Professional Documents
Culture Documents
Advance Stochastic Calculus (Abstracts) PDF
Advance Stochastic Calculus (Abstracts) PDF
Preface
International Conference Advanced Finance and Stochastics was held on June
2428, 2013 at Steklov Mathematical Institute in Moscow, Russia.
This book contains abstracts of the 51 papers presented at the Conference, which
cover areas such as stochastic control in finance, derivatives pricing and hedging,
portfolio selection, statistics of financial data, risk theory and others.
The Conference was organized by Steklov Mathematical Institute, Laboratory
for Structural Methods of Data Analysis in Predictive Modeling, Center for Structural Data Analysis and Optimization, and Institute for Information Transmission
Problems. Financial support was provided by the Government of the Russian Federation, grant ag.11.G34.31.0073.
We dedicate the hosting of this event to the tercentenary of Jacob Bernoullis
remarkable paper Ars Conjectandi, 1713.
Scientific Committee:
D. Belomestny, M. Schweizer, A. Shiryaev, V. Spokoiny
Organizing Committee:
A. Shiryaev, V. Spokoiny, A. Kuleshov, N. Beyer, E. Burnaev, B. Kashin,
A. Muravlev, A. Sergeev, T. Tolozova, P. Yaskov, M. Zhitlukhin, S. Zhulenev
Contents
Conference program
Plenary talks
13
15
Denis Belomestny
Developing long term yield curve models for low rate regimes . . . . . . . . . . . . . . .
16
M. A. H. Dempster
17
Ernst Eberlein
20
Paul Glasserman
21
Maria do Ros
ario Grossinho
25
Juri Hinz
26
Yuri Kabanov
27
Constantinos Kardaras
29
Masaaki Kijima
30
Dmitry Kramkov
31
Michael Markov
32
Ernesto Mordecki
35
Alexander Novikov
36
Martin Schweizer
37
Contents
38
Teruyoshi Suzuki
41
Lioudmila Vostrikova
Response to Paul A. Samuelson letters and papers on the Kelly capital growth
investment strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
45
William T. Ziemba
Contributed talks
47
49
Akhlaque Ahmad
Pricing foreign currency options under jumps diffusions and stochastic interest
rates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
50
Rehez Ahlip
51
Hamed Amini
52
Michail Anthropelos
54
Abel Cadenillas
55
Christa Cuchiero
57
Stefan Gerhold
58
Florence Guillaume
62
Contents
63
Andrey V. Makarenko
65
Gennady Martynov
67
Dmitry Muravey
69
Alexey Muravlev
71
Sergey Nadtochiy
72
Omer Onalan
73
Vladimir Panov
75
Deimante Rheinlaender
77
Thorsten Rheinlaender
81
Evelina Shamarova
82
Alexander Slastnikov
84
Sergei Sidorov
85
Tatiana Vasilyeva
87
Mikhail V. Zhitlukhin
Posters
89
91
Natalia V. Danilova
Contents
95
Georgy Mironenko
Pseudo binary differential evolution algorithm for cardinality constrained portfolio optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
97
Andrey Homchenko
99
Yerkin Kitapbayev
Author index
106
Conference program
Monday, 24 June
8:40 9:10
Registration
9:10 9:30
Opening
9:30 10:20
10:30 11:20
11:30 11:50
D. Rokhlin: On a generalized shadow price process in utility maximization problems under transaction costs
11:50 12:10
Coffee break
12:10 13:00
13:10 13:30
13:30 15:00
Lunch
15:00 15:50
16:00 16:20
T. Rheinl
ander: Hedging of barrier options via a general selfduality
16:20 16:40
Coffee break
16:40 17:30
17:40 18:00
18:00 20:00
Welcome reception
Conference program
Tuesday, 25 June
9:30 10:20
10:30 11:20
11:30 11:50
11:50 12:10
Coffee break
12:10 13:00
13:10 13:30
D. Rheinl
ander: Pricing and hedging variance swaps on a swap
rate
13:30 15:00
Lunch
15:00 15:50
16:00 16:20
16:20 16:40
Coffee break
16:40 17:00
17:00 17:20
17:20 17:40
17:40 18:00
10
Conference program
Wednesday, 26 June
9:30 10:20
10:30 11:20
11:30 12:20
12:30 14:00
Lunch
Conference program
Thursday, 27 June
9:30 10:20
10:30 11:20
L. Vostrikova: Semimartingale models with additional information and their application in mathematical finance
11:30 11:50
R. Ahlip: Pricing foreign currency options under jumps diffusions and stochastic interest rates
11:50 12:10
Coffee break
12:10 12:40
Poster session
12:40 13:30
13:30 15:00
Lunch
15:00 15:50
16:00 16:20
A. Slastnikov: Optimization of credit policy of bank and the government guarantees in a model of investment in a risky project
16:20 16:40
Coffee break
16:40 17:30
17:40 18:00
S. Sidorov: GARCH Model with jumps augmented with news analytics data
18:00 18:20
11
12
Conference program
Friday, 28 June
9:30 10:20
10:30 11:20
11:30 11:50
11:50 12:10
Coffee break
12:10 12:30
A. Ahmad: Option pricing via stochastic volatility models: impact of correlation structure on option prices
12:30 12:50
12:50 13:10
13:10 13:30
13:30 15:00
Lunch
15:00 15:20
O.
Subdiffusive OrnsteinUhlenbeck processes and applications to finance
15:20 15:40
15:40 16:00
16:00 16:20
Coffee break
16:20 16:40
16:40 17:00
17:00 17:10
Closing
17:10 19:00
Farewell drinks
Plenary talks
Plenary talks
15
Email: [email protected]
16
Plenary talks
Developing long term yield curve models for low rate regimes
M. A. H. Dempster
University of Cambridge and Cambridge Systems Associates Limited, UK
After a brief discussion of alternative approaches to long term yield curve modelling, this talk will first discuss typical applications of these models to structured
product pricing, investment and asset liability management. It will go on to evaluate existing models and their drawbacks before developing and testing a nonlinear
3 factor model appropriate for todays low rate environments which is based on a
posthumously published suggestion of Fisher Black. The talk will conclude with
a description of current research into computationally intensive calibration techniques for this model and a more complex model requiring the EM algorithm which
involves iterative Kalman filtering and maximum likelihood parameter estimation.
Progress to date and remaining challenges will be described.
References
[1] Dempster et al. (2006). Managing guarantees. Journal of Portfolio Management 32.2
245-256.
[2] Medova et al. (2008). Individual asset-liability management. Quantitative Finance 8.9
547-560.
[3] Dempster & Medova (2011). Asset liability management for individual households.
British Actuarial Journal 16.2 405-464 (with discussion of the Sessional Meeting of the
Institute of Actuaries, London 22.2.10).
[4] Dempster & Medova (2011). Planning for retirement: Asset liability management for
individuals. In: Asset Liability Management Handbook, Mitra & Schwaiger, eds. Palgrave Macmillan 409-432.
[5] Dempster, Mitra & Pflug (2009). Quantitative Fund Management. Chapman & Hall /
CRC
[6] Dempster, Medova & Villaverde (2010). Long term interest rates and consol bond
valuation. Journal of Asset Management 11.2-3 113-135.
[7] Dempster & Carton de Wiart (2011). Wavelet optimized valuation of financial derivatives. International Journal of Theoretical and Applied Finance 14.7 1113-1137.
[8] Bertocchi, Consigli & Dempster (2011). Stochastic Optimization Methods in Finance
and Energy. Springer.
[9] Evans, Dempster & Medova (2012). Developing a practical yield curve model: An
odyssey. To appear in New Developments in Macro-Finance Yield Curves, J Chadha,
A Durre, M Joyce & L Sarnio, eds., Cambridge University Press (2013). Available on
SSRN
Email: [email protected]
Plenary talks
17
In classical economic theory the law of one price prevails and market participants
trade freely in both directions at the same price. This approach is appropriate for
highly liquid markets such as e.g. stock exchanges. In the absence of perfect
liquidity the law of one price should be replaced by a two price valuation theory
where market participants continue to trade freely with the market but the terms of
trade now depend on the direction of the trade. Typical examples of markets with
poor liquidity are over the counter (OTC) markets or some markets for corporate
bonds.
We develop here a static as well as a continuous time theory for two price
economies. The two prices are termed bid and ask or lower and upper price but they
should not be confused with the vast literature where bid-ask spreads are related
to transaction costs or other frictions involved in modeling financial markets. The
two prices arise on account of an exposure to residual risk that results from the
absence of sufficient liquidity and cannot be eliminated. The prices are designed
to make this exposure acceptable. Acceptability as a strict mathematical term
is modeled by requiring positive expectations under a whole host M of test or
scenario probabilities Q as described for example in Artzner, Delbaen, Eber and
Heath (1999). As a consequence the bid or lower price of a cash flow which is
represented by a random variable X is given by the infimum of test valuations
b(X) = inf E Q [X]
QM
while the ask or upper price turns out to be the supremum of the same set of test
valuations
a(X) = sup E Q [X].
QM
The resulting pricing operators are nonlinear operators on the space of random variables, with the lower price being concave and the upper price convex. In
particular the upper price of a portfolio of financial instruments (risks) is smaller
than the sum of the prices of the components of the portfolio while the lower price
is similarly higher. Under the appropriate assumptions the computation of these
nonlinear expectations can be operationalized. The lower price can be expressed
as an expectation which is computed after distorting the risk distribution function
by composing it with a concave distribution function on the unit interval. The
price is then represented in the form
Z
b(X) =
xd(FX (x)).
Email: [email protected]
18
Plenary talks
(0 x 1, 0).
Plenary talks
19
[5] Eberlein, E., Madan, D., Pistorius, M., Yor, M.: Bid and ask prices as non-linear
continuous time G-expectations based on distortions. Preprint, University of Freiburg
(2013).
[6] Peng, S.: G-expectation, G-Brownian motion and related stochastic calculus of Ito
type. In Benth, F. E., Di Nunno, G., Lindstrm, T.; ksendal, B., Zhang, T. (eds.):
Stochastic Analysis and Applications: The Abel Symposium 2005, Proceedings of the
Second Abel Symposium, Oslo, July 29August 4, 2005, held in honor of Kiyosi Ito,
Springer, pp. 541568, (2007).
20
Plenary talks
Behzad Nouri
Contingent capital is debt issued by a bank that converts to equity when the
bank nears financial distress. This type of security has been proposed as a potential
solution to the problem of banks that are too big too fail, providing a private
sector alternative to costly government bail-outs.
The biggest challenge to implementation is the selection of the trigger for conversion from debt to equity. Some have proposed triggers based on market prices,
such as a decline in the banks own stock price. However, using the market price of
a firms equity to trigger a change in the firms capital structure creates a question
of internal consistency because the value of the equity itself depends on the firms
capital structure.
We analyze the general problem of existence and uniqueness of equilibrium values for a firms liabilities in this context, meaning values consistent with a marketprice trigger. The liquidity of the triggering security, as measured by trading
frequency, has important implications for this problem. In a static or discretetime formulation of the problem, multiple solutions are possible. In contrast, we
show that the possibility of multiple equilibria can largely be ruled out in continuous time. Continuous-time trading allows prices to fully adjust in anticipation of
reaching a trigger.
Put abstractly, the problem we consider is the following. Given two martingales
U and V on a time interval [0, T ], when does there exist a (unique) third martingale
S, such that S coincides with U the first time S crosses a specified barrier, and
ST = VT if S never crosses the barrier. We interpret U and V as the prices of
claims on post-conversion and no-conversion variants of a firm, and we seek an
equilibrium price process S in which conversion is triggered by a barrier crossing
by S itself. (This is analogous to pricing a self-referential barrier option in which
the barrier is applied to the price of the barrier option itself, rather than to the
underlying asset.) We impose conditions on U and V that lead to a unique solution
and interpret these conditions in terms of contract features.
Within our general framework, existence of an equilibrium is ensured through
appropriate positioning of the trigger level; in the case of contingent capital with
a stock price trigger, we need the trigger to be sufficiently high. More generally,
if the conversion is to be triggered by a decline in the market price of a claim,
then the key condition we need is that the no-conversion price be higher than the
post-conversion price when either is above the trigger. Put differently, we require
that the trigger be sufficiently high to ensure that this holds. For the design of
contingent capital with a stock price trigger, this condition may be interpreted
to mean that conversion should be disadvantageous to shareholders. Our results
apply as well to other types of changes in capital structure and triggers based on
debt values as well as equity values.
Email: [email protected]
Plenary talks
21
M. R. Grossinho
u(0, x) = g(x) in Rd ,
(1)
i, j = 1, . . . , d,
with real coefficients (written with the usual summation convention), f and g are
given real-valued functions, and T (0, ) is a constant. We assume that operator
/t L is uniformly parabolic, and allow the growth in the spatial variables of the
first and second-order coefficients in L (linear and quadratic growth, respectively),
and of the data f and g (polynomial growth).
