The upsurge of artificial intelligence was brought to the peak by alphago. However, there is a period of prosperity and stability in the discipline development history of artificial intelligence. A technological breakthrough will bring unimaginable prosperity in a certain period of time, and then the scientific development will be faster than before, but it needs to be understood that it will not be a technological breakthrough every day.
There are a lot of hot discussions about artificial intelligence, and there are many routines to play. It is worth believing that only real players will laugh to the end, whether in the prosperity or stability of artificial intelligence. Take a look at the mainstream playing methods of artificial intelligence and finance of real players:
Automatic report generation: investment banking, securities research
In the investment banking and securities research business of investment banks, a large number of fixed format reports are involved, such as some chapters of the prospectus, research reports, and investment letters of intent. Writing these reports requires a large number of junior employees of investment banks to list, sort out and copy paste data repeatedly for a long time.
By searching the keyword “format” in the regulations of the new third board, it is concluded that there are 189 regulations of the new third board on the requirements of announcement format. As for the transfer instructions in the new third board market, a large number of contents can be generated by templates. For example, previous share changes can be automated through industrial and commercial data integration, and financial statements can also be automated with accounting statistics. Automation can not only improve efficiency, but also verify the consistency of data. The official feedback of the share transfer system has mentioned many times that the accounting data are inconsistent with the data in the transfer instruction. One feedback and feedback often take more than half a month, and the auxiliary verification of the machine is very necessary. The natural language processing (NLP) in artificial intelligence can just liberate the employees of securities companies and investment banks from these boring jobs, generate more valuable judgments and insights, and comprehensively improve the efficiency of the financial market.
At present, automatic report generation mainly uses two technologies in natural language processing (NLP):
Natural language understanding (NLU): digest and understand daily discourse and transform it into a structure that can be processed by machines;
Natural language generation (NLG): express the structured data split by the machine in natural sentences that people can understand.
We can regard these two kinds of technical understanding as the splitting, processing and assembling of the raw material of daily dialogue into an understandable natural sentence – the final product.
However, three steps need to be completed by using the above technology to truly generate the report:
1,Processing massive heterogeneous data
Digest the annual report that investment bank analysts need to read, real-time news and data from Bloomberg news, industry analysis report, legal notice and other resources. The pictures and tables in the text need to be analyzed by OCR (optical character recognition) and other technologies.
This process involves using the commonly used knowledge extraction and entity Association in the knowledge map to extract its key logic backbone, and embedding the key information into the pre-designed report template in combination with factors such as event location.
After processing massive heterogeneous data and analysis data, we can produce news, brokerage analysis and Research Report, listing prospectus, enterprise annual report, fixed increase announcement, and even the investment proposal required by fund researchers to open the daily morning club can be generated in a similar way. Users only need to select the template that meets their needs, determine the theme and key information, and the presentation form of the report to generate the basic content. Moreover, investment bank analysts can proofread and manually edit again, add valuable views and conclusions, and improve the accuracy of the report.
AI assisted: quantitative transactions
Quantitative trading has been aided by machines since very early. Analysts write simple functions, design some indicators and observe the data distribution, which only use the machine as an arithmetic unit. Until the rise of machine learning in recent years, data can be quickly and massively analyzed, fitted and predicted. People gradually connect artificial intelligence with quantitative trading more and more closely. It can even be said that the three sub fields of artificial intelligence (machine learning, natural language processing and knowledge map) run through quantitative trading.
Machine learning: inferring models from numbers
Quantitative transaction analysts model the financial and transaction data, analyze the significant characteristics, and use traditional machine learning algorithms such as regression analysis to predict the transaction strategy. This method has two main disadvantages. One is that the data is not rich enough and is limited to transaction data. More importantly, it is limited by feature engineering. The quality of the model depends on the sensitivity of analysts to the data. In addition, one approach is to imitate the behavior of experts, select specific experts in a certain field, copy their decision-making process, and import a repeatable computing framework.
Natural language processing: grasping market dynamics
When quantitative transaction analysts found the limitations of the digital speculation model, they began to consider introducing rich texts in news, policies and social networks, and using natural language processing technology to analyze, process unstructured data structurally, and explore clues affecting market changes.
Knowledge map: reduce the interference of black swan event to prediction
Machine learning and natural language processing technology often fail to predict when some accidents (such as “black swan” event) occur, such as 911, circuit breaker mechanism, short selling ban and so on. Artificial intelligence systems have not encountered these situations and cannot learn relevant patterns from historical data. At this time, if artificial intelligence is allowed to manage assets, there will be a great risk.
In addition, machine learning is good at discovering the correlation between data rather than causality. A famous example is that as early as 1990, hedge fund first quadrant found that butter produced in Bangladesh, cheese produced in the United States and the number of sheep in Bangladesh had a statistical correlation of more than 99% with the S & P 500 index in the 10 years since 1983. After 1993, this relationship disappeared inexplicably. This is because the self-learning machine can not distinguish the false correlation. At this time, it needs the knowledge base (rules) set by experts to avoid the occurrence of this false correlation.
