Game theory in Stock trade, portfolio

optimization
Ashutosh Kumar(11168)1
1 ashukr@iitk.ac.in

ABSTRACT
Stock market is one of the most lucrative career choice for gifted mathematicians and economists, so
it comes at a no surprise that a lot of work has been done on the application of game theory to stock
market . Game theory with its vast arena of approaches that are simple and rationale in nature have
been extensively used as an indispensable tool by Quantitative Analysts. This paper tries to find the
development of game theory as a tool to optimize stock portfolios.

Keywords:

Game theory, Stocks, Market, Portfolio Optimization, Bayesian Nash Equilibrium

1 INTRODUCTION
Stock market is a place for sellers and buyers to meet and exchange stocks, bonds etc. under the supervision of a regulatory body i.e. the Exchange. Its a hub of people crunching numbers trying to figure out
what is the market gonna do next. But its a difficult job, the market is volatile and a good way to get some
idea is from historical data. With regression analysis or more advanced artificial intelligent networks like
Neural Network, CNN we can do that up to a certain extent. But the historical data is not sufficient, it fails
to measure the sentiment of the market. As we will discover later in this paper, in a stock market the health
of the sock may depend more on the rumors flying about the company shutting down than its balance sheet.
The expected earnings from a stock is measured by the expected value of return and its volatility is
measured by its variance called risk. Return on an stock is the sum of dividend received and change in
market price. Here we assume that only return possible is the change in the market price. Similarly, risk
is the off-chance of producing negative returns. The riskier the stock the more likely is a negative return.
1.1 Risk and Returns
Consider a single stock with an expected return of 12%. If there exists a risk free investment like a bank
deposit with an interest rate of 12%, then anyone would be better off to invest in the bank as this entitles
him to the same return with no risk. It may also be possible that some people may choose the bank even if
the interest rate offered is 10%, less than 12% offered from the stock. This is called Risk Aversion. No
rational person would undertake a risk without adequate compensation. The minimum difference in the
expected compensation between a risky asset, like stocks here, and the compensation from a risk-less
asset to make a rational person indifferent between the two is called Risk Premium. Hence risk premium
can be considered as the compensation to investors for holding on to riskier investments. If the expected
return of an stock falls below the sum of risk free return and risk premium, then no rational person will
buy that stock. Please bear in mind that risk premium is a personal construct and may vary from person to
person, with a high value for those who are more risk-averse. We will talk of stocks only with a further
assumption that all the prices and cost, that in future too, have been reduced to their present value.
1.2 Portfolio
According to Investopedia, A portfolio is a grouping of financial assets such as stocks, bonds and cash
equivalents, as well as their mutual, exchange-traded and closed-fund counterparts. Portfolios are held
directly by investors and/or managed by financial professionals. Portfolio are more attractive to stand
alone stocks. The reason will become clear with a bit of math ahead.

Consider there are n stocks in a portfolio with weight wi and expected return Ei . The expected return
of the portfolio is the weighted average of the expected return of the stocks. But due to the correlation
between the stock the risk is less than both the individual risk and the weighted risk of the stocks. Hence
their is a sharp decrease in the risk with a small decrease in expected return. This property of portfolio, to
provide average return with a lower than average risk, makes it more attractive.
1.3 Risk of a stock in a portfolio
The next question that arises is that how do we calculate the risk added by a particular stock to a portfolio.
The risk of a stock in a portfolio is measured by the ratio of its volatility to that of an average stock on the
market, called . If = 1, then the stock is as volatile as an average stock, if this value is greater than 1,
then the stock is more volatile than an average stock. Conversely,if the value is less than 1 then the stock
is less volatile than the market. If a stock with greater than 1 is added to a portfolio with = 1, then it
increases the volatility of the portfolio and vice-verse. can be calculated by running a regression on the
past data of returns on the market.