Multidimensional PDE problems arise in Financial Mathematics and in Mathematical Physics. We are mainly motivated by the application to a large class of
stochastic models in Financial Mathematics comprising the non path-dependent
options, with fixed exercise, written on multiple assets (basket options, exchange
options, compound options, European options on future contracts and foreignexchange, and others), and also to a particular type of path-dependent options:
the Asian options (see, e.g., [14]).
Let us consider the stochastic modeling of a multi-asset financial option of European type under the framework of a general version of Black-Scholes model, where
the vector of asset appreciation rates and the volatility matrix are taken time and
space-dependent. Owing to a Feynman-Kac type formula, pricing this option can
be reduced to solving the Cauchy problem (with terminal condition) for a degenerate second-order linear parabolic PDE of nondivergent type, with null term and
unbounded coefficients (see, e.g., [14]). Therefore, alternatively to approximating the option price with probabilistic numerical methods, we can approximate
the solution of the corresponding PDE problem with the use of non-probabilistic
techniques.
When problem (1) is considered in connection with the Black-Scholes modeling
of a financial option, we have that the growth of the vector SDE coefficients in the
underlying financial model is appropriately matched. Also, by setting the problem
with this generality, we cover the general case where the asset appreciation rate
vector and the volatility matrix are taken time and space-dependent. Finally, by
imposing weak conditions on the initial data g, we allow the financial derivative
Email: [email protected]
22
Plenary talks
For the time discretisation, we refer to [8] where it is investigated the approximation of a
general linear evolution equation problem which the PDE problem (1) can be cast into.
2
The references for Purtukhias works can be found in [10].
3
We refer to [19] for a brief summary of the methods history, and also for the references of
the seminal works by R. Courant, K. O. Friedrichs and H. Lewy, and further major contributions
by many others.
4
For the application of the finite-difference method to financial option pricing, we refer to the
review paper [2] for the references of the original publications by M. Brennan and E. S. Schwartz,
and further major research.
Plenary talks
23
24
Plenary talks
[7] Evans, L. C., 1998: Partial Differential Equations. Graduate Studies in Mathematics
19, American Mathematical Society, U.S.A.
[8] Goncalves, F. F., M. R. Grossinho, and E. Morais, 2012: Discretisation of abstract linear evolution equations of parabolic type. Adv. Difference Equ., 2012:14.
doi:10.1186/1687-1847-2012-14
[9] Goncalves, F. F., 2007: Numerical approximation of partial differential equations
arising in financial option pricing. Ph.D. Thesis. University of Edinburgh, U.K.
[10] Gy
ongy, I., and N. V. Krylov, 1990: Stochastic partial differential equations with
unbounded coefficients and applications I. Stochastics, 32, 5391.
[11] Gy
ongy, I., and N. V. Krylov, 2005: An accelerated splitting-up method for parabolic
equations. SIAM J. Math. Anal., 37, 10701097.
[12] Gy
ongy, I., and N. V. Krylov, 2009a: First derivatives estimates for finite-difference
schemes. Math. Comp., 78, 20192046.
[13] Gy
ongy, I., and N. V. Krylov, 2009b: Higher order derivative estimates for finitedifference schemes for linear elliptic and parabolic equations. Methods Appl. Anal.,
16(2), 187216.
[14] Lamberton, D., and B. Lapeyre, 1996: Introduction to Stochastic Calculus Applied to
Finance. Chapman and Hall, U.K.
[15] Lions, J.-L., and E. Magenes, 1968: Probl`emes aux Limites Non Homog`enes et Applications, Vol. 1 (in French). Dunod, Gauthier-Villars, Paris, France.
[16] L
otstedt, P., J. Persson, L. von Sydow, and J. Tysk, 2007: Spacetime adaptive finite
difference method for European multi-asset options. J. Comput. Math. Appl., 53(8),
11591180.
[17] Pooley, D. M., P. A Forsyth, and K. R. Vetzal, 2003: Numerical convergence properties
of option pricing PDEs with uncertain volatility. IMA J. Numer. Anal., 23, 241267.
[18] Tangman, D. Y., A. Gopaul, and M. Bhuruth, 2008: Numerical pricing of options using
high-order compact finite difference schemes. J. Comput. Appl. Math., 218, 270280.
[19] Thomee, V., 1990: Finite Difference Methods for Linear Parabolic Equations. Chapter
in: P. G. Ciarlet, and J. L. Lions, Eds., Finite Difference Methods (Part 1) Solution
of Equations in Rn (Part 1), Handbook of Numerical Analysis, Vol. I, North-Holland,
3196.
[20] Zisowsky, A., and M. Ehrhardt,, 2008: Discrete artificial boundary conditions for
nonlinear Schr
odinger equations. Math. Comput. Modell., 47(1112), 12641283.
Plenary talks
25
Email: [email protected]
26
Plenary talks
E. Lepinette3
Email: [email protected]
Plenary talks
27
1. Probabilistic set-up
Let E be a Polish space, modelling all possible states in an economy. Append
a point 4 to E, which will model a cemetery state. If : [0, ) 7 E {4}
is right-continuous, define () := inf{t R+ | (t) = }, with the interpretation
of the economys lifetime (or default). Define as the set of all right-continuous
: [0, ) 7 E {4} such that (0) E and (t) = 4 holds for all t [(), ),
and let F = (Ft )tR+ be the right-continuous augmentation of the smallest
filtration
W
that makes the coordinate process on adapted. Finally, set F := tR+ Ft .
Two probabilities P and Q on (, F) will be called equivalent prior to if for
any stopping time T and AT FT , Q[AT {T < }] = 0 holds if and only if
P[AT {T < }] = 0; denote such relationship via Q < P. Furthermore, agree
to call foretellable under Q if there exists a nondecreasing sequence (n )nN of
stopping times such that Q[n < , n N] = 1 and Q[limn n = ] = 1. The
next result, mathematically interesting in its own right, is of major importance
(albeit, in a tacit way) in the development of the financial model later.
Theorem. For any probability measure P on (, F), there exists Q < P such
that is foretellable under Q.
2. Financial set-up
The class of wealth processes available to an investor with (normalized) unit
initial capital will be defined in a rather abstract and general-encompassing way.
It contains as a special case any reasonable class of (potentially, constrained) nonnegative wealth processes resulting from frictionless trading that has appeared in
the literature, including the possibility of an infinite number of assets (such as bond
markets or equity markets with traded options), and allowing for overall default of
the economy. A set X of stochastic processes will be called a wealth-process set if:
Each X X is a nonnegative, adapted, right-continuous process with X0 = 1,
and Xt = 0 holds for all t .
1[,[ X .
X is fork-convex: for any s R+ , X X , any strictly positive X 0 X and
X 00 X , and any [0, 1]-valued Fs -measurable random variable s , the process
if 0 t < s,
Xt ,
0
0
00
00
R+ 3 t 7 s (Xs /Xs )Xt + (1 s )(Xs /Xs )Xt , if s t < , (FC)
0,
if t,
Email: [email protected]
28
Plenary talks
is also an element of X .
Fork-convexity corresponds to re-balancing: (FC) exactly describes the wealth generated when a financial agent invests according to X up to time s, and then reinvests
a fraction of the money in the wealth process described by X 0 and the remaining
fraction (1 ) in the wealth process described by X 00 .
3. Fundamental Theorem of Asset Pricing (FTAP) and superhedging
For a stopping time T and nonnegative optional process V , define
p(V, T ) := inf{x > 0 | > 0, X x, xX with P[XTx, < VT , T < ] < }.
Loosely speaking, p(V, T ) is the minimal capital required at time zero in order
to superhedge VT at time T with as high a probability as desired, provided that
default has not occurred by time T .
An arbitrage of the first kind in the market is a pair of T R+ and nonnegative
optional process V such that p(V, T ) = 0 and P[VT > 0, T < ] > 0 hold. Say that
condition NA1 holds if there are no opportunities for arbitrage of the first kind.
Denote by Q the collection of all Q < P such that all processes X X are
(nonnegative) supermartingales.
FTAP. In the previous set-up, condition NA1 is equivalent to Q =
6 . Under any of
the previous two (equivalent) conditions, for any stopping time T and nonnegative
optional process V , the following superhedging duality holds:
p(V, T ) = sup EQ [VT ; T < ].
QQ
4. Market completeness
For this last part, a precise mathematical statement will not be given, since it
would require quite involved definitions and notation; however, we shall discuss
below the main ideas in a somewhat precise manner.
In the setting of Section 2, assume further that X is closed in the semimartingale
topology, locally in time for probabilities Q < P under which is foretellable.
Then, under condition NA1 , one can obtain a variant of the second Fundamental Theorem of Asset Pricing, which provides an equivalence between the notion
of market completeness (basically, the ability to perfectly replicate bounded nonnegative positions using maximal wealth processes) with uniqueness of a maximal
probability Q Q. The latter qualification of maximality for Q Q means that
whenever Q Q is such that Q[AT {T < }] Q[AT {T < }] holds for any
stopping time T and AT FT , then Q = Q.
Plenary talks
29
Yuan Tian2
Email: [email protected]
30
Plenary talks
Email: [email protected]
Plenary talks
31
The subject of this talk is the application of dynamic analysis of financial time
series data in detecting of fraud and other suspicious and de-stabilizing activity
of financial institutions. Mr. Markov will present a powerful technique Dynamic
Style Analysis (DSA) that helps to detect hidden risks, leverage and even alert to
possible fraud using only financial performance data.
Hedge fund industry has grown rapidly over the past decade to over $2 trillion
in assets and over 8,000 funds. At the same time, despite significant efforts to
regulate the industry, the amount of information available to hedge fund investors
remains negligible as compared to traditional investment products such as mutual
funds. In most cases, the only available information on a hedge fund is a time series
of monthly performance numbers and a vague description of the strategy. Hedge
funds represent significant systemic risk: they amassed significant amounts of assets
and through use of leverage and derivatives could destabilize world markets. In
addition, some hedge funds manipulate their performance data and unsuspected
investors become victims of outright fraud.
Practical aspects of the methodology will be discussed, e.g., parameter calibration, model selection, structural shift detection, etc.
The focus of the talk will be on case studies including:
- cases of fraud: a recent one (2012) in Japan where AIJ Investment Advisors
defrauded pension funds of billions of dollars in a sophisticated derivatives
strategy;
- detecting insider trading (Galleon hedge fund, 2009);
- massive arbitrage and rapid trading: Soros famous breaking of Bank of England (1992) and 2010 Flash Crash;
- extreme leverage: Long Term Capital (LCTM, 1998) where Federal Reserve
had to orchestrate a bailout of the hedge fund to avoid disruption of world
markets.
Email: [email protected]
32
Plenary talks
Usually in this case one has to somehow guess the solution and prove that this
guess in fact solves the optimization problem; we call this approach verification.
For example we can consider the papers [7], [8], [11], [10]. The second class consists
of general results, for wide classes of processes and reward functions. We call this
the theoretical approach. It typically include results about properties of the
solution. In this class we mention for example [2], [4], [1]. But these two classes not
always meet, as frequently in concrete problems the assumptions of the theoretical
studies are not fulfilled, and, what is more important, many of these theoretical
studies do not provide concrete ways to find solutions. In what concerns the first
approach, a usual procedure is to apply the principle of smooth fit, that generally
leads to the solution of two equations: the continuous fit equation and the smooth
fit equation. Once these equations are solved, a verification procedure is needed
in order to prove that the candidate is the effective solution of the problem (see
chapter IV in [9]). This approach, when an explicit solution can be found, is very
effective. In what concerns the second approach, maybe the most important result
is Dynkins characterization of the solution of the value function V as the least
-excessive (or -superharmonic) majorant of the payoff function g [2]. Other ways
of classifying approaches in order to study optimal stopping problems include the
martingale-Markovian dichotomy as exposed in [9].
Our departing point, inscribed in the Markovian approach, is Dynkins characterization of the optimal stopping problem solution. Dynkins characterization [2]
states that, if the reward function is lower semi-continuous, V is the value function
of the non-discounted optimal stopping problem with reward g if and only if V is
the least excessive function such that V (x) g(x) for all x I. Applying this
result for the killed process Y , and taking into account the relation between X
and Y , we obtain that V , the value function of the problem with discount , is
characterized as the least -excessive majorant of g.
The second step uses Rieszs decomposition of an -excessive function. We
recall this decomposition in our context (see [5, 6, 3]). A function u : I R
is -excessive if and only if there exist a non-negative Radon measure and an
-harmonic function such that
Z
u(x) =
G (x, y)(dy) + (-harmonic function).
(2)
(`,r)
Email: [email protected]
Plenary talks
33
where L is the infinitesimal generator and m(dy) is the speed measure of the diffusion. Assuming that the stopping region has the form I {x x }, and taking
into account that V is -harmonic in the continuation region and V = g in the
stopping region we obtain as a suitable candidate to be the representing measure
if y < x ,
0,
(4)
(dy) = kx (dy),
if y = x ,
( L)g(y)m(dy), if y > x ,
Based on these considerations, we present some theoretical results and some new
examples.
