Home > Alumni Views > Alumni Views | QIAO Ye: Projections on the trends of AI and the stock market (HOPE keynote speech at the Innovation and Entrepreneurship Conference)

Alumni Views | QIAO Ye: Projections on the trends of AI and the stock market (HOPE keynote speech at the Innovation and Entrepreneurship Conference)

Mon, Oct 26, 2020

First published by Tongji University Alumni Association November 12, 2019

How could we see the true face of the ever-changing stock market? Tongji alumni QIAO Ye has devoted most of his time in developing artificial intelligence that could be used in quantitative investment. He tries to sort out logical connections and patterns that are beyond normal human think patterns, to overcome greed and fear that accompany any stock investment decision, to foresee a definitive future through every ups-and-downs of the stock market by building and using machines with deep learning capability.

Speaker Profile

QIAO Ye

Class of 2006, MBA Program, Tongji University

Class of 2019, Ph. D candidate, AI research discipline, Computer Science Department

CEO, Tongliang Zhineng

Hello, everyone! I’m Qiao Ye, and I specialize in AI-facilitated quantitative investment. We know the Chinese stock market has been slacking in the past two years, however, AI-facilitated investments have yielded satisfactory results. I’m honored today to have the opportunity to discuss with you some of the AI applications we have used in stock market investment.

Recognize the Dunning-Kruger Effect and Overcome Greed and Fear

Let’s take a quiz. Those who have invested in the stock market and those who are ready to make an investment, please rate your investment ability in one of four levels—high, above-average, below-average, and low. According to statistics, most stock investors would rate themselves above average, in other words, most people believe they are better at making investment decisions than average people. This is a psychological effect called the “Dunning-Kruger Effect”, it is a mishandling of self-perception. Most people would overestimate themselves while underestimating the people around them. If all investors have better than average abilities, then should there be no one left to lose money in the stock market?

Let’s take a look at the real distribution of investor profit/loss, 10% of investors could make a profit, 20% could make even, 70% would lose money. In a nutshell, it is tremendously difficult for 90% of the investment crowd to make money, and such is the truth. Most investors have strong self-confidence in their trading abilities, and they conduct buying and selling of stocks on a high frequency. Let’s review the data from 2017, the turnover generated by retail investors accounted for 82% of the overall trading volumes, but only 9% of the profits from those tradings went to retail investors. 82% turnover for merely 9% of the profits.

Why is it so difficult for retail investors to make money? Because making money from the stock investment would demand the investors to forgo their common sense. Warren Buffet has once made a famous quote about “Greed and Fear”. We are all common folks, and when we are common folks, we follow the common sense. Buffet’s quote requires an extraordinary investor to keep a sense of fear when everyone else is greedy, and when everyone else is in panic, the investor should show his greed. Such demand goes against any normal human think pattern, I think only an investment guru or a robot could stick to such requirements.

Artificial Intelligence, the Master of Portfolio Investment

How would an investment guru or a robot overcome greed and fear when committing to stock investment? The answers lie with how they perceive the question in hand. I will focus on two major issues, one is called portfolio investment, and the other is called go with the flow.

The keys to portfolio investment are probability distribution and risk divergence, and the whole idea is to lower risks while maintaining a certain level of investment return. Some of you might ask, investment profit is always in proportion to risk taken, how could there be such a free lunch? There is. An economist named Markowitz has won the Nobel Prize for his research on assets portfolio management models. The extremely popular internet celebrity Ray Dalio of Bridgewater has also mentioned divergent investment in his best-selling book, and he called such model the Holy Grail of investment. The question is, what kind of distribution or divergence would be considered the reasonable kind for asset portfolios?

We all know George Soros. Another great figure just like him is a mathematics professor at the State University of New York at Stony Brook, James Simons. Simons run the Medallion fund with an annualized return of 35%. Most Wall Street hedge funds would look up to the Medallion as their model. Simons attached great importance to portfolio management. His fund will buy in 3,000 stocks in a single day and conduct over 10,000 trading actions. Investments made at such diversity and such quantity are meant to be executed by AI, not by man. The companies run by Simons like to hire scientists who are adept at linguistics and language recognition, and their research focus is AI models such as the Hidden Markov Model.

It is tremendously difficult to ask a single investor to familiarize himself with various investment-level assets. The same is true that we didn’t see many people who are both masters of chess and Chinese chess at the same time, but AI can be both. AlphaZero is an all-capable chess master. Let me make a brief introduction of the Alpha family, the first generation is called AlphaGo, and it is solely dedicated to studying the game of Go. The second generation is called AlphaGo Zero, what’s the difference between the two? The first generation requires manual inputs of human chess plays into the AI, and it takes a long time to train the AI. The second generation no longer requires human inputs, and two Ais battled each other. In about 40 days, the second generation learned how to beat the first generation. The third generation is called AlphaZero, and it studies more than just the game of Go, and it learns how to play multiple types of chess games. The fourth generation is called AlphaStar, it studies how to play computer games such as Starcraft.

