In this video, we’ll look at how platform lending like those done on LendingClub or Funding Circle actually work. Another less accurate name for this is peer-to-peer lending or P2P lending, which we’ll use interchangeably with platform lending for now and distinguish between these in the next video. While we’re discussing these platforms, try to relate these new concepts to the credit analytic methods that were discussed in the previous module. So you can see how these are applied basically the same way, but in new platform settings. Again, let’s start with an example. I wanted to think about the last time you applied for a loan, a mortgage or a credit card.
Take a minute to think about the process that you went through and the stakeholders that you dealt with. There’s you, the borrower, and on the other side, the lender who provided the capital, and the bank which either lends directly to you or serves as a bridge between you and the ultimate lender. Now, let’s take a look at the processes that you went through. First, you most likely filled out an application, either online or in person. In the application, you would have to give some personal financial information about yourself, like your income, debt levels, assets, and your social security number to get your credit scores. Once the bank receives the application, it’ll do some credit analysis.
Based on the information that you filled out, it’ll essentially run the credit model that we discussed to predict your probability of defaulting on that loan. It will then make a decision based on the probability of whether to actually give you the loan, and if so, how much interest rate it will charge you for that loan? If the loan is approved and you agree to take it, then it goes into the underwriting stage, which is just a fancy name for actually putting the loan terms into a legal contract. Depending on the loan type and the market, registering that contract with the appropriate regulators.
Finally, you get the money and the loan will either be financed internally in the bank or more likely be packaged and sold to investors. Either directly or securitized in the form of asset-backed or mortgage-backed securities. But the bank’s job is not over, because as you start paying back, it’s still have to service the account by collecting your payments and doing the periodic paperwork. If you don’t pay back, they try to collect the money from you. This could be outsourced and done by different bank or different institution. This is a generic lending process in a nutshell. Now, where do online platforms fit in? If you think about it, the banks are already a platform that connects the borrower with a lender.
However, there are some functions that the bank does better than others, and there are other functions that are better served by tech companies. What are these? Well, the banks are pretty good at generating loans and maintaining a service in the accounts. These are their core competencies and should be retained in any case. However, most banks are not naturally data analytics companies. The steps before the underwriting are essentially data analytics. This is where online lending platforms come in. As tech companies, data analytics and user interface are their bread and butter.
So they could come in and take over the functions of doing the interface to take the loan applications and capture the data, and providing the analytics to dig through the data and generate the default probability predictions. This is essentially what platform lending is. Platforms take over the front-end functions of customer acquisition and data processing, then use the banks to help with underwriting and servicing the loans. The revenue on the loans will be split between the two parties. The platform would usually connect front-end user interface and back-end servicing functions using an application programming interface or API. Now, let’s zoom into the platform and look at how these functions are actually done.
For example, let’s look at LendingClub, one of the earliest online lending platforms. First, you can fill out a loan application directly on the platform’s website. It would essentially be the same application that you fill out in a bank. The application gathers financial data about you in order to generate the variables and ratios they’ll be fed into the platform’s credit models. These models are essentially the ones that we talked about in the last module. They could be logistic regression, clustering, or other machine learning algorithms. Bottom line is, using historical data that it has, the platform would have estimated the key parameters of the model.
So when your variables come in, the model will be able to generate a prediction about your likelihood of defaulting on that loan, which is the key output of this step. The next step should be relatively new to you. If you browse through the loans on LendingClub, you’ll see that each loan has a grade, like A, B, C, etc. This is because it’s more efficient to group the loans into grades and give the same interest rate to all loans in that grade, rather than giving a different interest rate to each applicant based on their predicted probability of default. This would also avoid potential legal problems like interest rate discrimination. This practice would also avoid potential legal problems like interest rate discrimination.
This process is usually done using a clustering algorithm, which takes the predicted probability of default, which is different for each person and cluster them into a set of buckets, where within each bucket, the default probabilities are close enough. We can denote these buckets as A, B, C, and so on. If you want more granularity, you can also cluster again and divide each bucket into sub buckets like A1, A2, etc. Finally, each of these buckets will be assigned an interest rate, which again will be a function of the predicted default risk of that bucket,
the loan characteristics like how long the loans are for, as well as macroeconomic conditions like the current interest rate environment. The output of this step is an individualized interest rate for each bucket. Because these are just data analytics, they could be done in a matter of seconds. So if you are an applicant, say you applied for a $10,000 loan, and a couple of seconds the platform would have done the prediction and placed your application into the appropriate bucket, and obtain the associated interest rate. Say the application falls into the B2 bucket with an annual interest rate of 7.3 percent.
This will quickly be communicated back to the applicant, who will get a message saying, “Look, we got you a great interest rate of 7.3 percent, would you wish to proceed?” If you say, “Okay,” then your application together with a great data will be listed on the platform’s website for the investors to browse. Let’s now transition to the investor side of the platform. Here’s the loan from the last slide, a $10,000 loan in B2 grade, with an interest rate of 7.3 percent. Now, the key concept here is that the platform is not selling the entire loan to a single investor. Instead, it does a little bit of “securitization,” where it breaks up the loan into little chunks called notes.
For example, here, the $10,000 loan will be broken up into 400 notes of $25 each. If you go to say LendingClub’s website, you’ll see exactly that. You can choose to invest in increments of $25 as opposed to buying the entire loan. Now, what’s the purpose of this? In addition to making the loans more accessible to small investors, it also serves an important diversification purpose. In that investing across many loans reduces idiosyncratic default risks. To see this, let’s have another B2 loan in the platform by somebody else. Say, you’re an investor with $50 to invest. If you buy one loan with that $50, and the borrower defaulted because he’s hit by lightening, then you’re out of the entire amount.
Instead, because idiosyncratic default risks like being hit by lightning are not correlated across investors and don’t hit everyone at the same time. If you buy two loans with $25 each, these idiosyncratic risks will be diluted and reduced.
So from an investor perspective, based on a simple statistical principle called the law of large numbers, is imperative to invest in a large number of loans, ideally more than 30 in the same bucket as opposed to concentrating the investment in one or fewer. With other loans on the platform, these would allow an investor to construct a diversified portfolio that are customized to get the desired risk exposure without exposing the investor to unnecessary idiosyncratic risks. For example, you could build a portfolio of 200 grade E8 notes, 100 C4 notes, and 100 B2 notes. Now, the next step is underwriting.
If enough investor buy the loan so that it’s fully funded, the platform will send the data to its partner bank to write the contract and generate the paperwork. This would take awhile, so you’ll usually have to wait a few weeks before you see the underwritten loan showing up in your portfolio. Finally, once you have the portfolio, you need to get paid with interest and principal payments. Again, this servicing function will be outsourced to the partner bank, which will handle the payment collection and transmitting the payments to your bank accounts. The key thing here to note though, is that neither the platform nor the partner bank is assuming any default risk here.
Because the loan is directly sold from the borrower to the investor, the investor, there’s all risk associated with that loan. These steps constitute the mechanism of how the platform connects the borrower directly to the lender.