It seems that financial markets are always characterized by great uncertainty. Therefore, finance is often considered as a paradise for adventurers. Financial predators smile at each other, set off the bloody hurricane of the market, and then earn excess profits from people’s fear. But in fact, they are not always able to control the situation and instead often shoot themselves in the foot, and even suddenly quit the market.
The other side of finance is often like the silent flow of water. It can help people to improve their material and spiritual life or self-advancement.
Whether ups and downs or silent improvement, behind it is the handling and response to the complex logic of both capital and information flow. The sense of financial acquisition has increasingly become an important social proposition. How could financial services no longer be a “big players’ game” and better for hundreds of millions of ordinary people?
Artificial intelligence may be the best tool to deal with such massive amounts of information and uncertainties. Its innovative breakthroughs and practices in the financial sector are bringing about the dawn of inclusive finance.
In 1994, the United States was in its golden age. It was the only superpower in the world, thriving and prosperous. That summer, the FIFA World Cup just ended in Los Angeles, and the flames of new technology were rising in Silicon Valley. Netscape browsers soon became popular around the world. China just got access to the Internet through a full-functional 64K line. In an effort to contain the financial bubble, the Federal Reserve, under the leadership of Alan Greenspan, began to raise interest rates sharply, but bond markets did not realize that the Fed was entering a cycle of rate hikes. The global financial crisis did not come until thirteen years later; in 1994, Wall Street was flourishing, with countless people from all over the world pouring in to find their dreams.
Before the start of summer vacation, I received a letter of internship from a Dow Jones subsidiary; the job was involved with financial information-processing systems.
I studied information management during my college life. After graduation, I studied in the United States and obtained a master’s degree in computer science at the University at Buffalo, the State University of New York. The internship was a perfect combination of information management with computers. For those three years, I worked with financial news every day. Then I participated in the design of real-time financial-information system of the Wall Street Journal online version and became a senior engineer in Infoseek, the internationally renowned Internet company. While dealing with financial information, I watched the commercial battle in Silicon Valley through the Wall Street Journal and started thinking about how to deal with the problem of information cheating. Soon after, I put forward the idea of “super-chain analysis” technology and applied for a patent, which laid the foundation for the future engine development.
At that time, I did not realize that one day, with the development of artificial-intelligence technology, the change of the machine’s role in the financial system would be so fundamental today. It would penetrate deeply into various aspects, such as financial-information processing, data analysis, risk control, credit reporting, smart investment (robo-advisors), smart customer acquisition, and quantitative investment.
David Shawn, the king of quantitative investment, said, “Finance is a wonderful business of information processing.” That’s why Zhu Guang, CEO of Du Xiaoman Financial, said, “The most revolutionary changes will occur at least in the financial field, because artificial intelligence is the ultimate in the cycle of data collection, analysis, and processing.”
The machine’s penetration in various financial services is essentially the result of continuous improvement of financial-information processing capabilities offered by machine learning. They include integrated user portraits and construction of risk-control models in the field of credit and antifraud, mined investment-decision factors, and the matching clients and individualized investment portfolios.
With the continuous advancement of natural-language recognition and information database technology, even pure financial-information processing has undergone substantial changes. The intern walking toward us is a robot, and its distinctive way of working is first visually reflected in the generation of a financial-analysis report.
A financial-information system is probably the most complicated and boring information system. A share transfer agreement contains more than two hundred pages, and there are a large number of annual reports, semi-annual reports, research reports, announcements, feedback, and due-diligence results. We don’t know how many industry analysts are working to finish reading the information before they make decisions. Maybe, it is not that they are not diligent enough, but that reading this information is unrealistic.
Yang Xiaojing, former general manager of Baidu Data Finance, said about the industry that during the 1990s, it took about ten hours for a fund manager to read the daily market research report, news, transaction data, etc., which is equal to the workload of two days. In 2010, after the outbreak of mobile data, it took about ten months for him to absorb the information generated on the market every day. In 2016, it took about twenty years for the same fund manager to read all the information on the market that day, which is equivalent to his or her entire career. Therefore, fund managers urgently needed the assistance of advanced intelligent technologies, such as Baidu’s natural-language processing technology.
