Raphael Steigmeier, managing director at VT Wealth Management, reports that 0.5% of shares contribute three quarters of the performance. Artificial intelligence outperformed by 50% during the coronavirus crash
Raphael Steigmeier:”We know that unexpected shocks remain possible, whether it is an OPEC decision or a pandemic. Stress scenarios, like the last few days, show that our model is solid.”
Raphael Steigmeier, managing director at VT Wealth Management, reports that 0.5% of shares contribute three quarters of the performance. Artificial intelligence outperformed by 50% during the coronavirus crash.
Raphael Steigmeier has scheduled his first game at the age of 12. Today,in his role as manager and architect of artificial intelligence at VT Wealth Management in Zurich, he combines the traditional management of a private bank with new technologies.
VT Wealth Management is a wealth management company, with more than one billion francs under management. It was created in 2008 by Thomas Fedier, now Chairman of the Board of Directors,and since 2013 it has been run by the latter’s son, Sacha Fedier..
How have you weathered the 2020 crash so far? The current situation and the economic consequences are worrying.. But we remained rather relaxed in the face of falling titres markets, under the resilience of our stock selection. .
VT Wealth Management’s strategy combines traditional wealth management with the use of new technologies. Customer management is individualized and personalized.. The use of artificial intelligence and algorithms is subtle and non-bureaucratic. That’s why we fortunately don’t have stocks that we wouldn’t really like to hold in our portfolios. Our securities have become more affordable because of the decline in the market,without them becoming really cheap. If they had become, it would mean that they do not have the quality they are looking for..
While the global equity index fell by 32%, our investment fund gave up only 26% and our long/short strategy even gained 18%, achieving an out performance of almost 50% compared to the n’a index..
Is this strong out performance the result of chance? No, we can rule it out. Our approach,and the resulting model, has been extensively tested and designed generically. This means that he has been trained to recognize general trends, but that he is not troubled by low nuances. In short, our system is not designed to recognize a poodle and distinguish it from a German shepherd. . But he is able to see that a poodle and a German shepherd are both dogs.. From the moment it was designed, we gave up optimizing it down to the smallest detail.. I don’t know why. Critics of our type of model accuse it of an excess of conditions to be fulfilled so that it fits exactly with the historical behavior of the markets. The grievance is justified to the extent that many systems completely lose their relevance as soon as a new event, a black swan, occurs. We preferred to build it in such a way that it behaves well in all situations, including unforeseen and unprecedented events. .
Is it a system that can improve without human intervention? ? Yes, yes. It incorporates a lot of artificial intelligence. .
What do you mean by that?? For example, humans can easily recognize the difference between a dog and a cat. But if I show these two animals to my computer, he won’t make the distinction and he won’t be able to identify them. By making it available 100 million images of dogs and cats, we can train the computer so that it understands the difference. We have done the same with equities so that our model distinguishes between good and bad securities..
For decades, hedge funds have been trying to build such models, but accidents follow one another, like LTCM in 1998. Aren’t black swans inherently unpredictable?? Absolutely. Events such as the coronavirus outbreak were not predictable.. The key is to understand how the model is built.. If it is generic in nature, it will behave better in the event of an unexpected event than a specific model such as the LTCM model. The latter, based on the assumption of a convergence of interest rates in Europe, was not bad in itself.. But the profit margin was so modest that the fund needed huge debt to get a reasonable profit. It only takes a little for such a system to be destabilized.. That is what happened. What’s your approach? From a particularly large universe of data, it is enough to identify a small number of actions that are,with a high probability, better than the others.. Bessenbilder and Al, in August 2019, presented research work over the period from 1990 to 2018 which proved that only 40.5% of the 61,100 securities analysed offered a return higher than government bonds. . So 0.5% of the total companies are responsible for three quarters of the performance. The Swiss equity investor can have a passive strategy, with an ETF, or active and select for example five securities without knowing if they are the best. Our model aims to recognize these five”best” values.
We know that unexpected shocks remain possible, whether it is an OPEC decision or a pandemic. Stress scenarios, like the last few days, show that our model is solid..
