0% found this document useful (0 votes)
77 views38 pages

Dimensional Analysis, Scaling, and Zero-Intelligence Modeling For Financial Markets

Dimensional analysis, scaling, and zero-intelligence modeling can be applied to understand financial markets. Dimensional analysis allows relationships between differently sized systems to be understood using consistent units of measurement. Zero-intelligence modeling examines how simple rules and constraints shape outcomes without assuming strategic behavior. For financial markets, this includes examining the continuous double auction mechanism where buyers and sellers continuously place limit orders that are matched against market orders. Key market outcomes like bid-ask spread, volatility, market impact, and liquidity can be analyzed at different scales using these approaches.

Uploaded by

Hmt Nmsl
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
77 views38 pages

Dimensional Analysis, Scaling, and Zero-Intelligence Modeling For Financial Markets

Dimensional analysis, scaling, and zero-intelligence modeling can be applied to understand financial markets. Dimensional analysis allows relationships between differently sized systems to be understood using consistent units of measurement. Zero-intelligence modeling examines how simple rules and constraints shape outcomes without assuming strategic behavior. For financial markets, this includes examining the continuous double auction mechanism where buyers and sellers continuously place limit orders that are matched against market orders. Key market outcomes like bid-ask spread, volatility, market impact, and liquidity can be analyzed at different scales using these approaches.

Uploaded by

Hmt Nmsl
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 38

Dimensional analysis, scaling, and

zero-intelligence modeling for


nancial markets
Eric Smith (SFI)
based on work with
Doyne Farmer (SFI)
Supriya Krishnamurthy (Swedish Inst. Comp. Sci)
Laci Gillemot (Budabest U. Technology and Economics)
Giulia Iori (City U. London)
Martin Shubik (Cowles Foundation, Yale)
Paolo Patelli (LANL)
Marcus Daniels (LANL)
Outline

Where and why can a scientic point of


view contribute to economic understanding?

Dimensional analysis and scaling

Zero-intelligence modeling

A worked example: the continuous double


auction of nance

Compelling open problems


I. Relation of science to economics

A little history and ideology of current


mainstream economics

Formal foundations and their aws

Entry points for scientic methodology


The intellectual homage of neoclassical
economic theory
..every individual necessarily labours to render the annual revenue of the society
as great as he can. He generally, indeed, neither intends to promote the public
interest, nor knows how much he is promoting it. By preferring the support of
domestic to that of foreign industry, he intends only his own security; and by
directing that industry in such a manner as its produce may be of the greatest
value, he intends only his own gain, and he is in this, as in many other cases,
led by an invisible hand to promote an end which was no part of his intention.
Nor is it always the worse for the society that it was no part of it. By pursuing his
own interest he frequently promotes that of the society more effectually than
when he really intends to promote it. I have never known much good done
by those who affected to trade for the public good.
From An Inquiry into the Nature and Causes of the Wealth of Nations by Adam Smith (1776)
Objectives of (much of) modern
mathematical economics

Formal concept of optimal resource allocation

Identication of nal goals of trade


(~implies equilibrium)

Proof that optimal equilibria are dened by


the uncoordinated desires of self-regarding
individuals (mathematical existence theorem)
Foundations of neoclassical economic theorizing

Complete, costless contracts

All consequences known and bargained for

All contracts, costless, instantaneous, guaranteed

Rationality := Global self-fullling prophecy

Strategic actions by everyone as good as given

Non-constructive existence proof for equilibria

Utilitarian preferences (usually self-regarding)

Partial order on all possible states of the world

All outcomes over all time have present value


Amounts to a process-free, institution-free world
(No pig farms)
(No accountants, courts, govt)
(Mutual best response)
(Multiple, uncomputable)
(Known false many ways)
Some cautions that go with criticism

Beware the Laplacians:

People are not simply machines

No reason economic order will look like physical order

Beware the econophysicists

Renaming mathematical models from physics would not


count as good theory even in physics

Empirical validation of social science is tricky


Opportunities for science in economics

Institutions are by nature mechanistic

Based on rules and constraints

Opportunity for structural and process analysis (analysis of


market function, design of auctions, etc.)

Zero Intelligence as a pure model of institutional


constraint

Constraints can shape both action and reasoning

Errors in ZI models can be among the best


indicators of behavioral regularity

Provides better focus for proper behavioral science


What would a scientic economics look like?

