I am pleased to introduce the latest issue of The Journal of Computational Finance, which has, as a recurring theme, the development of methodologies for novel extensions of classical problems. Deviating from the well-studied setting adds realism to the models (eg, by including practical features of real markets or by relaxing assumptions on perfect observations), but it also leads to computational challenges. The papers in this issue find creative ways of making these more complex models tractable.
In the issue’s first paper, “Deep learning for efficient frontier calculation in finance”, Xavier Warin proposes and tests different formulations of mean–variance-type portfolio allocation problems with neural network parametrization of the optimal feedback strategy. The approach allows him to trace out the efficient frontier, and it is shown to perform well in a variety of settings, including market constraints, stochastic volatilities and with multiple assets.
Julien Vaes and Raphael Hauser generalize a traditional trade execution model in “Optimal trade execution with uncertain volume target”, the second paper in this issue, by augmenting price uncertainty with uncertainty in the volume target. Avoiding expensive dynamic programming by a judicious problem formulation, their paper analyzes the properties of optimal strategies and demonstrates numerically their superior performance compared with standard approaches.
In our third paper, “A general firm value model under partial information”, Cheikh Mbaye, Abass Sagna and Frédéric Vrins consider generic structural credit models, but under the premise that the true firm value process is unobservable. Using a filtering approach and a numerical approximation by recursive quantization, they are able to solve the partial information version of this classical problem efficiently.
Finally, “Subsampling and other considerations for efficient risk estimation in large portfolios”, the issue’s fourth paper, finds Michael B. Giles and Abdul-Lateef Haji-Ali discussing computationally tractable estimators of risk measures for large portfolios. A subsampling strategy can lead to a complexity that is independent of the portfolio size and can be adapted to the heterogeneity of the portfolio constituents and interdependencies between them.
I hope you find inspiration for your own work in these papers.
The author puts forward a means to calculate the efficient frontier in the Mean-Variance and Mean-CVaR portfolio optimization problems using deep neural network algorithms.
This paper demonstrates that risk-averse traders can benefit from delaying trades using a model that accounts for volume uncertainty.
The authors propose a general structural default model combining enhanced economic relevance and affordable computational complexity.
The authors apply multilevel Monte Carlo simulation to the problems inherent in computing risk measures of a financial portfolio with large numbers of derivatives.