flood of data and identify the drivers of returns that can be used to estimate future price developments. Big data technologies have great potential in the financial sector, as computer algorithms are ideal for processing huge volumes of data and detecting patterns in data sets at high speed.
Deriving insights from data
State-of-the-art machine learning algorithms are able to independently scour financial market data and uncover hidden correlations between business metrics and future returns without human assistance. This process can be divided into four steps:
- Database of equities
The basis for all machine learning algorithms is an extensive database covering decades of history for all equities worldwide and all available company data (valuation, profitability, growth, analyst estimates, risk indicators, momentum, etc.). The more data the learning algorithm has available for training, the better and more stable the achieved return forecasts will be.
- Data preparation
In order for the algorithms to be able to learn efficiently, the raw data is generally enriched and further prepared with supplementary key figures. This can be done, for example, by transforming or neutralising the data.
- Training of machine learning algorithms
In this step, the algorithm learns independently from the prepared data. It analyses which combination of company key figures leads to above-average returns and determines the characteristics of equities with poor performance. Based on these findings, the algorithm then automatically derives complex models that can be used to estimate future price developments for all equities worldwide. These comprehensive forecasts are hardly possible for humans, as a large number of analysts would be needed to make daily return forecasts for thousands of equities.
- Aggregation
To further increase the probability of return forecasts materialising, different types of algorithms are combined. In this way, models can be linked that are specialised in short-term (tactical) and long-term (strategic) forecasts or have their strength in upwards or downwards trending markets. Dynamic aggregation of different prediction models leads to more stable and reliable forecasts.
Artificial intelligence – innovative technology for financial investments
Equity strategies based on machine learning algorithms tap into sources of alpha that were previously hidden from humans due to their high complexity.
These strategies adapt quickly and dynamically to changing market conditions and can generate additional returns even in difficult phases – such as the IT bubble, the financial crisis or the coronavirus pandemic. The excess returns determined in back-testing are promising and also only slightly correlated with the returns of traditional equity funds that pursue a fundamental approach. Compared to systematic investment strategies, which usually have permanent style orientations such as value or quality, strategies based on machine learning do not have permanent style bets.
So far, there are few publicly available equity products that use machine learning algorithms for stock selection. An investment solution will be added in autumn 2021: Swisscanto will issue artificial intelligence equity certificates for the first time, the selection process of which is based entirely on artificial intelligence. With this innovation, access to previously hidden sources of alpha is open to all investors.