Richer conceptual substructure between


empiricism and theory

Abstractions should attach quantitative


consequences to partial problem
specications

Believe models when the paths that led to


them have excluded everything else

Look for notions of universality as the


justication for simple models
(Warning! speculation and editorializing)
II. Dimensional Analysis and scaling

Two principles

Equations must be homogeneous in dimension

Quantities with the required dimensions control scaling

Two consequences

Can guess the sizes of observables

Can relate differently-sized cases to a single model


Example: how fast do you walk?

A pure dimensional analysis

You want to know a speed

You have a property of height

You are walking in earth gravity

Only one combination has correct dimensions

Could break this down in more detail

Your leg is a pendulum

Pendulum is characterized by frequency

Only one combination works dimensionally

Speed is ticking frequency times length



speed
time
length
g acceleration
time
2
length

l
frequency

All walkers great and small

Walking is a scale-invariant process

All walkers share gravity, they differ


mostly in their characteristic lengths

Dimensional homogeneity can be used


to generate scale factors relating
different walkers to each other

III. Zero-Intelligence (ZI) modeling: a pure


formulation of institutional constraint

We dont know what intelligence or


rationality are; but we do know how to
exclude them

An example: the minority game

ZI modeling can be a foundation to which


more formal models of learning are added
The minimal Minority Game: the original
zero-intelligence model

A population of N players (usually odd) choose


one move from the set (0,1), independently and
simultaneously.

Each player whose choice was the minority in


the population is awarded a payoff (wins).

Moves have no intrinsic value

The number of constellations of winners is


enormous (frustration)

A model of the non-rational component of


purely speculative stock trading
Minority Games: a basis
for learning models

Each agent remembers


outcomes of M prior
rounds of play

A set of random lookup-


tables (particular to each
agent) provides next
moves in response to each
possible history (2
M
)

Agents learn by choosing


the table with best
performance so far

A phase transition occurs in


z = 2
M
/ N

Most satisfaction at critical


N=11
25
101
1001
slope=!1
10
!3
10
!2
10
!1
10
0
10
1
10
2
10
3
10
4
10
!2
10
!1
10
0
10
1
10
2
Figure 6: s=2
z
N
/
2
^
v
e
d
t
S
From R. Savit, Y. Li, R. Riolo, Physica A 276 (2000): 265
Symmetric
Unpredictable
Ordered
non-symmetric
Predictable
Variance in the number of winners
IV. Worked example of dimensional analysis and ZI:
the continuous double auction of nancial markets

The basis of all continuous-trading nancial


markets today

Solves problem of matching asynchronously-


arriving orders to buy and sell

Heavily institutionalized and high volume


(~5-7 billion dollars / day in 1999)

Complete, good-quality data


Mostly from Smith, E., J. D. Farmer, L. Gillemot, and S. Krishnamurthy. "Statistical
Theory of the Continuous Double Auction." Quant. Fin. 3(6) (2003): 481-514
The Continuous Double Auction (CDA) mechanism

Two kinds of orders

Market (v)

Limit (p,v)

Limit orders (LO)


accumulate

Market orders (MO)


clear immediately

LO at single price
clear in order of
arrival
buy
market
orders
sell
market
orders
price
s
h
a
r
e
s
bids
offers
spread
Observables of the CDA

Bid-ask spread (responsible for costs)

Volatility (responsible for risk and prot)

Market impact (responsible for costs and risk)

Liquidity (resistance to market impact)

Depth (responsible for stability)


The bid-ask spread

Bid = best offered buying price

Ask = best offered selling price

Spread = ask - bid

Rapid small buy-sell alterations will pay the average


spread per pair of transactions

Nonzero spread is the leading source of transaction


costs

Nonzero spread is the regulator against divergent


volume of trading
price
bid ask
spread
The volatility

Midprice:
m = (ask + bid) / 2

Only midprice motion


creates risk + prot
opportunities in the
presence of a spread

Has traditionally been


modeled as a random
walk (Bachelier 1900)

Diffusion constant of
the random walk is
called the volatility
<
[
m
(
t
+

)

-

m
(
t
)
]
2
>
0
2.0
4.0 6.0 8.0
time (events)
0
1E-4
Midprice motion is a good
approximation to diffusion
Courtesy Paolo Patelli
Market Impact

