WP134. Dynamic Arbitrage Gaps For Financial Assets
WP134. Dynamic Arbitrage Gaps For Financial Assets
WP134. Dynamic Arbitrage Gaps For Financial Assets
Abstract
In this paper we are concerned with the existence of a dynamic arbitrage gap that
evolves out of an adjustment process for disequilibrium prices, within a complex
dynamics framework which takes into account the market microstructure and
transactions costs.
Although this gap exhibits non linear and chaotic behavior, it doesn’t preclude
effective arbitrage transactions from taking place in real markets. Moreover, it may
explain much better those factors which usually impede actual perfect arbitrage.
Besides, this dynamic arbitrage gap depends upon a truly financial gap that accounts
for unexpected events and superior information on the professional dealers´side.
In this way, we can learn much more about dynamical adjustment processes from
financial assets, making the arbitrage gap instrumental to set about real arbitrage
positions. Finally, the dynamic arbitrage gap could become useful when coping with
financial crisis as far as some basic parameters´range of values for which the dynamics
becomes chaotic could be measured in advance.
1.- INTRODUCTION
1
Market microstructure and non linear dynamics have aroused deserved expectations from
both practitioners and academicians, for the last fifteen years. Moreover, not only the
dealer’s management of selling and buying orders and the specialist’s activity, but the
regulatory framework and contracting costs which seems to pervade markets as well, have
become current issues in theory and practice. Furthermore, environmental signals and
unexpected events have been given due attention in modelling price adjustments process.
Not surprisingly, a growing financial literature has so far become less involved with the so
called equilibrium prices convention, paying due attention to disequilibrium values,
transaction prices and fundamental values. Outstanding research running along this path has
been carried out by, among others, Cohen-Maier-Schwartz-Whitcomb (16) (17), Garbade-
Silber (24), Benston-Smith (12), Blume-Siegel (13), Demsetz (22), Levy-Livingston (28),
and mainly Beja-Goldman (9) or Beja (10),(11), and also Apreda (1) (2) (4) which are
quoted in the References section. We are truly indebted to Beja’s dynamic approach.
Getting onto the subject, our concern here is with Complex Dynamics. Briefly stated,
economic dynamics is the systematic study of economic change. When the patterns of
change can be translated by finite paths, convergent paths, or balanced settings, we deal with
the so called simple dynamics. But if patterns of change can be translated by non-periodic
paths, overlapping waves, switching regimes, or structural changes, then we face complex
dynamics. Although research on this matter is a promising academic field, it seems to be only
at the outset either in Financial or in Economic issues. Suitable references for mathematical
foundations are found in Devaney (23), Hirsch-Smale (27) and Li-Yorke’s paper (29).
Seminal work has been carried out by Professor Day(19) (20) (21). Chaotic Dynamic
Systems, and Chaotic Dynamic Processes applied to Finance are treated in Apreda (3) (4)
(6); and a preliminary development of this paper’s contents can be found in Apreda (1).
Computation and Econometrics problems are dealt with in Brock (14). Recent work in non-
linear dynamics to Finance is surveyed in Cuthbertson (18)
Before going into our proposal, we would like to stress an increasing empirical support to
the following statements, drawn mainly from the foregoing sources:
• An adjustment process takes its time, and frequently is not convergent, and most of the
time price sequences follow either stochastic processes or non linear complex behavior
which includes chaotic itineraries within a range of fluctuation not necessarily wide or
even fixed.
2
setting up or improving financial institutions, in measuring financial performance, or
helping markets to run more smoothly.
Briefly, what we want to do in this paper can be broken down into the following steps:
q Then, we present a dynamic setting with professional traders in the context of a market
transactional structure.
q Finally, a complex dynamics will evolve that allows for disequilibrium stages, which we
track by means of a dynamic arbitrage gap. This gap conveys to non linear dynamics and
likely chaotic patterns. Furthermore, it is directly related to transactions costs and
market microstructure.
Let us assume that the investment horizon is [t ; T] . By E[Pt (T)] we mean the expected
price at the moment T of a tradeable financial asset, assessed at time t. By Pt (t) we mean the
current transaction price at the moment t, assessed at time t.
We will denote k(t) the rate of transaction costs assessed at time t, for the moment t. By the
same token, k(T) will denote the rate of transaction costs assessed at time t, for the moment
T. That is to say, respectively, entrance and exit transaction costs. The wealth relative net of
transaction costs follows from:
Now we can proceed along with a differential rate for the transaction costs:
3
[1 - k(T)] / [1 + k(t)] = 1 / [1 + k t ]
to get
Definition 1:
Remarks:
• we could have done [1 - k(T)] / [1 + k(t)] = [1 + g t ] for instance, but we picked up the other
way so as to factor in the nominal return in terms of the net return and the transaction costs gap.
