Past Mathematical and Computational Finance Internal Seminar

In recent years, limit order books have been adopted as the pricing mechanism in more than half of the world's financial markets. Thanks to recent technological advances, traders around the globe also now have real-time access to limit order book trading platforms and can develop trading strategies that make use of this "ultimate microscopic level of description". In this talk I will briefly describe the limit order book trade-matching mechanism, and explain how the extra flexibility it provides has vastly impacted the problem of how a market participant should optimally behave in a given set of circumstances. I will then discuss the findings from my recent statistical analysis of real limit order book data for spot trades of 3 highly liquid currency pairs (namely, EUR/USD, GBP/USD, and EUR/GBP) on a large electronic trading platform during May and June 2010, and discuss how a number of my findings highlight weaknesses in current models of limit order books.

We analyse the effect of a natural change to the time variable on the convergence of the Crank-Nicholson scheme when applied to the solution of the heat equation with Dirac delta function initial conditions. In the original variables, the scheme is known to diverge as the time step is reduced with the ratio (lambda) of the time step to space step held constant - the value of lambda controls how fast the divergence occurs. After introducing the square root of time variable we prove that the numerical scheme for the transformed PDE now always converges and that lambda controls the order of convergence, quadratic convergence being achieved for lambda below a critical value. Numerical results indicate that the time change used with an appropriate value of lambda also results in quadratic convergence for the calculation of gamma for a European call option without the need for Rannacher start-up steps. Finally, some results and analysis are presented for the effect of the time change on the calculation of the option value and greeks for the American put calculated by the penalty method with Crank-Nicholson time-stepping.

NB: EXTRA SEMINAR THIS WEEK

Executives compensated with stock options generally receive grants periodically and so on

any given date, may have a portfolio of options of differing strikes and maturities on their

company’s stock. Non-transferability and trading restrictions in the company stock result in the executive facing unhedgeable risk. We employ exponential utility indifference pricing to analyse the optimal exercise thresholds for each option, option values and cost of the options to shareholders. Portfolio interaction effects mean that each of these differ, depending on the composition of the remainder of the portfolio. In particular, the cost to shareholders of an option portfolio is lowered relative to its cost computed on a per-option basis. The model can explain a number of empirical observations - which options are attractive to exercise first, how exercise changes following a new grant, and early exercise.

Joint work with Jia Sun and Elizabeth Whalley (WBS).

We consider a portfolio optimisation problem on infinite horizon when

the investment policy satisfies the drawdown constraint, which is the

wealth process of an investor is always above a threshold given as a

function of the past maximum of the wealth process. The preferences are

given by a utility function and investor aims to maximise an asymptotic

growth rate of her expected utility of wealth. This problem was firstly

considered by Grossman and Zhou [3] and solved for a Black-Scholes

market and linear drawdown constraint.

The main contribution of the paper is an equivalence result: the

constrained problem with utility U and drawdown function w has the same

value function as the unconstrained problem with utility UoF, where

function F is given explicitly in terms of w. This work was inspired by

ideas from [2], whose results are a special case of our work. We show

that the connection between constrained and unconstrained problems holds

for a much more general setup than their paper, i.e. a general

semimartingale market, larger class of utility functions and drawdown

function which is not necessarily linear. The paper greatly simplifies

previous approaches using the tools of Azema-Yor processes developed in

[1]. In fact we show that the optimal wealth process for constrained

problem can be found as an explicit Azema-Yor transformation of the

optimal wealth process for the unconstrained problem.

We further provide examples with explicit solution for complete and

incomplete markets.

[1] Carraro, L., Karoui, N. E., and Obloj, J. On Azema-Yor processes,

their optimal properties and the Bachelier-Drawdown equation, to appear in

Annals of Probability, 2011.

[2] Cvitanic, J., and Karatzas, I. On portfolio optimization under

drawdown constraints. IMA Volumes in Mathematics and Its Applications

65(3), 1994, 35-45

[3] Grossman, S. J., and Zhou, Z. Optimal investment strategies for

controlling drawdowns. Mathematical Finance 3(3), 1993, 241-276

In this work, we study equilibrium solutions for a LQ

control problem with state-dependent terms in the objective, which

destroy the time-consisitence of a pre-commited optimal solution.

We get a sufficient condition for equilibrium by a system of

stochastic differential equations. When the coefficients in the

problem are all deterministic, we find an explicit equilibrium

for general LQ control problem. For the mean-variance portfolio

selection in a complete financial market, we also get an explicit

equilibrium with random coefficient of the financial.

Cubature on Wiener space" is a numerical method for the weak

approximation of SDEs. After an introduction to this method we present

some cases when the method is computationally expensive, and highlight

some techniques that improve the tractability. In particular, we adapt

the Multilevel Monte-Carlo framework and extend the Milstein-scheme

based version of Mike Giles to higher dimensional and higher degree cases.

: Recently strict local martingales have been used to model

exchange rates. In such models, put-call parity does not hold if one

assumes minimal superreplicating costs as contingent claim prices. I

will illustrate how put-call parity can be restored by changing the

definition of a contingent claim price.

More precisely, I will discuss a change of numeraire technique when the

underlying is only a local martingale. Then, the new measure is not

necessarily equivalent to the old measure. If one now defines the price

of a contingent claim as the minimal superreplicating costs under both

measures, then put-call parity holds. I will discuss properties of this

new pricing operator.

To illustrate this techniques, I will discuss the class of "Quadratic

Normal Volatility" models, which have drawn much attention in the

financial industry due to their analytic tractability and flexibility.

This talk is based on joint work with Peter Carr and Travis Fisher.

When modelling biochemical reactions within cells, it is vitally important to take into account the effect of intrinsic noise in the system, due to the small copy numbers of some of the chemical species. Deterministic systems can give vastly different types of behaviour for the same parameter sets of reaction rates as their stochastic analogues, giving us an incorrect view of the bifurcation behaviour.

The stochastic description of this problem gives rise to a multi-dimensional Markov jump process, which can be approximated by a system of stochastic differential equations. Long-time behaviour of the process can be better understood by looking at the steady-state solution of the corresponding Fokker-Planck equation.

In this talk we consider a new finite element method which uses simulated trajectories of the Markov-jump process to inform the choice of mesh in order to approximate this invariant distribution. The method has been implemented for systems in 3 dimensions, but we shall also consider systems of higher dimension.

1pm Kawei Wang

\newline Title: A Model of Behavioral Consumption in Contnuous Time

\newline Abstract: Inspired by Jin and Zhou (2008), we try to construct a model

of consumption within the framework of Prospect Theory and Cumulative

Prospect Theory in continuous time.

\newline

\newline

1.20 Rasmus Wissmann

\newline Title: A Principal Component Analysis-based Approach for High-Dimensional PDEs in Derivative Pricing

\newline Abstract: Complex derivatives, such as multi asset and path dependent options,

often lead to high-dimensional problems. These are generally hard to

tackle with numerical PDE methods, because the computational effort

necessary increases exponentially with the number of dimensions. We

investigate a Principal Component Analysis-based approach that aims to

make the high-dimensional problem tractable by splitting it into a

number of low-dimensional ones. This is done via a diagonalization of

the PDE according to the eigenvectors of the covariance matrix and a

subsequent Taylor-like approximation. This idea was first introduced by

Reisinger and Wittum for the basic case of a vanilla option on a basket

of stocks [1]. We aim to extend the approach to more complex derivatives

and markets as well as to develop higher order versions. In this talk we

will present the basic ideas, initial results for the example of a

ratchet cap under the LIBOR Market Model and the current plans for

further research.

[1] C. Reisinger and G. Wittum, Efficient Hierarchical Approximation of

High-Dimensional Option Pricing Problems, SIAM Journal of Scientific

Computing, 2007:29

\newline

\newline

1.40 Pedro Vitoria

\newline Title: Infinitesimal Mean-Variance and Forward Utility

\newline Abstract: Mean-Variance, introduced by Markowitz in his seminal paper of 1952, is

a classic criterion in Portfolio Theory that is still predominantly used

today in real investment practice. In the academic literature, a number of

interesting results have been produced in continuous-time version of this

model.

In my talk, I will establish a link between the multi-period

Mean-Variance model and its continuous-time limit. A key feature of the

results is that, under suitable but mild technical conditions, it

captures the results of Forward Utility, thus establishing an important

link between Mean-Variance and forward utility maximisation.

The first half of this seminar will discuss the hedging problem faced by a large sports betting agent who has to risk-manage an unwanted position in a bet on the simultaneous outcome of multiple football matches, by trading in moderately liquid simple bets on individual results. The resulting mathematical framework is that of a coupled system of multi-dimensional HJB equations.

This leads to the wider question of the numerical approximation of such problems. Dynamic programming with PDEs, while very accurate in low dimensions, becomes practically intractable as the dimensionality increases. Monte Carlo methods, while robust for computing linear expectations in high dimensions, are not per se well suited to dynamic programming. This leaves high-dimensional stochastic control problems to be considered computationally infeasible in general.

In the second half of the seminar, we will outline ongoing work in this area by sparse grid techniques and asymptotic expansions, the former exploiting smoothness of the value function, the latter a fast decay in the importance of principal components. We hope to instigate a discussion of other possible approaches including e.g. BSDEs.

The aim of this work is to show how to derive the electricity price from models for the

underlying construction of the bid-stack. We start with modelling the behaviour of power

generators and in particular the bids that they submit for power supply. By modelling

the distribution of the bids and the evolution of the underlying price drivers, that is

the fuels used for the generation of power, we can construct an spede which models the

evolution of the bids. By solving this SPDE and integrating it up we can construct a

bid-stack model which evolves in time. If we then specify an exogenous demand process

it is possible to recover a model for the electricity price itself.