Average price of an order always


worse than the starting best price

Return depends only on price ratios

No prot if impact > return

Expect impact to scale


exponentially, or else there is an
incentive to split or join orders

Yet it doesnt
price

original
ask
average price
you pay
Farmer, J. D., P. Patelli, and I. I. Zovko. "The Predicitive Power of Zero
Intelligence in Financial Markets" PNAS USA 102(11) (2005): 2254-2259
Observed regularities of the market impact

Power laws with exponents


0.25 - 0.5 are common

Corresponds to a power-law
distribution of limit orders

Market impact is integral of


the depth
8uy orders
0.26

p
m
0
.
2
0
.
4
0
.
6
0
.
8
l0
-l
l0
-2
l0
0
Order size
Sell orders
0.23

p
m
0
.
2
0
.
4
0
.
6
0
.
8
l0
-l
l0
-2
l0
0
Order size
price
n ~ p
3/4
~ p
l/4
Courtesy Paolo Patelli
Liquidity

Liquidity = resistance of price


to change in response to
market orders

More standing orders should


mean more liquidity

Coefcient of the power law of


market impact is a natural
liquidity measure

For impact power ~0.5 this is


just the slope of the distribution
price

original
ask
average price
you pay
price

original
ask
average price
you pay

Depth of the limit-order book

Clearly asymptotic order


density determines where
(concave) power-law scaling
ends and linear scaling begins
(and gives the coefcient)

Order density (shares / price)


is termed depth (n)

Dimensional analysis of the continuous ows

Summarize the order-


placement processes by a
collection of continuous
rates and rate densities

Limit order placement

Limit order deletion

Market order arrival


price

s
h
a
r
e

d
e
n
s
i
t
y
buy
market
orders
sell
market
orders
price
s
h
a
r
e
s
bids
offers
spread

These are complete for shares, price, and time


Dimensional analysis of granularities

Orders arrive and are removed


in typical-sized chunks

Prices are delimited in ticks

Both units are redundant with


dimension provided by ow
variables

Redundancy creates the


possibility of functions of non-
dimensional variables

Here I will take the ow variables to dene the classical


scaling dimensions, and treat the discreteness parameters
as the source of nondimensional corrections
Classical scaling and dimensional analysis

Flow variables dene


characteristic scales for
fundamental properties

Create guesses for


observables based on their
dimensions
Scaling collapse of different instances

Non-dimensionalized variables relate cases


differing only by scale to a universal model

Price (coordinate)

Share depth (variable)

Spread (observable)

Diffusivity (observable)
How well does classical scaling work?
Try
collapsing
the market
impact four
different
ways
Farmer, J. D., P. Patelli,
and I. I. Zovko. "The
Predicitive Power of
Zero Intelligence in
Financial Markets"
PNAS USA 102(11)
(2005): 2254-2259
Classical scaling predictions for the spread

Prediction for the spread


from the ow variables is

Broad selection of stocks


from the London Stock
Exchange agree with this
prediction to overall scale
log(predicted s)
l
o
g
(
m
e
a
s
u
r
e
d

s
)
-3.0 -2.5
-3.0
-2.5
Farmer, J. D., P. Patelli, and I. I. Zovko. "The Predicitive Power of Zero
Intelligence in Financial Markets" PNAS USA 102(11) (2005): 2254-2259
Discreteness parameters determine possible
importance of dimensionless ratios

Tick size

Order chunk size


Overall scale is respected, but the discreteness parameter matters
0 1 2 3 4
0
1



Nondimensional spread versus e Depth prole versus e
0 1 2
0
1
2
3

Correction of the spread from the classical guess

Recall scale factor and non-


dimensionalization of prices

Classically we would expect


non-dimensionalized spread
to take value

From simulations get


small correction with
epsilon

Comparisons to data
are remarkably good
0.00 0.05 0.10 0.15 0.20 0.25
0
.
0
0
.
2
0
.
4
0
.
6
0
.
8
simulation f()
s: London Stock Lxchange (vOD)
n
o
n
d
i
m
e
n
s
i
o
n
a
l
i
z
e
d

s
p
r
e
a
d

The volatility

Classical scaling of the diffusivity

Simulations show different short-


term and long-term corrections

Data scale with approximately the


short-term correction, with no
apparent qualitative change
AZN
BARC
CW.
GLXO
LLOY
ORA
PRU
RTR
SB.
SHEL
VOD
log(simulated diffusivity)
-6.8 -6.6 -6.4 -6.2 -6.0
l
o
g
(
m
e
a
s
u
r
e
d

d
i
f
f
u
s
i
v
i
t
y
)
-6.8
-6.6
-6.4
-6.2
-6.0
0 0.1 0.2 0.3 0.4 0.5
0
0.1
0.2
0.3
0.4
0.5
0.6
<
[
m
(
t
+