By E[[ rwt ], we mean the expected rate of return from any financial asset, assessed at the
moment “t” for period [t ; T] in an ideal market, such as it is provided by a qualifying
valuation model.
Taking into account transaction costs, it is worth trying arbitrage whenever the discrepancy
between the theoretical price and the current price covers up transaction costs and leaves
something to take advantage of. To see this, let us make
4
Definition 2
This rate, when being different from zero, triggers off plausible non-zero arbitrage positions.
Still, these positions don’t need to be attainable ones, unless costs of transaction may be
covered and microstructure features could be coped with, as we are going to see in the
following paragraphs.
Remarks :
< 1 + E[[rw t ] > . < 1 + k t > . < 1 + bt > = < 1 + E[[r t ] > = < 1 + E[ rt, N ]>
>. <1 + k t >
and then,
• E[rw t ] doesn’t convey transaction costs in most valuation models. In this sense, is “net” of transaction
costs.
We are going to link effective returns to continuous returns and prices along (t ;T). Current
spot and expected futures prices assessed at “t” will be denoted by pt (t) and E[pt (T)],
respectively. On the other hand, model or “equilibrium” prices assessed at “t” will be
denoted by wt (t) and E[wt (T)], respectively. For “rt ” we will understand the tradeable
financial asset expected continuous return from the tradeable financial asset, and for “rwt ”
the modelled financial asset expected continuous return from the modelled financial asset. So
we get:
5
If we put
< 1 + k t > = exp [ - K(t) (T - t)]
But the expected future price from the traded asset should likely be assessed by the same
model used to assess the future price of the ideal asset. That is to say:
E [ pt (T) ] = E [ wt (T) ]
Hence:
Remarks:
a) By W(t) = Wt (t) we mean the fundamental value assessed at “t” for moment “t”, brought about
by a valuation model.
b) We put
< 1 + bt > = exp[[ - β (t) . (T - t )]]
bearing in mind the gap will narrow as far as the market arbitrages evolves trying to close the gap.
Indeed, our dynamic approach may shed light onto Goldman-Sosin’s measure of inefficiency in a
market [ Goldman-Sosin (26) ]
MI = E [ ( ln pf t - ln pp t ) 2 ]
where pf is a price in a fully informed economy and pp stands for a price in a partially informed
economy. We are going to make a new reference to this issue (although departing from this setting
towards an arbitrage approach) in part 6 which will be devoted to price adjustment.
c) We put
6
< 1 + kt > = exp[[ - K (t) . (T - t )]]
just to get W(t) and K(t) with the same sign in [4]
Definition 3:
We see that the dynamic arbitrage gap defines a temporal path which depends on traded
prices, model values , and transaction costs through the investment period. For a unit period
horizon:
Remark:
β (t) = rt - rwt + kt
Two different approaches stand a chance of modelling the dynamic arbitrage gap behavior:
the deterministic and the stochastic ones. We deal with the deterministic approach through
this paper, and have coped with the stochastic model elsewhere; on this account, see Apreda
(2) (6) (8).
b) framing a dynamic setting with professional traders and market transactional structure;
7
c) disequilibrium prices, embedded in a complex dynamics framework, will enable us to
prove two lemmas on the existence of a dynamic gap, its relationship with the market
microstructure and transaction costs, showing that such an arbitrage gap also conveys a
likely chaotic behavior.
• Information flows allow agents for successful arbitrages, most of the time.
Remarks:
b) Beja (9) compares an equilibrium price with the current price, and Goldan-Sossin (26) compare a
price in a completely informed economy with a price in a partially informed economy. We depart
from these approaches because we feel more sensible to take an “adequately” informed economy, that
is to say, an economy where there is no serious hindrance to arbitrage, as a proxy of equilibrium or
perfect information in real markets.
c) Bounded rationality is used in Herbert Simon’s sense (as found in his work “Administrative
Behavior”, The Free Press, fourth edition, 1997). A very interesting economic development of this
concept was made by Reiter for markets out of equilibrium . Reiter (31).
• Market microstructure hampers transactions, because of lags, breadth and depth market,
regulations, contracting and transaction costs.
8
• There are agents with superior information (“superior traders”) who claim competitive
advantages to play in the market. Cuthbertson in (18) surveys relevant research on this
behavior.