Today, much industry news and structured key information are processed by using Baidu Financial’s intelligent financial-information analysis system, which structures all the key information or reads the annual report of listed companies and then forms an analysis report—within minutes.
In this entire process, the machine is equivalent to a junior analyst of a financial institution, or even an intern who undertakes all the basic work. The working logic of this machine intern is similar to the process of extracting keywords and recombining them.
The machine can instantly read massive heterogeneous data such as announcements, financial statements, official releases, social platforms, securities quotes, real-time news, industry analysis reports, etc., of all listed companies. Technical tools such as OCR (optical character recognition) are used on pictures and forms in the text. Then, the extraction of key organizational information is made, which basically finds relationships, such as the industry upstream and downstream relationship, supply-chain relationship, stock-rights change history, fixed increase and significant asset recombination, data cross-validation between multiple financial statements, etc., forming and presenting a knowledge map about these complex relationships.
One step further, once analysts select a template that meets the requirements and determine the subject, the machine can generate a basic report text. Before the final output, analysts can manually check the accuracy of the report and add their unique personal analysis and conclusions, so that a format standard—and even illustrated—financial analysis report is done.
This potential intern is obviously not going to stop at the stage of simply processing information. Since it has already passed through the hall and into the inner chamber, it will certainly go further.
The robot at first opens the door to enter the core area of traditional finance—credit rating.
On the afternoon of July 18, 2016, Baidu announced an investment in ZestFinance, a US financial technology company, which also got an investment from Jingdong Group. Being preferred by China’s two major Internet giants makes ZestFinance, a data credit company that only served one hundred thousand Americans, known more to Chinese people.
The Los Angeles–based financial technology company uses machine-learning techniques to assess the credit risk index of personal loans. Its founders are Douglas Merrill, a former chief information officer and vice president of engineering at Google, and Sean Budd, a former head of First Capital International Group.
In the United States, ZestFinance challenges the industry giant FICO, which accounts for about 99 percent of the US credit-scoring market and the credit-scoring market in most developed countries.
ZestFinance believes that “all data are credit data.” Whereas FICO’s credit score only contains dozens of variables of the lender, ZestFinance’s model is based on massive social-network data and unstructured data. It contains nearly ten thousand variables and forms an independent credit score based on big-data mining. The efficiency can be increased up to 90 percent as compared to the conventional credit-rating system. ZestFinance claims to be able to analyze more than ten thousand pieces of raw information data for each credit applicant within five seconds and to derive more than seventy thousand indicators that can measure the behavior.
Prior to investing in ZestFinance, Baidu announced, at the Baidu Alliance Summit, that “artificial intelligence will have a transformative impact on finance, and it can truly upgrade credit information.” Baidu emphasized that “now Baidu’s finance unit can decide to make an educational loan within one second.” Underlying the “instant check” is the big-data risk control based on machine learning, which is a small test for improving the efficiency of credit services and increasing the coverage of financial services.
In general, like traditional financial institutions, big-data risk control also results in two lists: whitelist-based for credit and blacklist-based for antifraud. The latter is often cloaked in mystery because of the purpose of antifraud. For example, Palantir, the world’s fourth-largest unicorn, an artificial intelligence company founded by Peter Thiel, not only helps the US security ministry to fight against terrorism but also is recognized all over the world for discovering Bernard Madoff’s Ponzi scheme after combining and fully mining records and data of the past forty years.
But, we are more willing to tell the story of artificial intelligence in mainstream financial risk control.
According to the data of the People’s Bank of China, as of September 2015, 370 million out of 870 million ordinary people included in the credit system of the People’s Bank of China had credit records; personal credit reports and personal credit score can be designed for 275 million people. This means that there are still about 500 million in China who do not have any credit history and who are blocked from the threshold of traditional financial services.
