Risk looms under the upsurge of "AI+finance" in the future

Original: 21st Century Business Herald

Image source: Generated by Unbounded AI‌

The current wave of artificial intelligence applications sparked by Chat GPT extends to the financial industry.

Recently, China Industrial Securities Global Fund launched an AI fund trading robot - intelligent trader "Xingbao", becoming the first domestic fund company to apply AI technology to the field of fund trading.

In fact, under the tide of AI, from AI avatars of brokerage analysts, AI quantitative investment to AI fund trading robots... Since this year, many financial institutions such as brokerages and fund companies have launched investment scenarios. AI products. At the same time, financial software service providers represented by Hang Seng Electronics are also promoting the implementation of AI investment research products.

Since 2016, securities companies have successively launched intelligent applications such as robo-advisory services and digital innovation laboratories. It is the general trend to deeply integrate AI technology with operations, risk management, customer service and other businesses. At the same time, fund companies are also making the same attempt, and have even integrated AI methods into the transaction process. The birth of Chat GPT has accelerated this process.

"For the application of AI, the industry as a whole is still in the initial stage of exploration. The industry is thinking from various directions how to use AI technology to improve its own operational efficiency and effectiveness. Among them, the more active companies have begun to try to launch related products, with success. There are many cases, but there are also many failure cases.” A senior industry insider pointed out to the 21st Century Business Herald reporter.

Regarding the combination of AI and transactions, the person believes that capital transactions have extremely high requirements for security, stability, and accuracy. Therefore, applications such as AI capital transactions still need to guard against hidden risks in data, algorithms, and computing power. **

AI+brokers: need to take into account compliance

As we all know, the launch of ChatGPT marks that the development of AI has entered a new era of general artificial intelligence (AGI). AI applications such as dialogue, writing, and Vincent diagrams based on "big models" have also begun to affect the business of securities companies.

In May of this year, the AI avatar of a brokerage analyst entered the public eye for the first time.

China Merchants Securities created an AI digital avatar for Gu Jia, its chief media analyst. According to the official introduction, Gu Jia's AI avatar can appear in roadshows, press conferences, research report interpretations, analyst conference calls, and wherever customers need it.

It is understood that in recent years, China Merchants Securities has comprehensively built an AI system, and its digital employee assistant case has been introduced into the "Securities Company Digital Transformation Practice Report and Case Collection (2022)".

At the same time, some brokerages are seeking external cooperation to explore AI applications applicable to brokerage business.

For example, on May 18, Soochow Securities and Tonghuashun officially signed a contract. The two parties will jointly establish an AI research institute to jointly develop Soochow Securities - a large model of the securities industry.

China Galaxy stated on the investor interaction platform on April 4 that the company has cooperated with a number of AI companies with market competitiveness, and has made great achievements in smart marketing, smart investment, smart customer service, smart risk control, smart documents, identity AI technology is used in fields such as recognition, and the latest AI technology will continue to be followed up in the future to expand the application scenarios and application fields of AI technology.

There are also financial software service providers targeting the "AI + brokerage" scenario. Recently, it has been reported that Hang Seng Electronics may launch a new digital intelligence financial product positioned in AI investment research.

In addition, Overseas Columbia University recently launched FinGPT, a financial large-scale model product, jointly with New York University Shanghai.

In overseas markets, investment banks have applied the latest GPT products to wealth management business. In March of this year, after OpenAI released GPT-4, Morgan Stanley stated that it has used GPT-4 technology to convert all think tank content into a format that is easier to use and operate. Morgan Stanley had previously tested the tool with 300 advisors and plans to roll it out broadly in the coming months, it is reported.

The application of AI technology can not only help various business departments of securities companies improve work efficiency, but also tap into broader wealth management needs.

Hou Yanjun, general manager of Houshi Tiancheng Investment, believes that AI will have a wide range of applications in the financial field in the future in terms of data mining, algorithms, customer service, and work efficiency.

However, some people in the industry pointed out that the actual implementation of some AI technologies still needs to be explored. In addition, how securities companies take into account compliance when exploring AI applications is one of the common problems faced by the industry. For example, virtual digital humans such as AI clones of analysts are still in a "regulatory gap". **

AI+ Fund Trading: Preventing Three Major Risks

In contrast, for fund companies with a relatively single business model, their demands for AI are more focused. In recent years, some fund companies have combined AI with risk control, research, customer service or assisted decision-making. The latest trend is that fund companies have begun to apply AI technology to the field of capital transactions and inquiry links.

Recently, Industrial Securities Global Fund has launched an AI fund trading robot - intelligent trader "Xingbao". At present, "Xingbao" has been officially launched on the Qtrade platform.

According to reports, AI traders can not only actively initiate question confirmation through the identification and extraction of key elements and understanding of context logic, extract deep-level intentions in real time, actively distribute questions, and quickly obtain counterparty intentions, but also communicate through continuous question-and-answer , after a series of inquiry and negotiation process, complete the collection of the counterparty’s inquiry demand, and feed back the inquiry status to the trader in real time, obtain the final matching transaction feedback to the trader, and complete the transaction after confirming with the counterparty.

In March of this year, the "Xing Xiaoer" AI bond trading robot independently developed by Industrial Fund was also launched. The company became the first public fund company to launch an intelligent inquiry robot on the iDeal platform of the foreign exchange trading center.