An important byproduct of our approach has to do with the smooth fit principle.
Our results are independent of this principle, but they give sufficient conditions in
order to guarantee it. Our approach is also applicable to certain classes of processes
with jumps.
All the presented results are joint work with Fabian Crocce.
References
[1] El Karoui, N. and Lepeltier, JP and Millet, A., A probabilistic approach of the reduite,
Probability and Mathematical Statistics, 13, 97121, 1992
[2] E. B. Dynkin, Optimal choice of the stopping moment of a Markov process, Dokl. Akad.
Nauk SSSR, Doklady Akademii Nauk SSSR, 150, 1963, 238240.
[3] E. B. Dynkin, The exit space of a Markov process, Uspehi Mat. Nauk, Akademiya Nauk
SSSR i Moskovskoe Matematicheskoe Obshchestvo. Uspekhi Matematicheskikh Nauk,
24, 1969, 4 (148), 89152
[4] On Stefans problem and optimal stopping rules for Markov processes, Grigelionis, B.
I. and Shiryaev, A. N., Theory of Probability & Its Applications, 11, 4, 541558, 1966.
[5] Hiroshi Kunita and Takesi Watanabe, Markov processes and Martin boundaries, Bull.
Amer. Math. Soc., Bulletin of the American Mathematical Society, 69, 1963, 386391.
[6] Hiroshi Kunita and Takesi Watanabe, Markov processes and Martin boundaries. I,
Illinois J. Math., Illinois Journal of Mathematics, 9, 1965, 485526.
[7] H.P. McKean Jr, Appendix: A free boundary problem for the heat equation arising
from a problem in mathematical economics, Industrial Management Review, 6, 2, 32
39, 1965.
[8] Merton, Robert C., Theory of rational option pricing, Bell J. Econom. and Management
Sci., The Rand Journal of Economics, 4, 1973, 141183,
[9] Peskir, Goran and Shiryaev, Albert, Optimal stopping and free-boundary problems,
Lectures in Mathematics ETH Z
urich, Birkhauser Verlag, Basel, 2006,
34
Plenary talks
[10] Shepp, L. A. and Shiryaev, A. N., The Russian option: reduced regret, Ann. Appl.
Probab., The Annals of Applied Probability, 3, 1993, 3, 631640.
[11] Taylor, Howard M., Optimal stopping in a Markov process, Ann. Math. Statist.,
Annals of Mathematical Statistics, 39, 1968, 13331344.
Plenary talks
35
Nino Kordzakhia
1,2
36
Plenary talks
Martin Schweizer2
ETH Z
urich and Swiss Finance Institute, Z
urich, Switzerland
Email: [email protected]
Plenary talks
37
Yan Dolinsky1,2
The original transport problem is to optimally move a pile of soil to an excavation. Mathematically, given two measures of equal mass, we look for an optimal
map that takes one measure to the other one and also minimizes a given cost functional. Kantorovich relaxed this problem by considering a measure whose marginals
agree with given two measures instead of a bijection. This generalization linearizes
the problem. Hence, allows for an easy existence result and enables one to identify
its convex dual.
In robust hedging problems, we are also given two measures. Namely, the initial and the final distributions of a stock process. We then construct an optimal
connection. In general, however, the cost functional depends on the whole path of
this connection and not simply on the final value. Hence, one needs to consider
processes instead of simply the maps S. The probability distribution of this process
has prescribed marginals at final and initial times. Thus, it is in direct analogy
with the Kantorovich measure. But, financial considerations restrict the process
to be a martingale Interestingly, the dual also has a financial interpretation as a
robust hedging (super-replication) problem.
In this talk, we prove an analogue of Kantorovich duality: the minimal superreplication cost in the robust setting is given as the supremum of the expectations
of the contingent claim over all martingale measures with a given marginal at the
maturity. The related papers are [1, 2].
Acknowledgements. Research partly supported by the European Research Council under the grant 228053-FiRM, by the ETH Foundation and by the Swiss Finance
Institute.
References
[1] Y. Dolinsky and H.M. Soner, Martingale optimal transport and robust hedging in continuous time, submitted to Probability Theory and Related Fields, (2012).
[2] Y. Dolinsky and H.M. Soner, Robust Hedging with Proportional Transaction Costs,
math archive, 1302.0590v1, (2013).
Email: [email protected]
38
Plenary talks
Teruyoshi Suzuki2
1. Model Setup
Our model is an extension of [1], [3], [2] and other related papers by introducing
default costs.
Suppose that there are totally n firms that cross-hold their equities and debts
and that the debts have a seniority structure with at most m priorities. Let bki
be the face value of k-th subordinated debt issued by firm i, k = 1, . . . , m and
m1
bk = (bk1 , . . . , bkm )> . That is, bm
with
i is paid with first priority at maturity, bi
second, etc.
The cross-holding structure of k-th debts is described by the n n matrix
Mk = (Mijk )ni,j=1 . More concretely, firm i owns a proportion Mijk [0, 1] of the
k-th debt issued by firm j (Mii = 0). If Mk is a substochastic matrix, it means
that a part of k-th debts is owned by outside investors. The equities are also
cross-held with the structure M0 = (Mij0 )ni,j=1 .
Let ei be the business asset of firm i and e = (e1 . . . , en )> . The total asset
(payment resource for debts) of firm i, denoted by ai , is written as
ai = e i +
m X
n
X
Mij` rj` ,
(1)
`=0 j=1
where rik is the payoff of firm is k-th debt at maturity for k = 1, . . . , m and ri0 is
the payoff of firm is equity (rk = (r1k , . . . , rnk )> ).
A key assumption in this study is that a default at maturity accompanies some
liquidation costs. More concretely, if firm i cannot fully pay back all the debts at
maturity, the payment resource of firm i is reduced by the proportion i [0, 1].
We write = diag(i )ni=1 .
Define
dki =
m
X
b`i
(2)
`=k+1
and dk = (dk1 , . . . , dkn )> . The clearing payment vector r R(m+1)n is naturally
defined by the following equations:
ri0 = ai d0i +
(3)
Email: [email protected]
Plenary talks
39
and
rik = 1{ai d0 } bk1 + 1{ai (r)<d0 } bki (1 i )ai dki
i
(4)
for k = 1, . . . , m and i = 1, . . . , n.
2. Existence of Clearing Vectors
To study the clearing payment vector, let us define the function f : R(m+1)n
R(m+1)n by
f 0 (r) = a(r) d0
+
(5)
+
(6)
for k = 1, . . . , m, where
a(r) =e +
m
X
M` r` ,
(7)
`=0
1{a1 (r)<d01 }
D(r) =
O
..
(8)
1{an (r)<d0n }
r00
0 1
= (I M )
e+
m
X
!
` `
M b d
`=1
(9)
+
(10)
and rh = f (rh1 ) for h 1. We can verify that {rh } is a monotonically nonincreasing sequence and bounded below. The limit r is what we want to obtain.
Alternatively, we can prove the existence of a clearing vector with the sequence
{rh } defined by r0 = 0 and rh = f (rh1 ). The proof is completed by noticing that
{rh } is a non-decreasing sequence and bounded above, indicating the existence of
a limit r. We can also show that r 6= r in general.
40
Plenary talks
(11)
(m+1)n
for any r R+
with f (r) = r. Similarly the clearing payment vector r is the
least clearing vector in the sense that
rr
(m+1)n
(12)
Intuitively, the computational iteration {rh } echoes the fictitious default algorithm proposed by [1]. To see that, let the sequence of matrices {Dh } be given
by
D0 = O,
Dh = D(rh ).
(13)
Dh = D(rh ).
(14)
and calculate the sequence {rh }. The survival firms are determined by the matrix
D(r). We call the process of {Dh } the fictitious survival algorithm.
3. Numerical Results
If any.
Acknowledgements. The authors are grateful to the financial support of the
Japanese Ministry of Education, Culture, Sports, Science and Technology (MEXT)
Grand in Aid for Scientific Research (B) #23310098.
References
[1] Eisenberg, L. and T. H. Noe (2001), Systemic Risk in Financial Systems, Management
Science, 47(2), 236249.
[2] Fischer, T. (2012), No-Arbitrage Pricing under Systemic Risk: Accounting for CrossOwnership, forthcoming in Mathematical Finance.
[3] Suzuki, T. (2002), Valuing Corporate Debt: The Effect of Cross-Holdings of Stock
and Debt, Journal of the Operations Research Society of Japan, 45(2), 123144.
Plenary talks
41
Semi-martingale models are widely used for pricing and hedging of financial assets. Namely, the price of risky asset can be represented by Dolan-Dade exponential
S = (St )t0 of some semi-martingale X:
St = S0 E(X)t .
In many cases the answer on the questions what is the price of the option related
with S, and what is the optimal strategy, depend on the information which an
investor can have. Initiated by Baudoin (2003), and investigated in a number of
papers ( see, for example, Gasbarra, Valkeila, Vostrikova (2004), Hillairet, Jiao
(2010,2011)) these questions begin to be very important in Mathematical Finance.
According to the quantity of information of investors, we will distinguish three
type of them, namely, non-informed, partially informed and perfectly informed
agents. Since a semi-martingale is always given on filtered probability space, we
will model the information of an investor by enlargement of the filtration. Namely,
we suppose that this information is given by an additional random variable or
a random process = (t )t0 . Then, non-informed agents will work on initial
probability space equipped with natural filtration F = (Ft )t0 of the process X
where
\
Ft =
(Xu , u s).
s>t
The partially informed agents will work on enlarged probability space with pro = (Ft )t0 where
gressively enlarged filtration F
\
Ft =
(Xu , u s) (u , u s).
s>t
And, finally, the perfectly informed agents will use again enlarged probability space
with initially enlarged filtration G = (Gt )t0 where
\
Gt =
(Xu , u s) (u , u T )
s>t
and T is time horizon. It should be pointed out that the enlargement of the
filtration leads to incomplete markets even in the case when initial market was
complete.
We give now a several examples of the situation presented above. The first one
concerns default models. The modelling of defaultable world suppose to introduce
Email: [email protected]
42
Plenary talks
a default time , which can be simply a random time, related with some economical reasons such as structural changes, increasing of the price of raw material, or
political changes, as change of the regime (see for instance, Eliott, Jeanblanc, Yor
(2001), El Karoui, Jeanblanc, Jiao (2009)).
The next example concerns so called change-point models, i.e. the models with
different behaviour before and after some random time. It should be noticed that
change-point problems have a long history, probably beginning with the papers
of Page (1955) in an a-posteriori setting, and of Shiryaev (1963) in a quickest
detection setting. The problem was later considered in many papers and books,
and was often related to a quickest detection approach. We underline, that not
only quickest detection approach is interesting in financial mathematics, and this
fact is related, for instance, with the investigation of the models with changing of
the regime (see Cawston, Vostrikova (2011)) , in particular with the models with
random dividends ( see Gapeev, jeanblanc (2010)).
We will concentrate ourselves on the third example, which concerns the indifference pricing. Namely, in the real financial market investors can held traded risky
assets of maturity time T and receive some particular derivatives such as contingent
claims offering some pay-off at maturity time T 0 > T > 0. It can happen that the
assets related with contingent claims can not be traded since the trading is difficult
or impossible for investor because of lack of liquidity or legal restrictions. In this
situation the investor would like maximize expected utility of total wealth and at
the same time reduce the risk due to the uncertainty of pay-off of the contingent
claim. In such situations the utility indifference pricing become to be a main tool
for option pricing.
To be more precise, let us suppose that our market consists on non-risky asset
Bt = B0 exp(rt), where r is interest rate, and two risky assets
t
St = S0 E(X)t , St = S0 E(X)
are semi-martingales with jumps X > 1, X
> 1, and E
where X and X
is Dolean-Dade exponential. The investor can trade S and at the same time he
has a European type claim on S given by g(ST 0 ) where g is some real-valued Borel
function. Let us denote by the set of self-financing predictable strategies. Then,
for utility function U and initial capital x, the optimal expected utility related with
S will be
Z T
VT (x) = sup E U x +
s dSs
VT (x, g) = sup E U x +
s dSs + g(ST 0 )
As known, the indifference price pbT for buyer of the option g(ST 0 ) is a solution to
the equation
VT (x pbT , g) = VT (x)
Plenary talks
43
and it is an amount of money which the investor would be willing to pay today for
the right to receive the claim and such that he is no worse off in expected utility
terms then he would have been without the claim. The indifference price for the
seller psT of the option is a solution to the equation
VT (x + pbT , g) = VT (x)
and it is an amount of money which the seller of the option would be willing to
receive in counterpart of the option in order to preserve his own optimal utility.
The optimal utility of assets containing the options highly depends on the level
More precisely, the investor can be nonof information of the investor about S.
informed, partially informed or perfectly informed agent and the level of information changes the class mentioned in previous formulas. Namely, a non-informed
agent can maximize his expected utility taking the strategies only from the set of
self-financing predictable strategies with respect to the natural filtration F of X.
