The banner of this company is "bringing Google algorithm into the field of credit reporting". It uses machine learning and big data technology to create a set of credit scoring methods different from the traditional model, in which the applied data variables are hundreds of times that of the traditional model.
ZestFinance was established on 20 10. Douglas Merrill, founder of Zest Finance, is a former information director and vice president of engineering at Google. Shawn Budde, another founder from the financial sector, was in charge of credit business at Capital One.
Caixin reporter interviewed Merrill, the founder of ZestFinance, and he thought that "ZestFinance can completely replace the algorithm used by banks now."
However, Chen Jian, president of FICO China, said that this is impossible. In the United States, there are more than 65,438+0,000 local credit reporting agencies serving consumers, which basically belong to the three major credit reporting companies. These three credit companies have databases covering the whole country, including the credit records of more than 65.438+0.7 billion consumers. After collecting a large number of personal credit data, the three major credit companies have to go through complex model calculations to form credit products. The calculation methods and models currently used by these three credit companies all come from the same company, FICO, which is called "the big boss behind the scenes".
More and more startups like ZestFinance covet the traditional Wall Street territory. And its momentum, like dominoes, is constantly pushing the key to the traditional financial industry.
Data turns waste into treasure
ZestFinance official website has a saying: "All data are credit data." This sentence just condenses the work done by ZestFinance-turning thousands of data into treasure and applying it to credit scoring.
In just four years, this company has successively obtained financing as high as $654.38+$200 million, and the investors behind IT are all famous IT venture capitalists, including FlyBridge, GRP, LightSpeed and Matrix.
At present, the credit scores used by most financial institutions in the United States come from FICO's model algorithm. FICO's position in the American credit information system has never wavered since 1960s.
In the United States, after finishing by the three major credit reporting companies and scoring by FICO, massive credit reporting data has become neat and beautiful reports and scores in the range of 325-900. Users can only buy reports, or they can buy reports+scores in packages.
Ewing, an academician of Chinese Academy of Sciences and professor of Peking University, told Caixin that FICO's scoring model is second to none, but it is not perfect. FICO's credit score refers to less than 50 data variables, so many people can "model arbitrage" to increase their credit score after finding out the variables FICO is concerned about. For example, a person can borrow and return books from the library repeatedly every day to "brush credits".
"In view of FICO's shortcomings, ZestFinance redesigned a set of credit evaluation models. Compared with FICO's less than 50 reference variables, ZestFinance refers to tens of thousands of data variables and uses nonlinear and more cutting-edge technologies for analysis, thus preventing the phenomenon of' model arbitrage' and evaluating consumer credit risks more accurately. " Ewing said.
In fact, ZestFinance is far beyond the bounds of FICO 50 variables. In the team of 65 people in Los Angeles, most of them are data scientists. They have developed several machine learning analysis models, which use tens of thousands of data variables. Thousands of data variables are just raw information data. Based on these data, the model can get more than 70 thousand indicators that can judge credit behavior. It takes less than 3 seconds for the model to run these indicators.
The so-called machine learning, that is, the core of artificial intelligence, is to let computers simulate or realize human learning behavior to acquire new knowledge and skills, and constantly improve themselves in data accumulation. Hilbert, one of ZestFinance's models, is a successful case of commercial application of machine learning, which allows machines to undertake data analysis of 70,000 indicators, find logical relationships and constantly improve themselves. Humans only need to make some logical analysis and judgment according to the results.
"For many years, American financial institutions have been using 50 data variables to decide whether to provide credit to customers. The problem is that many people don't have complete credit records, which leads them to continue to be rejected by traditional credit, "Merrill Lynch said. "At ZestFinance, we analyze tens of thousands of data variables and use a wider range of data to predict customers' risks more accurately. "
There are also a wide range of data types: a person's web browsing record, mobile phone payment record and supermarket shopping list can all be important reference basis, and even whether users use uppercase letters or lowercase letters when filling out credit application forms can become data variables.
"A lot of data can serve credit. For example, the time an applicant stays on our website can reflect his caution and sincerity in applying for a loan. " Merrill said.