"In the past, some institutions have been trying to combine AI with transactions, and more to explore artificial intelligence through data algorithms or business rules. In recent months, with the rise of large language model capabilities represented by ChatGPT, this type of exploration It has once again become a hot spot in the industry." The above-mentioned senior industry insider pointed out to the reporter of 21st Century Business Herald.

In his view, in the next few months or even years, institutions will continue to increase their attempts to combine AI with transactions.

This is mainly because, "AI technology, especially the new generation of large language model technology, has an impact on most industries. In the field of trading, AI technology and algorithms are used to continuously deepen the details of investment transactions, such as robot inquiry, etc. , freeing people from simple, repetitive or even general-sense work scenarios, so that traders can focus more on professional digging.”

However, he also said that AI currently performs better in general applications, but trading is a very professional field. Due to the limitations of data sources, training scenarios and other factors, there are not many successful cases. . In addition to technological breakthroughs, the successful application of AI requires a large number of business scene integration. Only through the polishing and in-depth refinement of scene technology integration can AI technology be truly implemented**.

It is worth mentioning that capital transactions have extremely high requirements for elements such as security, stability, and accuracy. In the process of combining with AI technology, it is still necessary to guard against hidden risks in data, algorithms, and computing power.

The above-mentioned senior industry insiders mentioned in detail that, first of all, due to the extremely high requirements for data security and accuracy in capital transactions, the importance of data foundation ranks first in the process of applying AI technology. AI itself cannot solve the problem of data accuracy, so the basic data governance work is very solid. We must first ensure a high-quality data foundation, including data consistency, accuracy, security, and so on.

Secondly, in terms of algorithms, many things can be done through AI artificial intelligence, but at present, AI technology algorithms often do things with mathematical probability. Artificial intelligence has not yet developed to be as intelligent as humans, so wrong results will appear. For example, the ChatGPT large model performs well most of the time, but in some cases, the answer is unreliable. However, capital transactions have a very low tolerance for errors, and the probability of one in ten thousand is not allowed to occur. In this case, special attention needs to be paid to how to review or multi-layer recognition the results generated by AI.

Thirdly, in terms of computing power, the new generation of artificial intelligence represented by large language models has high requirements for computing power, with parameters ranging from tens of billions to hundreds of billions, and a lot of training is required. For fund companies, if they directly borrow the public large-scale model capabilities of Baidu and HKUST Xunfei, it will involve data security issues. If they deploy and train independently, they will also face the problem of input-output ratio. How to do a good job in public computing? The balance between computing power and private computing power is a problem that all companies in the industry must face.

In addition, Yang Delong, chief economist of Qianhai Open Source Fund, said that when applying AI to transactions, the risk that needs to be avoided lies in the compliance of transactions, such as whether such transactions take into account the fairness of transactions while taking into account efficiency. At the same time, we must also pay attention to prevent systemic risks. For example, the United States has a large proportion of quantitative investment, and there may be a stampede event caused by quantitative trading, causing the Dow to drop by 1,000 points in an instant. These risks need to be taken seriously.

**Can AI replace active fund managers? **

On the other hand, the market is more concerned about whether AI can replace fund managers in the future and conduct independent investment?

"In quantitative investment, AI technology has been widely used in investment decision-making, but there are relatively few cases where subjective bulls, especially value investment, use this technology." Zhishan Investment Fund Manager He Li pointed out , in order to achieve AI investment in value investment strategy, it is necessary to have in-depth value investment ability + advanced AI technology and understand the technology. The combination of the two will work. At the same time, a lot of costs may be spent on research and development in this process.

"chat GPT has brought AI into the limelight, but specific AI technologies such as machine learning have already been widely used in quantification, so the application of AI in the field of quantification and investment is no longer in the initial stage, and now it has matured. However, the latest algorithms and computing power are still in the process of constant iteration. In the long run, AI is one of the good investment aids, but in the future, AI is unlikely to completely replace human investment, but AI investment may replace You don’t know how to invest with AI.” Hu Bo, FOF fund manager of Rongzhi Investment, pointed out.

Yang Delong also mentioned that it is feasible for fund companies to apply AI technology to participate in some auxiliary work in investment research, because AI applications can indeed save a lot of manpower, and also have some advantages that manual labor does not have, such as strong data computing capabilities and so on. But it is unlikely to completely replace human labor. After all, there are still many aspects of the capital market that require manual judgment, so AI cannot completely replace manual labor.

According to the above-mentioned senior industry insiders, whether AI technology can be successfully applied to investment decision-making depends on three aspects: First, technical feasibility, which fields AI can empower, and which links AI technology is relatively mature; Social feasibility, in the investment decision-making process, in which fields investment managers and researchers are more willing to use AI technology, which links are easier to achieve AI empowerment; the third is the input-output ratio, many AI applications are innovative projects, there is a probability of failure , must consider the input-output ratio. **

"Considering these three aspects, at present, when investing in research, AI technology is relatively mature for sorting out research records, refining information content, and writing research reports. It is expected that investment researchers are more willing to use AI technology to improve themselves. The work efficiency is high, and the cost investment is not too large. These can be prioritized to form intelligent auxiliary tools to empower investment research and increase efficiency. The specific investment decision depends more on the personal experience and ability of the fund manager. AI technology is not mature, and investment It is difficult to determine the willingness of personnel and the degree of integration, and it is even more difficult to measure the ratio of input to output, so the participation of AI in it is much slower. However, quantitative investment is expected to be the first field to use AI capabilities for decision-making.” He said. express.