Allowing for the fact that economic agents really carry over their buying and selling at
transaction prices instead of theoretical ones (such as those provided by qualifying valuation
models), with imperfect adjustment mechanisms, we can write:
where the excess of demand splits up into two components, the first of them taking the place
of that excess demand which would follow if the market belonged to an adequately informed
economy. That is why, within such an economy,
ε ( f, P(t) )
ε ( ∆ , P(t) )
measures the gap between the effective excess demand and the fundamental one.
Which price would be a fair one for the fundamental demand? We will denote it by means of
W(t) + K(t)
ε ( f, W(t) + K(t) ) = 0
9
[8] ε ( f, P(t) ) = g1 [ P(t) - W(t) - K(t) ]
where g1 is an increasing monotonous function. This price adjustment takes into account
transaction costs. Not surprisingly, within square brackets we find the dynamic arbitrage
gap.
Remarks:
a) Although most of the markets participants behave as “smart money” or rational agents, recent
academic research has included a subset of “irrational” or “noise” traders who don’t quote prices
equal to fundamental values, and still survive in the market, in spite of arbitrage ( See
Cuthbertson (18) ). Allowing for this “heterogenous traders”, the rational agents must take into
account what the former would do in the future and behave accordingly.
b) At this moment, it is worth recalling what Stahl and Fisher wrote in their work about stability
with disequilibrium awareness (31): “Agents in Walrasian world formulate demands again and
again taking prices as given and paying no attention to the fact that they will often not be able to
complete their planned transaction”.
Whereas equilibrium excess demand shows positions wished by agents if prices were those
attainable by arbitrage, ε ( ∆ , P(t) ) is an excess demand that takes place only when
disequilibrium comes up to breed speculation, at transaction prices. It is an environmental
excess demand. Here we are taking advantage of Beja (8).
It is this disequilibrium which prompts the demand for the financial asset on the grounds of
future price ex0pectations, but also push up demand for other alternative assets. Hence,
this addtional excess demand can’t be explained by the fundamental one.
An unexpected event, as the arrival of new information, will bring about an intertemporal
sequence of price changes, matching the sequence of excess demand variations which come
up as long as that information is spread over among the market participants. It is an striking
fact that the process not only can be stochastic [ Apreda (2),(6) ] but non linear and chaotic,
as well .
Now, we are ready to blend in this additional excess demand into an adjustment adaptive
process. In order to do that, let us introduce two new variables:
10
q The speculator’s forecast about the continuous return trend from the asset, in terms of a
rate net of transaction costs, rsp (t) , plus the continuous rate of transaction costs K(t)
q(t)
Remarks:
a) Opportunities to trade on different terms are a feature of markets out of equilibrium as Reiter
pointed out in (31), adding that information’s role has two main aspects: first, the institutional
structure of the market determines the information that agents may get from the market process,
which can be called the structural aspect. Second, the restricted capacity of economic agents to
handle information, and this can be called the bounded rationality aspect. By the way, that’s why
some agents would have incentives to become mediators in such a market.
b) The speculator’s forecast doesn’t necessarily coincides with E [r(t)], as assessed by a general
agent, or an arbitrageur, as long as the speculator takes advantage of superior information and
the market microstructure. A telling portfolio management approach to superior information
within a transactions costs setting can be found in Levy-Livingston (27).
c) Capelin and Leahy (14) made the following daring remarks in an article published in May 1996:
• “ Frictions prevent trade and therefore impede the acquisition of information. There is no reason
to view the price as a sufficient statistic for the state of the market. The volume of trade will
provide important additional information.”
• “Trading costs provide an explanation for the commonly observed combination of sharp
contractions and slow expansions; they provide an explanation for downward price rigidity. It is
a mistery that prices have so little explanatory power in many markets.”
The foregoing statements enable us to plug these relations into [ 3 ] and set up a dynamics:
11
[9] dP(t) /dt = H {g1 [ P(t) - W(t) - K(t) ] + g2 [ rsp (t) + K(t) - q(t) ] }
Assuming standard regularity conditions, developing the function H in Taylor Series and
picking up the first order approximation:
[ 10 ] dP(t) /dt = a · [ P(t) - W(t) - K(t) ] + b · [ rsp (t) + K(t) - q(t) ] + o(dt) /
dt
[ 11 ] dP(t) /dt ≈ ∆ P(t) / ∆ t = a · [ P(t) - W(t) - K(t) ] + b · [ rsp (t) + K(t) - q(t)
]
p(t)
( 1 + ν ) · p(t)
sensitizes itself to supply and demand shifts, calling for a parametric pattern as
12
taking any deviation from the benchmark µ = 1, as a market’s breadth measure; see Garbade
(23). This embeds the excess demand in a market microstructure setting.
Adopting a simple relationship between price changes and the excess demand, Samuelson-
sort, it follows:
Remarks:
a) Firstly, the increase in the supply of the financial asset translates an inventory change:
b) In real life, the mediator doesn’t know neither the supply-demand structure, nor the clearing price.
The relevant variable here, for him, is how his own inventory changes allowing him to mark up or
down his transaction price; see Day (17). The adjustment comes up as:
Now he can estimate λ , as prices change and adjustement through excess demand takes place,
As a consequence of [13], changes in prices come from excess demand with respect to
spread structure, adjustment velocity in terms of excess demand and, finally, the breadth of
13
the market. This embeds the price change in a market microstructure setting and takes into
account transaction costs directly related to the mediator.
d) Preventing negative prices, we arrive at the expected price as function of the breadth of
the market, adjustment velocity, mark-up and excess demand:
θ (p(t)) determines an orbit as long as “t” evolves, for each initial condition p(t0), and the
mapping can be considered a dynamical system. [ Apreda (3) ] [ Hirsch, Smale (27) ]
It has been proved by Day (19) that stability analysis applied to relation [ 14]] leads to the
following strong statement:
The relative adjustement in normal markets shows every feature attending simple dynamics and complex
dynamics. In general, convergence towards an only competitive equilibrium, convergence to periodic cycles,
chaotic topology, strongly chaotic trajectories and, at last, self-destruction.
Further, it has also been proved that for normal demand and supply functions, both mediator
tatonnement and relative mediator tatonnement displays the following features:
• for a wide range of parameters values, the process converge to a market clearing equilibrium;
• for a wide range of parameter values, the process exhibits cyclic or chaotic price sequences.
The reader can find the stability and relative stability analysis fully expanded in the Appendix
(see section 9)
8.- COMPLEX DYNAMICS WITH DISEQUILIBRIUM PRICES
After encompassing both the dynamic adjustment behavior and the complex-dynamics with
professional traders, we are ready for dealing with the following lemma.
Lemma 1
Within the framework of the dynamical model for prices introduced in part 6, and the
complex dynamics with mediator developed in part 7, it follows:
14
[ 15 ] Max { -1 ; [ λ . ε ( p(t) ) ] ÷ [ Max { D((1+ν
ν ) p(t) ) ; S( p(t) ) } ] }
Proof:
Step 1: In a discrete and deterministic framework, the total return is translated by:
Applying the relative adjustment we left to develop in the Appendix (paragraph 9.3) it
follows:
Step 2 : Providing standard regularity conditions are fulfilled, as we saw in part 2, we can
take advantage of [ 12]]
Recalling that
P(t) = ln p(t)
it follows
Furthermore,
15
a · [ P(t) - W(t) - K(t) ] ∆ t + b · [ rsp (t) + K(t) - q(t) ] ∆ t
However, the nature of this gap is far from simple. In the framework of a complex
deterministic dynamics, it matches current prices trajectories against those of the
fundamental prices. But these trajectories can be periodical or chaotical. [ In the framework
of stochastic process it can be regarded as a brownian one, as proved in Apreda (2) (6). ]
Coming back to the departure relation and substituting the dynamic arbitrage dynamic gap
for W(t) - P(t) - K(t) , and taking unit intervals:
Lemma 2
Let us take a financial asset in a market with mediator. Then, there is a dynamic
arbitrage gap between the model value and the transaction price, which locally depends
on the market structure, the transaction costs and a financial gap.
Proof:
16
a · β (t) + b · [ rsp (t) + K(t) - q(t) ]
The left hand expression stands for the transactional structure of the market through the
parameter λ, the professional trader mark-up, and the excess demand. The expression
should be regarded as a financial gap between the asset forecast return and the oportunity
cost from alternative investments in the market.
Remark:
If we raised the question whether we can track the arbitrage gap along stochastic trajectories, the
answer would be affirmative, as we have proved in other research paper, making use of stochastic
differential equations; see Apreda (2),(6).
7.1.- CONSEQUENCES
• Changes in the parameters are the most usual feature for a model to explain what is
going on in real markets.
• It is a foremost consequence from non linear dynamics that changes in parameters value
give rise to bifurcations periodic trajectories and, for certain range of values, chaotic
trajectories.
• The presence of chaos doesn’t convey the idea of non tractability. On the contrary, there
are analytical tools that may lead to the range of parameter values which take to chaotic
paths. ( See Cuthbertson (18) or Day (20) (21) )
• By the foregoing remarks, simulation models could help us into the management of the
dynamic arbitrage gap in the range of parameters values that lead to chaos.
