Digital Lending Trouble
September 9, 2018
What is common between an airline, a ride & delivery company, a payment company, and an e-commerce company? You might have guessed that it’s lending. In fact, there is a whole new wave of digital lending propositions with the premise that anybody can lend. Anybody who has a lot of customers, data on them along with engagement to give them a loan. One of the most common reasons to lend is to finance their products/services or finance sellers on a marketplace. This wave was started by the US, the UK, and Chinese lending companies that we fondly referred to as digital lending FinTechs (including but not limited to P2P lenders).
These digital lending platforms started on a very strong note. Their business models were based on their underwriting algorithms, with their alternate data inputs, being able to predict the delinquency rates of certain borrowers with a higher degree of accuracy than the traditional credit scoring models. A higher level of defaults was expected from the target audience (thin file or new to credit), and supposedly built-in into the pricing, with the interest rates being charged, shocking almost everyone except the borrowers. Investors flowed in, with their deep pockets, either buying up equity in the platforms or lapping up the connected Asset-Backed Securities (ABSs). This included the banks that were not willing to lend to these borrowers directly, safe in the assumption that the securitization process would reduce the risk to acceptable levels. The risk-adjusted returns were simply too good to pass up, especially in the low-interest environments in the developed world.
What the business models did not factor in were increased competition (they were supposed to be in a safe space that banks were not interested in, and where their algos were supposed to act as an entry barrier to anyone else), any regulatory costs (they were operating in a grey area, and doing a public good at the same time), and their algos going wrong. Not to mention borrowers defaulting in big numbers. Lesson learned – Underwriting risk is a complex animal.
The crazy-high valuations of these startups inevitably ended up attracting everyone’s attention. Who doesn’t want to pocket some dollars for something as high growth as online lending? And in the process, raise their valuations. All you needed was (it seemed) alternate data and an awesome algorithm, and you were in business. It resulted in a number of new players entering the field, each in its own supposedly niche space, but in reality, targeting the same set of people.
In the economic markets, excess supply has a way of causing a downward spiral which ends up killing businesses. The investor pressure for dizzying growth rates, at a time when the competition was heating up, led to regular lowering of the bar in terms of the credit scores and more competitive pricing. The unlimited pool of new to credit (thin file) customers provided VC-friendly valuations for digital lending companies. As defaults rose in late 2015 and early 2016, the risk-adjusted returns took a beating for the investors (except that there were only mounting losses, not positive returns to set off the increasing risk). The interest rates being charged were no longer reflective of the higher risk profile.
In Q2 2015, the loss rate of Grade E loans for Prosper went as high as 18.07%. In Q3 2015, OnDeck reported a 6.4% default rate (close to subprime default rates). Wonga had a 6.6% default rate in 2015 (it finally fell into administration in August this year).
Delinquency rates for LendingClub, Prosper loans issued 2013 to 2016 are rising year over year.
As the loss rates kept increasing, the platforms realized that their models were not as robust as they expected. Any algorithm/AI/alternate data-driven model is only as good as the data and presumptions that go into its formation. Something was not right. They had to start tweaking the models. Lending club is at its 5th model iteration and Prosper at 7th. Whether the model should have been tested and tweaked before being let loose on borrowers and investors, is a debatable question, but the failing models did not sit well with the investors. Investors fled, valuations tanked, and regulators got interested.
Lesson learned: Risk Management is very different from software.
One of the essential elements of the expected success of these models was the lack of legacy costs of banks, including the costs involved in following regulations. As regulators started getting jittery about the platforms’ exponential growth rates, with people falling into debt traps, and with outright fraud, a regulatory crackdown began. Countries like China, US, and Kenya, which were the bedrock of digital lending, started imposing tough rules, interest rate caps, limiting the number of players, forcing players to revise their business models or banning the activity outright. In the US, banks were suddenly facing the prospect of their investments in these loans facing the same regulations as their direct lending portfolios. Among other implications, this meant that these platforms now needed to bear additional costs to bring their compliance standards up to bank levels, if they wanted the investments to continue to flow in.
The platforms backed off. With them falling out of favor with investors, borrowers (some also sued), as well as regulators, with returns falling and loss rates rising, the best strategy was to scale back. Things cooled down.
Then came the second wave. It’s here and now upon us. Although it’s global in nature, Asia (outside of China) is where we are seeing it clearly with a totally new set of players such as marketplaces that were starting their own lending operations. These loans were based on REAL profiles and data/numbers about consumers and sellers. As the Alibabas and the Amazons of the world grew, anybody can lend has become the latest mantra. This time, all you needed was a platform that was connecting buyers and sellers, their respective information, and a B2B partner/vendor/group company who would supply you with tools such as a credit scoring model. What could possibly go wrong?
Enter AirAsia, Grab, Stripe, Razorpay, Flipkart, Truecaller, etc. Across consumer and SMB segments, digital lending propositions from these companies are sprouting rapidly. They have tons of data on these users and are beginning to offer seamless digital lending experiences. The use cases are very interesting I must say.
There are some early successes but many failures such as Uber’s continuing struggle. It switched from Uber Financing to partnering with select lenders in select cities in the US to enable financing for its drivers.
The larger question here remain: are these two models underestimating the skills required for pricing in the risk and for collecting the debt? Skills that banks have honed over decades? Although they are still not great at it. Can these skills really be replaced by algorithms? Or do these entities need to take a step back, build these skills (either on their own or with partnerships with banks) and then enter the market? One would think the latter. Lending is not as easy as one may think. It requires a deep understanding of risk management and how underwriting should be done. It requires you to go through many cycles and see how it works before you can see hockey stick growth. Also, how do you collect the money digitally even when you lend digitally?
The jury is still out what is going to happen with these second wave of anyone can lend companies. But new to credit (thin file) customers are happy. They finally have many options now. I fear that as an industry, we may end up making loans that may not be paid back. The defaults will increase. Delinquency can become a problem if risk management is not done right. And my bigger worry is that we might repeat what happened in China with the P2P debacle that led to even more weird things such as a new desperate collection system with trouble written all over it. We came across an online platform in China that has become known not just for controversial practices allowed on its website for extending loans, but in its collection process as well. The platform, Jiedaibao, which connects lenders to borrowers, has started giving the general public access to debtors’ personal information (including phone number, address, ID number) to enable them to act as collecting agents for outstanding loans from lenders registering on the website. It won’t be preposterous to suggest that it is just a matter of time before things get out of hand.