arises from heavy weighting of extremely unlikely
events.
Expectation maximization is a poor choice for modelling rational
behaviour. So what is ``rational''?
Maximizing expectation is too risky. Instead, might want to
minimize risk of eventual ruin.
If r<1 then will never gambler will never lose all wealth
so let's define ruin as a net loss Wt < W0. (Could also
define ruin by granularity of money (eg. 1 penny) with the same
conclusions.)
Goal is to minimize
P(Wt(r) < W0) (for t large).
For large t wealth distribution is roughly log-normal (because
random walk on log-scale; will be discussed later).
Solution becomes clear from the probability distribution itself
(with v and D expanded in terms of p and r).
Minimize risk by choosing
So, in this interpretation, ``rational'' behaviour is to never
gamble unless p=1. Too safe?
Agent values utility rather than wealth U(W).
Utility is increasing, concave function (
0$ */ ?>
,
). Literature suggests particulars of
utility function largely irrelevant.
A popular choice in finance is exponential utility
Ue(W)=-e-a
W, or equivalently
Wgoal can be interpreted as maximum conceivable wealth
or goal wealth (determines riskyness). (For finite system Wgoal
must be not be greater than all available wealth.)
Consider a single iteration. Goal is to maximize expected utility
Solution: optimal investment fraction r* is
r* changes with each iteration as Wt changes. Decreases as
wealth increases to Wgoal.
Non-trivial solution for 1/2<p<1.
Gamblers are still ``irrational'' because they will always gamble
their entire wealth (r* =1) when the chance of winning is greater
than
Also common in the literature is the generalized Kelly
utility [7]
k=1 utility equivalent to
because derivatives
the same. Absolute value doesn't matter for
optimization.
Kelly [8] originally hypothesized just the logarithmic
form. Was generalized to
later.
The advantage over previous utility is that there is no arbitrary
cut-off wealth Wgoal, but there is a parameter k (the Kelly
parameter). Meaning will become clear.
Again, goal is to maximize expected utility
Kelly parameter k is ``riskyness''. k<1 = risk-adverse,
k>1 = risk-prone. 1/k is ``risk aversion''.
Kelly utility is more ``rational'' because r* =1 iff p=1.
Perhaps using the expectation value is an unfortunate choice.
Often the median value is a more typical realization. Then a rational
goal might be to try and optimize the median value of the future wealth.
Median Wmed is defined as point with equal probability of
greater or lesser values:
To derive the median value we must recognize that the wealth
Wt follows a multiplicative random walk
where
is distributed via
Use log-scale to get additive noise
Random (biased) walk so, after many iterations, P(h,t)
approaches a Gaussian distribution with drift velocity v and
dispersion D
Median (and average) of h linear in time
hmed = v t.
Median of wealth is
Goal is to maximize median (typical) wealth Wmed w.r.t. r
Same solution we saw for the Kelly utility function (k=1).
Optimizing Kelly utility equivalent to optimizing median value.
For exponential utility, must update r* with each iteration.
If we instead update Wgoal via
Wgoal = fgoal Wt for some
multiplier fgoal then r* constant.
Can map exponential utility Ue onto Kelly utility Uk by
equating r* yielding
Aside: regardless of riskyness k, for p>1/2, fgoal is
bounded in order that agent remain ``rational'' (r*<1)
All approaches explored (neglecting trivial solutions) so far
reduce to two models: Ue (Wgoal=constant) and Uk.
Simulation shows iterated wealth of two exponential utilities
(Wgoal = 1/2 and 2) and three Kelly utilities (k = 1/2, 1 (log)
and 2). Used p=0.6 and W0=1 and all agents used same history of
wins/losses.
All but risky k=2 Kelly utility performed well over short term
Fixed Wgoal exponential utilities underperform (as expected)
over long time. They can also crash (Wt=0) if wealth gets too low
(not seen in this realization).
k=1 (log) has best performance long-term but safe k=1/2 also
good.
Asked question ``How do we model rational agents?''
Looked at gambler playing a ``double-or-nothing'' type game.
Tricky because multiplicative process.
Maximizing expectation too risky.
Minimizing risk too safe.
Common (exponential) utility can be too risky (and contains
arbitrary scale).
Generalized Kelly utility favourable.
Maximizing median value equivalent to (original) Kelly utility.
Median and expected values can be very different in
multiplicative processes.
Simulations suggest optimizing median value best. But smaller
Kelly number can be just as good (and safer) on short time-scales.
Maslov and Zhang [11] proved k=1 is on borderline
of riskiness for similar model.
Exercise: Prove Kelly utility (k=1) optimal as
[12].
This document was generated using the LaTeX2HTML translator Version 98.1p1 release (March 2nd, 1998)
Copyright © 1993, 1994, 1995, 1996, 1997, Nikos Drakos, Computer Based Learning Unit, University of Leeds.
The command line arguments were:
latex2html -dir /home/blok/public_html/p510gambler -split 0 p510gambler.
The translation was initiated by Hendrik Blok on 1998-12-08