Through its immense capacity for data collection and analysis, artificial intelligence (AI) has quickly become a tool of choice for portfolio managers. And the Caisse de dépôt et placement du Québec (CDPQ) did not miss the opportunity that presented itself, quite the contrary.
Posted yesterday at 12:00 p.m.
Jean Gagnon
Special collaboration
Already 25 billion of its depositors’ money is in global stock market portfolios which have been the subject of systematic management for two years using AI tools. But does this approach carry additional risks compared to more traditional management?
“We use our own algorithms, and our results exceed those of our benchmark index,” assures Jean-François Bérubé, vice-president, quantitative strategies and data science at CDPQ.
The benchmark index for CDPQ stock portfolios covers an investment universe of 1,500 companies. And the first step is to establish a forecast on what the stock market performance of each of these companies will be, explains Jean-François Bérubé.
AI provides the Caisse with 75,000 data points every day. This involves classic financial data, i.e. all that appearing in the detailed financial balance sheets of companies, but also data on market behavior such as prices, volume and volatility of each of these securities, and finally several sentiment indicators. investors.
Once the forecasts have been established, we move on to the second stage, namely the construction of portfolios.
In each portfolio, we will ensure great diversification, control all risks, such as geographic risks, in order to avoid unpleasant surprises, and risks linked to the size of the companies.
Jean-François Bérubé, vice-president, quantitative strategies and data science, CDPQ
The analysis that is made of each company in the systematic portfolios is uniform or equivalent, but given the very large amount of data it includes, it will be less in-depth than that for the Caisse’s other traditional stock market portfolios. “To compensate for this factor, we take a large number of positions, around 300 per portfolio, in order to reduce the risk given that our convictions will necessarily be lower,” explains Jean-François Bérubé. Traditional portfolios have a smaller scope, but the analysis is more in-depth.
A large human component
When we talk about AI, we immediately think of heavy use of IT equipment, but many people work in what we call the “systematic fundamental management group” at the Caisse.
“It’s not magic,” assures the vice-president. Their daily life is certainly interesting. First there are those who follow up. How is the performance? Why do we have such results? Does the model do what we want? There are also those who negotiate. All they have to look at is the impact on portfolios, trading costs and fees, performance and above all how we can be better. How will our models today be even better tomorrow? It’s a lot of ongoing research. It keeps a lot of our people busy,” explains Mr. Bérubé.
On the lookout for generative AI
The Caisse de dépôt wants to be at the cutting edge of all the new features that can improve portfolio management. Generative AI seems destined to provide immense possibilities. For example, the ability to quantify textual data, or to transform information from text into numbers and use it in the algorithm.
The calls from company executives following the telephone calls that accompany the disclosure of quarterly results are a good example. Generative AI helps examine the complexity of language. It is assumed that a management team with simpler language generally performs better. “And since the performance gap is sufficiently large between the simplest and most complex languages, this data is part of the 75,000 pieces of data that we look at,” says Jean-François Bérubé.