Relying on the huge data foundation and the image-processing technology realized by artificial intelligence, Baidu Finance and other enterprises are quietly changing the problem of online processing of risk control and dropping the previously superior financial services to tend to those in need with no credit history.
For example, Li Liang, who has previously studied indoor design for four years in college, recently started searching on the Internet for UI (user interface) courses, training schools, and paying tuition fees by installments. He hopes to enter a large Internet company after learning these courses. However, high tuition fees in educational institutions have become his first barrier.
Li Liang didn’t know that there are still many people who are searching for the same keywords on Baidu at the same time, although at this moment they have not directly interacted with Baidu Finance. The group requirements of these people were collected in Baidu big-data risk control in the form of data and classified in a certain group portrait through machine learning, so they have the corresponding credit judgment.
After comparing several educational institutions, Li Liang finally chose a training school to learn UI and decided to try Baidu Umoney, which was recommended by the teacher, to pay the tuition on installment loans. He then completed the loan application by filling in identifiable personal information and taking a personal image in just a few minutes via a mobile phone.
Baidu’s risk-control strategy system responds quickly. With the support of the user portrait and image-recognition technology, the information of Li Liang is collected, processed, and analyzed, and the data-field result is sent to the risk-control platform, so as to complete the credit process. After a short wait, Li Liang received the text message for the first loan approval in his life.
After a few months of UI course study, Li Liang decided to learn VI (visual design) to get better prepared for a future job. This time, he was pleasantly surprised; because of the good repayment record and stable consumption record, the machine has expanded and improved his loan quota and credit-payment scenarios.
More important, Li Liang’s first credit record designed by the machine can help him to enjoy more comprehensive and better financial services in more financial institutions apart from Baidu financial system and bid farewell to the previous lack of a credit score.
Zhu Guang once said, “In our society, who loans to the young people with no favoritism and supports them in the critical climbing stage of their lives? Now, the answer may be ‘machines.’” When the machine completes digitizing its financial services for people, nothing can stop it from surging in the financial kingdom.
Warren Buffett, who has always been less interested in technology investment, probably wouldn’t have thought that someone would name a smart investment-consulting software after him. The software, which pays tribute to the investment guru, is Warren, a cloud-based financial software designed by smart-cloud investment company Kensho Technologies (Kensho in Japanese describes Zen’s clearness, meaning to see the essence in a phenomenon). It basically analyzes the impact of specific events (from natural disasters to election results) on the market by using big data and machine learning and presents results through easy-to-understand knowledge maps.
The software initially shocked Wall Street at its release, and many even telephoned Kensho’s founder, Daniel Nadler, and called him a “traitor.” On Wall Street, no matter whether artificial intelligence or any other gorgeous technology is used, it is normal to make money silently, but opening it to public and universalism must be a considered treason and heresy.
The control and processing of financial information itself are viewed as a monopolistic business. Bloomberg and Reuters estimate that long-term monopoly of financial data has a market capitalization of $26 billion. More and more users of Warren are breaking this situation.
Another company, Hedgeable, was founded to overthrow the Wall Street monopoly. Its founders, Michael and Matthew Kane, are twin brothers who have served the world’s top hedge fund, Bridgewater Associates, as well as Morgan Stanley, the most famous investment bank and another Wall Street giant. Because the twins were increasingly tired of Wall Street because it only serves the world’s richest people, they quit to create Hedgeable in New York, trying to provide investment-advisory services to the general public with the support of artificial-intelligence technology.
If investment consultants for Americans are commonplace, then, for the Chinese people who have accumulated wealth for more than thirty years, the service providers need to be popularized in China.
At the end of 2016, Wu Xiaobo, a famous financial writer, conducted a “consumption survey on the new middle class” and found that the class, which includes about 180 million people, is generally anxious about wealth preservation.
The high-net-worth people have always enjoyed private consultants in financial institutions. But who will defend the wallets of “new middle-class” and ease their wealth anxiety?