17
• We hope this model to shift conventional wisdom ground on how to understand much
better real financial markets and to cope with financial crises.
8.- CONCLUSIONS
a) We have produced a dynamic model for disequilibrium prices behavior closely intertwined
with the market microstructure.
b) In this context, it has been proved the existence of a dynamic arbitrage gap.
c) The dynamic arbitrage gap is brought about with an explicit relationship among the
market transactional microstructure, the transaction costs system and a financial gap.
On the other hand, when there are no more arbitrage opportunities because not
even transaction costs are broken even, this doesn’t mean that convergence
towards equilibrium has been reached, as “zero-arbitrage” model suppose,
because the trajectory of observable prices could be chaotic, in spite of the gap
width.
e) Modelling the dynamic arbitrage gap as a deterministic system seems to be more realistic
so as to understand the financial assets price mechanism in a deeper way. Besides, the quasi-
stochastic behavior in prices that comes out of the chaotic trajectories, should make the
model suitable for simulation trials, and this is a promising field for applied research..
9.- APPENDIX
9.1.-LOCAL STABILITY:
Recalling [ 8]] :
18
Taking derivatives with respect to price:
θ ’ ( p(t) ) = 1 + µ · λ · { ( 1 +ν
ν ) · D’ [ (1+ν
ν ) p(t) ] - S’ ( p(t) ) }
As Richard Day (17) (18) has proved, based on Li -Yorke seminal article (28), the most remarkable
stability characteristics are:
There is a range of parameter values where the dynamic process shows price trajectories which are
cyclical or chaotical.
which is increasing monotonic and allows to treat the dynamics in terms of rates of return:
[ µ ·λ
λ ·{
{ D[[(1+ν
ν ) p(t)]] - S(p(t)) } ] / Max{
{ µ D[[(1+ν
ν ) p(t)]] ; µ S( p(t) ) }
We have translated the porcentual change in prices by a multiple of the excess demand, in terms of
the long position of the market ( the greater of demand or supply as the long side). Now we are ready
to face the difference equation that holds in the price adjustment. Firstly, as
p(t+1) / p(t) =
1 + [ µ ·λ
λ ·{
{ D[[(1+ν
ν ) p(t)]] - S(p(t)) } ] / Max{
{ µ D[[(1+ν
ν ) p(t)]] ; µ S( p(t) ) }
we can proceed to
19
p(t+1) =
{ 0, p(t) + p(t) . [ λ ·{
Max{ { D[[(1+ν
ν )p(t)]] -S(p(t))}
} ] ÷ Max{
{ D[[(1+ν
ν )p(t)]] ; S(p(t)) }
θ (p(t)) =
The function splits itself into two branches as long as the observable prices be less or bigger than the
fundamental (equilibrium) price p(ê).
a) if observable prices are less than p(ê), then the demand is bigger than the supply:
b) if observable prices are bigger than p(ê), then the demand is less than the supply:
θ (p(t)) = Max{
{ 0, p(t) + p(t) . { [ λ . { D[[(1+ν
ν )p(t)]] -S(p(t))}
} ] ÷ S(p(t)) }
θ (p(t)) = Max{
{ 0, p(t) - p(t) . λ + λ . p(t) .{
{ D[[(1+ν
ν )p(t)]] ÷ S(p(t)) }
θ (p(t)) = Max{
{ 0, p(t) . ( 1 - λ ) + λ . p(t) .{
{ D[[(1+ν
ν )p(t)]] ÷ S(p(t)) }
Stability, in a) excludes negative prices which could attend b) is look for through:
20
but there are elasticities within this expression. Hence:
10.- REFERENCES
1.- Apreda, Rodolfo
Complex Financial Dynamics for Arbitrage Gaps
Paper presented at the 34 Annual Meeting, Eastern Financial Association
Williamsburg, Virginia, April 1994
7- Apreda, Rodolfo
Budgetary Gaps and All-in-Cost Gaps
Research Paper, Department of Finance, UADE, Buenos Aires, 1995
21
11.- Beja, Avraham; Hakansson
Dynamic Market Processes and the Rewards to up-to-date information
The Journal of Finance, Vol.32, Nº 2, May 1977
16.- Cohen, Kalman; Hawawini, Gabriel; Maier, Steven;Schwartz, Robert; Whitcomb, David
Implications of Microstructure Theory for Empirical Research on Stock Price Behavior
The Journal of Finance, Vol. XXXV, Nº 2 , May 1980
22.- Demsetz, H
The Cost of Transacting
Quarterly Journal of Economics, 1968, Volume 82; Number 1; pp 33-53
22
Addison-Wesley, New York, 1989
23