Recently, as Huawei, Baidu and other large manufacturers have made high-profile statements about financial models, the wave of financial models is hot.
In this context, the "Big Language Model Empowers Asset Management High-quality Development Summit" jointly organized by Tonglian Data Co., Ltd. (hereinafter referred to as Tonglian Data) and Shenzhen Investment Fund Association was held in Shenzhen on September 22, and experts and scholars from financial institutions, technology companies and research institutes had a heated discussion on the theory and application, development and trend of big language model.
Jiang Long, General Manager of Tonglian Data, said in his opening speech that with the development of AI and large language models, the role of technology in empowering the high-quality development of asset management and wealth management will become more and more. Of course, there is still a lot of room for development of large model technology itself in the future, and large model technology may not be the last stop of general artificial intelligence technology, and there is still a very broad space for AI-enabled asset management and wealth management.
Virtual experts and intelligent agents
What can big models do, or should do, in the field of asset management?
In this regard, Ma Ben, global managing partner of McKinsey and head of asset management and wealth management in Greater China, said that we are standing on the cusp of the era of "AI of all things", "If the wave of mobile Internet since 2006 has greatly accelerated the speed and convenience of information interaction, then generative AI will fundamentally change the way of human-computer interaction and the way of extracting, analyzing, summarizing, learning and generating global information."
He believes that generative AI has a wide range of applications in the field of asset management and has great potential; "We believe that the most promising applications are intelligent research assistants to improve investment and research efficiency, and virtual industry experts to enhance industry research professionalism. In addition, generative AI is also of great value to asset management companies to strengthen marketing and post-investment accompanying content automation and personalized generation."
He further summarized the potential application of large models in the asset management field into three key points: the first is virtual assistant, the second is virtual expert, and the third is intelligent customer service.
"Virtual assistants around the asset management value chain can empower researchers, investment managers, channel and institutional sales, and marketing teams to help them improve efficiency and reduce the proportion of low-value work in overall working time," said Ma.
As for virtual experts, Ma Ben believes that "the vertical application of large models will be fully rolled out in various industries in the next 2 to 3 years, when virtual experts based on generative AI will be produced in all walks of life, which will bring more possibilities for asset management companies to obtain information and analyze information in the future."
Ma Ben also believes that virtual experts will also have greater development in wealth management, especially in the field of investment advisory, for example, asset management companies will not only provide customers with products in channel customer service, fund investment advisory, pension and other businesses in the future, but also need to provide investment advisory empowerment services and even investment advice. "In the future, the application of virtual investment advisory experts will be very critical in the process of extending the business model of asset management companies from asset management to wealth management."
The last one is intelligent customer service. He said that intelligent customer service can directly face customers, including channel customers, C-end customers, etc. "At present, some global asset management companies are accelerating the upgrade of intelligent customer service robots, thereby greatly improving the efficiency of channel customer post-investment services and marketing. For example, the intelligent customer service robot can automatically generate targeted post-investment accompanying content based on the different situations of the customer's position profit and loss. ”
Jiang Long, general manager of Tonglian Data, also said in an interview with the media that from the perspective of the penetration of AI into research and analysis work, its participation is getting higher and higher, which is beyond doubt. Over time, AI can do more and more. It is worth noting that although people do less in proportion, it will be the most valuable part. In this sense, it can be said that the proportion of participation in science and technology is rising, while the proportion of human labor is declining, but the value of human labor is increasing. With the help of AI, people have more time and energy to do more creative things.
Xu Danqing, general manager of Tonglian Data Intelligent Investment and Research Business Center, also said that the big language model is a paradigm subversion and innovation for many industries; "The same is true for the asset management industry, the whole paradigm is a trinity: one is the information system, the second is the model system, and the third is the decision-making or action system. We believe that in the entire field of intelligent asset management, the three keywords of this system should be information, data and opinions. Therefore, now is the stage of focusing on AI investment research assistants, we will use the ability of large language models to re-implement a lot of investment and research scenarios, here is mainly in the search for information, read research reports, check data, write reviews and do reviews and other aspects to provide assistant-level service support. ”
Impact on quantitative finance
Recently, there has been a lot of discussion about quantitative trading in the market, and in the financial industry, the relationship between quantitative finance and technology is also the closest and most direct. Therefore, with the advent of the era of AI big models, the impact it will have on quantitative finance has also attracted the attention of the market.
Sun Dongning, a researcher (senior officer) at Pengcheng Lab, said that because he has been doing quantitative research, he feels that in the wave of artificial intelligence big models, quantitative investment has a certain landing space: "The first is data collection and management; The second is market analysis and forecasting; The third is factor development, especially text-related factor development; The fourth is combined construction; The fifth is risk control. When assembling construction, we need to determine which style factors should be turned on or off, or how much exposure should be opened. If there is a large model that can provide market movements and control style judgments, it will help us build strategies and optimize combinations. ”
In addition, when talking about the enlightenment of large models to quantitative finance, Sun Dongning believes that the transformation of model structure from RNN family to Transformer greatly improves computing efficiency and long-range memory, and the transformation of training paradigm from supervised learning to self-supervised learning greatly expands the use of non-labeled data, making large-scale training possible, and capabilities emerge with scale. For quantitative finance in the era of large models, text data has important value from market monitoring, factor development, style recognition, portfolio optimization and even risk exposure management, and the paradigm of large models and Prompt fine-tuning provides efficient and convenient tools for processing text data and reducing the labor cost and time cost of text processing, so that unstructured text data is no longer different, and greatly reduces the cost of field adaptation.
"In terms of quantitative fundamentals, transactional strategies are gradually crowded, the scale of support is limited, and the quantitative private equity transformation is medium and low-frequency; Low-frequency transactions require a deeper understanding of macro, market, industry, and company, and low-cost processing of text data makes the data more multi-dimensional, organically combining operational data, news and public opinion, and analyst data. Sun Dongning believes that the big model provides the possibility for people and systems (machines) to deeply integrate and become decision-making processes for each other. In terms of quantitative investment, some cutting-edge researchers are exploring the joint training of cross-variety and multi-time scale data, establishing a comprehensive time series forecasting large model, and providing a basic foundation for specific vertical fine-tuning models.
However, at a time when financial big models are attracting much attention, a practical problem they face is that the asset management field is a very rigorous field. So, in this context, whether it is a technology company or an asset management company, what should be paid attention to in the use of big language models?
In this regard, Ma Ben stressed that to use a large model, we must use its advantages: "First of all, it is not omnipotent, and at present, its degree of intelligence in universal information is high; Second, the current large model is more suitable for summarizing the information of past stock, and it is still in the exploratory stage in predicting the validity of the future. Third, in areas where compliance risks are relatively large and sensitive, there should be corresponding risk control measures for the application of large models."
Ma Ben further said that model risk management is the focus of many financial institutions around the world, and they have generally established large-scale professional model risk management teams.
"The big model of the future is a multi-model architecture, some public, some localized. In the future large model framework, the requirements for model risk management will become higher and higher. He said.