Smart investment, which is also known as robot investment and smart financial management, usually refers to the process by which a computer, based on artificial intelligence and big data, provides users (of different risk preferences and investment requirements) with algorithm-based investment-management advice, to help investors to make personalized asset and investment decisions and to realize the optimization of personal asset allocation on different risk preferences and investment requirements.
Smart investment advisors generally follow certain steps. First, they understand the requirements of investors; that is, they clearly understand the key data of investors themselves and their family as a whole. For example, the investor’s life stage, income level, historical investment experience, and preferences need to be considered. In general, the richer the investor and the finer the granularity of the portrait, the more accurate the understanding of the investor would be.
Financial management means wealth protection, long-term investment, and asset allocation; it is also a way of life planning. Therefore, machines must describe in detail the life pursuits of investors, such as buying a house, buying a car, studying, parenting, and pensioning, in order to match the corresponding investment cycle and examine the return-on-investment expectations.
Next, the machine will examine the investor’s appetite for risk. Age, stage of the career, income structure, and expenditures all are used to determine the investor’s threshold on taking risks. The risk preferences expressed in a face-to-face talk with investment advisors often deviate greatly from the investor’s true thoughts. This requires investment advisors to have a professional and meticulous communication with investors, so as to gain clear insight into the real risk threshold, which is a costly process. However, machines identify customer risk preferences through big data and can dynamically adjust in real time according to market conditions, drawing on the investor’s risk-preference curve, which greatly reduces communication cost, and the price investors have to pay.
After fully grasping the basic situation of investors, it is necessary to select the most suitable asset-allocation combination among various financial products according to specific customer characteristics. Therefore, while completing the investor’s portrait, the machine advisors must also carefully understand the financial products, such as, the basic assets behind the optimal investment target—for example, asset characteristics, volatility, price to earnings, stability, and the correlation between multiple assets.
After understanding both sides, the machine must consider how to make a combination of matches, instead of a single match. This process requires strong computing power and efficient deep-learning algorithms. This is why a technology company like Baidu can cut into this field and swim like a fish in water.
Finally, there must be asset monitoring and control. As the market adjusts the asset portfolio, the machine needs to continuously update its investment plan to match user requirements.
The emergence of smart investment advisors has changed the way financial management agencies interact with customers. They can truly understand the investor’s preferences, so that resources can be allocated through customized production. More important, with the help of artificial-intelligence technology, smart investment advisors can provide personalized and exclusive financial management solutions to the ordinary middle class at a low cost. Generally, there are no specific investment thresholds for smart investment companies. The management fee is only about 0.15–0.35 percent. The larger the user’s investment amount, the lower the rate charged. This fee is equivalent to only about 10 percent of the cost of a human investment consultant. Some smart investment companies can reduce the cost to as low as 5 percent of the cost of human consultants, or can even do it for free.
In addition, machine advisors can effectively avoid human weaknesses. In the tactical asset allocation, such as stock investment, once general investors feel trapped, they can only wait for the price to go up. Once the income is obtained, they cannot get the money in pocket fast enough. Machine advisors cannot be affected by emotions. After setting profit points and stop-loss points, they can be executed automatically and strictly without any greed or fear.
Furthermore, machine advisors have unlimited energy. They can provide customers with personalized and modular 24/7 service through uninterrupted and reliable digital asset allocation and put forward service plans in time, based on the investor’s demand preference and the market change. This ensures that machine investment advisors can communicate with countless customers on the phone or PC at the same time.
UBS Wealth Management, the world’s largest and most famous private wealth-management institution, had about 4,250 wealth-management consultants around the world at the end of 2015, but it has to serve approximately 4.5 million individuals and businesses with a service coverage ratio of over 1:1,000. The efficiency of the machine will obviously be higher.
In recent years, smart investment has developed rapidly in the United States. In 2012, the US smart investment industry was not started yet. By the year 2014, the asset management scale of the smart investment industry reached $14 billion. Several companies, including Wealthfront, Betterment, and Hedgeable, have developed in the field of smart investment. Kensho, which provides investment-decision information services, is another smart investment company. KPMG estimates that by 2020 the asset management of US smart investment will reach about $2.2 trillion.