2 PORTFOLIO SELECTION MODELS

There are three popular selection models:
i)Mean-Variance (MV) model: Maximize the mean/variance value
ii)Minimax Model: Game Theory zero sum game
iii)Mean Absolute Deviation (MAD) model: measure of the variability
2.1 Mean Variance Model
First presented by H. Markowitz, it basically is an mean-variance optimization model. Putting to layman
terms, it says to maximize the expected return for a given risk. Thus this model tries to maximize Sharpe
ratio where Sharpe ratio is the excess expected return (Expected return - risk free return) per unit of total
risk. Due to quadratic dimensions of its co-variance matrix that makes a solution difficult to obtain, this is
not used nowadays.
2.2 Minimax Model
Originally formulated for two player zero sum game, it attempts to minimize the maximum loss from all
plays, Minimax criterion. Alternatively it can be explained as the maximization of minimum gain from
any play, Maximin criterion.
Young (1998) proposed this model first. According to his paper, suppose there are n stocks for a T time period.
y j t = After a time period t, return on every dollar invested.
T
yj = Average Return on stock j = 1/T t=1
y jt
w j = Portfolio allocation to stock j
y pt = Return on portfolio in time period t = j=1 w j y j t
E p = Average Return on portfolio
M p = Minimum return on portfolio = mint y pt
As mentioned earlier, the minimax model minimizes the maximum loss or alternatively maximizes the
minimum gain, i.e.
max M p

M,M p

(1)

subject to
N

w j y jt 0,t = 1, ...t, T

(2)

j=1
N

w j W,

(3)

w j 0, j = 1, ..., N.

(4)

j=1

2/7

Constraint on M p makes sure that it is greater than the portfolio return and hence M p is being maximized.
These sets of equations will give the value of maximum minimum return, or the minimum maximum loss.
2.3 MAD
This model was developed by Konno and Yamazaki and was presented in 1991. This model combines the
advantages of MV model with outdoing its dependence on quadratic solvers. It minimizes the portfolio
returns mean absolute deviation for a given expected return. As it is linear in nature, it scores major
brownie points over MV model.

3 LITERATURE SURVEY
Game theory has ever increasingly spread its influence in the financial markets. Wiszniewska-Matyszkiel
(2005) tried to model stock market as a dynamic game with a continuum of players. This concept was
first introduced by Aumann (1964) and Aumann (1966). Before these articles, the games couldnt model
the insignificance of the action of a single player on the market. These games as a continuum of players
were further studied by Mas-Colell (1984), Balder (1995) and Wieczorek (2004). The same in dynamic
setting were explored by Wiszniewska-Matyszkiel (2000) and Wiszniewska-Matyszkiel (2001).
Another popular perspective is to the view the market as evolutionary and hence try to find the
evolutionary stable strategy. Cai et al. (2008) tries to model the market as a continuous double auction
in a homogeneous population. A continuous double auction allows both buyers and sellers to exchange
offers simultaneously and continuously throughout the auction. Smith (1962, 1965) conducted many
an experiments to document the behavior of the players in the game. Smith (1962) claimed that even
with limited information available, the continuous dynamic game can model the market very efficiently.
Another breakthrough was that it didnt need a very large number of players contrary to the classical theory.
There are multiple think tanks working on separate theories to use game theory to optimize a portfolio
and maximize profit thereof. gto (1988) tried to build a game theoretic model to optimize portfolios for a
wide variety of payoff functions in a single or multiple play of the stock market. He postulated that if the
payoff of the two players satisfied that the primitive -game has pure optimal strategies W 1 = W 2 =1 if
and only if (1) 0 exists and
((t 1)/t 0 (1) (t) (1) (t 1) 0 (1)