At the same time, a partially informed agent can built his optimal strategy using
the set of self-financing predictable strategies with respect to the progressively en with the process X.
Finally, a perfectly informed agent can use
larged filtration F
the self-financing predictable strategies with respect to initially enlarged filtration
G with ST 0 .
The first part of our talk is devoted to the general results about the maximisation
of utility for semi-martingale models depending on a random factor. As previously
let us introduce the total utility with the option g():
Z T
V (x, g) = sup EP U x +
s dSs () + g()
0
(G)
and then integrate the solution with respect to . It should be noticed that the
investigation of semi-martingales depending of a random parameter leads, in general, to deep measurability problems as it was pointed out in Stricker, Yor (1978).
44
Plenary talks
Plenary talks
45
The Kelly Capital Growth Investment Strategy (KCGIS) is to maximize the expected utility of final wealth with a logarithmic utility function. This approach
dates to Bernoullis 1738 suggestion of log as the utility function arguing that
marginal utility was proportional to the reciprocal of current wealth. In 1956 Kelly
showed that static expected log maximization yields the maximum asymptotic long
run growth. Later, others added more good properties such as minimizing the time
to large asymptotic goals, maximizing the median, and being ahead on average
for the first period. But there are bad properties as well such as extremely large
bets for short term favorable investment situations because the ArrowPratt risk
aversion index is near zero. Paul Samuelson was a critic of this approach and here
we discuss his various points sent in letters to Ziemba and papers reprinted in the
recent book, MacLean, Thorp and Ziemba (2011). Samuelsons opposition has prevented many finance academics and professionals from using and suggesting Kelly
strategies to students. For example, Ziemba was asked to explain this to Fidelity
Investments, a major Boston investment firm close to and influenced by Samuelson
at MIT. I agree that these points of Samuelson are theoretically correct and respond
to theory. I argue that they all make sense and caution users of this approach to be
careful and understand the true characteristics of these investments including ways
to lower the investment exposure. While Samuelsons objections help us understand
the theory better, they do not detract from numerous valuable applications, some
of which are discussed here.
Email: [email protected]
Contributed talks
Contributed talks
49
Email: [email protected]
50
Contributed talks
Marek Rutkowski2
Foreign exchange options are studied in the Heston stochastic volatility model
for the exchange rate which includes jumps in both the spot exchange rate and
volatility dynamics combined with the Cox , Ingersoll and Ross dynamics for the
domestic and foreign stochastic interest rates. The instantaneous volatility is correlated with the dynamics of the exchange rate return, whereas the domestic and
foreign short-term rates are assumed to be independent of the dynamics of the
exchange rate. The main result furnishes a semi-analytical formula for the price
of the foreign exchange European call option. The FX options pricing formula is
derived using the probabilistic approach, which leads to explicit expressions for
conditional characteristic functions. We argue that the model examined in this paper is the only analytically tractable version of the foreign exchange market model
which includes jumps in the Heston stochastic volatility model and the exchange
rate with the CIR dynamics for interest rates.
Acknowledgements. The research of M. Rutkowski was supported by the ARC
Discovery Project DP120100895.
Email: [email protected]
Contributed talks
51
Damir Filipovic1
Andreea Minca2
The reform of the functioning of over the counter (OTC) derivatives markets lies
at the core of the Dodd-Frank Wall Street Reform and Consumer Protection Act of
2010. Among the regulations is that the majority of OTC derivatives, of the order
of hundreds of trillions of US dollars in terms of notional, should be centrally cleared
so as to insure financial stability. The Basel Committee for Banking Supervision,
European and UK regulators have enacted similar proposals.
Introducing a central clearing counterparty (CCP) modifies the intermediation
structure of the market: any financial obligation between members of the CCP is
now intermediated by the CCP, while part of the members liquidity is transferred
to the CCP in form of guarantee fund contributions. A CCP would therefore
increase exposure concentration in the market and the critical issue is whether
this is accompanied with proper capitalization of the CCP, proper guarantee fund
requirements and proper management of the guarantee fund.
In this work we study financial networks in a stochastic framework. We measure
systemic risk in terms of a coherent valuation principle. The framework allows us
to examine the effects on systemic risk and price contagion of multilateral clearing
via a central clearing counterparty. We build on the framework introduced by [3]
and use a network representation of the OTC market to analyze contagion effects
without and with central clearing, while accounting for liquidation costs. We prove
existence and uniqueness of an interbank payment equilibrium in conjunction with
the price impact on external assets. We find that a CCP not always reduces
systemic risk and provide sufficient conditions for the latter to hold. We also
propose an optimal capitalization of a CCP based on game theoretic arguments.
A real world calibrated numerical study illustrates our findings.
References
[1] H. Amini, D. Filipovic, and A. Minca. Systemic risk with central counterparty clearing.
2013.
[2] H. Amini, R. Cont, and A. Minca. Resilience to contagion in financial networks. Mathematical Finance, to appear.
[3] L. Eisenberg and T. H. Noe. Systemic risk in financial systems. Management Science,
47(2):236249, 2001.
Email: [email protected]
52
Contributed talks
Michael Kupper2
Antonis Papapantoleon3
2
3
(1)
[0,0 ],hp R
sup
Z
s
E Us h (PT F ) +
,hs R
Email: [email protected]
s dSs
(2)
Contributed talks
53
where (St )t[0,T ] is the stock price process (exogenously priced). We further assume that the market is driven by a vector of stochastic factors modelled by an
d-dimensional Levy process in the following way: St = S0 E(hu1 , Xt i) and the consumers random demand at time T is given by T (P ) = 0 (P ) + hu2 , XT i, where
ui Rd , i = 1, 2.
The central goal of this paper is to determine the spot and forward prices of
the commodity that makes the forward and the spot market clear out. Given the
optimization problems (1) and (2), the equilibrium price of the forward contract
on this commodity is the one that makes the forward market equilibrate, that is
the price F that solves the equation hp (F ) = hs (F ). The corresponding price P0 is
the equilibrium spot price.
A List of Contributions
The main results of this work can be summarized in the following list:
The existence and the uniqueness of the equilibrium spot and forward prices
are proved when agents utility functions are exponential.
We extensively analyze two market model examples; one with continuous
stochastic factors and one with jumps. In both cases, we get closed-form
solutions of the equilibrium prices and discuss the findings. These formulas
allow us for instance to identify which parameters have an upward impact on
the commodity spot price.
A dynamic version of the this equilibrium problem is also established and
developed.
54
Contributed talks
Ricardo Huaman2
Email: [email protected]
Contributed talks
55
Josef Teichmann2
ETH Z
urich, Switzerland
56
Contributed talks
[4] J. Jacod. Asymptotic properties of realized power variations and related functionals of
semimartingales. Stochastic Process. Appl., 118(4):517559, 2008.
[5] J. Jacod and P. Protter. Discretization of processes, volume 67 of Stochastic Modelling
and Applied Probability. Springer, Heidelberg, 2012.
[6] P. Malliavin and M. E. Mancino. Fourier series method for measurement of multivariate
volatilities. Finance Stoch., 6(1):4961, 2002.
[7] P. Malliavin and M. Mancino. A Fourier transform method for nonparametric estimation of multivariate volatility. The Annals of Statistics, 37(4):19832010, 2009.
[8] V. Todorov and G. Tauchen. The realized Laplace transforms of volatility. Econometrica, 80: 1105-1127, 2012
[9] J. H. C. Woerner. Power and multipower variation: inference for high frequency data. In
A. N. Shiryaev, M. R. Grossinho, P. E. Oliveira, and M. L. Esquivel, editors, Stochastic
Finance, pages 343364. Springer US, 2006.
Contributed talks
57
We aim at understanding the typical shape of a local volatility surface, by focusing on its extremes (w.r.t. strike and maturity). The asymptotic behavior is
governed by a saddle-point based formula, akin to Lees moment formula for implied volatility. Applications include local vol parametrization design and assessing
volatility model risk. Secondly, it is well known that local vol models cannot deal
with jumps in the underlying. We propose a simple regularization procedure as a
remedy. Its validity is related to recent work of Yor et al. on Kellerers theorem
from the theory of peacocks (processus croissants pour lordre convexe). The talk
is based on joint work with S. De Marco, P. Friz, and M. Yor.
Email: [email protected]
58
Contributed talks
Wim Schoutens
K.U.Leuven, Belgium
The calibration of a model on a given market situation is a critical and important part of any derivative pricing and risk management exercise. Traditionally,
one solves the so-called inverse problem, which consists of finding the parameter
set that is compatible with the observed market price of a set of liquidly traded
derivatives. Typically, a perfect match is not plausible and one looks for an optimal match. More precisely, one minimizes the distance between the model and
the market prices of benchmark instruments by using a search algorithm. Most
commonly, practitioners are minimizing the root mean square error between model
and market vanilla prices or between model and market implied volatilities. Although the root mean square objective function is the current industry practice,
there exist other alternatives just as suitable. The optimal parameter set typically
strongly depends on the choice of the objective function, leading to significantly
different prices for the more exotic and structured products. Besides the calibration
risk issue, it is well known and documented that several additional problems can
arise with the standard calibration methodology. Firstly, one faces the problem
of selecting an appropriate starting value for the search algorithm used. Indeed,
the objective function to be minimized is typically a non-convex function of the
model parameter set and can thus have several local minima, making the solution
of the standard calibration problem dependent on the initial parameter set, which
is taken as starting value of the optimization algorithm and on the sophistication
of the numerical search performed. Further, one has the typical related problem
of finding a local minimum instead of the global minimum. Also, a calibration
exercise can be quite time consuming, especially if the number of parameters to be
calibrated is becoming large.
Hence, we provide a new calibration formalism which consists of matching the
moments of the asset log-return process with those inferred from liquid market
data. In particular, we derive a model independent formula for the moments of the
asset log-return distribution function by expanding power returns as a weighted
sum of vanilla option payoffs. The new calibration methodology rests on closedform formulae only: it is shown that, for a model with N parameters, the moment
matching calibration problem reduces to a system of N algebraic equations which
give directly the optimal parameter set in terms of the market implied standardized
moments of order 2 to order N and avoids thus the delicate choice of a particular
objective function. For the numerical study, we first work out different popular
exponential Levy models and illustrate how the new methodology outperforms the
current market standard ones in terms of both the computation time and the quality of the fit. We then consider exponential Levy models with piecewise constant
Email: [email protected]
Contributed talks
59
parameters between successive quoted option maturities and extend the moment
matching market implied calibration procedure to take into account the term structure characteristic of these models. More particularly, we propose a bootstrapping
moment matching calibration. This sequential calibration arises naturally due to
the additive property of cumulants of independent random variables and consists
in solving M independent moment matching systems of N equations, where M
denotes the number of quoted maturities.
The new calibration formalism provides thus an appealing alternative to the
standard calibration problem since, for the Levy models under investigation, the
method is not requiring any search algorithm and hence any starting value for the
model parameters, it is almost instantaneously delivering the matching parameters
and it avoids local minima problems. The new methodology can also be used as
a preliminary calibration aimed at delivering appropriate market implied starting
values or prior model for the standard inverse problem.
References
[1] Guillaume, F. and Schoutens, W (2013). A moment matching market implied calibration. To appear in Quantitative Finance.
[2] Guillaume, F. and Schoutens, W (2013). A bootstrapping market implied moment
matchingcalibration for models with time-dependent parameters. Working paper.
60
Contributed talks
1. Kramkov and Schachermayer in their seminal paper [3] studied the problem of
maximizing the expected utility of terminal wealth in the framework of a general
incomplete semimartingale model of a financial market. One of the main results
of [3], Theorem 2.1, says that the Fenchel conjugate to the value function of the
utility maximization problem is the value function of a dual problem, and this
dual problem is a minimization problem over the set of the terminal values of
supermartingale densities (deflators). This result was proved in two steps. At
the first step an abstract version of this result, Theorem 3.1, was proved. Its
statement deals with an abstract market model and says that the minimum in the
dual problem has to be taken over a certain set D. In a semimartingale model,
this set D is larger than the set of the terminal values of supermartingale densities.
At the second step it was shown that the value functions of the corresponding
minimization problems over these two sets coincide. An analysis of the proof of the
second step shows that it is based on Theorem 5.5 in Delbaen and Schachermayer
[1] that says that the superhedging price of any nonnegative contingent claim can
be computed as the supremum of its expectations over the set of all equivalent
-martingale measures.
The aim of this paper is to understand on an abstract level what is the mechanism responsible for this connection.
2. Let (, F , P) be a probability space. Denote by L0 the space of all (equivalence
classes of) real-valued random variables. L0 is equipped with the convergence in
probability, and bar means the closure with respect to this convergence. L0+ is the
cone in L0 consisting of nonnegative random variables.
