Current location - Trademark Inquiry Complete Network - Futures platform - How to evaluate Qlib, an AI quantitative investment platform developed by Microsoft Research Institute?
How to evaluate Qlib, an AI quantitative investment platform developed by Microsoft Research Institute?
Quantopian was so popular that Point72 invested in him and SteveCohen's resources were used by him, but it went bankrupt this year. The reason is that the profit model of the quantitative platform is problematic. First, unprofessional. Second, because it is unprofessional, people who use it do not make money. Cubist owned by Point72 is very profitable, but Cubist will not give infra to Quantopian, because infra that can make money is a scarce resource. So Quantopian's framework is very amateur for professionals. It is precisely because amateurs and professionals don't need it, and amateurs can't make money with Quantopian, so they can't share it with the platform, so such a platform can't be profitable.

There are precedents for Microsoft executives to quantify. Former Microsoft COOKevinTurner used to be CEO of CitadelSecurities, but he didn't make much achievements. Finally, Griffin appointed Zhao Peng and Zhao Peng promoted CitSec.

Therefore, it is extremely difficult for programmers to invest directly across borders without knowing the routine. It is not because programmers are not skilled, but mainly because they don't understand the investment and research system. A good programmer can become a good QR only after training. For example, what kind of person is james simons? He has long wanted to be a stock trader. A group of scientists have been working hard, but there is no progress, and there has been no progress for several years. At that time, the stocks of PDT and Deshaw were much better than RenTech. Finally, RenTech did a good job in the strategic framework of statistical arbitrage for former PDT employees, and then another programmer who knew both structure and stock adjusted the details of the strategy to get such a great medal.

Take statistical arbitrage as an example to talk about why this QLib platform of Microsoft is doomed to failure. The core of statistical arbitrage strategy is signal. How to mine meaningful signals from various data sources and how to test the validity of signals are all statistical categories. The simple volume and price signal of A shares is still very good at present. As long as it is an organization that understands routines, it is no problem to score more than 30 points in the past two years. However, it is not so easy for US stocks to do statistical arbitrage. Most simple quantity and price signals are useless, and the grand medal returnonGMV can't score 10. A mature market, it is difficult to dig out some signals without understanding the market.

In the model, programmers with machine learning do have some advantages in parameter tuning, but how to deal with labels and how to engineer features may not be available to ordinary people.

In other aspects, how to control the style and how to use the algorithm to place an order requires practical experience. Without understanding the market, it is too difficult to build a printing machine by some fancy algorithms in machine learning. Nowadays, many private investors say that deep learning is so useful and will be useful, but it is definitely not so magical. In fact, everyone is doing the same thing, don't go too far. If deep learning is really so useful, who will talk about it everywhere? Quantification, a really useful thing, will eventually be known by peers, but few people will talk big in public.

To sum up, quantitative strategies involve statistics, data mining, trading, market understanding, machine learning and many other aspects. It is not so easy to stir up the market by a set of machine learning algorithm library and an optimizer.