Daniele Grassi, CEO and Co-Founder, Axyon.AI
By now, most investment managers are aware of the benefits artificial intelligence (AI) can bring their business. Using sophisticated algorithms, asset and hedge fund managers can save significant time building investment strategies by using these tools to analyse vast data sets in secondsto enhance their decision-making processes.
This ability to analyse and digest huge data sets, which can span across investment universes, languages, orcountries, has been a breakthrough for many funds. It has streamlined a process which would take a human being days, or even weeks.
But more data doesn’t always mean better results. And vitally, all businesses using machine learning or AI technologies to support investment decisions must ensure proper data preparation to guarantee the best outcomes.
Quality over quantity
AI works with data which could be laid out in a different way fromthe data sets used by humans. While the benefits of harnessing both AI and manual processes to feed into strategiesare substantial, it is important that managers ensure that both the data used by their team, and the data sets used by their AI provider, are up-to-date, correct, and sensible –otherwise,they may risk inaccurate results.
As an example, one important step in the investment process is to build an AI model using point-in-time data, which means themodel is fed only with historical data as recorded on each historical moment (i.e. not revised ex-post), whichmitigates look-ahead bias risk. This way, we make sure the strategy’s historical performance is unaffected by data which could artificially cause the results to look better.
It’s therefore essential that managers carefully evaluate the quality of the data theyinvest in. Purchasing raw data costs money, which is then followed by the additional costs of time and resources spent to clean this data for use.
Safety equals success
Data is at the centre of every AI engine. Although uncovering gold is an exciting concept, in our field, choosing the right data, verifying its reliability, and checking it is continuously up-to-date are probably the most critical steps of the whole workflow.By using AI algorithms based on reliable data, financial services companies can integrate reliablepredictions into their investment processes.
It’s important to note that AI algorithms are not magic; they are mathematical functions that work to minimise a cost function by doing the best they can with the data they have. Unlike humans, they will not act on “hunches”.
Explaining explainability
The explainability of AI models is a widely debated topic at an academic and industrial level, and several techniques have recently been proposed to address it. Highly effective techniques have been developed to identify which subsets of data have contributed most to the decision made by an AI model.
Complex machine learningsystems are, by nature, nearly impossible for humans to understand. But without some level of explainability, it is hard for human users to justify and trust the results and output created by machine learning algorithms. By using clear, accurate data sets and explainable algorithms, managers can be confident that the system is working as expected, and explain these to clients or the regulator should it be necessary.
Overall, it is clear that integrating AI signals into a fund’s investment processes can bring a great deal of success, both by streamlining existing processes and opening the door for the analysis of much larger data sets than would be possible by humans alone.
