The modelling of the short rate offers many advantages, with the models explored in this dissertation all offering closed-form, analytic formulae for bond prices and for options on bonds. Often, a vital primary condition is for a model to be calibrated to the initial term structure and to recover the bond prices observed in the market – that is, to be calibrated to the initial yield curve. Under the two exogenous models explored in this dissertation, the Hull-White and the CIR++, the effect of increasing the volatility parameter of the SDE increases the mean of the short rate. Increasing volatility of an SDE is a common approach to stress testing a model, as such, the consequences of bumping volatility in a calibrated model is a vital concern. The Hull-White model and CIR++ model were calibrated to market data, with the former being able to match the observed cap prices, while the latter failed, displaying an upper bound on cap prices. Investigating this, under CIR++ model, bond option prices are shown to not be straightforward increasing functions of the volatility parameter. In fact, for high volatility, bond option prices display an upper limit before decreasing, thus providing a limit to the level of cap prices too. This dissertation points to the reason residing in the underlying CIR model from which the CIR++ is based on, and the manner in which the model is extended

The Monte Carlo method (MC) is a common numerical technique used to approximate an expectation that does not have an analytical solution. For certain problems, MC can be inefficient. Many techniques exist to improve the efficiency of MC methods. The Multilevel Monte Carlo (ML) technique developed Giles (2008) is one such method. It relies on approximating the payoff at different levels of accuracy and using a telescoping sum of these approximations to compute the ML estimator. This dissertation summarises the ML technique and its implementation. To start with, the framework is applied to a European call option. Results show that the efficiency of the method is up to 13 times faster than crude MC. Then an American put option is priced within the ML framework using two pricing methods. The Least Squares Monte Carlo method (LSM) estimates an optimal exercise strategy at finitely many instances, and consequently a lower bound price for the option. The dual method finds an optimal martingale, and consequently an upper bound for the price. Although the pricing results are quite close to the corresponding crude MC method, the efficiency produces mixed results. The LSM method performs poorly within an ML framework, while the dual approach is enhanced.

A method to price American options under a stochastic volatility framework is introduced which is based on Rambharat and Brockwell (2010). We price American options under the Heston and Bates stochastic volatility models where volatility is assumed to be a latent process. The pricing algorithm is based on the least-squares Monte Carlo approach made popular by Longstaff and Schwartz (2001). Information about the volatility of the underlying asset is used to assist in solving the pricing problem. Since volatility is assumed to be a latent, a particle filter is used to estimate the filtering distribution of volatility. A summary vector is constructed which captures the essential features of the filtering distribution. At each time step before maturity, the elements of the summary vector and the current share price are used as explanatory variables in a regression function which estimates the continuation value of the option. Estimating the continuation value assists in finding the optimal time to exercise the option. This pricing approach is benchmarked against a method which assumes volatility is observable. Furthermore, our pricing approach is compared to simpler methods which do not use particle filtering. Results from our numerical experiments suggest the proposed approach produces accurate option prices.

Includes bibliographical references. / Liquidation strategies consider the problem of minimising transaction costs occurring in a portfolio liquidation. Transaction costs are the difference between current market value and the realised value after the liquidation. A strategy to follow to perform a liquidation is especially important to institutional investors due the large size of their trades. Large trades can have a significant effect on the price of a security which can impact the realised returns of the liquidation. These models solve for trading trajectories that maximise this. The models investigated do this in a mean-variance framework where the expected return of the strategy is constrained by its variance and the investors risk preference. Parameters used in liquidity functions are estimated for securities on the South African JSE Securities Exchange. The effects of security liquidity, volatility, stock correlation and length of liquidation horizon on the optimal strategy are investigated. There is little or no existing literature that attempts to model these functions in the South African market. Due to the smaller size of the South African market as well as the number of thinly traded shares compared to most markets studied in the literature, many securities are highly illiquid. We investigate relationships between firm size and daily traded value and these liquidity parameters. General rules are presented to help traders improve a liquidation strategy without the need to estimate all parameters needed to calculate an optimal strategy using one of these models.

This dissertation explores a key challenge of the financial industry — the efficient computation of sensitivities of financial instruments. The adjoint approach to solving affine recursion problems (ARPs) is presented as a solution to this challenge. A Monte Carlo setting is adopted and it is illustrated how computational efficiency in sensitivity calculation may be significantly improved via the pathwise derivatives method through adapting an adjoint approach. This is achieved through the reversal of the order of differentiation in the pathwise derivatives algorithm in comparison to the standard, intuitive ‘forward’ approach. The Libor market model (LMM) framework is selected for examples to demonstrate these computational savings, with varying degrees of complexity of the LMM explored, from a one-factor model with constant volatility to a full factor model with time homogeneous volatilities.

This dissertation focuses on the application of neural networks to financial model calibration. It provides an introduction to the mathematics of basic neural networks and training algorithms. Two simplified experiments based on the Black-Scholes and constant elasticity of variance models are used to demonstrate the potential usefulness of neural networks in calibration. In addition, the main experiment features the calibration of the Heston model using model-generated data. In the experiment, we show that the calibrated model parameters reprice a set of options to a mean relative implied volatility error of less than one per cent. The limitations and shortcomings of neural networks in model calibration are also investigated and discussed.

Unspanned stochastic volatility term structure models have gained popularity in the literature. This dissertation focuses on the challenges of implementing the simplest case – bivariate unspanned stochastic volatility models, where there is one state variable controlling the term structure, and one scaling the volatility. Specifically, we consider the Log-Affine Double Quadratic (1,1) model of Backwell (2017). In the class of affine term structure models, state variables are virtually always spanned and can therefore be inferred from bond yields. When fitting unspanned models, it is necessary to include option data, which adds further challenges. Because there are no analytical solutions in the LADQ (1,1) model, we show how options can be priced using an Alternating Direction Implicit finite difference scheme. We then implement an Unscented Kalman filter — a non-linear extension of the Kalman filter, which is a popular method for inferring state variable values — to recover the latent state variables from market observable data

In this dissertation, two efficient approaches for pricing European options on amortising swaps are explored. The first approach is to decompose the pricing of a European amortising swaption into a series of discount bond options, with an assumption that the interest rate follows a one-factor affine model. The second approach is using a one-dimensional numerical integral technique to approximate the price of European amortising swaption, with an assumption that the interest rate follows an additive two-factor affine model. The efficacy of the two methods was tested by making a comparison with the prices generated using Monte Carlo methods. Two methods were used to accelerate the convergence rate of the Monte Carlo model, a variance reduction method, namely the control variates technique and a method of using deterministic low-discrepancy sequences (also called quasi-Monte Carlo methods).

This dissertation considers the concept of potential future exposure, and how initial margin can be used to mitigate it. In addition to this, the cost of implementing initial margin is estimated, and some of the difficulties associated with it are addressed. The two primary techniques for calculating initial margin considered are nested Monte Carlo, and Gaussian Least Squares Monte Carlo. These two techniques are compared for effectiveness. It is shown that the nested Monte Carlo technique performs well under numerous conditions, and that the Gaussian Least Squares Monte Carlo relies on particular model and instrument characteristics.

Unspanned stochastic volatility (USV) models have gained popularity in the literature. USV models contain at least one source of volatility-related risk that cannot be hedged with bonds, referred to as the unspanned volatility factor(s). Bivariate USV models are the simplest case, comprising of one state variable controlling the term structure and the other controlling unspanned volatility. This dissertation focuses on pricing with two particular bivariate USV models: the Log-Affine Double Quadratic (1,1) – or LADQ(1,1) – model of Backwell (2017) and the LinearRational Square Root (1,1) – or LRSQ(1,1) – model of Filipovic´ et al. (2017). For the LADQ(1,1) model, we fully outline how an Alternating Directional Implicit finite difference scheme can be used to price options and implement the scheme to price caplets. For the LRSQ(1,1) model, we illustrate a semi-analytical Fourierbased method originally designed by Filipovic´ et al. (2017) for pricing swaptions, but adjust it to price caplets. Using the above numerical methods, we calibrate each (1,1) model to both the British-pound yield curve and caps market. Although we cannot achieve a close fit to the implied volatility surface, we find that the parameters in the LADQ(1,1) model have direct control over the qualitative features of the volatility skew, unlike the parameters within the LRSQ(1,1) model.