In China, smart investment companies that imitated foreign models began to appear in around 2015. During 2016, a large number of products with smart investment began to emerge. Although it seems that the market is hot, we have to admit that smart investing has been hyped. For example, some companies only recommend a certain fixed investment portfolio in the name of smart investment. In fact, they have long deviated from the original intention of smart investment in terms of risk allocation, investment objectives, investment ability, and willingness.
It requires comprehensive capabilities such as big data, machine learning, and financial insights to engage this industry, and the Chinese market environment calls for higher requirements. For example, the investment targets of foreign smart investment are mainly ETFs (transactional open index funds, often referred to as exchange-traded funds), which are passively managed funds. Correspondingly, the US market has nearly 1,600 ETFs, and the assets under management total $2.15 trillion. However, the total number of ETFs listed in China are about 130, and the total assets are nearly 472.9 billion yuan (about $66.8 billion, according to Wind Data Service, as of June 2016).
The shortage of index-type products has forced China’s smart investment to introduce active-management funds. However, many changes in the active-management fund are unpredictable. For example, changes in fund manager or fund-company strategy result in a huge change in the fund’s earnings, which is difficult to predict and simulate. This makes extremely high demands on the investment ability of smart investment companies.
In order to make a more accurate investor portrait, general smart investment companies will draw on the form of offline investment consultants and ask their customers to fill out the questionnaire. However, in China’s investment and financial-management market, the investment group is dominated by retail investors, which have a greater speculative mentality than investment psychology. This easily leads to risk-preference distortions during the customer interview. For example, when investors find that some high-risk assets have good historical returns, they may ignore the potential risks and fill in a falsely high-risk acceptance in the questionnaire. On this issue, big-data companies, which can carry out user portrait including indicators such as “investment risk is above 100,000 RMB” and risk assessment rather than using general evaluation indicators, have a very obvious advantage.
The United States has a type of third-party organization that can bring together all the accounts of investors. As long as the user clicks on the authorization, the smart investment company can obtain information such as all the user’s cash flow. For example, Pefin, a smart investment company, can manage all the accounts for investors under one platform, including savings accounts, bank consumer accounts, credit-card accounts, monthly payments, loans and investment accounts, etc. Pefin can use its analytical model to build a knowledge map of the investor’s current financial situation in no time.
However, there is no such account consolidation agency in China. To generate an investor portrait as comprehensive as possible, we also need the machine-learning portrait ability based on big data, in addition to questionnaires filled by users themselves.
The biggest challenge is investor education. In a market dominated by retail investors, the main problem is how to convince investors that machines can carry out professional asset allocation and provide them with good financial management services. Moreover, in case traditional investment education has not yet been completed, the education process is difficult. Yuan Yue of Du Xiaoman Financial’s Robo-Advisor team described the situation as skipping grades: “It’s like you want to go to junior high school even when you haven’t yet graduated from primary school.”
The “black box” of machine learning has also increased the difficulty of education. Traditional investment advisers can explain the decision logic to the customer after making an investment transaction, but it is rather difficult for machine advisors to explain their “ideas.”
In order to overcome this apparent defect, many smart investment companies choose to emphasize the professional factors behind their investment logic. An example is China Merchants Bank’s MachineGene Investment, which highlights its comprehensive grasp of many fund companies and their fund managers as an important investment model factor. At the other end, Hedgeable has created an investment-advisory section that allows registered investment advisors, financial advisors, certified financial planners, certified public accountants, lawyers, insurance agents, and other key players in the financial value chain use the Hedgeable platform to serve their clients.
Of course, the development of the Chinese smart investment industry depends on the gradual clarification of future regulatory policies. If the smart investment industry is unable to provide direct investment services to individual investors’ securities accounts due to license restrictions, then the best method is to export technical capabilities to institutions, which may produce a wider gap between institutional investors and public investors.
