Posted by Joaquin Delgado :: July 7, 2007 @ 9:43 am

In this second edition of our technical series I will try to answer some of the questions lingering from the previous posting as I explain further how we generate portfolio recommendations.

When using LendingMatchâ„¢, lenders are invited to input the amount they intend to lend and the level of risk they are willing to take on a scale of 1 to 5; 1 being the most conservative and 5 being the most aggressive.




As the user moves the slider, he or she can see a pie chart that shows the projected composition of his or her portfolio. A recommended portfolio that best matches the user’s risk tolerance is automatically created and presented back to the user.

As noted two weeks ago, LendingMatchâ„¢ also takes into account, in addition to the various levels of risk associated with each loan grade, the time left to close each loan request (promoting loans that will close soon), the amount necessary to close a loan, and the connections between lenders and borrowers (such as belonging to the same group(s) or network(s) in Facebook).




Now let’s dig into the recommendation process.

The Recommendation Process




The loan recommendation process can be broken down into 4 phases: Scoring, Diversification, Matching and Allocation. Let’s now describe each one of them:

• Scoring is done based on business rules (Lending Club’s credit policy) primarily taking into account the borrower’s credit score (FICO) and risk modifiers such as loan amount and debt-to-income ratio, and assigning a loan grade ranging from A to G.

• The diversification step pre-computes several optimal portfolio distributions (mix) for a series of user risk functions using the Markowitz Model and establishes the allocation in each loan grade. This translates into the risk sliding bar and the pie-chart with possible allocations among loan grades that ultimately translates into a requested number of loans from each grade, with comparable amounts allocated to each loan.

• For each loan grade, the system matches lenders to loans by degree of connection with the borrowers and generates a LendingMatch™ rank for each loan.

• Finally, a priority queue is created for each loan grade based on LendingMatch™ values and a number loans are selected to compose the portfolio according to the allocation computed by LendingMatch™, and are presented back to the user.

Markowitz and the Efficient Frontier

We use the Markowitz Model to analyze the construction and qualitative nature of a portfolio’s risk-return characteristics. By using Markowitz our system obtains a series of weights that determines the allocation over several loan grades, determining a portfolio that lies on the Efficient Frontier.

The Efficient Frontier is the collection of portfolios constructed from the given set of loans that have the lowest possible risk for a given level of expected return. (See Modern Portfolio Theory for more on Markowitz and the Efficient Frontier).

Note that the weights of each loan making up the portfolio may themselves be subject to constraints (for example, no one loan can have a weight of more than 20 percent or less the than 5 percent of the portfolio). This is especially useful when handling real-time inventory.

Note that to apply any of these models we need to predict how loans from each grade will perform, which we do, based on historical default data from the credit bureaus. We will talk about how precisely we measure risk and diversification in our next series.

Until next time - have a good weekend and a productive upcoming week.

-- Joaquin from Lending Club