We consider solution of stochastic storage problems through regression Monte Carlo (RMC) methods. Taking a statistical learning perspective, we develop the dynamic emulation algorithm (DEA) that unifies the different existing approaches in a single modular template. We then investigate the two central aspects of regression architecture and experimental design that constitute DEA. For the regression piece, we discuss various non-parametric approaches, in particular introducing the use of Gaussian process regression in the context of stochastic storage. For simulation design, we compare the performance of traditional design (grid discretization), against space-filling, and several adaptive alternatives. The overall DEA template is illustrated with multiple examples drawing from natural gas storage valuation and optimal control of back-up generator in a microgrid.
We investigate the predictability of several range-based stock volatility estimators, and compare them to the standard close-to-close estimator which is most commonly acknowledged as the volatility. The patterns of volatility changes are analyzed using LSTM recurrent neural networks, which are a state of the art method of sequence learning. We implement the analysis on all current constituents of the Dow Jones Industrial Average index, and report averaged evaluation results. We find that changes in the values of range-based estimators are more predictable than that of the estimator using daily closing values only.
We describe a robust calibration algorithm of a set of SSVI slices (i.e. a set of 3 SSVI parameters $\theta, \rho, \varphi$ attached to each option maturity available on the market), which grants that these slices are free of Butterfly and Calendar-Spread arbitrage. Given such a set of consistent SSVI parameters, we show that the most natural interpolation/extrapolation of the parameters provides a full continuous volatility surface free of arbitrage. The numerical implementation is straightforward, robust and quick, yielding an effective, parsimonious solution to the smile problem, which has the potential to become a benchmark one.
The Monte Carlo pathwise sensitivities approach is well established for smooth payoff functions. In this work, we present a new Monte Carlo algorithm that is able to calculate the pathwise sensitivities for discontinuous payoff functions. Our main tool is to combine the one-step survival idea of Glasserman and Staum with the stable differentiation approach of Alm, Harrach, Harrach and Keller. As an application we use the derived results for a two-dimensional calibration of a CoCo-Bond, which we model with different types of discretely monitored barrier options.
This article presents a generic model for pricing financial derivatives subject to counterparty credit risk. Both unilateral and bilateral types of credit risks are considered. Our study shows that credit risk should be modeled as American style options in most cases, which require a backward induction valuation. To correct a common mistake in the literature, we emphasize that the market value of a defaultable derivative is actually a risky value rather than a risk-free value. Credit value adjustment (CVA) is also elaborated. A practical framework is developed for pricing defaultable derivatives and calculating their CVAs at a portfolio level.
The least squares Monte Carlo algorithm has become a popular tool for solving stochastic control problems, mainly due to its ability to handle multiple stochastic state variables. However, a configuration that remains challenging is when dealing with high dimensional controls, due to the exponential complexity of grid search methods or the mathematical difficulty of deriving and solving high-dimensional first-order optimality conditions. This paper proposes an efficient least squares Monte Carlo algorithm that is suitable for handling high-dimensional controls. In particular, we first approximate backward recursive dynamic programs on a coarse grid of controls, then use a combined technique of local control regression and adaptive refinement grids to improve the optimal control estimates. We numerically show that the local control regression is more accurate and more efficient than global control regression, and that the overall computational runtime scales polynomially with respect to the dimension of the control. Finally, we further validate our method by solving a portfolio optimization problem with twelve risky assets with transaction cost, liquidity cost and market impact, which involves twelve exogenous risk factors (asset returns) and thirteen endogenous risk factors (portfolio value and asset prices subject to market impact).
The Constant Elasticity of Variance (CEV) model significantly outperforms the Black-Scholes (BS) model in forecasting both prices and options. Furthermore, the CEV model has a marked advantage in capturing basic empirical regularities such as: heteroscedasticity, the leverage effect, and the volatility smile. In fact, the performance of the CEV model is comparable to most stochastic volatility models, but it is considerable easier to implement and calibrate. Nevertheless, the standard CEV model solution, using the non-central chi-square approach, still presents high computational times, specially when: i) the maturity is small, ii) the volatility is low, or iii) the elasticity of the variance tends to zero. In this paper, a new numerical method for computing the CEV model is developed. This new approach is based on the semiclassical approximation of Feynman's path integral. Our simulations show that the method is efficient and accurate compared to the standard CEV solution considering the pricing of European call options.
We propose a fast and accurate numerical method for pricing European swaptions in multi-factor Gaussian term structure models. Our method can be used to accelerate the calibration of such models to the volatility surface. The pricing of an interest rate option in such a model involves evaluating a multi-dimensional integral of the payoff of the claim on a domain where the payoff is positive. In our method, we approximate the exercise boundary of the state space by a hyperplane tangent to the maximum probability point on the boundary and simplify the multi-dimensional integration into an analytical form. The maximum probability point can be determined using the gradient descent method. We demonstrate that our method is superior to previous methods by comparing the results to the price obtained by numerical integration.
We introduce a model for the short-term dynamics of financial assets based on an application to finance of quantum gauge theory, developing ideas of Ilinski. We present a numerical algorithm for the computation of the probability distribution of prices and compare the results with APPLE stocks prices and the S&P500 index.
This article presents a new model for valuing a credit default swap (CDS) contract that is affected by multiple credit risks of the buyer, seller and reference entity. We show that default dependency has a significant impact on asset pricing. In fact, correlated default risk is one of the most pervasive threats in financial markets. We also show that a fully collateralized CDS is not equivalent to a risk-free one. In other words, full collateralization cannot eliminate counterparty risk completely in the CDS market.
In this paper, the estimation problem for sparse reduced rank regression (SRRR) model is considered. The SRRR model is widely used for dimension reduction and variable selection with applications in signal processing, econometrics, etc. The problem is formulated to minimize the least squares loss with a sparsity-inducing penalty considering an orthogonality constraint. Convex sparsity-inducing functions have been used for SRRR in literature. In this work, a nonconvex function is proposed for better sparsity inducing. An efficient algorithm is developed based on the alternating minimization (or projection) method to solve the nonconvex optimization problem. Numerical simulations show that the proposed algorithm is much more efficient compared to the benchmark methods and the nonconvex function can result in a better estimation accuracy.
We develop a mixed least squares Monte Carlo-partial differential equation (LSMC-PDE) method for pricing Bermudan style options on assets whose volatility is stochastic. The algorithm is formulated for an arbitrary number of assets and driving processes and we prove the algorithm converges probabilistically. We also discuss two methods to greatly improve the algorithm's computational complexity. Our numerical examples focus on the single ($2d$) and multi-dimensional ($4d$) Heston model and we compare our hybrid algorithm with classical LSMC approaches. In both cases, we demonstrate that the hybrid algorithm has significantly lower variance than traditional LSMC. Moreover, for the $2d$ example, where it is possible to visualize, we demonstrate that the optimal exercise strategy from the hybrid algorithm is significantly more accurate compared to the one from the full LSMC when using a finite difference approach as a reference.
There are two possible ways of interpreting the seemingly stochastic nature of financial markets: the Efficient Market Hypothesis (EMH) and a set of stylized facts that drive the behavior of the markets. We show evidence for some of the stylized facts such as memory-like phenomena in price volatility in the short term, a power-law behavior and non-linear dependencies on the returns. Given this, we construct a model of the market using Markov chains. Then, we develop an algorithm that can be generalized for any N-symbol alphabet and K-length Markov chain. Using this tool, we are able to show that it's, at least, always better than a completely random model such as a Random Walk. The code is written in MATLAB and maintained in GitHub.