We consider an abstract market model described as a quadruple (, F , P, A ).
where A is a convex subset of L0+ . It is assumed also that A contains a random
variable such that P( ) = 1 for some > 0. A is interpreted as the set
of terminal wealths of an investor corresponding to all her strategies with initial
wealth 1. If the initial wealth is x > 0, then the corresponding set of terminal
wealth is xA .
Put A0 = A L0+ L0+ , C = A0 , D = { L0+ : E 1 for all A }.
Let B L0+ . A possible definition of the superhedging price of B is
(B) = inf{x > 0 : there is a A such that B x} = inf{x > 0 : B xA0 }
(here and below inf = +). Since we do not assume any kind of closedness of
Email: [email protected]
Contributed talks
61
Here the second equality follows from the bipolar theorem by Brannath and Schachermayer. Obviously, (B) (B) for all B L0+ . It is easy to check that
\
(B) = (B) for all B L0+
C
(A0 ).
>1
y R.
x > 0,
y0
y 0.
x>0
(1)
y
dQ/dP
for all Q P and y 0, and for every utility function U described above.
References
[1] F. Delbaen, W. Schachermayer (1998). The fundamental theorem of asset pricing for
unbounded stochastic processes. Mathematische Annalen 312:215-250.
[2] A. A. Gushchin (2011). Dual characterization of the value function in the robust utility
maximization problem. Theor. Probab. Appl. 55:611630.
[3] D. Kramkov, W. Schachermayer (1999). The asymptotic elasticity of utility functions
and optimal investment in incomplete markets. Ann. Appl. Probab. 9:904950.
62
Contributed talks
Oleksandr Zhylyevskyy2
2
We develop a new econometric technique the Generalized Method of Moments(GMM) under a non-localized common shock to consistently estimate parameters of a financial market model using only a single cross-section of stock
returns. Unlike the existing methods, the technique does not require observing a
long history of stock return data. Our technique is novel in that it accounts for
a non-localized common shock (e.g., market risk) affecting cross-sectional observations, which induces strong cross-sectional data dependence and renders standard
estimation methods inconsistent. Also, it differs from the popular two-pass regression method of Fama and MacBeth (1973), which is consistent as the time-series
length grows infinitely large. In contrast, our proposed GMM estimators are shown
to be consistent as the number of stocks in a cross-section grows infinitely large.
We also prove that the estimators are asymptotically mixed normal, which differentiates them from the regular, asymptotically normal GMM case, Hansen (1982).
Despite the asymptotic mixed normality, statistical inference on parameters can
still be conducted using conventional Wald tests. In addition, the overidentifying
restrictions (OIR) test is shown to retain its standard properties. Until now, crosssectional GMM estimation under a non-localized common shock has received only
tangential attention in Andrews (2003), where such a possibility is suggested but
not fully explored. To empirically illustrate the technique, we estimate a financial
model that is an adaptation of the classical ICAPM model of Merton (1973). The
model comprises a well-diversified market portfolio index and a cross-section of
stocks. The price of the index follows a geometric Brownian motion and is affected
by a single source of market risk. Individual stock prices follow geometric Brownian motions and depend on this same source of market risk, but are additionally
affected by stock-specific idiosyncratic risks. We discuss in detail the estimation of
an idiosyncratic volatility premium. Using cross-sections of daily, weekly, monthly,
quarterly, and annual U.S. stock returns from 20002011, we find that the premium
is positive on daily return data, but tends to be negative on monthly, quarterly,
and annual data.
References
[1] Fama, E.F., MacBeth, J.D., 1973. Risk, return, and equilibrium: Empirical tests. Journal of Political Economy 81, 607636.
[2] Andrews, D.W.K., 2003. Cross-section regression with common shocks. Cowles Foundation Discussion Paper 1428. Yale University.
[3] Merton, R.C., 1973. An intertemporal capital asset pricing model. Econometrica 41,
867887.
[4] Hansen, L.P., 1982. Large sample properties of generalized method of moments estimators. Econometrica 50, 10291054.
Email: [email protected]
Contributed talks
63
Modern financial and economic system extremely complex, and the creation
of accurate adequate models from first principles is very difficult. One of the
approaches to the study of these systems is based on the analysis of financial and
economic time series of the form:
{sk }K
k=1 ,
s S RN ,
n = 1, N ,
k K N,
k = 1, K.
(1)
Every k-th countdown can be associated with moment of time tk , at that tk+1 > tk ,
(n)
t T R. Variable sk may be interpreted as value of some financial indicator
on moment of time tk .
In recent times, in financial mathematics for the analysis of these time series
are increasingly using methods of statistical physics, nonlinear dynamics and chaos
theory. One of the most effective tools - this a symbolic dynamics, which allows you to explore a variety of complex phenomena in dynamical systems: chaos,
strange attractors, hyperbolic, structural stability, controllability, etc. In the author article [1] introduced by the finite T-alphabet for encoding shape of trajectories
(n) K+1
K
of {sk }K
k=1 in the space S K through the matching: {sk }k=0 {Tk |n }k=1 ,
sk( n)1
sk( n )
T7
sk( n)1
sk( n )
T4P
T3P
T1
T3N
sk( n1)
T5P
T2
(n)
T5N sk 1
T4N
sk( n )
sk( n)1
T6
T8P
T0
sk( n)1
T8N
64
Contributed talks
Central Bank of Russia [4]. The sample size of K = 3 545 counts. Data extraction
and processing was carried out in the program Wolfram Mathematica 9.
Among the results obtained. For specified 5-financial indicators are constructed
and analyzed their of symbolic TQ-images [1, 3]. The analysis revealed a difference
between the structure of the time series for the exchange rate USD/RUB and the
rest currency pairs. Over all possible combinations (v) the specified 5-currency
pairs, the analysis of T-synchronization [2, 3]. For some combinations found nonrandom higher values of integral level of synchronicity and the entropy of
SD spectra of synchronous
the structure of synchronous domains SD
E . In the H
SD
domains, detected long periods L
of synchronized exchange rate fluctuations.
For all relevant periods obtained their real date (for linking the to external events).
1
1 2 3 4 5 6 7 8 9 10 11121314151617181920 2122232425 26
USD
EUR
JPH
CHF
GBP
-1
10
-2
10
-3
10
10
15
20
25
0.8
v
0.6
H SD
10
0.4
0.2
0.0
15
20
400
20
SD
E
26
1
10
20
30 LSD 39
day
v 25
It is further planned multiscale CTQ-analysis of these financial indicators for
investigation of their temporal structure in order to study the mechanism and
causes of synchronicity.
10
15
20
References
[1] Makarenko A.V. Multidimensional dynamic processes studied by symbolic analysis in
velocity-curvature space // Computational Mathematics and Mathematical Physics,
2012, 52 (7), 1017-1028.
[2] Makarenko A.V. Measure of synchronism of multidimensional chaotic sequences based
on their symbolic representation in a T-alphabet // Technical Physics Letters, 2012,
38 (9), 804-808, arXiv:1212.2724.
[3] Makarenko A.V. The possibilities by symbolic analysis in velocity-curvature space:
TQ-bifurcation, symmetry, synchronization // School-Seminar Interaction of Mathematics and Physics, Moscow, August 2230, 2012, Steklov Mathematical Institute.
http://www.rdcn.org/data/report/p69 Presentation.pdf
[4] Bank of Russia. Foreign Currency Market.
http://www.cbr.ru/eng/currency base/dynamics.aspx
Contributed talks
65
Cram
er-von Mises test for Gauss processes
Martynov G. V.
Institute for Information Transmission Problems of the Russian Academy of Sciences, Russia
National Research University Higher School of Economics, Russia
One of the problems in the theory of the goodness-of-fit tests is the problem to
test if an observed random process S(t) on [0, 1] is the Gauss process with zero mean
and a covariance function KS (t, ), t, [0, 1]. This problem arises in particular
in applications of financial mathematics, when the assumption that the process
under study is a Gaussian process, is not deemed sufficient reasonable. This test
should be based on n realisations S1 (t), S2 (t), ..., Sn (t), t [0, 1], of S(t). The process S(t) and its realisations are considered here as the elements of the separable
Hilbert space L2 ([0, 1]). We choose as a basis for L2 ([0, 1]) the orthonormal basis
formed by eigenfunctions g1 (t), g2 (t), ... of the covariance operator with the kernel
KS (t, ). The processes S(t) and Si (t) can be represented in the form of expansion
in the mentioned basis as s = (s1 , s2 , s3 , ...) and si = (si1 , si2 , si3 , ...), correspondingly. The vector s has independent components with normal distributions. It can
be transformed to the random vector T = (T1 , T2 , T3 , ...), T [0, 1] , with the independent components, having the uniform distribution on [0, 1]. The observations
si can be transformed similarly to the observations Ti = (Ti1 , Ti2 , Ti3 , ...) of the
uniform distribution on [0, 1] . It can be introduced a distribution function
F (t) = F (t1 , t2 , t3 ...) = P {T1 t1 1 , T1 t1 1 , T1 t1 1 ...} = t1 1 t2 2 t3 3 ... .
Here, i > 1 should tend sufficiently quickly toward zero. Correspondingly, the
empirical distribution function can be introduced as
2
Fn (t) = Fn (t1 , t2 , t3 ...) = (1/n) ]{Ti : Ti1 t1 1 , Ti2
t2 , ...}.
p
The empirical process n (t) = (n)(Fn (t)F (t)), t [0, 1] , weakly converges to
the Gaussian process in L2 (L2 [0, 1]). This process has the zero mean and covariance
function
Y
Y
ti i vii .
K(t, v) =
min(ti i , vii )
i=1
i=1
n Y
X
Y
1
n2 = n
ITi,j <tj
ti i dt.
n
[0,1]
Z
i=1 j=1
i=1
Limit distribution of the Cramer-von Mises statistic is calculated using the methods
described in the papers listed in the bibliography. The statistic n2 can be calculated
by the Monte-Carlo method. In turn, the distribution of the statistic was calculated
Email: [email protected]
66
Contributed talks
also using the Monte Carlo method. We will present the estimated quantiles of the
distribution n2 with i = 1/i3 . The integration was carried out over the cube
[0, 1]10 . The percent points are P {n2 0.90} = 0.34 and P {n2 0.95} = 0.45. It
can be noted that corresponding percent points for the classical univariate Cramervon Mises statistic are 0.35 and 0.46.
Described theory can be directly applied also to testing the distribution uniformity on the unit multidimensional cube.
References
[1] Deheuvels, P., Martynov, G. (2003) Karhunen-Lo`eve expansions for weighted Wiener
processes and Brownian bridges via Bessel functions., Progress in Probability,
Birkh
auser, Basel/Switzerland, 55, 57-93.
[2] Martynov, G. V., (1979) The omega square tests, Nauka, Moscow, 80pp., (in Russian).
[3] Martynov, G. V., (1992) Statistical tests based on empirical processes and related
questions, J. Soviet. Math, 61, 2195 - 2271.
[4] Krivjakova E. N., Martynov G. V., and Tjurin J. N. (1977) The distribution of the
omega square statistics in the multivariate case. Theory of Probability and its Applications., V. 22, N 2, 415420.
Contributed talks
67
The Asian option (or average value option) is a special type of option contract.
The Asian options payoff is determined by the average underlying price over some
pre-set period of time. There are several types of Asian options: fixed strike (also
known as the average rate) Asian option payout based on the average price of the
underlying asset, and the floating strike the so-called floating rate) Asian option
payout (based on the average price and the spot price of the underlying asset. Note,
that in almost all works about the Asian option value in the Black Scholes model
the fixed strike options are considered. In this work we will consider both fixed and
floating strike options. The results of this work summarize the results of papers
[7], [8], [12] and [13]. This talk is about applying PDE methods for Asian options
value determination. The main method in this work is the PDE approach: it is
known that the value of an option is the expected value from the discount payout
function, and this expected value is the solution of the parabolic PDE boundary
problem. Using the Laplace transform, and the theory of hypergeometric and special functions we determine the Green function of the original problem. The report
will show that the Laplace transform of the basic equation can be reduced to the
Whittaker equation (The Schrdinger equation with Morse potential). Based on
this fact, we construct the Green function as the infinite integral of the product
of the first and second kind Whittaker functions (Confluent hypergeometric function). Using the properties of special functions we find two more representations
of the Green function: the infinite integral of the first kind Macdonald function
(Modified Bessel function) and the double infinite integral of elementary functions.
Then, the option value is the convolution product of the Green function and the
payout function. A big number of works originating from Boyle and Emmanuel [1]
is devoted to the Asian option value problem in the Black-Scholes model. There are
several directions dealing with this issue: the analytical methods, the estimates of
the value, the Monte-Carlo simulations and the PDE numerical methods. The estimates of the interested value are covered by Turnbull and Wakeman [2], Milevsky
and Posner[3] and other authors. The numerical approaches are considered Kemna
and Vorst[4], Rogers and Shi [5] , Vecer [6] and other authors. The first work about
analytical methods in the Asian option value problem was Yors [7] work published
in 1992. In this work the value was found as the triple infinite integral. Later,
in 1993 Geman and Yor[8] determined the Laplace transform of value, which can
be inverted numerically. This technique is illustrated by Fu[10], Craddock[11] and
other authors. In 2000 Dufresne determined the value as the infinite series of the
Laguerre polynomials. Finally, Linetsky[13] using the spectral expansion methods
Email: [email protected]
68
Contributed talks
found the value as the infinite integral of the Whittaker functions, which was estimated through a series of the same functions. We should mention that the Asian
option value problem is partially covered by Yor and Matsumoto [14], [15].
