Machine Learning: Portfolio Management

Portfolio choice is a non-trivial problem faced by economic agents. In a nutshell, it involves making a decision regarding how one should distribute one’s wealth…

Portfolio choice is a non-trivial problem faced by economic agents. In a nutshell, it involves making a decision regarding how one should distribute one’s wealth across multiple assets. Distinct assets offer their unique outcome possibilities. The two dimensions to be taken into account in the process of decision making are “expected return” and “risk”. The former characterizes the percentage of increase or decrease in a given investment i.e. the income generated by it over a time period such as a month, a quarter, or a year. The latter quantifies the uncertainty of the payoffs to the investor. Risk factors span beyond the marketplace, including political and economic events, and in some cases even weather has an impact on many commodities.

In order to evaluate portfolio risk, the scientist called James Kelly came up with a mathematical formula, naturally named the Kelly Criterion. Skipping on the technical details, the principle boils down to the following: the more comprehensive and of higher quality our information is, the lower our risk becomes.

You have probably heard about machine learning and potentially even wondered whether it could be utilised to benefit your business in addition to every day life. Great news is that the applications of this subset of Artificial Intelligence are not restricted to self-driving cars and automatic tagging of photos. This field can be brought into practice to portfolio choice in a twofold manner.

Firstly, using and tackling the ever-growing information from historical market data and financial valuations, we can model it to make predictions or forecasts. In the machine learning context, this means we use the available past data for the training process. If we apply deep learning algorithm, the model will learn how the different inputs, or feature, such as revenue growth rate and terminal growth rate, can influence a firm’s final value. It is worth noting the current power of supercomputers: tasks that would have taken weeks to solve several years ago now take up to several hours, so with accelerating speed of technology come faster results.

Secondly, we can use machine learning to avoid the common hazard of overfitting. The issue with overfitting often stems from the temptation for analysts to believe they have unravelled a relationship in the data after tweaking parameters or mistakenly assume spurious correlations as meaning causation. Machine learning enables us to bypass such problems by restricting the human involvement to set up the whole framework for investing. The system can then explore and discover an optimal investment strategy and perform direct allocation choices.

We believe that methods from artificial intelligence will become increasingly important in the field of investment management over the next years. The process of being able to spot an opportunity in the market and come up with an investment approach is both technical and creative. Through the use of machine learning algorithms and automating certain aspects of the strategies’ creation, asset managers will enhance their accuracy, efficiency, and potentially boost returns. 

Email us for a case study that aligns with your business: