AI stocks, winning percentage geometry？

□ Compared with most people, AI can indeed achieve a more rational investment state. Especially in the field of quantitative trading, AI will have a more stable output for trading strategies that need a lot of analysis and processing data in a short time.

□ However, AI can’t beat the best people in the market. Humans have their own advantages, such as processing a lot of unstructured information and judging a lot of non-quantitative behaviors and states.

□ Investing in this field can’t be achieved by throwing money, smashing machines and smashing equipment. These are not core competitiveness. The core still depends on cognitive depth, independent thinking and innovation, and even some beliefs.

□ ChatGPT can accelerate the realization of ideas through efficient data collection and processing capabilities, programming capabilities, text analysis capabilities, etc., but the idea itself is the key to investment.

Remember the world’s first ETF fund invested by artificial intelligence (AI) launched by Wall Street after AlphaGo defeated the world champion of Go? The financial sector’s "Alpha Dog" investment performance is not ideal, and it has not made the global investment managers lose their jobs. Nowadays, the more powerful "generative AI" represented by ChatGPT is born, which inevitably makes people wonder whether "AI stock trading can beat the market".

Even if we can’t "beat the market", some investors worry that AI stock trading will aggravate market volatility and make it more difficult for small and medium-sized investors to make profits.

How does ChatGPT predict the stock price

Since the world’s first stock market was born in Amsterdam, the Netherlands more than 400 years ago, it has been the dream of all investors to "beat the market"-accurately predict the stock price trend.

The price change of a stock is determined by many factors, among which there is a very complicated and nonlinear relationship. In the past, the parameters of AI model were small and could not represent complex market relations. The ChatGPT model is obviously different from the previous AI model for forecasting stock prices.

According to the interviewees, compared with the traditional AI model, the ChatGPT large-scale pre-training model has many advantages. In the field of stock price forecasting, the big model can handle a large number of heterogeneous data, such as stock trading data, macroeconomic data, corporate financial reports, etc., and can also handle unstructured data, such as news reports, social media information, etc., which enables the big model to capture market information from many aspects and improve the accuracy of forecasting.

Specifically, when making stock price forecast, investors often pay attention to four aspects: technical aspects, fundamentals, news and events, and market sentiment. Based on the above dimensions, the generative AI technology represented by ChatGPT has brought some new changes and potential application forms:

First, in terms of fundamental analysis, because AI technologies such as ChatGPT have made remarkable progress in natural language processing, it can better understand and deal with the complexity of human language. In investment, this ability can be used to analyze and understand text data such as financial statements and company announcements, thus providing more comprehensive and accurate information for investment decisions.

Secondly, ChatGPT is applied to Sentiment Analysis and market sentiment prediction, which can identify the emotions and emotions of market participants by analyzing social media, news and other contents, and help investors better understand the changes of market sentiment and predict its impact on stock prices and market trends. Recently, a study published by the School of Finance of the University of Florida shows that integrating ChatGPT into the investment model can predict the trend of the stock market. Its research method is to provide ChatGPT with a large number of news headlines and contents, so that ChatGPT can judge the impact of these events on the stock market with emotional analysis.

Thirdly, in the field of stock investment, there has always been a technical school, that is, judging the future stock price trend through the trend of K-line chart, which requires a lot of re-examination after the close, and image recognition technology can replace this work, that is, by giving AI a large number of K-line samples, and each sample has a classification label for future ups and downs, convolutional neural network technology automatically looks for features useful for future ups and downs classification from the K-line chart, and feature extraction and verification are automatically completed.

"Technologists often look for forms that break through new highs, including bottom features such as backsliding, arc bottom, and bottom volume. However, these forms lack strict validity tests, and they are often false breakthroughs in practice, and it is easy to fail to follow suit. AI technology is not the case. Instead of looking for these features, it automatically mines features from the pixel level and directly matches the classification results. " An Ningning, chief analyst of financial engineering of GF Securities, found that according to the rising probability value predicted by AI, the winning rate of the group with the highest score and the group with the lowest winning score is about 89%. However, this winning rate can only be rich enough in excess returns if the number of decisions is very frequent, that is, quantitative high-frequency trading.

"In the final analysis, ChatGPT is still a big language model, which will make language processing easier." Liu Xin, founder, chairman and CEO of Kuanrui Technology, said that ChatGPT, as a large language model, is more suitable for general use than a small language model. Generative text abstracts can accelerate investors’ analysis and understanding of research reports and papers. ChatGPT can generate some more direct and accurate analysis for public opinion analysis, emotional analysis and event-driven strategies.

Has been put into actual combat in quantitative trading

"We may stand on the eve of the greatest change of this era." This is the opening statement when Magic Square announced in April this year that it would concentrate its resources and strength on AI. Undoubtedly, seizing the AI highland has become the consensus of domestic quantitative private equity leaders. In their view, AI technology will become the core engine of the quantitative investment industry, and even subvert the technical pattern of the quantitative investment industry.

The technical iteration of quantifying investment is basically synchronized with the technical iteration of AI. Wang Xiong, founder investment director of Siyuan Quantification, said that the history of AI iteration can be summarized into four stages: 1.0 multi-factor stage based on linear regression; 2.0 high-frequency quantity and price factor mining stage based on machine learning; 3.0 end-to-end structured data mining stage based on deep learning; Quantization stage of 4.0 depth fundamentals based on general artificial intelligence.

Wang Xiong believes that compared with the traditional fundamental quantification based on financial statements, there are four major differences in the 4.0 stage:

First, the data sources are different. Traditional fundamental quantification mainly depends on the company’s financial statements for analysis; The deep fundamental quantification can obtain more detailed information and market sentiment by mining the company’s public information, including company announcements, analyst reports, social media and other unstructured data.

Second, timeliness and data frequency are different. The traditional so-called fundamental analysis based on financial statements has low frequency and weak timeliness, and most of the information has been digested by the market; However, deep fundamental quantification needs to deal with more and more timely and higher frequency fundamental information.

Third, the analysis methods are different. Traditional fundamental quantification mainly evaluates the company’s value through financial analysis methods; The deep fundamental quantification pays more attention to the influence of non-financial factors on company performance, and uses natural language processing and machine learning technology to analyze text information, so as to understand the relationship between market information and company performance.

Fourth, the modeling methods are different. Traditional fundamental quantification usually adopts traditional modeling methods such as linear regression or factor model; In the quantification of depth fundamentals, a deep learning model is adopted, which learns the laws and characteristics of data from a large number of unlabeled data, and automatically identifies complex relationships by simulating human thinking, and continuously improves the performance of the model through self-learning, self-upgrading and evolution.

In addition to fundamental quantification, GPT big model has another application in quantifying investment, that is, code generation and model reference to improve efficiency. "Simply put, quantitative investment requires standardized codes. Using ChatGPT will make it easier to generate some standardized codes, saving code generation, and quantitative investors can make adjustments on this basis." Liu Xin introduced.

Generally speaking, the application of AI technology will make the whole quantitative investment strategy iterative faster and more efficient. However, some quantitative private investors have suggested that quantitative investment is a comprehensive system engineering, and AI can effectively improve investment efficiency, but it can not completely replace human work, and can not be equated with quantitative models and quantitative strategies. Moreover, it is worth noting that there may be deviations and errors in the data source and algorithm of GPT, and the risks cannot be ignored.

The winning percentage geometry of generative AI stock trading

Whether the generative AI stock market can beat the market has always been a controversial topic.

Some people think that "the stock market is not an area that AI can win in essence"; There are also views that it is not impossible for "AI to beat the market" as long as technology continues to break through. However, the respondents agreed that "beating the market" is extremely difficult, and the stock market is a complex and uncertain system with no regularity.

Wang Xiong believes that it is difficult to beat the market simply by AI, but AI, as an auxiliary tool, can greatly improve the efficiency of information acquisition, analysis and decision-making. That is to say, "people’s correct and scientific investment concept +AI’s efficiency improvement" can beat the market, which has actually been verified by countless excellent quantitative private equity funds and will be verified in a longer period of time in the future.

Anning Ning holds the same view. In An Ningning’s view, the stock market is influenced by many factors, and the interaction between these factors makes it extremely difficult to predict the stock market, so it is not easy to "beat the market". However, AI, with its powerful ability of massive data processing and analysis, can assist investors in decision-making to a certain extent. To beat the market and find a relatively stable profit strategy, more empirical and in-depth research is needed.

Undeniably, one of the advantages of AI is that it can avoid human weaknesses, such as emotional and irrational behaviors. However, the essence of financial market transactions is still the game of people’s different emotions and mentality. ChatGPT may not be able to accurately grasp the changes of emotions and mentality of all kinds of trading people, and then make the best investment decisions.

Liu Xin said that compared with most human beings, AI can really reach a more rational investment state. Especially in the field of quantitative trading, AI will have a more stable output for trading strategies that need a lot of analysis and processing data in a short time. However, AI can’t beat the best people in the market. Human beings have their own advantages, such as processing a lot of unstructured information and judging a lot of non-quantitative behaviors and states. Generally speaking, the way of AI and the way of human will be different behavior types in the market.

"Investing in this field can’t be achieved by spending money, smashing machines and smashing equipment. These are not core competitiveness. The core still depends on cognitive depth, independent thinking and innovation, and even some beliefs and beliefs. Simply put, when you have a good investment idea, ChatGPT can accelerate the realization of this idea through efficient data collection and processing capabilities, programming capabilities, text analysis capabilities, etc., but the idea itself is the key to investment. " Wang Xiong said.

Will it aggravate market volatility?

The application of GPT and other AI technologies in stock market investment may not only bring convenience to trading, but also cause some potential risks. For example, institutions with advanced AI technology surpass ordinary investors in information acquisition and decision-making speed, will it lead to unfair market? Will the large-scale use of AI tools in quantitative trading lead to transaction convergence and increase market volatility?

"When extreme markets occur, the collective position adjustment of quantitative strategies will strengthen the market trend to some extent, which is also common in overseas markets." A person in charge of quantitative private placement believes that A shares have experienced the ultimate deduction of AI market, which has a significant effect on the liquidity extraction of the whole market, and the quantitative strategy has obviously strengthened this trend.

The person in charge of the frontier assets of the century also said that the risk of AI+ quantification strategy lies in the homogenization of strategy, because quantification is based on historical data to make a model, which is equivalent to everyone reading the same book. Finally, the conclusions drawn by everyone have some similarities, which will cause high strategic relevance and congestion. To put it simply, everyone uses similar large-scale investment tools to make short-term investments, which will invalidate the investment strategy and make it more and more difficult to earn income.

Some people in quantitative institutions have expressed different views. Wang Xiong believes that improving the efficiency of analysis and execution with AI tools does not necessarily increase the effect of market fluctuation. AI and quantification are just tools to help realize strategic thinking, and there are different types of strategies. Short-term trading strategies based on high-frequency volume and price are indeed likely to be more homogeneous, while those based on deep fundamentals are less homogeneous, and the same fundamental information may have different interpretations and low strategic relevance.

Moreover, quantitative positions are generally very scattered and have limited impact on individual stocks. On the whole, quantification makes money by looking for opportunities of market mispricing, and the long-term effect is to make market pricing more reasonable, and more to suppress fluctuations caused by irrational trading.

Liu Xin said that quantitative trading itself is a way to gain profits from irrational fluctuations in the market, which is actually to stabilize excessive irrational fluctuations in the market. Different investment models will make the market more mature and stable.

In addition to the confrontation of views, strengthening supervision to better regulate quantitative transactions has become an industry consensus. According to industry insiders, on the premise of meeting regulatory and compliance requirements, we should focus on the needs of the industry, promote data, computing power and algorithms in coordination, and grow together with the capital market in terms of improving transaction efficiency, stabilizing market liquidity, eliminating information asymmetry, and promoting effective pricing in the market. This is the social responsibility that quantitative practitioners should bear and the internal driving force for the vigorous development of the quantitative industry.