(5)

for all t>0, then any player would achieve the optimum value by choosing the conditional expected log
optimal portfolio.
Another theory in the making is the canonical coalition game theory for optimal portfolio selection.
Kocak (2014) tried to find the optimum of a portfolio for stocks with identical targets but different risk
capabilities by applying Coalition Game theory. Evolutionary game theory too has been put to use to
find the optimum portfolio. Evstigneev et al. (2008) examines a dynamic model of financial market with
endogenous asset prices. The model determines the short term equilibrium of demand and supply. An
assumption is made the traders use fixed-mix investment strategies and distribute their wealth in fixed
proportions. The result corroborate that of Kelly (1956) that investing in stocks according to their expected
relative returns is a evolutionary stable strategy.
A further study in the same field with an incomplete market was done by Hens and Schenk-Hoppe
(2005) It studies the evolution of wealth shares, in incomplete markets with short-lived assets, of portfolio
rules. It derives the necessary and sufficient conditions for evolutionary stable strategy and states that if
investors employ only simple strategies, i.e. (w) k , then the simple strategy * defined by,
k = E(Rk (w))

(6)

for k = 1,. . .,k is evolutionary stable. Furthermore, there is no other strategy that is evolutionary stable.
3/7

4 MODELING STOCK MARKET

4.1 As a three player game
Consider that the market is composed of just three people and on stock. Now each of them is trying to
buy/sell that stock. Depending on the demand of the stock, its price is going to fluctuate. If everyone
is trying to sell the stock, its price will go down and conversely if every one is trying to buy the stock
its price is gonna increase. Taking these things in mind we can say that the below Table 1 of payoffs is
justifiable. Player 1 is the row player, player 2 is the column player and player 3 buys in left table and

Buy
Sell

Buy
(2,2,2)
(0,3,3)
Buy

Sell
(3,0,3)
(3,3,0)

Buy
Sell

Buy
(3,3,0)
(3,0,3)
Sell

Sell
(0,3,3)
(2,2,2)

Table 1. The payoffs of different players.

sells in the right table. As we can, when everyone tries to buy the stock prices may go up but there is no
one to buy from. Similarly there is no change in their payoffs if everyone is trying to sell. Things get
interesting when two of then tries to either buy or sell and the third sells or buys respectively. In such
cases the two who take the same action are better off as they follow of the trend of the market and the third
is worse off. As we can easily observe this game has only two pure strategy Nash Equilibrium namely
(Buy,Buy,Buy) and (Sell,Sell, Sell) and thus every player wants to follow the market.
The above game is influenced by the Keynesian beauty contest developed by Keynes (1936). Keynes
created a fictional beauty contest in a newspaper, asking people to choose six of the faces they found
to be the most handsome or beautiful. He offered a lucrative prize to the winners who picked the most
popular faces. At first sight it looks like everyone chooses the most handsome picture at their discretion,
but this strategy is weakly dominated by the strategy to picks the faces that would most likely be picked
by most of the population. So everyone would vote on the basic of his/her knowledge of public opinion
rather than the attractiveness of the face to him/her. Taking it one step forward people, those who know
that people would be voting on the basic of their knowledge of public perception, will try to predict the
eventual outcome on the next reasoning level based on other agents rationale to vote thus. This loop goes
on and on till we reach an equilibrium.
It is not a case of choosing those [faces] that, to the best of ones judgment, are really the
prettiest, nor even those that average opinion genuinely thinks the prettiest. We have reached
the third degree where we devote our intelligences to anticipating what average opinion
expects the average opinion to be. And there are some, I believe, who practice the fourth, fifth
and higher degrees. (Keynes, General Theory of Employment, Interest and Money, 1936).
Hence, it doesnt matter to the players what are the characteristics of the stock but rather what are the
perception of other players towards that stock. This claim seems to satisfied by the payoff table, with
playing in the general direction of the market being a strictly dominant strategy for each player.
4.2 As a game between a Buyer and Seller of Stock
Most of the literature on the stock optimization looks at a one-player game against nature trying to
optimize their portfolio. An alternative measure to evaluate each stock exchange as a zero sum game
and try to find the optimum values for each player. Let there be two players, on of them has to choose
between whether to sell a stock and the other decides whether to buy the stock. We assume that the risk
premium to hold that particular stock is the same for both the players. The seller offers to sell the stock
at 1.25,1,0.75 times the value of the stock, which is assumed to be normalized to 1. Similarly the buyer
offers to buy the stock at 1.25,1,0.75 times the value of the stock.
A transaction occurs if the amount offered by the buyer equals or exceeds the amount asked for by the
seller. The stock is either a low-risk (low average return, low volatility) or a high risk(high expected return,
high volatility). Nature has two moves upmarket or downmarket. Let the probability of the upmarket
be and downmarket be 1-. For e.g. in a case of low risk stock the Nature can play upmarket with
4/7