)

-

m
(
t
)
]
2
>

/

p
C
2

Farmer, J. D., P. Patelli, and I. I. Zovko. "The Predicitive Power of Zero


Intelligence in Financial Markets" PNAS USA 102(11) (2005): 2254-2259
Learning from the limitations of ZI models

Simplest ZI model leads to innite range of


limit prices and number of orders (clearly silly)

Real order distributions are still very regular,


but not by any institutional rule

Candidate for a
regularity of behavior
From Mike, S., and J. D. Farmer. "An Empirical
Behavior Model of Price Formation." Santa Fe
Institute Working Paper 05-10-039
V. Some compelling problems in
income and wealth distribution

An economically central issue

At the heart of questions of inequality and social welfare

Sets wealth from capital ownership apart from wealth


from wage labor

Concerns whole-nation real productivity

Reects relation of nance to the real economy

The regularities are stunning and durable

Nobody even knows whether they are


institutional or behavioral
Income distribution in modern nations

Two scaling regimes

Income in wage
range is lognormal
or exponential

Income of wealthy
is power-law

Both distributions
are forms of
maximum-entropy
subject to different
constraints

Laws, customs, etc.


all get folded into
just four or ve
constants
10
3
10
4
10
5
10
6
10
7
10
8
10
!8
10
!7
10
!6
10
!5
10
!4
10
!3
10
!2
10
!1
10
0
Income in 1999 US dollars
C
u
m
u
l
a
t
i
v
e

p
r
o
b
a
b
i
l
i
t
y
US
Japan
Courtesy Makoto Nirei and Wataru Souma
Concluding thoughts

Theory is best constrained by empiricism

Institutions, behavior, the physical world, etc. are real and


accessible to experiment

Empiricism is best constrained by theory

Numbers dont become meaningful until you understand


(something about) the measurement system that generates them

Good methods tie falsiable consequences to


approximate descriptions

Dimensional and scaling analysis and ZI have proved


useful for some institutional process analysis
Some further reading

Farmer, J. D., D. E. Smith, and M. Shubik. "Is Economics the Next Physical Science?" Physics Today 58
(9) (2005): 37-42.

Farmer, J. D., P. Patelli, and I. I. Zovko. "The Predicitive Power of Zero Intelligence in Financial Markets"
PNAS USA 102(11) (2005): 2254-2259.

Lillo, F., S. Mike, and J. D. Farmer. "Theory for Long Memory in Supply and Demand." Phys. Rev. E 7106
(6 pt 2) (2005): 287-297.

Lillo, F., and J. D. Farmer. "The Key Role of Liquidity Fluctuations in Determining Large Price
Fluctuations." Fluctuations and Noise Lett. 5 (2005): L209-L216.

Challet, D., Marsili, M., and Zhang, Y.-C. Minority Games, Oxford U. Press, New York (2005)

Farmer, J. D., L. Gillemot, F. Lillo, S. Mike, and A. Sen. "What Really Causes Large Price Changes?"
Quant. Fin. 4(4) (2004): 383-397.

Smith, E., J. D. Farmer, L. Gillemot, and S. Krishnamurthy. "Statistical Theory of the Continuous Double
Auction." Quant. Fin. 3(6) (2003): 481-514.

Lillo, F., J. D. Farmer, and R. Mantegna. "Master curve for price impact function." Nature 421 (2003): 129.

Bouchaud, J.-P., and Potters, M. Theory of Financial Risk: From Statistical Physics to Risk Management,
Cambridge U. Press, New York (2000)

You might also like

pFad - Phonifier reborn

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.

Note: This service is not intended for secure transactions such as banking, social media, email, or purchasing. Use at your own risk. We assume no liability whatsoever for broken pages.


Alternative Proxies:

Alternative Proxy

pFad Proxy

pFad v3 Proxy

pFad v4 Proxy