Risk

In this article series, we present a machine learning-based approach to solving a common problem in financial modelling where one is faced with the task of estimating the value of a function which requires a significant amount of computation to evaluate. More specifically, a function that corresponds to a so-called nested simulation aimed at, for example, estimating a capital requirement for a financial institution or the risk associated with a structured product for a retail investor.

The modern wealth management industry still relies on the 50-year-old approaches to portfolio management, widely popularized by Markowitz's Modern Portfolio Theory (1952). Despite heavy criticism within the academic circles, the alternative methods remain undeservingly overlooked in practice. In the context of the modern leap for hyper-customization, we look into one of the alternatives to Modern Portfolio Theory in greater detail - the Utility-based approach.

The second part of the “Portfolio Construction”-series explores whether introducing parameter uncertainty to the model would improve the out-of-sample performance of the optimal portfolio. Additionally, the article proposes and tests two adjustments to regular utility optimisation.

There is a number of challenges associated with portfolio construction based on historical data. This three-part article series explores some of the most common issues attributed to the model-based portfolio optimization: the sensitivity to changes in data, large variations in portfolio weights and the bad out-of-sample performance.

As machine learning methods grow in use and popularity, we explore yet another dimension of wealth management that our experts consider fit for applying such frameworks. In this article, we deploy hierarchical clustering to find more consistent ways of predicting the relative future performance of funds.

Machine learning applications have become more prominent in the financial industry in recent years. Our new article series is exploring the benefits and challenges of using self-normalising neural networks (SNNs) for calculating liquidity risk. The first piece of the series introduces the main concepts used in the investigative case study for the Swedish bond market.

In the third and concluding article in the ALM using LMSC series, we focus on analyzing the optimal asset allocations in the context of changing asset classes as well as finding the optimal allocation by maximizing the risk-adjusted net asset value. The estimates based on the LSMC method are then compared to the estimates obtained from the full nested Monte Carlo method.

The second part of the series exploring the use of Least Squares Monte Carlo in Asset and Liability Management is focused on evaluation of accuracy and performance of this method in comparison to full nested Monte Carlo simulation benchmarks.

In the first part of the ”Asset and Liability Management using LSMC” article series, we outline an ALM framework based on a replicating portfolio approach along with a suitable financial objective. This ALM framework, albeit simplified, is constructed to provide a straightforward replication of the complex interactions between assets and liabilities. Moreover, a brief introduction to the LSMC method used to generate all underlying risk factors is presented.

This article will discuss why it is important to model credit indices and detail a number of different approaches to this problem.