References
[1] Boyle, P. P., D. Emanuel. 1980. Options on the general mean. Working paper, University
of British Columbia, Vancouver, British Columbia, Canada.
[2] Turnbull, S., L. Wakeman. 1992. A quick algorithm for pricing European average options. J. Financial Quant. Anal. 26 377 - 389.
[3] Milevsky, M. A., S. E. Posner. 1998. Asian options, the sum of lognormals, and the
reciprocal gamma distribution. J. Financial Quant.Anal. 33 409 - 422.
[4] Kemna, A., A. Vorst. 1990. A pricing method for options based on average values. J.
Banking Finance 14 113 - 129.
[5] Rogers, L. C. G., Z. Shi. 1995. The value of an Asian option. J. Appl.Probab. 32 1077
- 1088.
[6] Vecer, J. 2001. A new PDE approach for pricing arithmetic average Asian options. J.
Computational Finance 4(4) 105 - 113.
[7] Yor, M. 1992. On some exponential functionals of Brownian motion. Adv.Appl. Probab.
24 509 - 531.
[8] Geman, H., M. Yor. 1993. Bessel processes, Asian options and perpetuities. Math.
Finance 3 349 - 375.
[9] Eydeland, A., Geman, H. 1995. Domino effect. RISK 8(4) 65 - 67.
[10] Fu, M., D. Madan, T. Wang. 1998. Pricing Asian options: A comparison of analytical
and Monte Carlo methods. J. Computational Finance (2) 49 - 74.
[11] Craddock, M., D. Heath, E. Platen. 2000. Numerical inversion of Laplace transforms:
A survey of techniques with applications to derivatives pricing. J. Computational Finance 4(1) 57 - 81.
[12] Dufresne, D. 2000. Laguerre series for Asian and other options. Math. Finance 10 407
- 428.
[13] Linetsky V. 2004. Spectrac expansio for Asian (Average price) options Operations
Research 52, 6, 856 - 868.
[14] Yor M. Matsumoto H. Exponential functionals on brownian Motion I . Probability
Surveys, 2003, 2 ,312 - 347.
[15] Yor M. Matsumoto H. Exponential functionals on brownian Motion II . Probability
Surveys, 2005, 2 ,348 - 384.
Contributed talks
69
s, t 0.
It is well known that, for H (0, 1/2) (1/2, 1), such a process B H is neither a
semimartingale nor a Markov process (see, for example, [1]).
The problem of estimation of the parameters and H in the model (1) naturally
arises in a context of modeling of stock prices [2]. In this note we suppose that the
Hurst index H is known and is a random variable which is independent of B H
(we suppose that E || < ). Thus, we follow the Bayesian approach.
2. The problem of a sequential estimation of the drift in such a setting was
studied in [2]. Here we consider the problem of testing the hypothesis H1 , . . . , Hn ,
Hi : Ai , by a sequential observation of P
X. The sets A1 , . . . , An R are
supposed to be nonintersecting and such that i P ( Ai ) = 1.
Each testing procedure is represented by a decision rule = (, d ) which consists of a stopping time of the filtration (FtX )t0 , FtX = (Xs ; s t), and an
FX -measurable function d taking values 1, . . . , n. The time corresponds to
the moment of stopping the observation, and the value of d to the hypothesis
accepted.
With each
decision
rule
(,
d
)
we
associate
the
loss
function
R
()
=
E
c +
X
W (, d ) , where first component is the observation cost (proportional to the observation time) and second component is the (nonegative) penalty for making a
wrong decision. The problem consists in finding the decision rule = ( , d )
with the minimal average loss, i. e. such that
RX ( ) = inf RX (),
(2)
where the infimum is taken over all decision rules = (, d) with E < . In our
talk we present a general way to tackle such kind of problems.
Email: [email protected]
70
Contributed talks
with
cH =
(3 2H)
2H(3/2 H)3 (1/2 + H)
1/2
is a sufficient statistic in (2). In the case = 0 the process M (X) is the well-known
Molchan (or fundamental) martingale (see [3], [1]), but for 6= 0 it appears to be
a diffusion.
Next, we transform (2) to the similar problem but for standard Brownian motion
and nonlinear cost of observation. Finally, we reduce our task to the standard optimal stopping problem for Brownian motion. To investigate it one can use methods
from general theory (see, for example, [4], [5]).
Details and particular examples will be provided.
Acknowledgements. The research was supported by the Laboratory for Structural Methods of Data Analysis in Predictive Modeling (RF government grant,
No 11.G34.31.0073), the Russian Foundation for Basic Research (grant No 12-0131449-mol a) and the Ministry of Education and Science (ag. 14.740.11.1144).
References
[1] Yu. Mishura, Stochastic calculus for fractional Brownian motion and related processes,
Lecture Notes in Math., 1929, Springer, Berlin, 2008.
[2] U. C
etin, A. Novikov, A. N. Shiryaev, Bayesian sequential estimation of a drift of
fractional Brownian motion, to appear in Sequential Analysis, 2013.
[3] G. Molchan, Gaussian processes with spectra which are asymptotically equivalent to a
power of , Theory Prob. Appl. 14, 1969.
[4] Peskir G., Shiryaev A. Optimal stopping and free-boundary problems, Birkhauser, Basel,
2006.
[5] P. van Moerbeke, On optimal stopping and free-boundary problems. Arch. Rational
Mech. Anal., 60, 1976.
Contributed talks
71
The classical Reflection Principle is a technique that allows one to express the
joint distribution of a Brownian motion and its running maximum through the
distribution of the process itself. It relies on the specific symmetry and continuity
properties of a Brownian motion and, therefore, cannot be directly applied to an
arbitrary Markov process.
We show that, in fact, there exists a weak formulation of this method that allows
us to recover the same results on the joint distribution of a Brownian motion and
its running maximum. We call this method a Weak Reflection Principle and show
that it can be extended to a large class of Markov processes, which do not posses
any symmetry properties and are allowed to have jumps.
We demonstrate various applications of this technique in Finance, Computational Methods, Physics, and Biology. In particular, we show that the Weak Reflection Principle provides an exact solution to the problem of hedging Barrier
options with a semi-static position in European type claims. Our method allows us
to find such hedging strategies in the diffusion- and Levy-based models. In addition, we show how it can be used to establish robust static hedging strategies that
are model-independent. We illustrate the theory with numerical examples.
Email: [email protected]
72
Contributed talks
Contributed talks
73
Exponential functionals of L
evy processes
Vladimir Panov
University of Duisburg-Essen (Germany) and Premolab (MIPT, Russia)
(1)
and its terminal value A = limt At , where s is a Levy process. The integral A naturally arises in a wide variety of financial applications as a stationary
distribution of the generalized Ornstein-Uhlenbeck process,
Z t
s
t
e ds ,
V0 +
Vt = e
0
(2)
where Ts is a pure jump process with Levy measure . The aim of this study is to
statistically estimate the Levy triplet of s given by the observations of the integral
A . The algorithm presented below is based on the following recursive formula for
the moments of A , which is completely proved in [4]:
(s)
E As1
=
E [As ] ,
s
where s C is such that 0 < <(s) < sup z 0 : E[ez1 ] 1 , and (s)
Laplace exponent of the process , i.e., (s) := log E es1 . In particular
= 0, the Laplace exponent can be represented as
Z +
Z +
vx
(u + v) = c (u + v)
e
(dx) +
(dx),
u, v R,
0
(3)
is a
case
(4)
where (dx) = eux (dx). The formula (4) motivates the algorithm, which we
describe below.
In what follows, we suppose that N observations Y1 , ..., YN of the integral A
are given.
b [As ] =
1. Estimate
the As for s = u + v, where u is fixed and v varies by E
PN
s /N.
Y
i
i=1
Email: [email protected]
74
Contributed talks
=sE
b As1 /E
b [As ] . This estimate is based on (3).
2. Estimate (s) by (s)
R +
0
(5)
(dx) by
c<(s) .
a
:= mean <(s)
(6)
(7)
6. Estimate by
(x) =
1 ux
e
2
(8)
Contributed talks
75
76
Contributed talks
Contributed talks
77
78
Contributed talks
difference to the put written on the modified price process can then be hedged
dynamically, and we provide a representation of the resulting hedge portfolio.
Starting from the general representation, we then derive the dynamic hedging portfolios by Malliavin calculus, in particular the Clark-Ocone formula. In a
stochastic volatility context, this necessarily involves higher greeks. Such an approach has been pioneered for European options in the Heston model in [1]. Here we
adapt this approach to our specific situation, i.e. hedging of a time-dependent put
option written on the modified price process under the measure R, and generalise
it to our general stochastic volatility framework.
This is joint work with Elisa Al
os and Zhanyu Chen.
References
[1] E. Al`
os, A decomposition formula for option prices in the Heston model and applications
to option pricing approximation. Finance and Stochastics, 16 (2012), pp. 403-422
[2] P. Carr, and A. Chou, Breaking barriers, Risk 10 (1997), pp. 139-145
[3] P. Carr, and R. Lee, Put-call symmetry: extensions and applications, Mathematical
Finance, 19 (2009), pp. 523-560
[4] I. Molchanov, and M. Schmutz, Multivariate extension of put-call symmetry. SIAM
Journal of Financial Mathematics, 1 (2010), pp. 398-426
[5] T. Rheinl
ander, and M. Schmutz, Self-dual continuous processes. Stochastic Processes
and their Applications, 123 (2013), pp. 17651779
[6] T. Rheinl
ander, and M. Schmutz, Quasi self-dual exponential Levy-processes. ArXiv:
1201.5132v1 (2012)
[7] M. Tehranchi, Symmetric martingales and symmetric smiles. Stochastic Processes and
their Applications, 119 (2009), pp. 37853797
Contributed talks
79
T
X
(S t (t ) S t (t )+ ), 1 = 0, T = 0; t := t t1 .
t=0
Here S t S t are the bid and ask prices of a risky asset (stock) and t is the
number of stock units in traders portfolio at time moment t, x+ = max{x, 0},
x = max{x, 0}. Note that for the frictionless model
PT (S = S = S) this formula
shapes to the customary form: 1 + ( S)T := 1 + t=1 t1 St .
All processes are adapted to a filtration (Ft )Tt=1 , F1 = {, }, FT = F on a
probabilty space (, F, P). It is assumed that S t , S t Lq (Ft ) for some q [1, ]
1
Y, where
and t L (Ft ). We allow portfolio constraints of the form (t )Tt=0
QT 1
Y is a convex subset of t=0 L (Ft ).
A functional : Lq (FT ) 7 [, ] is called monotone if (X) (Y ) whenever X Y , X, Y Lq (FT ) and quasiconcave if
(1 X + 2 Y ) min{(X), (Y )}
for all X, Y Lq (FT ), 1 + 2 = 1, i 0. We admit that traders utility is
represented by a monotone quasiconcave functional .
Following [4] we call an adapted process S [S, S] a shadow price if
S := sup{(1 + ( S)T ) : Y} = := sup{(XT ()) : Y}.
Thus, it is impossible to outperform the optimal utility value related to the
market with friction by trading at a frictionless shadow price. So, a shadow price
can be interpreted as a least favorable frictionless price from traders point of view.
2. The existence of a shadow price for the case of finite (and a more traditional
utility functional) was proved in [4]. In general a shadow price need not exist: see
examples given in [1], [2], [5]. However, it was shown in [5] that the existence of a
generalized shadow price process S is guaranteed under rather weak assumptions.
This process corresponds to the relaxed utility functional.
To be precise, put
(S, ) = (1 + ( S)T )
and t = (Lq (Ft ), Lp (Ft )), where 1/p + 1/q = 1. So, t is the weak topology of
Lq for q [1, ) and the weak-star topology of L . In any case the set
[S t , S t ] = {St Lq (Ft ) : S t St S t }
Email: [email protected]
80
Contributed talks
Q
is t -compact. Consider the vector space Tt=0 Lq (Ft ) with the product topology
b ) the -lower semicontinuous envelope (relaxation) of (, )
and denote by (,
Q
as a function on [S, S] = Tt=0 [S t , S t ]:
b
(S,
) =
sup
inf (S 0 , ).
0
V N (S) S V
Here N (S) is the neighbourhood base at S of the topology , restricted to [S, S].
b ) is the largest -lower semicontinuous function majorized by
As is known, (,
(, ).
b
Consider the optimization problem for the relaxed functional :
b
bS = sup{(S,
) : Y}.
We call S a generalized shadow price if
bS = .
