中文
Home Research All Content

The Application of AI Technology in Silicon Valley


The Rapid Development of Artificial Intelligence

The rapid development of artificial intelligence (AI) has produced many breakthroughs, including in the financial sector, specifically in credit assessment, investment, and personal financial management. AI is particularly well suited to applications in the financial industry because the financial sector deals with large volumes of data, the field needs to be able to standardize and pattern specific processes. The volatility of globalization has caused changes to trading markets, and the proliferation of sensors such as smart devices has also produced a large amount of unstructured data. Risk pricing in the financial sector can use data and algorithms to do efficient assessments, which can effectively reduce the cost of financial institutions, and mine more valuable information to help the market participants make decisions at the same time.

In recent years, Silicon Valley, which is one of the global innovation centers, has generated many fintech startups with AI as their core technology. We visited companies that were established between 2016 and 2017, to discover how the latest AI technology has improved or reshaped the original fintech landscape.

 

The Application of AI Technology in Lending Industry

The rise of smart devices has generated a large amount of information around users, which has created a lot of data mining value for assessing personal credit risk. At the same time, traditional FICO credit scores remain deficient. The FICO score depends on the individual's reported credit history. (The five indicators considered by the FICO score include: (1) Repayment history, including repayment information and negative public records of various credit accounts, accounting for 35% of the total score; (2) Credit arrears, accounting for 30%; (3) credit history, accounting for 15%; (4) newly opened credit accounts, accounting for 10%; and (5) credit types in use, accounting for 10%.) The results in the FICO score cause some problems in actual use. Firstly, those who lack a credit history, such as recent immigrants or young people, cannot be assessed. Secondly, for people with existing FICO scores, FICO score assumptions and accuracy can be questionable. For example, if the user suffers a short-term economic blow and loses their house, but if they retain their job, they should be found to have a stable credit score, yet the FICO score will consider them to be less reliable. FICO data iteration is also slow, which leads to some users with higher scores also having a violation status. In recent years, the FICO score has increasingly become a reference for financial institutions rather than a decisive indicator. For individual users, the organization collects their information and uses their internal risk control model to evaluate it. Some companies no longer even use FICO. For example, SoFi, the largest online lending company in Silicon Valley, announced in January 2016 that it would no longer use the FICO score.

 

The use of AI algorithms can also improve personal user credit portraits. A job paper from MIT uses a classification and regression decision tree algorithm to classify and return consumer credit historical data and consumer data from 2009-2012 to calculate credit score for consumers. It also compares the result with the one that derives credit data from the bureau in traditional methods. The following figure shows that the two models recognize people with good credit and poor credit positions, but for people with moderate credit records, the former can predict the expected behavior of the group more accurately.

image.png

Source: Consumer Credit Risk Models via Machine-Learning Algorithms Amir E. Khandani†, Adlar J. Kim‡, and Andrew W. Lo§

Speaking of the fastest-growing deep learning technology in recent years, information that may have been judged by the loan officer as irrelevant variables, such as address information, APP using habits and so on, may become useful variables through mining and integration to point to the user's repayment willingness or repayment ability related variables.

In Silicon Valley, we visited Upgrade, and Random Forest Capital, both of which use AI technology to evaluate users' credit. While Upgrade's method has already been used in China, Upgrade claims to have launched a 2.0 version of the evaluation model. Firstly, the model uses geographic data, which can be combined with macroeconomic factors to reflect the economic situation in different places. Secondly, the platform is connected to the user's checking account and analyzes the user's free cash flow to get the user's payment habits, such as if there are overdue, penalties or missing payment dates. In this way, Upgrade has applied the method of analyzing the cash flow of SMEs to personal risk control.

Founded in 2016, Random Forest Capital is a San Francisco-based cross-platform machine learning, data engineering, and investment management company. Random Forest states that the existing underwriting methods are expensive, inefficient and inaccurate, and cannot accurately assess the risks of these debts. Random Forest uses cross-platform machine learning algorithms to price bonds, which significantly improves accuracy and efficiency while also resolving potential conflicts of interest between investors and borrowers. Because much of the data that the platform obtains are categorical data, the company utilizes a tree-like algorithm, and a boosting algorithm has also been proved useful.

Because of the protection of personal data and privacy in the United States, many financial institutions need to mine as much information about the subject as possible without sacrificing personal privacy.

The Application of AI Technology in the Investment Industry

AI technology has been used in the investment industry for some time. Artificial Neural Network (ANN) is one of the algorithms that people consider to be useful. In comparison with a linear regression model, ANN modeling can better handle financial market uncertainty, mine nonlinear relationships in the data and effectively deal with noises in large data sets. What's more attractive is that ANN modeling can update existing models by training for new data and can respond quickly in a rapidly changing market.

