Saturday, 28 October 2017

Could AI make us all rich?


During the past week I found an interesting article about an ETF that is managed by AI. The ETF is all about IBM's Watson-platform managing the portfolio by analyzing one million market signals and news articles from 6000 companies on a daily basis. It uses a portfolio of 60-70 companies that Watson selects by optimizing lower risk and capital appreciation.

The concept is so fascinating that I couldn't help myself thinking it further and write about this. Could this kind of AI make human portfolio managers unemployed some day? Could this kind of ETF be more profitable in the long-run because it has no human error in portfolio optimization?

For these questions I had to consult my AI-expert (who we can call with an alias Kolari for example) so that I could have e better perception about an AI as a portfolio manager.

It is true that the AI is free of human emotion. It doesn't experience greed or fear, the two most common feelings in the stock-market. The AI doesn't make human errors in picking companies and it doesn't sell the equities in panic in case of a market crash. Similarly, it doesn't buy stocks that are hyped by third party analysts if it sees that the stocks are overvalued or present a risk that isn't valued to the price of the stock. The stock-prices are calculated by robots and transactions are made by robots (expect in NYSE), so why wouldn't stock picking made by a robot?

As stated in the article, in theory the AI should become more efficient and more "smart" as time goes by. The AI is placed to the context of financial markets, where by analyzing data continuously, it learns more and more about the movements of the financial markets. It will become more capable of making right decisions when it has figured out what inputs create wrong decisions that produce loss and what inputs create decisions that generate profit. It is also more efficient to analyze news, reports and social media and AI requires less time to process them than a human portfolio manager.

This has been witnessed before. In the video-game world several AI's have been created to observe and learn from the best players of certain strategy-based video-games. When the AI has observed millions and millions of games and transactions from the world's best players, it can copy those strategies in order to beat other players and to learn even more from the context in which it operates. It can also copy itself hundreds of millions of times to be able to play against itself in millions of different servers. Handy, am I right?

This is where the first question arise. In what transactions will the AI actually learn while managing a portfolio? In theory it can copy the other "human players", the investors in the market but they certainly make mistakes some time. Copying them would make the AI repeat the same mistakes. However, the AI is still able to learn from them.

What Kolari suggested is that this AI should be thought perhaps as a more sophisticated analytics. In terms of that the analytics should work just fine in certain circumstances. However, even with modern technology it is almost impossible to know in which way the stocks are going. Furthermore, this kind of technology would base also partly on historical information such as decisions made in the past and on what circumstances the stocks have gone up. This means that every single thing that affects them should be inputs in this system. But we all know that historical information cannot be used as a certain way of predicting where the stocks are heading.


Perhaps the biggest reason why I'm being careful on the hype for AI-investing, is the existence of unexpected shocks. A great amount of uncertainties exist in the stock market. Have you ever wondered why the stocks start sky-rocketing if a purchase offer or a fusion notice are announced? It is because no-one could expect that these events could happen. Usually the biggest volatility in the stock-market occur because of either positive or negative surprises. I highly doubt that even an AI could expect unexpected circumstances no matter how much information is gathered. The point is that the right information couldn't exist at all before the unexpected surprise.

Still, the AI portfolio management could work and it could be the next revolution in the financial markets. It would make the AI strategy transparent for all investors, not just hedge funds that might have use this kind of technology before. However, the development of this AI could take years or even decades before it would become efficient enough and making the minimum amount of errors in portfolio optimization. We don't have much data since this AI started trading on October the 18th, only a week ago. I'm following the ETF regularly and if I'm crazy, I might put a couple of euros to it and see where it is taking me.

But I'm not yet fully convinced. A guess is more proficient when it is backed with mathematics, but in the end of the day, it is still just a guess.

Text: SW
Pictures don't belong to me

No comments:

Post a Comment