Archive

Going beyond Facebook has sparked quite a lot of press and the interest of more users, most of whom are unfamiliar with P2P Lending. Along with the opening of our website, we have also received several requests to expose raw data and performance statistics about loans and lenders. We believe this is very important in order to maintain full transparency and allow users to see the benefits of Lending Club for themselves. While we work on ways of exposing this information in a private, secure and efficient manner, we have already taken some steps in this direction. Here are some datasets we have made available:

Top Lenders by Performance

The Lender Rankings page contains a list of the top 100 performing portfolios (one per lender) containing at least $50, in order of estimated return on investment. Lenders are identified by screen name to protect their privacy. The table displays the following information:

• Portfolio rank

• Lender screen name

• Portfolio name

• Amount of initial investment

• Estimated return on investment

Click to Enlarge Image

Calculation of estimated return on investment

The estimated return on investment is calculated by taking the average interest rate of a portfolio and deducting the Lending Club servicing fee, defaults and late loans expected to default.

• The average interest rate excludes the origination fee for each loan (origination fee varies according to each loan grade)

• The late loan amount is calculated by deducting monthly payments already made from the principal

• Loans more than 1 month late are estimated to default at a 50% clip

• Default losses (projected defaults for late loans and actual defaults) are calculated using a 90% clip (10% recovery rate) for default loans

• Paid/repurchased loans are included in the current average interest rate calculation. This may change in the future

• The late/default rates are projected out to one year

• Lending Club servicing fees are 1% for all loans

• All loans are included in the calculation for late/default rates in the formula regardless of age

Formula:

Estimated ROI = Average Interest Rate - (Loss due to Late Loans - Loss due to Default Loans) - Lending Club Servicing Fee

Where:

• Loss due to Late Loans = Sum(50% * (unpaid percentage) * (interest rate of the late loan))

• Loss due to Default Loans = Sum(90% * (unpaid percentage) * (interest rate of the default loan))

• Lending Club Servicing Fee = 1% / 3 years = 0.33%

For those that want to use this information as input to other computations, we also expose this information via a downloadable XML file (available here) that gets updated daily.

Member Map

Available directly on our homepage, this feature displays information about current lenders, borrowers and issued loans plotted on a Google Map. Members are identified by screen name, and location is based on zip code only (with a certain level of Geocode randomization) to ensure members’ privacy. When an icon is clicked, a call-out displays the following information:

Borrowers / Issued Loans	Lenders
• Screen name	• Screen name
• Title / Link to the loan page	• Amount of portfolio initial investment
• Amount of the Loan	• Location (city, state)
• Amount left to fund
• Location (city, state)
Click to Enlarge Image	Click to Enlarge Image

This information is also available in XML format for external consumption here, and the file is updated on a daily basis.

As noted, we will be releasing additional data in the near future. Meanwhile, we hope the above information will be helpful to those who are evaluating Lending Club.

Better Rates. Together.

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 LendingMatch™, 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, LendingMatch™ 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

LendingMatch™: Diversification and Matching

Diversification in finance involves spreading your money around into many types and numbers of investments. When it comes to Lending Club, we offer various loan classes, ranging from A to G for lenders to choose from. At any given time a lender may decide to lend a given amount of money which can be allocated across a number of selected loans from different classes, thus forming a diversified loan portfolio.

The goal of LendingMatch™ is to help the user build a loan portfolio in such a way that their allocations are optimal with respect to the specified lender's risk/reward utility function, and match their "profile" and "risk criteria" for each portfolio while keeping the overall loan portfolio "diversified". This is generally considered a search and matching as well as a loan allocation optimization problem.

We provide two routes for lenders to build their loan portfolio: a) via a recommendation that takes into account your return goals and generates a suggested portfolio and b) build your portfolio a-la-carte with browse and search functionalities. You can also start with a recommendation and then continue a-la-carte.

In both cases the system generates a LendingMatch™ rank for each loan that has both a personalization component (specific to the lender and lender-borrower pair, including affinities and connections through social networks) as well as a loan component that considers other factors such as remaining amount to fund and time to close.

This rank is used to select loans in the recommendation process as well as to display results to the user when they search or browse the loan inventory. At the same time the system has to be able to optimize the processing of loans with respect to the existing inventory to ensure proper money flow between lenders and borrowers.

Some immediate questions that come to mind are:

• How does LendingMatch™ consider the "connections" between the lender and borrower?
• What is the "optimal" portfolio that can be constructed from a given set of loans which have the lowest risk for a given value of the expected return?
• How many loans should the portfolio(s) contain in order to be considered "diversified”?
• Can a lender concentrate their funds on a few loans and still be diversified?
• How is “concentration” measured?

Stay tuned for a deeper dive into these questions!

Joaquin