Theorem 1. Let be monotone and quasiconcave, S t , S t Lq (Ft ), t = 0, . . . , T
for some q [1, ]. Then there exists a generalized shadow price S [S, S] and
the following minimax relations hold true:
b
= sup (XT ()) = sup inf (S,
) = sup inf (S, )
Y S[S,S]
inf
Y S[S,S]
b
b , ) =
sup (S,
) = sup (S
bS .
S[S,S] Y
Theorem 1 was deduced in [5] from the intersection theorem of [3] (Theorem 3).
If there exists an optimal solution of the utility maximization problem for the
market with friction, then a pair ( , S ), where S is a generalized shadow price,
b
is exactly a saddle point of the relaxed utility function (S,
) [5] (Theorem 2.3).
If the original utility function (S, ) is already lower semicontinuous in S in
an appropriate topology, Theorem 1 implies the existence of a shadow price. We
b but a generalized shadow price S is in fact a
present an example where 6= ,
shadow price, and examples, where a generalized shadow price exists and a shadow
price does not. In addition, we discuss the connection between shadow prices and
duality theory.
References
[1] Benedetti G., Campi L., Kallsen J., Muhle-Karbe J. On the existence of shadow prices.
Finance Stoch. (2013, to appear), doi:10.1007/s00780-012-0201-4.
[2] Czichowsky C., Muhle-Karbe J., Schachermayer W. Transaction costs, shadow prices,
and connections to duality. Preprint (2012), arXiv:1205.4643 [q-fin.PM], 20 pages.
[3] Ha C.-W. Minimax and fixed point theorems. Math. Ann. 248, 7377 (1980).
[4] Kallsen J., Muhle-Karbe J. Existence of shadow prices in finite probability spaces.
Math. Meth. Oper. Res. 73, 251262 (2011).
[5] Rokhlin D.B. On the game interpretation of a shadow price process in utility
maximization problems under transaction costs. Finance Stoch. (2013, to appear),
doi:10.1007/s00780-013-0206-7.
Contributed talks
81
Rui Sa Pereira
Centro de Matem
atica da Universidade do Porto, Portugal
Email: [email protected]
82
Contributed talks
Alexander Slastnikov
There is a project that, in a certain period after investing (lag of fixed assets)
can bring some (random) income flow. The project is risky, i.e. with probability
q, 0 < q < 1 it fails (remains unrealized) and investor defaults, and with probability
1 q it starts acting.
It is assumed that the required investment I are credited by bank on certain
terms (loan agreement). These terms include loan amount and duration, repayment
of the loan principal, and interest. These elements make up the credit policy K of
bank (in relation to given investment project).
If the project became operational, the repayment of loan and accrued interest
on it starts after the lag, but in case of investors default the loan is not returned.
This in turn leads to higher interest rates.
In order to attract investment in risky projects, there is a mechanism of government guarantees. This means that if the project fails and investor does not return
the credit, the bank receives compensation from the Government (reimbursement
of credit) as a part T of issued loan. This mechanism allows, in particular, to
reduce the interest rate on the loan.
The tax system is represented in the model by coefficient of the tax burden,
which is a share of tax payments in profits.
The aim of the investor is to solve an investment timing problem, i.e. on the
base of the observed information on current market prices and the forecast of the
future flow of profits from the project to choose the investment time in such a way
that the expected net income value of the project (NPV), discounted to zero (base)
time will be maximal.
This optimal time of investment , under certain assumptions about the process
of profits t (see [1, 2]) is specified by a threshold that depends on the credit policy
of the bank.
Knowing the dependence of investor behavior on credit terms, the bank chooses
the credit policy, maximizing the expected discounted profits from the project,
equal to the difference between the expected return on the loan (from the investor
and the state) and the amount of the loan. This optimal credit policy is a function
of the part of guaranteed reimbursement of credit.
Knowing the optimal credit policy of the bank and the corresponding optimal
investor behavior as a function of the part of the reimbursement, the government
determines the optimal part of a loan reimbursement so that appropriate budgetary
effect will be maximal. We define the budgetary effect as the difference between the
expected tax revenue from the project and the expected costs of the government
on reimbursement of the bank loans (at the optimal behavior of the investor).
Email: [email protected]
Contributed talks
83
So, an optimal investment time, optimal credit policy and optimal government
guarantees can be viewed as Stackelberg equilibrium point in a three-player game.
Assuming that the profit flow is described by a geometric Brownian motion,
we can obtain the explicit form of the optimal credit policy of the bank and the
optimal part of the guaranteed reimbursement of credit.
In this case it is shown that the optimal part of the reimbursement is inversely
proportional to the risk q and increases in the coefficient of tax burden. As for
the dependence of the optimal part of the reimbursement on the volatility of the
project, it is determined by the value of the tax burden. For small values of the tax
burden (not exceeding a certain level), the optimal part decreases in volatility, and
at higher values (above this level) increases. It is also shown that the optimal (from
the banks point of view) interest on the loan decreases linearly on the part of the
reimbursement, and under the optimal part increases both in risk and volatility of
the project, and decreases in tax burden.
References
[1] V.I. Arkin, A.D. Slastnikov. A Variational Approach to an Optimal Stopping Problems
for Diffusion Processes. Theory of Probability and Its Applications, 53 (3), 467-480,
2009.
[2] V.I. Arkin, A.D. Slastnikov. Threshold Stopping Rules for Diffusion Processes and
Stefans Problem. Doklady Mathematics, 86 (2), 626-629, 2012.
84
Contributed talks
V. A. Balash1
A. R. Faizliev1
E. Korobov1
P. Date2
Brunel University, UK
The work presented here tries to evaluate the impact of news on stock volatility
through a small empirical study on augmented GARCHJumps models. While
news analytics tools became more popular among investors as indicated in [1],
there are not so much research works studying quantitative impact of news on
stock volatility. It is worth to be mentioned the pioneering works [3] and [2]. In
the paper of [3] firm-specific announcements were used as a proxy for information
flows. It was shown that there exists a positive and significant impact of the arrival
rate of the selected news variable on the conditional variance of stock returns on
the Australian Stock Exchange in a GARCH framework. They split all their press
releases into different categories according to their subject. In the second of the
papers the author examines impact of news releases on index volatility. In the
paper [4] was shown that the GARCH(1,1) model augmented with volume does
remove GARCH and ARCH effects for the most of the FTSE100 companies, while
the GARCH(1,1) model augmented with news intensity has difficulties in removing
the impact of log return on volatility.
Based on empirical evidences for some of FTSE100 companies, it will be examined two GARCH models with jumps. First we consider the well-known GARCH
model with jumps proposed in [5]. Then we introduced the GARCH-Jumps model
augmented with news intensity and obtained some empirical results. The main
assumption of the model is that jump intensity might change over time and that
jump intensity depends linearly on the number of news. It is not clear whether
news adds any value to a jump-GARCH model. However, the comparison of the
values of log likelihood shows that the GARCH-Jumps model augmented with news
intensity performs slightly better than pure GARCH or the GARCH model with
Jumps.
Acknowledgements. The work was supported by RFBR, grant 13-01-00175.
References
[1] Mitra G., Mitra L., editors. (2011). The Handbook of News Analytics in Finance. John
Wiley & Sons.
[2] Janssen G. (2004). Public information arrival and volatility persistence in financial
markets. The European Journal of Finance. Vol. 10, pp. 177197.
[3] Kalev P. S., Liu W.-M., Pham P. K., Jarnecic E. (2004). Public information arrival and
volatility of intraday stock returns. Journal of banking and Finance. Vol. 28 (6), pp.
14471467.
[4] Sidorov S., Date P., Balash V. (2013). Using news analytics data in GARCH models.
Applied Econometrics. Publishing House SINERGIA PRESS. Vol. 29 (1), pp. 82-96.
[5] Maheu J. M., McCurdy T. H. (2004). News arrival, jump dynamics, and volatility
components for individual stock returns. Journal of Finance. Vol. 59 (2), pp. 755793.
Email: [email protected]
Contributed talks
85
Olessia Vasilyeva2
D
usseldorf, Germany
(1)
where t is current time 0 < t < T , S(t) is asset price 0 < S < , T is expiry
date, r is the interest rate, is volatility of the market prices. The inhomogeneity
is given by
r K, if 0 < S S (t)
.
f = f (S, t) =
0 if S > S (t)
The final time condition is
P (S, T ) = (S)
and the free boundary is determined by
P (S , t) = K S ,
P
|S=S = 1.
S
We take the Mellin transform and for the price P (S, t) they get
Z Z
rK T c+i v (S (x))v 1 2 h(v)(xt)
P (S, t) = p(S, t) +
S
e2
dv dx.
2i t ci
v
(2)
Substituting S = S (t) they get the following integral equation for the free boundary
Z Z
rK T c+i 1 S (t) v 1 2 h(v)(xt)
86
Contributed talks
Then we use the Newtons method for the calculation of free boundary S ( ).
The Figure 1 shows the isolines of the function of the American put option as the
smooth conjunction of the family of the strict vertical isolines with the curve linear
isolines.
Figure 1: Isolines of the space for the function p(s, ) with different choices for
sigma option with K = 45; = 0, 3; T = 0, 5833
Contributed talks
87
We consider a model of stock prices which initially increase but start to decrease
after an unknown moment of time. The aim is to detect this moment and to
sell the stock maximizing the gain. Applying methods of the theory of quickest
changepoint detection, we show that the optimal detection rule consists in observing
the posterior probability process (or, equivalently, the ShiryaevRoberts statistics),
and selling the stock as soon as this process exceeds a time-dependent threshold.
We apply the result to real market data and show that it provides good performance.
The research was supported by the Laboratory for Structural Methods of Data
Analysis in Predictive Modeling (RF government grant, No 11.G34.31.0073).
This is a joint work with A. N. Shiryaev and W. T. Ziemba.
Email: [email protected]
Posters
Posters
91
Danilova N. V.
21
t + 1 Wt , t [0, ],
S0 exp r1
2
St =
2
M exp r2 2 (t ) + 2 Wt , t (, T ],
2
and
Bt =
B0 er1 t , t [0, ],
B er2 (t ) , t (, T ].
Theorem 1 (see [2]). The fair price of the European call option fT = max(ST
K, 0) is C = E C( ) where C( ) given by
d( ) + 2 ( )
B0
d( )
C( ) = S0
K
( )
BT
( )
with
d( ) = ln
K
S0
2
r1 1
2
Email: [email protected]
q
22
r2
(T ) , ( ) = 21 + 22 (T ).
2
92
Posters
Z x
t2
1
e 2 dt is the distribution function of the standard normal
Here (x) =
2
law and E denotes the expectation with respect to martingale measure P .
Theorem 2 (see [2]). Let c > S0 and r1 21 /2d > 0. Then has the distribution:
g(x),
0 x < T,
Z T
p (x) = 1
g(x)dx, x = T,
0,
x > T,
where
r
g(x) =
and
1
a=
1
b ab 1
b
1
e
ax +
3 exp
2
2
x
x2
21
r1
d ,
2
1
b=
ln
1
c
S0
.
So,
T
C(x)g(x)dx + C(T ) 1
E C( ) =
0
g(x)dx .
exp
2 ln
ln
0tT
c
S0
r1
21
21
2
2
r1 21 d T
1 T
2
d
ln Sc0 r1 21 d T
1 T
c
S0
At adequate selection of parameters of model the fair price will be less BlackScholes price.
References
[1] Shiryaev A.N. The bases of stochastic mathematical finance// M.:FASIS, 1998.
T.1.Facts,models.512 p. T.2.Theory.544 p.
[2] Belyavsky G.I., Danilova N.V. The diffusion models with stochastic switching of parameters. Calculations and financial applications// Lambert Academic Publishing, 2012.
122 p.
Posters
93
In this work we study the problem of utility maximization for a financial investor.
We consider a model of financial market, which consists of one asset and has a finite
maturity T . The price of the asset is modeled with a semimartingale S. The capital
of the investors portfolio has the form X = x + H S, where constant x denotes the
initial wealth and H is a predictable S-integrable process, which specifies amount
of the asset held in the portfolio. We assume that all capitals are strictly positive
X (x) = {Xt > 0 : X0 = x}. The investor has a concave utility function U . His
main aim is to maximize the expected terminal utility EU (XT ), and we can denote
the value function of this problem
u(x) = sup E[U (XT )].
XX (x)
We can also consider the dual problem here. Let V (y) = supx>0 (U (x)xy), y > 0.
Then the value function of the dual problem is
v(y) =
inf
Y Y(y)
E V (YT ).