How are your models built? Our models are based on traditional risk factors (criteria based on dividends, momentum, size, value, quality, growth). We add artificial intelligence so that the computer discovers for itself the best possible combination of titles. A company that pays a high dividend may not be as attractive as another company that offers a lower return for the same level of profit, but makes more investments to increase its long-term growth. Classical theory would lead me to favor the company with a high dividend, but the computer could choose the second if it considers that it offers the best opportunities.
In addition, our approach learns to distinguish between industries. We carried out monthly analyses for fifteen years for more than 60 data (profitmargin, taxes, profit growth, etc.) within 5000 companies. We collected an estimated 70 million data to deduce a company’s likely future behaviour compared to that of other companies..
Do other managers work in the same way? To my knowledge, very little.. Some hedge funds used a similar approach in the 1980s, such as Rebellion Research, but in Switzerland the subject is just beginning to emerge.. Having said that, I stress that we are not a hedge fund, but traditional managers who use new technologies.
You are based on data on fifteen years of the life of each company.. Don’t you omit much of its culture and history? We don’t integrate older data, but the computer learns from its errors. The current data are the most important. In the 1990s, for example, banks and insurance companies traded at a price equivalent to three times their intrinsic value. . That has change dat the latest since the financial crisis. These securities are now trading at their intrinsic value. The system notices these changes.
Your model is based on the Fama Nobel Prize theory and market efficiency, but behavioural finance has en challenged this work. How do you take that into account?? Yes, I do. That’s why, pour if someone tells me they can predict everything 100%, I’ll tell them with a 100% probability that they’re wrong. Behavioural finance shows that investors can over-react or misjudge events. If a company makes a profit, the stock may go up. But this information may not be fully integrated from day one.
The evolution of a share price contains information on what we do not know, for example the trends of the oil market. . Since last fall, we have been negative about oil stocks because of the signals sent by our system..
VT Wealth Management, with 21 employees,does not rely on a research team the size of a large bank.. But if I receive the recommendations that accompany the studies of these institutions,I remain quite skeptical,because I do not know how the analyst came to his conclusion. The tone of the employee sometimes differs from his final recommendation. I sit because of the bank’s business relationship? I’m wary of this type of filter..
Is your use of the term n’est “artificial intelligence” not a marketing exercise? No, it’s not. It is only with the use of artificial intelligence, big data and conventional algorithms that we can simultaneously analyze thousands of actions. We then manage to prioritize them, like bank analysts, by “buy,””conserve” and”sell.” Unlike the bank analyst, however, we have no potential conflict of interest. Our computer has no emotion, so our ratings are consistent.. These allow en us, as asset managers, to put into practice active management and to have a clear opinion on equities, which sets us apart from our competitors..
What are the most advantageous stocks today? In the United States,we avoid oil companies like Schlumberger, Chevron, Occidental Petroleum, big banks like JP Morgan, Morgan Stanley. On the other hand,we like United Health, Microsoft, Visa, Mastercard, so fintech rather than investment banks. . In Switzerland, Nestlé, Schindler, Bucher, Sonova,Logitech, Conzetta or Straumann are for purchase..
Hedge funds suffer from their high fees and poor performance. Do they still have a future? We are not a hedge fund, but this branch of finance has been around for almost a century and has been going through several crises. The goal of hedge funds is to contribute to the stability of performance. Their interest lies not in their possible annual outperformance but in their ability not to suffer sharp declines.. Suppose I only have two shares, one bullish position (long) and one for sale (short). I can’t win if both go up the same.. Better like today’s highly volatile markets. The performance-related fee structure is only practiced if the objective is achieved. Our long/short product outperforms by 50%. No one is obliged to buy it..
How do you integrate the GSS? I consider environmental, social and governance (ESG) criteria from a performance perspective. There is a link between a company’s data count and its performance. In emerging markets,the contribution of the ESG approach comes mainly from the G (governance)and less from environmental and social criteria because it avoids scandals. Our selection does not include a durability ▅ filter, but we can take this into account at the request of a customer. .
“It is only with the use of artificial intelligence, big data and conventional algorithms that we can simultaneously analyze thousands of actions”»