Weinan believes that credit record is a strong variable. In the absence of strong variables, we can refer to various weak variables. When these weak variables are combined, they can form strong variables to serve credit risk control. "For example, children are a source of family expenses, so if we can infer the age of the borrower's children, we can predict his consumption cycle: babies have fixed expenses such as milk powder, and students have to pay tuition fees every September. As long as he can avoid his main expenses, he can control bad debts. "
Ewing said that in China, due to the short history of the credit reporting industry and the lack of sufficient credit data, many weak variable data can be used to predict a person's repayment situation. At present, many people in academic circles are also conducting similar research.
For the development and utilization of these "weak variables", Chen Jian also agreed that "the value of mining data is an inevitable trend, and the development of big data will change with each passing day." However, he said that FICO first digs value from Internet data. "Bank credit card transactions obtain data in real time and identify risks through analysis. FICO was invented more than a decade ago, and now more than 90% of banks in developed markets are using FICO. "
Fight for FICO
Indeed, ZestFinance and FICO cannot be mentioned in the same breath at present. FICO occupies 99% of the credit scoring market in the United States and most of the credit scoring markets in developed countries, while ZestFinance currently only serves 654.38+million Americans.
In China, FICO currently has a team of 80 people and has established cooperation with 15 commercial banks, more than 30 city commercial banks and rural commercial banks. ZestFinance has no business outside the United States, but Merrill Lynch told Caixin that it is currently negotiating cooperation with several China financial institutions.
But from the perspective of future development, it seems that new things can always win more favor. Facing the birth of new credit scoring companies such as ZestFinance, the mainstream media in the United States reported one after another-The Economist magazine wrote: "Compared with the traditional scoring method, ZestFinance has reduced the default rate by 40%." The American Consumer News and Business Channel said, "ZestFinance keeps people without accounts from being shut out."
All these voices seem to point to Pheko.
Merrill Lynch said that ZestFinance adopted a completely different technology from FICO. FICO is based on the "logistic regression" model created in 1950s, when there were not many data variables available for reference. However, with the advent of the Internet era, data began to explode, and Pheko's scoring method has not changed. Merrill, a former Google man, introduced Google algorithm into the field of credit reporting and walked at the forefront of technology. "ZestFinance can completely replace the algorithm used by banks now." Merrill said confidently.
Pheko said that he was very embarrassed by the doubts from the outside world. Chen Jian said that the outside world actually lacks understanding of FICO. FICO has more than one algorithm, but hundreds. There are nearly 200 algorithm patents registered in the United States alone. Different data variables and quantities are used in different data scenarios.
Chen Jian believes that the more data variables the better.
"FICO credit score actually has more than 65,438+0,000 candidate variables, but each score only uses dozens of variables." Chen Jian said that it is naive to think that the more variables, the better the model. From the statistical point of view, on the one hand, model calculation should grasp the essential law, on the other hand, it should avoid over-fitting.
"Too many variables will cause the problem of over-fitting. It's like making a pair of shoes. Your feet 100% fit, but no one else can put them on. FICO does not make a pair of shoes for one person, but for the whole society. If some variables are not suitable for everyone, they are not suitable for the model. " Chen Jian said.
According to the research results of Fitch Ratings, the influence of FICO score is declining. Now all banks in the United States have their own models, and they will use their own models to run the original credit data. FICO score is only one of the reference variables. For example, Wachovia's FICO score reference ratio has dropped to zero.
In this regard, Chen Jian believes that this is only an individual phenomenon. "As far as I know, 99% of the asset portfolio of the US banking industry is still based on FICO. It is desirable to take out 1% to test new things, but this is not the mainstream. "
Chen Jian said that technology serves the industry, and credit scoring is not a fantasy in an ivory tower, but the accumulation of deep roots in the industry. At present, 99% banks in the United States use FICO's scoring system, and its profound accumulation is unmatched by other companies.
Chen Jian made no secret of his confidence in FICO: "In developed markets, FICO has become an entity part of financial management, and no one wants to remove their original arms and replace them with a pair of high-tech plastic arms."
Serve people who don't have accounts.
"Financial inclusion" is becoming a new word, which means that people without bank accounts or bad credit records can enjoy financial services fairly.
Ajay Banga, CEO of MasterCard, recently said in a proposal on financial inclusion that there are currently 2.5 billion adults in the world who do not enjoy financial services, most of whom are women and young people, and some people live in rural areas. In the United States, there are currently 44 million people without bank accounts. "Therefore, financial inclusion needs to be launched in all countries, not just developing countries."
Merrill Lynch said that ZestFinance is to solve the loan problem for those who don't have bank accounts and have bad credit records.