Knowledge map is essentially a semantic network. It is a relational network composed of graph based data structure and connecting with different kinds of entities according to the rules designed by experts. Knowledge map provides the ability to analyze problems from the perspective of “relationship”. In the financial field, rules can be experts’ understanding of the industry, investment logic and risk control. The relationship can be the upstream and downstream, cooperation, competitors, subsidiaries and parent companies, investment and benchmarking of the enterprise, the employment relationship between executives and enterprises, or the logical relationship between industries. The entities are investment institutions, investors and enterprises, They represent their knowledge map, so as to carry out more in-depth knowledge reasoning.
At present, the application of knowledge map in finance mostly lies in risk control and credit investigation. Risk control based on big data needs to integrate data from different sources (structured and unstructured), which can detect inconsistencies in the data. For example, the borrower Zhang San and the borrower Li Si fill in the telephone number of the same company, However, the company filled in by Zhang San is completely different from that filled in by Li Si, which has become a risk point and requires the auditor’s special attention.
Taking the investment relationship as an example, the knowledge map can link up the whole equity evolution, and easily show which PE Institutions entered in which year, what the entry price is, and whether there are gambling terms. This information can not only judge the current valuation of the institution, the future development of the company (the rhythm of the company’s growth), but also see the investment preference of PE Institutions, How the investment logic changes and develops.
At present, knowledge atlas has not formed a large-scale application in industry. Even if some enterprises try to develop in this direction, many are still in the research stage. We believe that the difficulty lies in how to establish a set of cooperation mode with institutions in specific fields, how to turn the cooperation into an easily programmable interface, so that domain experts can model the industry logic in a very simple way through the system, and their logic can be verified in real time through the system to further update it, Only through repeated iterations between experts and machines to form a closed loop can we serve users well.
Financial search engine: Securities Research
When conducting research work, securities companies / private fund researchers need to collect a large amount of information, reorganize and analyze the contents, such as upstream and downstream analysis, benchmarking enterprise research, competitor research, enterprise highlight / risk point analysis, etc. However, at present, the auxiliary research software used by most securities analysts, such as Bloomberg data terminal, only solves the problem of basic data, without considering the problem of information overload. This makes researchers unable to find the most accurate and valuable information in the face of a large number of basic data and explosive information, and can not improve their work efficiency.
The core technology behind the financial search engine is a high-quality knowledge map and a large number of business rules to help realize Association, attribute search and short-range relationship discovery. The exploration engine, such as faceted browser, also provides a man-machine cooperation interface based on the knowledge map, so that people can easily record, iterate and reuse the data exploration process. In addition, the recommendation system and push system are also very useful to help financial users focus on key data and save time and effort for pre investment discovery and post investment tracking.
Semantic search is to provide different types of queries (such as enterprises, funds, events, etc.), such as the impact of the Chilean earthquake on copper futures, the impact of the Middle East crisis on the overall money market, etc. Then the information is sliced and aggregated to provide visual elements for overview, such as the average market value and financing P / E ratio of fixed growth related to film and television media. Semantic search gives complex queries to users, such as looking for upstream enterprises of VR. When the search cannot provide accurate upstream information, it will recommend the enterprise of camera to users, and provide a convenient interactive interface to users for some complex filtering.
Smart investment advisor: Wealth Management
Traditional investment advisers need to stand in the perspective of investors and help investors carry out portfolio management that conforms to their risk preference characteristics and adapts to the market performance in a specific period. These tasks need to be completed by a large number of expensive manual methods, so wealth management services have virtually raised the entry threshold and are only opened for high net worth people.
But now, the robot advisor is managing your portfolio with minimal human intervention. You can manage your assets by a row of computers, and you don’t have to be a high net worth person. In addition, intelligent investment advisers are using artificial intelligence technology to draw wealth portraits of a large number of customers with a more powerful computer model, and using artificial intelligence algorithms to provide tailored asset management investment solutions for each customer.
Intelligent investment advisers make artificial intelligence technology no longer far away from the crowd, which really enables every ordinary person to enjoy the benefits brought by intelligent financial technology companies. It also makes many people who once thought that “artificial intelligence is out of reach” realize that intelligent financial companies can not only serve professionals in the financial industry, but also create value for the people related to the whole business society.
When artificial intelligence is no longer new, the four mainstream playing methods of the combination of artificial intelligence and finance, such as automatic report generation of investment banks and securities research, artificial intelligence assisted quantitative trading, financial search engine securities research and intelligent investment consultant wealth management, let us see that the combination of Finance and artificial intelligence will become infinite possibility of Intelligent Finance in the future.
Intelligent finance is providing a large number of auxiliary decision-making tools in a man-machine combination way, so that investors can easily obtain the support of data and analysis in the process of forming a logical chain, so as to find the work that machines are not good at, so as to greatly improve the work efficiency