Since 2004, the scale of the U.S. domestic quantitative funds has continued to increase. From 2004 to 2016, the total size of quantitative fund assets increased rapidly from US$30 billion to US$300 billion.  Well-known quantitative funds include Two Sigma, DE Shaw, and Citadel.

image.png

Compared with traditional quantitative investment funds, the big advantage of AI funds is that they can better evade human-made errors, handle more substantial amounts of data in a short period, and dynamically update the parameters and the model itself makes the algorithm more flexible and adaptable.  Because of this, AI funds may surpass the performance of static and traditional quantitative models in the long term. Compared with conventional quantitative models, AI fund's algorithms are more flexible—they even set up some randomness, which makes their algorithmic correlation weaker than traditional quantitative models.

The following figure shows the study of EUREKAHEDGE in January 2017. The blue line shows the income of AI quantitative funds, the purple line shows the income of traditional funds, the green line is the income of exponential funds, and the red line is the income of traditional quantitative funds.

image.png

It seems that since 2010, AI funds have seen higher returns compared to any other type of funds.

 

In addition, from the correlation matrix of the table below, the correlation between AI quantitative investment funds and the other funds is low, and the correlation between AI quantitative investment funds and general hedge funds is even negative.

image.png

Pit.AI is a company that uses AI to develop hedge fund trading strategies. Yves-Laurent Kom Samo hopes to shift hedge funds from manually driven to machine driven. The company increases the profitability of the entire company by saving labor costs, which is the most expensive in the industry. Pit.AI only charges fees that are related to earnings from investors and does not charge fund management fees. Pit.AI and investors will divide the revenues only when the fund performs better. The income share ratio is 3:7, and Pit.AI takes 30% of the revenue. In the future, Pit.AI hopes to turn the charging model into a curve. When income is lower, the charge will be reduced accordingly. When income increases, the charge will increase.

Pit.AI believes that the business model of hiring many traders, generating many trading algorithms and therefore forming competition does not necessarily produce diversified algorithms. People who create these strategies tend to use similar ideas, so their algorithms usually have strong similarities and correlations. These highly correlated algorithms do not adequately diversify transaction risks.

There are two main differences between Pit.AI and traditional machine learning. Firstly, each algorithm will always run, with the goal to find a better trading strategy. Secondly, Pit.AI’s algorithms retain randomness, so that different instances of the same algorithm will perform differently—generating different algorithms instead of generating the same algorithms.

AI funds not only have brought about changes in the operations, cost structure and income forms of fund companies but also have an impact on clients. For instance, in robo-advisor companies, the former fund management is limited by cost, and fund managers dominate technology, so individual investors can’t invest in personalization because their shares are too small. The introduction of AI technology makes it possible to adjust positions based on the actual situation of individual users in the future. It can genuinely change individualized positions based on the user's risk tolerance, income expectation, and personal values.

 

The Application of AI Technology in Personal Financial Management

With the rise of awareness and demand for personal financial management, more and more software has begun to use AI technology to conduct financial management for individual users. The software provides financial advice for users based on their goals and helps the users to achieve their targets by analyzing the users' expenses and other related data. Founded in Silicon Valley, Olivia AI is such a company, which is committed to creating an intimate personal financial managing assistant so that each user can receive personalized, professional and caring services in personal finance. By combining the expertise of AI technology, financial management, and behavioral economics, Olivia AI serves its customers with a chat robot image. Through learning the users' financial habits and spending habits, Olivia AI manages its users' fund accounts in a unified way and provides financial management and consumption plans according to the users' characteristics.

Many startups in this area have recently targeted young people's bad financial habits (no deposit) and the potential for unrestrained spending (Moonlight). Such startups also seek to take advantage of a more user-friendly interface, fuller service functions, and new interaction and motivation incentives to cultivate users' financial management habits. For instance, Digit provides automated deposits, and Qapital uses IFTTT as a savings incentive. (IFTTT is a new network service platform that can decide whether to execute the next commandthrough the conditions of different platforms. Getting its name from the slogan "if this then that," IFTTT reacts to the response of network services through other network services. Users can set some rules to save, for example, to save changes after each spending; IFTTT has thousands of possible rules and can bundle with a wide range of software to set goals, such as if it rains, IFTTT deducts a little money as for tourism fund on a sunny day.)

As one of the innovation centers in science and technology, Silicon Valley sees lots of companies that have applied AI technology to all the fields in finance. However, we are delighted to find that the fintech innovation in China has not lagged innovation in the U.S.. Many companies in Silicon Valley have benchmarked AI technology applications in China. In the future, we expect that AI technology will bring more significant changes to the financial sector and serve more people who are not reached by traditional financial services.