Utility maximization and the dual problem were considered by Kramkov and
Schachermayer in [3]. They indicated several properties for the solutions of these
problems and connections between them. Some relations were also found for functions u and v. This research was made on general assumptions. In our work
we show that in particular cases an explicit form and additional properties can
be found for the solutions. To make this possible the process S is assumed to
be a stochastic exponential of a Levy process L, L > 1, which is completely
determined by its triplet (b, c, ). Besides, we consider only the following utility
functions U (x) = ln x and U (x) = xp /p. Kallsen [4] showed an explicit form for
the solutions under several conditions, one of them de facto required the solution
of the dual problem to be an EMM. In our previous works [1, 2] we solved this
problem for the logarithmic utility for all possible triplets when L is not monotone,
which is equivalent to the absence of arbitrage and can be stated in terms of the
Levy triplet, see [1]. No other restrictions were placed. The solution of the primary
problem has the form E(L), where is a constant. There are only three options
for the solution of the dual problem and they can be characterized in terms of the
triplet:
1. Y is the density process of an equivalent martingale measure.
Email: m y [email protected]
94
Posters
Posters
95
Dmitry B. Rokhlin
1. We consider a firm whose capital X is affected by an exogenous (macroeconomic) random factor Y , and assume that (X, Y ) evolve according to the system
of stochastic differential equations
dXt = ((Xt , Yt , t ) ct )dt + (Xt , Yt , t )dWt ,
p
ct ,
dYt =
b(Yt )dt +
b(Yt ) dWt + 1 2 dW
X0 = x,
Y0 = y,
1 2
= 0, x > 0,
(1)
(2)
96
Posters
More precisely, with the use of Bellmans optimality principle it can be shown that
v is a viscosity solution of the above problem.
Denote by v the upper semicontinuous envelope of v. Under the assumption
inf {1,...,n} 2 (0, y, ) > 0 with the use of methods of [2] (Proposition 1.1) we
deduce that v (0, y) = 0. Based on this fact and following the argumentation of
[9] (Theorem 6.21), we prove a comparison theorem, which leads to the following
result.
Theorem 1. Assume that inf {1,...,n} 2 (0, y, ) > 0. Then the value function v
is a unique continuous viscosity solution of (1), (2). If, moreover, for any bounded
domain G and any {1, . . . , n} there exists a constant > 0 such that
2 (x, y, )12 + 2(x, y, )b
(y)1 2 +
b2 (y)22 (12 + 22 ),
(x, y) G, R2 ,
Posters
97
ij xi xj min,
i=1 j=1
subject to
N
X
i xi = ,
i=1
N
X
xi = 1, li i xi ui i , i = 1, . . . , N,
i=1
N
X
i = K, i {0, 1}, i = 1, . . . , N.
i=1
98
Posters
[2] Chang T.-J., Yang S.-C., Chang K.-J. (2009). Portfolio optimization problems in different risk measures using genetic algorithm. Expert Systems with Applications. Vol.
36, pp. 1052910537.
[3] Woodside-Oriakhi M., Lucas C., Beasley J.E. (2011). Heuristic algorithms for the cardinality constrained efficient frontier. European Journal of Operational Research, Vol.
213 (3), pp. 538550.
[4] Markowitz H. (1952). Portfolio selection. Journal of Finance. Vol. 8, pp. 7791.
[5] Storn R., Price K. (1997). Differential evolution A simple and efficient heuristic for
global optimization over continuous spaces. Journal of Global Optimization. Vol. 11
(4), pp. 341359.
[6] Changshou D., Thomas W., Bingyan Z. (2012). Pseudo Binary Differential Evolution
Algorithm. Journal of Computational Information Systems. Vol. 8, pp. 24252436.
Posters
99
Yerkin Kitapbayev
Swing options are particular financial derivatives that may be described as American options with multiple exercises. They are widely traded in the energy markets
and usually the options underlying asset is the price of a given commodity. In
mathematical terms the price of an option of this kind is described by the value
function of an optimal stopping problem with multiple stopping times.
We assume that the dynamics of the price is a geometric Brownian motion and
study the Swing put option on finite time horizon T > 0 with a strike K > 0
and two exercise rights. An important parameter of this problem is the so-called
refraction period > 0. It can be interpreted as the minimal period that the
options seller needs to deliver a new portion of asset. If the holder exercises its
first right, then the second optimal exercise time is an optimal stopping time for
a standard American put option. Therefore the double optimal stopping problem
reduces to a single optimal stopping problem.
Using the local time-space calculus [5] we derive a closed form expression for the
value function in terms of the optimal stopping boundaries for both exercise rights
and show that the optimal stopping boundaries themselves can be characterised as
the unique solution of nonlinear integral equations. These integral equations are
then evaluated numerically.
References
[1] Bender, C. (2013). Dual pricing of multi-exercise options under volume constraints.
Finance and Stoch. 15 (126).
[2] Carmona, R. and Dayanik, N. (2008). Optimal multiple stopping of linear diffusions.
Math. of Operations Research. 33 (446460).
[3] Carmona, R. and Touzi, N. (2008). Optimal multiple stopping and valuation of
Swing options. Math. Finance 18 (239268).
[4] Kobylanski, M., Quenez, M. and Rouy-Mironescu, E. (2011). Optimal multiple
stopping time problem. Annals of Appl. Prob. 21 (1365-1399).
[5] Peskir, G. (2005). A change-of-variable formula with local time on curves. J. Theoret.
Probab. 18 (499535).
Email: [email protected]
100
Posters
Poissonian Stochastic Index process (PSI - process ) is defined by a subordination of the random sequence (n ), n = 0, 1, . . . , to an independent Poissonian
process (s), s 0, by the following way (s) = (s) . We mainly focus on the
case when (n ) consists of i.i.d. random variables. We consider (j (s)), j = 1, 2, . . .
the sequence of independent copies of the process .
Basic Theorem. Let (n ), n = 0, 1, . . ., be i.i.d. rvs from the domain of attraction to a symmetrical -stable law with the characteristic function Eexp{it0 }
= exp{|t| }, t R, (0, 2].
Then the following convergence of the finite dimensional distributions takes place
as N , s 0,
N (s)
N
X
1
N 1/
j (s) U (s),
s 0.
(1)
j=1
In the particular case of = 2 the right part in (1) is the standard Gaussian
Ornstein-Uhlenbeck process with the viscosity coefficient > 0 which is explicitly
equal to an intensity coefficient of the leading Poisson process . Constructive
description of the limit of (1) for arbitrary (0, 2] is based on the following embedding one-dimensional time s into the two-dimensional euclidian plane
(R.L.Wolpert, M.S.Taqqu (2005)). Such kind embedding Wolpert and Taqqu call
as Upstairs Representation.
Let consider the Wiener-Chentsov random field on B = [0, 1] (, +) 3
(t, s): it is the symmetrical -stable, (0, 2], white noise dZ (t, s) with the
scattering Lebesgue measure. Assume that Z (t, s) is normalized by such a way
that for all A B : |A| = 1,
Z
d
dZ (t, s)=L (1) ,
A
where L (v), v [0, 1], is the symmetrical -stable Levi process. For simplicity
here we assume that the Poissonian intensity = 1. Since the upstairs representation is a case of the moving average representation, so we have to introduce the
following 2-dimensional moving kernel, s 0,
n
o
As = (v, r) : r s; v e(sr) ; (v, r) B .
(2)
Email: [email protected]
Posters
101
dZ (v, u) .
(3)
U (s) =
As
102
Posters
1. Copula models widely use for modeling dependencies between random variables. Their applications include such fields as analysis of financial data and actuarial calculations. Copula function may be obtained by inversion from known joint
distribution[1]. In this work we find the joint distribution of Grubbs test statistics
and obtain a copula function from this bivariate distribution.
2. Let X1 , X2 , . . . , Xn1 , Xn be a random sample from a normal N (a, 2 ) distribution. Grubbs proposed the standardized maximum and minimum [2]:
Tn(1) = (X min {Xi })/S,
1in
1in
where X =
1
n
Pn
i=1 Xi
and S 2 =
(1)
1
n1
Pn
i=1 (Xi
X)2 .
(1)
(1)
0, t 1n , n 2;
Rt (1)
n
Fn1 (gn (x))fTn (x)dx,
< t) =
1, t > n1
, n 2;
n
1
n
<t
n1
,
n
n 3;
n4
2
n1
n2
n
2
/
1
x
;
2
2
(n 1)2
s
. n1
n
n
2
x
1
x .
gn (x) =
n1
n2
(n 1)2
1
where fTn (x) =
n1
(1)
(1)
(2)
Posters
103
F (t ), t1 n1
n
n 2
(1)
n1
Fn (t1 ), t2 n ;
Rt2
n (t1 , t2 ) =
n
n1 (n (t1 , x), gn (x)) fTn (x)dx, (t1 , t2 ) n ;
, t2 < n1
,
0, (t1 , t2 )
/ n , t1 < n1
n
n
where n (u, v) = (u +
.r
v
n1 )
n1
n2
n
v2
(n1)2
, |v| <
(4)
n1
;
n
functions gn (x) and fTn (x) may be calculated with using (2) and (1);
104
Posters
We consider the portfolio selection problem over an infinite time-horizon involving investors time preference. The issue of stochastic optimality is also discussed.
Consider a self-financed portfolio consisting of n assets. The price vector of risky
assets follows a multidimensional geometric Brownian motion with parameters
(i)
(i)
(t , t ), i = 1, n 1 and rt is the return on the riskfree (the nth) asset. Denote
by Xt the investors total wealth at time t, starting with initial wealth x > 0 and
ignoring the transaction costs. We assume (see, e.g., [1]) that Xt , t 0 is an
Rvalued stochastic process defined on a complete probability space {, F, P} by
dXt = rt Xt dt + Bt Ut dt + Ut0 Ct dwt , X0 = x ,
(i)
(i)
where Bt = t rt , i = 1, n 1 , Ct = diag(t ), i = 1, n 1 ;
(1)
(i)
{wt }t=0 is an n 1-dimensional standard Wiener process; Ut , t 0 is an admissible control, i.e. an Ft = {ws , s t}adapted n 1-dimensional process such
(i)
that there exists a solution to (1). Note that Ut defines the amount invested
in the ith risky asset. Let us denote by U the set of admissible controls. We
R
dont impose any specific restrictions on Ut such as EkUt k2 dt < , mean-square
0
stability of the corresponding process Xt , etc., which are common for portfolio
selection problems over the infinite-time horizon, see e.g. [2].
It is desired to determine the investment strategy Ut for tracking some investordefined reference portfolio Vt0 . The reference portfolio is riskless and described by
dVt0 = t Vt0 dt , V00 = x . Notice that its natural to assume t > rt . The quadratic
objective functional, see [3], measures the total loss from deviations of the actual
portfolio and also the costs of portfolio control:
T
JT (U ) =
(2)
Email: [email protected]
U U
(3)
Posters
105
(4)
s
R
1
0
the following
Theorem 1. Assume D and P. Then the solution to problem (5) is given by
p
1 Bt0 (t Xt t Vt0 + mt / ft ) ,
Ut = R
(5)
t
the optimal portfolio process is governed by
1 Bt0 t )Xt dt + Bt R
1 Bt0 lt dt + (lt t Xt )Bt R
1 Ct dwt , (6)
dXt = (rt Bt R
t
t
t
Author index
Ahlip R., 50
Akhlaque Ahmad, 49
Amini H., 51
Angelis T., 99
Anthropelos M., 52
Arkin V., 82
Balash V. A., 84
Belomestny D., 15
Belyavsky G. I., 91
Cadenillas A., 54
Choulli T., 36
Cuchiero C., 55
Danilova N. V., 91
Date P., 84
Dempster M. A. H., 16
Dolinsky Y., 37
Eberlein E., 17
Faizliev A. R., 84
Filipovic D., 51
Gerhold S., 57
Glasserman P., 20
Goncalves F. F., 21
Grossinho M. R., 21
Guillaume F., 58
Gushchin A. A., 60
Hinz J., 25
Homchenko A., 97
Huam
an R., 54
Ivanov M. Y., 93
Kabanov Yu., 26
Kardaras C., 27
Khovansky S., 62
Kijima M., 29
Kitapbayev Y., 99
Kordzakhia N., 35
Korobov E., 84
Kramkov D., 30
Kupper M., 52
Lepinette E., 26
Makarenko A. V., 63
Markov M., 31
Martynov G., 65
Minca A., 51
Mironenko G., 95
Mordecki E., 32
Muravey D., 67
Muravlev A., 69
Nadtochiy S., 71
Nishide K., 38
Nouri B., 20
Novikov A., 35
Onalan O., 72
Palamarchuk E., 104
Panov V., 73
Papapantoleon A., 52
Pereira R. S., 81
Rheinlaender D., 75
Rheinlaender T., 77
Rokhlin D. B., 79, 95
Rusakov O. V., 100
Rutkowski M., 50
Schoutens W., 58
Schweizer M., 36
Shamarova E., 81
Shiryaeva L. K., 102
Sidorov S., 84, 97
Slastnikov A., 82
Soner H. M., 37
Suzuki T., 38
Teichmann J., 55
Tian Y., 29
Vasilyeva O., 85
Vasilyeva T., 85
Vostrikova L., 41
Zhitlukhin M. V., 87
Zhylyevskyy O., 62
Ziemba W. T., 45