a probability of 2/3 and downmarket with a probability 1/3. For the high risk stock Nature can play
downmarket with a probability 2/3 and upmarket for 1/3. The player dont know the move of the Nature
and play without knowledge of the trend of the market. It has been assumed that the price of the stock
reached 1.25 in an upmarket and 0.75 in the downmarket. Furthermore a risk premium of 0.1 has been
taken i.e. if the seller sells the stock, he gains 0.1 for doing away with his uncertainty while the buyer
loses 0.1 for the uncertainty regarding the future of the stock.
An extensive form of the game is shown in the Figure 1 below. A normal form of the game is given in

Figure 1. An extensive form of the game. Player 1 is the buyer and player 2 is the seller
Table2. The value of each column in the normal form can be calculated by the following.

1.25
1
0.75

1.25
1
(-0.1,0.35) (0.15,0.1)
(0,0.25)
(0.15,0.1)
(0,0.25)
(0,0.25)
Upmarket

0.75
(0.4,-0.15)
(0.4,-0.15)
(0.4,-0.15)

1.25
1
0.75

1.25
1
(-0.6,0.35) (-0.35,0.1)
(0,-0.25)
(-0.35,0.1)
(0,-0.25)
(0,-0.25)
Downmarket

0.75
(-0.1,-0.15)
(-0.1,-0.15)
(-0.1,-0.15)

Table 2. Game in the normal form.

Let Pstock = Change in the price of the stock. Base price is 1.
Ps = Difference in the price asked by the seller and the base price
Pstock = Price of the stock, Ps = Price asked by the seller, Pb = Price asked by the buyer.
If Ps Pb , then the payoffs are:
(Pstock Pseller , Ps + )

(7)

If Ps < Pb , then the payoffs are:

(0, Pstock )

(8)

4.2.1 Pure strategy equilibrium

The upmarket game has a (1,1.25) and (0.75,1.25) as pure strategy Nash Equilibrium. This makes sense
as if the players know that the market is going to soar in future, the seller would sell at a higher price.
Though surprisingly, the buyer doesnt buy the stocks, the reason is that the seller will only sell the stock
at 1.25, but this doesnt cover the risk premium of player 1 as he is paying 1.25 for an uncertain asset
5/7

worth 1.25 now but whose values could take a dip later. The downmarket game has (0.75,0.75) as the NE
as both know that the value of the stock has fallen and hence would only bid or ask for 0.75. One should
note that if Pstock > 0.35 here, then there would be a unique Nash equilibrium to the upmarket game
namely (1.25,1.25) for this time player 1 is getting adequately compensated for the risk he has taken by
buying the stock.
4.3 Bayesian Nash Equilibrium
A point to note is that in an upmarket selling at 1.25 for player 2 and buying at 1 for player 1 are weakly
dominant strategies while in a downmarket buying at 0.75 for player 1 is a weakly dominating strategy.
This is analogous to that stocks should be sold at greater than par value and bought at below par value to
maximize profit. Furthermore, player 2 is always better off by making a sale in a downmarket
Each player has nine moves. To make our lives easier we will just calculate the expected payoffs
for each of the possibility assuming that = 1/2 and that both players can only use either 1.25 or 0.75.
The expected payoffs are shown in Table3. From the table we can ascertain that the only pure strategy

1.25,1.25
1.25,0.75
0.75,1.25
0.75,0.75

1.25,1.25
(-0.35,0.35)
(-0.05,0.05)
(-0.3,0.3)
(0,0)

