The evolving landscape of digital lending is turning traditional bank loans and credit analysis into a distant memory. The rise of big data and technological progress have led to alternatives or augmentations to Fair Isaac Corp.’s proprietary FICO score, the dominant credit score used to vet consumers’ creditworthiness. Now, a bill is making its way through the U.S. legislative process that would require Ginnie Mae and Fannie Mae to consider credit scores beyond FICO. Although these proposals are focused on mortgages, one can infer that alternatives to FICO are welcome across the board, including consumer loans. And we now have the technical means to deliver.
When it comes to consumer lending, lenders have traditionally relied upon a loan applicant’s FICO credit score obtained through a credit bureau such as Experian or Equifax to help determine an applicant’s creditworthiness. These three-digit scores are derived using a proprietary formula that uses data like payment history, credit history length, and credit line amounts. The lower the score, the less likely an applicant is to secure a loan. The exact formula is a trade secret, known only to Fair Isaac Corp. Hence, we are already relying on a proprietary “black box” to make a credit decision. We’ll get back to that point later when we discuss machine learning algorithms. Enter digital lending.
Digital Lending Creates a New Way to Vet Applicants
New credit models are based on the proposition that the old ways of approving applicants based on FICO credit score alone do not paint a complete picture of an applicant’s creditworthiness. The proliferation of new data points about consumers provides a wealth of raw data ready for analysis.
With the use of machine learning algorithms, and more broadly artificial intelligence, new models are looking at hundreds, and thousands of other data points, and not all are related to traditional financial risk. Enhanced use of personal information may include educational history, employment history, and even seemingly non-financial information such as bedtime, website browsing patterns, spelling on loan applications, social media data, and even messaging patterns.
While using big data could muddy the waters by creating more confusion than clarity, artificial intelligence could have a big impact on how alternative lenders perform.
Artificial Intelligence Streamlines Sales and Strategy
Savvy digital lending startups are testing the waters with machine learning to make underwriting decisions and enhance their loans. Machine learning algorithms can help to determine if applicants are telling the truth about income by looking at past employment history and comparing it to similar applicants. However, this technology can also favor the applicant by finding hidden patterns.
This data collection is advantageous for people with insufficient credit history, low incomes, and young borrowers who are typically charged with higher interest rates if they obtain credit at all. These methods may also appeal to mortgage companies looking to automate less risky applicants through a similar process.
Yet several challenges exist with these new credit models:
- First is what we’ll call a slow rinse and repeat the cycle. Machine learning algorithms, like humans, learn by doing and repeating while making correctional adjustments along the way. Economic credit cycles can last 5-7 years. Even if we back test a model using historical data, how do we know it will work in the future? A cliche in finance is that “past performance is not indicative of future returns.” It may take a long time to prove that a model is right or wrong because the model itself, like an inexperienced loan officer, hasn’t seen enough credit cycles.
- Second, models need to explain their black box to gain trust. FICO gets away with being a “black box,” but artificial intelligence cannot. Even if a model works, humans need to have some kind of explanation to feel comfortable with the output.
- Bank regulators need to know what’s going on. Fair credit regulators require that lenders keep records for the reason that credit was denied. The applicant has a right to inquire about why they were rejected. Disclosing the reason for a rejection is easy to do with an old-fashioned credit scorecard, based on a transparent point system. But what would regulators or auditors do with machine learning model outputs? For now, the practical answer is to run a traditional model as a backup whenever a machine learning model rejects an applicant, and hope that they both give the same answer! If so, record the traditional model’s output as the reason for credit denial.
For now, the most beneficial result of machine learning is the ability to detect consumer fraud by analyzing customer behavior with baseline data of ordinary customers and singling out outliers, such as how much time people spend considering application questions, reading contracts, or looking at pricing options. This filter alone leads to more accurate underwriting decisions, which, in turn, reduces defaults for lenders and lowers interest rates for consumers.
Blockchain Changes the Future of Funding
Peer-to-peer lending platforms originated out of one simplistic idea, one peer borrower asks for a loan, and another peer lender will decide to fund the loan. Both parties benefited by “cutting out the middleman” – the borrow paid a rate lower than that of a traditional loan, and the lender received a rate higher than that of a traditional savings account. But the peer-to-peer, or “people helping people” model, changed as large lending companies and institutional investors entered the space to become lenders, and, in institutional parlance, “buy loans” in bulk. Moreover, peers lack the expertise or ability to perform proper credit risk on other peers. Peer-to-peer became institutional-to-peer.
However, what if blockchain or distributed ledger technology could return us to the original concept of peer-to-peer lending? Blockchain-based solutions are currently developing identity and reputation models. With blockchain, an entire loan process can live online. Many parties share a record of transactions and supporting documents eliminating the need for intermediaries and third parties. Once I transfer the ownership to you, it’s done. I no longer have it. Currently, we can transfer ownership, but we need someone to record the transfer.
Eliminating the Need for Third-Party Risk Managers
With distributed ledgers, you can create a smart contract on a public utility blockchain without the need for a third party to execute the contract. This allows you to build a low-cost, high-trust platform that didn’t exist before. A handful of startups are designing platforms offering secured loans on a blockchain for those that are holding digital assets for the long term. Cryptocurrency investors will be able to earn interest on their holdings while the digital lender uses them as collateral for consumer loans.
Others are testing mechanisms for collateralized lending based on the value being stored in a smart contract on a blockchain. Collateral could be a security, a bond, a property, a title, data, or gold. The asset must have been digitized and recorded on a blockchain. For instance, the Perth Mint, Australia’s official bullion mint, announced plans to issue cryptocurrency backed by gold.
Mehul Agarwal is a Customer Success, Business Development and Marketing expert working with both users and creators of technologies to achieve their Engineering & Technology goals.
Mehul has worked with several startups, mid-size and large companies from the Valley and outside especially in the FinTech, Medical Devices, Connected Devices amongst others. He has generated over $45 Mil in revenue in the last couple years building one of the largest customer accounts for one of the companies he has worked with.
He mentors startups from around the world around Sales, Strategy, Growth & Marketing both as part of accelerator programs and independent companies.
On the education front, he has a bachelor’s and master’s degree in Economics from Pune University and also a Master’s in Customer Relationship Management from Symbiosis University.