In the case where there is just one fuel type being used there is an explicit formula for

the price. If the SDEs for the underlying bid prices are Ornstein-Uhlenbeck processes,

then the electricity price will be similar to this in that it will have a mean reverting

character. With this price we investigate the prices of spark spreads and swing options.

In the case of multiple fuel drivers we obtain a more complex expression for the price

as the inversion of the bid stack cannot be used to give an explicit formula. We derive a

general form for an SDE for the electricity price.

We also show that other variations lead to similar, though still not tractable expressions

for the price.

We consider a multidimensional structural credit model, where each company follows a jump-diffusion process and is connected with other companies via global factors. We assume that a company can default both expectedly, due to the diffusion part, and unexpectedly, due to the jump part, by a sudden fall in a company's value as a result of a global shock. To price CDOs efficiently, we use ideas, developed by Bush et al.

for diffusion processes, where the joint density of the portfolio is approximated by a limit of the empirical measure of asset values in the basket. We extend the method to jump-diffusion settings. In order to check if our model is flexible enough, we calibrate it to CDO spreads from pre-crisis and crisis periods.

For both data sets, our model fits the observed spreads well, and what is important, the estimated parameters have economically convincing values.

We also study the convergence of our method to basic Monte Carlo and conclude that for a CDO, that typically consists of 125 companies, the method gives close results to basic Monte Carlo."

In this talk I want to ask how to create a coherent mathematical framework for pricing and hedging which starts with the information available in the market and does not assume a given probabilistic setup. This calls for re-definition of notions of arbitrage and trading and, subsequently, for a ``probability-free first fundamental theorem of asset pricing". The new setup should also link with a classical approach if our uncertainty about the model vanishes and we are convinced a particular probabilistic structure holds. I explore some recent results but, predominantly, I present the resulting open questions and problems. It is an ``internal talk" which does not necessarily present one paper but rather wants to engage into a discussion. Ideas for the talk come in particular from joint works with Alex Cox and Mark Davis.

In this talk we will present results concerning the large scale long time behaviour of particles moving in a periodic (random) velocity field subject to molecular diffusion. The particle can be considered massless (passive tracer) or not (inertial particle). Under appropriate assumptions for the velocity field the large scale long time behavior of the particle is described by a Brownian motion with an effective diffusivity matrix K.

We then present some numerical algorithms concerning the calculation of the effective diffusivity in the limit of vanishing molecular diffusion (stochastic geometric integrators). Time permitting we will discuss the case where the driving noise is no longer white but colored and study the effects of this change to the effective diffusivity matrix.

Much mathematical work has gone into the creation of time-consistent nonlinear expectations. When we think of implementing these, various problems arise and destroy the beautiful consistency properties we have worked so hard to create. One of these problems is to do with horizon dependence, in particular, where a portfolio's value is considered at a time t+m, where t is the present time and m is a fixed horizon.

In this talk we shall discuss various notions of time consistency and the corresponding solution concepts. In particular, we shall focus on notions which pay attention to the space of available policies, allowing for commitment devices and non-markovian restrictions. We shall see that, for any time-consistent nonlinear expectation, there is a notion of time consistency which is satisfied by the moving horizon problem.

We consider an optimal stopping problem arising in connection with the exercise of an executive stock option by an agent with inside information.

The agent is assumed to have noisy information on the terminal value of the stock, does not trade the stock or outside securities, and maximises the expected discounted payoff over all stopping times with regard to an enlarged filtration which includes the inside information. This leads to a stopping problem governed by a time-inhomogeneous diffusion and a call-type reward. Using stochastic flow ideas we establish properties of the value function (monotonicity, convexity in the log-stock price), conditions under which the option value exhibits time decay, and derive the smooth fit condition for the solution to the free boundary problem governing the maximum expected reward. From this we derive the early exercise decomposition of the value function. The resulting integral equation for the unknown exercise boundary is solved numerically and this shows that the insider may exercise the option before maturity, in situations when an agent without the privileged information may not.

In portfolio management, there are specific strategies for trading between two assets that are cointegrated. These are commonly referred to as pairs-trading or spread-trading strategies. In this paper, we provide a theoretical framework for portfolio choice that justifies the choice of such strategies. For this, we consider a continuous-time error correction model to model the cointegrated price processes and analyze the problem of maximizing the expected utility of terminal wealth, for logarithmic and power utilities. We obtain and justify an extra no-arbitrage condition on the market parameters with which one obtains decomposition results for the optimal pairs-trading portfolio strategies.

Optimisation problems involving objective functions defined on function spaces often have a natural interpretation as a variational problem, leading to a solution approach via calculus of variations. An equally natural alternative approach is to approximate the function space by a finite-dimensional subspace and use standard nonlinear optimisation techniques. The second approach is often easier to use, as the occurrence of absolute value terms and inequality constraints poses no technical problem, while the calculus of variations approach becomes very involved. We argue our case by example of two applications in mathematical finance: the computation of optimal execution rates, and pre-computed trade volume curves for high frequency trading.