It is still impossible for a smart investment service to accurately say it benefits the public as a whole. It can only serve the middle class of a limited size in addition to the high net-worth group or serve as a technical exporter to an investment institution. This is determined by the zero-sum rules of the capital market. That is, when one person makes money, the other person must lose money. Until the machine discovers a better strategy through algorithms and procedures, the rational approach is to follow Wall Street’s rule for “making money silently.” If this strategy is widely used by the public, the rate of return would inevitably be greatly reduced and eventually fail. This is the paradox of smart investment.
In the case of Wealthfront, the chief investment officer is Burton Malkiel, author of A Random Walk Down Wall Street; the passive investment philosophy he advocated in this book is that since the market cannot be defeated consistently, then the individual should simply invest in it.
Wealthfront follows this philosophy and chooses an ETF, a passive investment vehicle that tracks the index, as an investment to achieve long-term and stable returns. But, as an open strategy, this also means there is no high returns. As a matter of fact, the wealth managed by these smart companies such as Wealthfront and Betterment is about $3 billion to $5 billion, which is far from the traditional, mega-asset management companies, such as BlackRock, that can easily manage trillions of dollars.
Kensho finally merged with Wall Street investment bank giants Goldman Sachs ($15 million in financing), GV (formerly Google Ventures), and New Enterprise Associates ($10 million in financing), etc. Its valuable data-processing technology was eventually closed to the public, being confined to the mysterious circle of Wall Street, and the revolution it spurred ended abruptly.
Some of the world’s top asset-management companies such as Charles Schwab, Fidelity Investments, and Vanguard and international investment banks such as Goldman Sachs, JP Morgan, and UBS Wealth Management are also joining the smart investment field by investing in mergers and acquisitions or self-built platforms.
The big players at the investment table are waking up before the arrival of the intelligent revolution in the financial sector. In the future, maybe only a technology giant with strong artificial-intelligence technology and massive data can compete and cooperate with it.
In November 2016, the US presidential election was in full swing. At the same time, another battle in the investment market was taking place.
Analysts of Credit Suisse found that CTA (commodity trading advisor, or a futures investment) funds with assets under management of about $330 billion, which specialize in both quantization and long and short strategies, were gradually turning to short positions. The short positions for US stocks soared to the recent highest level. On the other hand, long/short equity funds with assets under management of more than $215 billion set a new high for nine-month positions against US stocks.
Most of the investment transactions on both sides were carried out through algorithms and models by using the computers. The media defined it as a “robot war.”
Since most of the transactions for quantitative investments are done by using computers through various models and algorithms, many people understand them as artificial-intelligence interventions in the investment market. But in fact, quantitative investment is just used to find the risk-free arbitrage opportunities in the market with the support of powerful computing. It can only be categorized as trading strategy, which mostly has nothing to do with artificial intelligence.
True smart investment is still data-driven. Regardless of the iteration of algorithm, creation, and the ingenious logical relationship design, the financial algorithm model still needs to be facilitated by a large number of multidimensional data sets that meet the requirement of the model, such as economic, social, and industry-specific changes, to verify the model’s feasibility and precision.
An open big-data environment is crucial for smart investment or financial-information analysis. Because the cost of acquiring digital data from the physical world is extremely high, most of the companies do not have their own big-data resources, and there can be no intelligent investment analysis and decision-making to speak of.
Baidu and Google’s large, multidimensional, big-data resources of search data and map data are rich resources for the mining of financial-data features from the Internet, which is different from the other traditional financial-system data. Together with leading artificial-intelligence technology, it has endowed search engine companies with the best resources to break into the financial-investment field.
With resources, there will be “miners.” The alchemic process for big data is generally like this: After determining the data source, the entire network data will be integrated at high speed with mature big-data technology; then the system optimizes data-operation efficiency and maintains the integrity of network data. Next, advanced machine learning, artificial intelligence, big-data analysis, and other similar technologies analyze and process the massive data and explore the personalized characteristics of financial assets.