1.25,0.75
(-0.1,0.1)
(-0.1,0.1)
(-0.05,0.05)
(-0.05,0.05)

0.75,1
(-0.1,0.1)
(0.2,-0.2)
(-0.1,0.1)
(0.2,-0.2)

0.75, 0.75
(0.15,-0.15)
(0.15,-0.15)
(0.15,-0.15)
(0.15,-0.15)

Table 3. Expected pay offs of the players

Bayesian Nash Equilibrium is to play ((0.75,0.75),(1.25,0.75)). This checks out as player 1 has no benefits
from buying this stock and player too will like to dump his stock at the highest value he could get. If
the same table is made for all the possible actions, then we get ((0.75,0.75,0.75),(1.25,1,0.75)) as the
Bayesian pure strategy Nash Equilibrium.

5 INFERENCE
A lot of text can be found the application of game theory for stock market portfolio optimization.
Though the models can be very complicated mathematically, the underlying concept is simple enough to
understand.

REFERENCES
(1988). Game-theoretic optimal portfolios. Management Science, 34(6):724733.
Aumann, R. J. (1964). Markets with a Continuum of Traders. Econometrica, (32):3950.
Aumann, R. J. (1966). Existence of Competitive Equilibrium in Markets with Continuum of Traders.
Econometrica, (34):117.
Balder, E. J. (1995). A Unifying Approach to Existence of Nash Equilibria. International Journal of
Game Theory, (24):7994.
Cai, K., Niu, J., and Parsons, S. (2008). Adaptive Agents and Multi-Agent Systems III. Adaptation and
Multi-Agent Learning: 5th, 6th, and 7th European Symposium, ALAMAS 2005-2007 on Adaptive
and Learning Agents and Multi-Agent Systems, Revised Selected Papers, chapter Using Evolutionary
Game-Theory to Analyse the Performance of Trading Strategies in a Continuous Double Auction
Market, pages 4459. Springer Berlin Heidelberg, Berlin, Heidelberg.
Evstigneev, I. V., Hens, T., and Schenk-Hoppe, K. R. (2008). Globally evolutionarily stable portfolio
rules. Journal of Economic Theory, 140(1):197 228.
Hens, T. and Schenk-Hoppe, K. R. (2005). Evolutionary stability of portfolio rules in incomplete markets.
Journal of Mathematical Economics, 41(12):43 66. Special Issue on Evolutionary Finance.
Kelly, J.L., J. (1956). A New Interpretation of Information Rate. Bell System Technical Journal,
(35:4):917926.
Keynes, J. M. (1936). The General Theory of Employment, Interest and Money, chapter Chapter 12. New
York: Oxford University Press.
6/7

Kocak, H. (2014). Canonical Coalition Game Theory for Optimal Portfolio Selection. Asian Economic
and Financial Review, (4(9)):12541259.
Mas-Colell, A. (1984). On the Theorem of Schmeidler. Journal of Mathematical Economics, (13):201
206.
Smith, V. L. (1962). An experimental study of competitive market behaviour. Journal of Political
Economy, (70(2)):111137.
Smith, V. L. (1965). Experimental auction markets and the Walrasian hypothesis. Journal of Political
Economy, (73(4)):387393.
Wieczorek, A. (2004). Large Games with Only Small Players and Finite Strategy Sets. Applicationes
Mathematicae, (31):7996.
Wiszniewska-Matyszkiel, A. (2000). Modelling the Tragedy of the Commons, Game Theory and
Applications. Applicationes Mathematicae, (5):162187.
Wiszniewska-Matyszkiel, A. (2001). The Tragedy of the Commons Modelled by Large Games. Annals of
the International Society of Dynamic Games, (6):323345.
Wiszniewska-Matyszkiel, A. (2005). Stock market as a dynamic game with continuum of players. MPRA
Paper 32982, University Library of Munich, Germany.

7/7