So, the machine beating a human at a game is not a big deal. I believe the truly meaningful story from AlphaGo is machine learning. The rapid development of machine learning has narrowed the gap between different levels of human intelligence. In the past, the teacher must master the knowledge before it can be passed down to students. As we make a stock investment now, we all use computers, but if the investors did not understand the principles or technicalities of stock investment, the computers are useless. With the birth of machine learning, we can delegate AI to study the things we do not yet understand, and AI could find logical connections that are beyond human think patterns.

Further still, I used to believe that only those born with special talents could truly understand investment gurus such as Buffet and Simons. People with no special talents such as myself, we learned years after years and still didn’t make progress, and I almost gave up. It is AlphaGo that gave a sense of hope. I believe with the emergence of AI, it can help me study investment gurus, help me catch up or even surpass those who are born more intelligent. There was an earlier research project called deep blue, and the program is solely dedicated to studying how to play chess. The developer conquered the world chess champion after the launch of Deep Blue, but the fact remained that the developer himself is a world-class master of chess. The developers of AlphaGo also used a computer to beat world champions, but the developers are amateurs at chess.

Go with the Flow: Image Recognition and Big Data

Secondly, let’s talk about going with the flow. Chinese stock investors favor the stock-picking approach, especially picking out the “bull market stocks”, but if an investor spent all his time pick individual stocks, or even dreams about finding a bullish stock during a bearish market, that would be going against the flow. During a bearish market, go with the flow dictates the investors either make no investment or short the market. Retail investors prefer the stock-picking approach when professional institutional investors pay more attention to market timing.

The Wave Principle is an easy-to-use approach to do market timing, but counting waves is a mountainous task which some would say a thousand people could have a thousand versions of wave counts. We are trying to use AI-based image recognition technology to count waves, and we believe it could help our investment go with the flow.

Imaging technology has achieved rapid growth in recent years. If we line up six mugshots in a single slide, would anyone be able to tell the difference between real persons and virtual persons? In fact, they can all be virtual, yet every single one of them looks real. We can hardly identify virtual persons by the naked eye. We have found a picture of a baby Elon Musk, but it turns out the whole video clip was made with AI.

The two technical tools used to generate virtual persons and simulated videos were based on the same technology called the Generative Adversarial Networks (GAN). It is one of the unsupervised learning models with great potential. This technology has already been widely applied to making videos.

We also trained AI to conduct Bitcoin trading. AI research requires a huge amount of data input, in other words, if we can not acquire big data then we can not perform AI research. Some of our researchers on Bitcoin trading has run into this headache, they find the training of AI to be highly inefficient due to lack of sufficient Bitcoin trading data. With GAN, we first train the AI to generate virtual trading diagrams, then we use these diagrams to train the AI to trade. The accuracy of AI trading could become higher and higher as long as we can provide continuous data inputs.

As mentioned earlier, retail investors generate 82% of the total trading volume in China, and they are the determining factor of the stock market trend. We have developed another system to track retail investor behavior through search engines. Data related to retail investors are enormous and fragmented, and the only way to collect such data is through search engines. A few years back, some research institutions used web scraping algorithms to collect internet sentiment, but it was difficult for them to yield results due to the limitations of computing power and algorithms. With the rapid growth of AI, a technology called Convolutional Neural Networks has sprung to life. It allows us to achieve much greater computing speed in exchange for a small sacrifice of computing accuracy.

The field of AI research is like a battleground for algorithms, just like restaurants, whose superiority is determined by the chef’s culinary talents. If the chefs’ culinary levels remain the same, then the restaurant with better ingredients will produce better food than the other. By the same comparison, algorithms are key to AI research, and so is the innovative quality of the original data. One of Google’s engineers developed a system to track Trump’s tweets, and then paring these data with the performance of the stock market. At first, I thought it was flimsy at best, but now it appears the algorithm is right on spot.

They use Google and we use Baidu, and we have an actual running algorithm that has been put into work. The idea is that we use Baidu to track the daily change of the number of searches of keywords such as “stock account opening”. What does this tell us? It tells us the speed and number of new account openings, which is directly linked to the ups-and-downs of the stock market. In 2015, the Chinese stock market fell from bullish to bearish, and the data generated by this algorithm function very well as an alarming signal. Using search engine data to analyze the Sheep-flock Effect of the retail investors gave us a very practical market timing system. We put this system through the investment competition organized by Guosen Securities this year, and we won the monthly championship of 2019.

At last, I would like to briefly touch on the topic of where stock investment AI will go. We have two pictures taken at the exact same spot, the Fifth Avenue of New York City, one was in 1900 and we can see horse-drawn wagons during the Easter Parade, and another one was in 1913 and the wagons are but gone, replaced by automobiles. It may sound incredible that retail investors could use AI to make a stock investment right now. But if we look back 20 years, in 1999, personal computers and high-speed network were rare, it was considered avant-garde for an investor just to check his investment on a computer console. Now, every investor is equipped with personal computers, and maybe in a not so distant future, AI stock investment will replace the current way, same as automobiles replacing the wagons. With the rapid development of AI, perhaps someday in the future, every stock investor could have his/her stock investing AlphaGo. Investment is always accompanied by risks, and you should take caution when making investment decisions. Thank you all!

(The content is transferred from the official WeChat account of Tongji Alumni Association)

X Thank you for your interest in Master of Global Management, Tongji University!