On January 17th the ACATIS AI Global Equities was rebalanced. What is new?
The investment process has not changed since the rebalancing of July 2019: Our AI machine analyses almost 4,000 companies of developed countries with a market capitalisation of at least EUR 1 billion. Companies from the real estate and financial sectors are excluded. A graphical neural network is then trained to select the best companies of each sector to create an equally weighted portfolio, consisting of 50 stocks, long-only, with cash close to 0%. The sector and regional weights of the MSCI World Index are thereby reproduced as good as possible to minimise the exposure to currency and sector effects.
While the trades are executed by humans, the stock picking is performed entirely by the AI model and not overruled by humans.
Our AI model is a graphical neural network which consists of interconnected nodes. Each node corresponds to a company from the same sector and is described by its fundamental properties/features (sales, profits, book value, assets, etc.). During the learning process, 20 companies from the same sector are randomly selected. Then, the graphical network is trained to identify the outperformer over several time spans; based on the fundamental data of the company. By repeating the random selection, several hundred thousand different constellations are generated, providing a rich source of information. Eventually, an ensemble of models is trained, by repeating the entire procedure several times.
Once the models have been trained, a company such as Apple for example, competes against other companies of the IT sector and the AI model predicts a probability of whether Apple is likely to outperform its peers or not. By correctly aggregating the results of many different constellations, the winners of a sector are identified.
In the new portfolio only 2 companies remained (Inter Parfums and Interpublic Group of Companies), 48 companies were exchanged. FAANG stocks are still not represented. Only 14 companies of the portfolio are represented in the MSCI World Index. The remaining 36 companies exceed the MSCI World universe.
Due to the portfolio construction described above, we have a maximum over-weight/underweight of 4.3% in certain regions and a maximum overweight/underweight of 3.6% in some sectors compared to the MSCI World Index, mainly due to the exclusion of financials and real estate.
From a factor perspective, the new portfolio looks like the old portfolio. The AI focuses on small companies: on average, a stock is just under USD 8 billion in size [compared to USD 28 billion in the MSCI World Index]. In addition, the new portfolio continues to focus on value. According to factors such as price-to-earnings (P/E ratio), price-to-book (P/B ratio) or price-to-cash flow (P/C ratio), the portfolio is on average cheaper than the average in the MSCI World Index. Other characteristics of the portfolio are that the dividend yields are lower, debt is lower and liquidity is better than the average of the MSCI World Index.
Our disclaimer applies to the report.