For example, each day Baidu generates approximately 20 million searches directly related to stock names or stock codes. The search volume of a stock tends to be highly positively correlated with the stock price trend, with an average correlation of 0.7 or more. The search volume represents the degree to which investors are focused on a stock. Assisted by public opinion, the information can help buyers and sellers determine when to enter the market and when not to. Searching data can yield a gold mine of information for making investment decisions.
If searching is a manifestation of subjective intent, then with sufficient data, people’s potential requests and interests can be extracted. The special user structure of A-share market (mostly retail investors) and operational characteristics (customary speculation) can make search data an excellent indicator. Therefore, when the search data assists market selection, then we can effectively observe market changes.
Search data can be further explored and refined. For example, Baidu Map has already marked more than three thousand industrial parks and more than four thousand commercial districts in China. Observation of these industrial parks and commercial districts can help us to effectively judge the change in population flow in a region, business center, scenic spot, or even city.
The intelligent system can even refine the knowledge map of an enterprise and automatically update it in real time by machine learning. Therefore, with the help of rich data, it becomes an easy task for the investor to monitor the changes in the operation of the invested enterprise in real time with a broader view and more timely perspective and obtain in-time reference for investment decisions.
Traditional financial logic, investment logic, asset-management capabilities, and highly correlated real-time data will undoubtedly greatly improve the accuracy and forward-thinking of investment judgments.
In the financial sector, every 1 percent increase in efficiency and reduction in risk means huge wealth gains.
We need to respect the market—its complexity is far beyond human imagination. Geniuses have created relatively simple and perfect models to help us to understand the financial world abstractly, but in the process of breaking down the models, something is lost. These models are relatively static, and as time progresses, the models may become less precise. In the research of Preqin, a London-based consulting agency, the benefits of typical systemic funds were not as good as those operated by managers.
The machine-investment battle before the US presidential election in 2016 mentioned earlier may have been exaggerated by the media. As Buffett said, “Investing is not a game where the guy with the 160 IQ beats the guy with a 130 IQ.”
Regardless of the algorithms and models, we must respect the financial laws and investment logic. Volatility is not caused by machine investment but by changes in market expectations. Humans make insights and decisions behind everything, and this will not change, at least in the foreseeable future.
At present, many artificial-intelligence investment companies claim that their system’s stock trading is completely detached from human intervention. For example, Babak Hodjat, a chief scientist and CEO of Sentient, which Li Ka-shing bought into, announced that “our system allows the fund to automatically adjust the risk level.” Ben Goertzel, the founder of Aidyia, a hedge-fund company that analyzes the US stock market using artificial intelligence, is more confident about his system. “If we all die, it would keep trading.”
But it is quite difficult to imagine a completely machine-controlled investment market. If most of the funds use artificial intelligence entirely, we may get results that we don’t want. On one hand, under precise calculation, the rhythm and target of machine investment would be increasingly similar; the market fluctuations would get smaller and smaller, becoming extremely boring. On the other hand, since the investment objects of machine selection are more and more rigorous, the seeds of market imbalance are buried at the same time.
There is a scene in Liu Cixin’s short story “The Mirror” in which everything in the future of human society can be accurately calculated and predicted. The development of human society will become stagnant, and ultimately civilization will be destroyed.
Such imagination may be gloomy, but it gives us the understanding that investment combines the essential desires of human beings and also includes the cultural characteristics that promote the continuous development of human society. The investment process is far more significant than digital growth.
Regarding the relationship between artificial-intelligence investment and the humans’ investment decisions, Zhang Xuyang, vice president of Du Xiaoman Financial, explained his views at a seminar:
Investment is a combination of technology, art, and philosophy. Baidu’s cutting-edge big data and artificial-intelligence technologies can solve some technical problems. However, sometimes investment is artistic; otherwise, there would be no classical investment master like Buffett. Analysts will have different interpretations and methods for one market signal. For example, through Baidu big-data analysis, we found that barbers and teachers have begun to get interested in stocks. For this information, some analysts may think this is a sell signal, while other may think this is a buy signal. This is actually a perception of the improvement of investment experience. In the process of judgment making, investing is more like something perceivable but indescribable.
