An overview of Machine Learning in the Investment World

Artificial Intelligence (AI) is a term that is now commonly used across the globe but it is often used as a substitute for Machine Learning (ML), a subset of AI, that is often the real driver revolutionising many industries. None more so than the finance industry and equity investment strategies are a perfect example. Here is an overview of what ML is and how it can benefit trading strategies. 

What is Machine Learning? 

Machine learning (ML) is the process of using mathematical models on data to learn and predict outcomes. It enables algorithms to continue learning and improving on their own, based on experience. Supervised ML in particular, uses historical data to learn where signals exist, and outputs a predictive model based on a defined target. In Kvasir’s case, this could be a prediction of stock movement up or down for example. 

ML algorithms are classified into 4 mainstream types: 

– supervised learning, which uses a training set that is labelled to learn to predict target outcomes; 

– unsupervised learning, which analyses unlabelled data using clustering and association techniques. This is often used to find patterns and similarities and can help in a number of applications; 

– semi supervised learning, which uses a small amount of labelled and a large amount of unlabelled data; 

– reinforcement learning where positive and negative actions are used methodically to reward or penalise certain behaviours.

Deep learning is a subset of ML and also a buzzword in many industries right now. Deep learning uses multiple layers of small units called neurons that can identify signals in very complex and vast amounts of data. Those neural networks attempt to simulate the behaviour of the brain. Due to its sensitivity, deep learning requires a large data history and granularity of data, which means it can only be successfully applied in certain applications where data like this exists. The more layers, the more refined and optimised the accuracy of the outcomes. Therefore deep learning has some great benefits in the investment world when looking at large unstructured datasets such as news data, but is less useful for smaller structured datasets, such as daily price data series of securities where other ML techniques are more appropriate.  

The benefits of ML in alternative investment modelling

Many users of ML in the alternatives space often apply it in non-critical areas or in discrete parts of the overall investment process. This limits the benefits of using ML compared to when it is applied across the entire investment process in an end-to-end fashion. Using ML to build an end-to-end investment strategy:

– Removes human bias from the decision-making process which can be particularly beneficial at times when market regimes are shifting. Model selection is one such process that is often still done in a discretionary way, which especially in finance poses the risk of over-fitting. 

– Allow strategies to be applied at scale as it is able to look at thousands of stocks at once, across many dimensions. This simply isn’t possible in more traditional approaches andfunds where there is human discretion. 

– Allows many more data points to be assessed including alternative data sets outside of fundamentals and price data.

– Allows for an overall better optimised investment process compared to systems where areas such as signal engineering, forecasting, portfolio optimisation and execution optimisation are performed in isolation.

By removing human subjectivity from the decision making process a fund can evolve its strategy and adapt in changing environments, with predictions being made on a huge volume of stocks in just a few minutes. If the appropriate ML techniques are used correctly within the investment process, investment firms can not only increase potential returns but also reduce risk over time with diversification and portfolio optimisation techniques. 

Like some of the most successful quant firms out there, Kvasir has spent many years building a sophisticated AI driven approach to take advantage of these benefits and will continue to innovate as new datasets and market regimes appear.