"My initial inspiration came from my sister-in-law." Merrill recalled to Caixin reporter that his sister-in-law wanted to borrow money for a pair of car tires, but the bank refused because she didn't have enough credit history. "Later, I lent her money. If I don't lend her money, she will have to apply for a' payday loan'. "
Merrill Lynch's "payday loan" refers to a small personal loan applied for two weeks before payday. Borrowers only need to provide proof of income or government relief, and promise to repay after paying wages. If the principal and interest of the loan cannot be paid off at maturity, an extension can be proposed. However, the interest rate of this kind of loan is extremely high, with the interest of $65,438+05 per $65,438+000 and the annualized interest rate as high as 400%. In contrast, the annualized rate of credit cards is only 12%-30%.
In recent years, especially after the financial crisis, Wall Street and American regulators have been turning their attention to "payday loans", which are considered as high-risk loans, but they have been repeatedly banned. 2065438+On June 5, 2004, a group of borrowers appealed to American regulators, pointing out that it was unfair for regulators to classify these borrowers as "reputation risk". According to the content of the lawsuit, more than 80 prime banking companies in the United States were ordered by the regulatory authorities to suspend their relations with these borrowers.
These special borrowers have also aroused the sympathy of the mainstream people. Americans spontaneously organize groups to promote the financial inclusion of people without accounts.
"ZestFinance's mission is to create transparent and fair credit scores for borrowers who have no bank accounts or bad credit records." Merrill Lynch said that through thousands of data variables, everyone can have a fair credit evaluation.
In addition, ZestFinance also has an important component, namely ZestCash loan platform.
ZestCash is similar to a microfinance company. Its main business is to provide small loans to people who don't have bank accounts or have bad credit records. 90% of ZestCash's loans are used to buy daily necessities, such as car maintenance and medical insurance.
Merrill Lynch said that ZestFinance mainly helps people with bad credit records get loans in two ways: one is to provide loans directly from ZestCash; One is to let financial institutions that use ZestFinance scoring system issue loans to them through ZestFinance scoring results. "So far, we have helped more than 654.38 million Americans without bank accounts or bad credit records get loans."
It is worth mentioning that ZestFinance has not led to a high bad debt rate, because its target customers are "risk groups". Merrill Lynch said that the default rate of loans currently obtained with ZestFinance is 50% lower than that of bank's "payday loans". "In other words, with ZestFinance algorithm,' payday loan' can save half the cost."
Competition and risk
After the financial crisis, bank credit has become more cautious, and it people in Silicon Valley are constantly tasting the sweetness of financial cake. P2P lending platforms such as Prosper and Lending Club came into being, and microfinance companies such as ZestCash also flourished. Companies including Zebit, Avantcredit, Kreditech and DemystData have taken a good look at the shortcomings of bank credit. The common feature of these companies is to use big data for credit analysis, and most of them have their own online credit platforms.
Lending Stream Peer-to-Peer Lending Platform built by Zebit can get a semi-annual personal credit loan of 50- 1500 USD within 4 minutes.
Avantcredit's slogan is "Applying for a loan from here will not affect your FICO credit score". The company is also a self-built credit system, and different people's scores give different interest rates.
Kreditech is located in Hamburg, Germany Two confident IT people use big data analysis to evaluate the probability of borrowers paying back money. They do not require customers to provide credit certificates, and can provide small loans within 500 euros within 15 minutes. Similar to ZestCash, Kreditech wants users to provide as much information as possible, including whether the user's loan application was sent with an iPad or an old computer, the probability of input error, the frequency of using the Cancel button, and so on.
The above-mentioned companies have been favored by venture capitalists, such as Kreditech20 13, which received $9 million in Series A investment in September, and Avantcredit20 13, which received $20 million in Series B investment in August.
Chen Jian believes that such innovative companies and traditional FICO do not conflict with bank credit and can be used as a supplement to the traditional market.
Of course, such companies can't do whatever they want, and they have to be regulated by the United States. Among them, the Equal Credit Opportunity Law passed by 1975 stipulates that loans must be granted to all applicants with reliable credit, regardless of race, religious belief, gender, marital status, age and other personal characteristics. However, with the blowout of big data on the Internet, this information has been included in the variable calculation together with social network information by companies such as ZestFinance. In addition, since all the collection of credit data must be approved by me, this way of collecting a large amount of data will also face the risk of infringing consumers' privacy.