In some fields, where our current artificial-intelligence technologies can make breakthroughs, such as image recognition, speech recognition, natural semantic understanding, user portraits, algorithms, and assisted decision making, clear signals can be extracted, and the machine can make judgments after self-learning. But, the most critical part in investment decision-making is unclear. It is something in meaning but not in language. Our current technology is powerless to understand such things, so as a result there is no way to make insights and judgments for people.
After AlphaGo defeated Lee Sedol, many people thought that the machine can make investments on behalf of the human. But in fact, the game of investment is completely different from Go. Go has a closed and well-informed game environment, whereas investment involves human’s irrationality and has an open environment. So, in this case, technology needs to be upgraded a lot in order to replace the human and make decisions. This process needs at least ten years.
In fact, the artificial intelligence algorithms used for investment are similar. These intelligence algorithms are based on logistic regression and causal analysis. Later, with deep neural networks, there was a so-called gradient decision tree; finally, there was a genetic algorithm. However, the advancement of these algorithms has not yet gone beyond the scope of relevance analysis. Only some short-term memory can be realized. After all, the degree of response of the human brain has not yet been reached. Machines can perform well on some repetitive and recyclable investment decisions in the investment market. However, manual intervention is always required for markets with certain deficiencies, or for markets that are flawed in the artistic level.
I think it is still hard for machines to replace the human at the artistic level, at least so far. Behind our smart investment, we also need a team that can maintain an algorithm. The logic of this algorithm must be constantly adjusted to adapt to the arrangements in different investment environments.
The third level of investment is philosophy and self-discipline. That is, why do we make investments? We must have a rule to stop loss and make profit. In this aspect, machines may perform better than humans. In fact, it is easy to stop loss or profit through the machine, because people are inevitably influenced by mood swings, greed, and confidence. For example, people would always think, I’m different from others; unlike the last time, this time I can escape with someone picking me up. I can make this shot because I know that although the bubble is big, I am confident I won’t be the last one. But, history is often repeated. And through the machine’s algorithm, you can set it up to admit defeat, or stop profit at some point; there may be a process of thesis-antithesis-synthesis.
However, having machines replace people at the artistic level is difficult. Our smart investment requires a team-maintenance algorithm. The logic of this algorithm must be constantly adjusted to adapt to different investment environments.
In short, humans are still the most crucial determinant in the financial investment market, especially when the integration of finance and artificial intelligence is accelerating.
Just as a large number of physicists and mathematicians poured into Wall Street and brought revolutionary changes, nowadays, the cross-border flow and integration of artificial intelligence and financial talent is emerging in technical companies like Baidu.
“Robin said welcome back from Silicon Valley. So here I am,” Wu Jianmin, who previously worked at Microsoft Research Institute, said about why he came to Baidu. He is mainly engaged in research on smart customer acquisition and other related research in Baidu Finance. He and more artificial-intelligence experts solve financial problems by technology. Zhang Xuyang, who manages financial management, and Huang Shuang are responsible for consumer credit, and other professionals from traditional financial institutions are those who raise and define questions.
This cross-border combination of talent is an inspiring innovation. We will wait and see how the convergence of artificial intelligence and financial service will revolutionize the financial-service industry in terms of identity authentication, big-data risk control, and smart investment. For individuals, artificial-intelligence financial instruments are necessary. Five to ten years from now, when we go to banks, securities companies, insurance companies, and other institutions to get financial management, credit, and financial services, artificial intelligence will be working in the background. Zhu Guang called the future state of financial technology “AI inside.” We hope that our own technology, data, and capabilities will support all financial institutions in China to reduce the impact of financial uncertainty and to give full play to finance to help realize a better life and the dream of inclusive finance.