Greetings,

US consumers are confident about their spending due to higher take-home pay after the tax reform, a period of decades-low unemployment, and continued economic expansion. Consumer confidence clocked in near its highs this past week at 95.3. Retail sales rose across categories boosting US GDP growth, fueled by expansion in consumer credit.

Real wage growth remains a weak point. Real wages have not kept pace with GDP growth and raises the specter of a stagflationary scenario. With inflation picking up recently, the Labor Department’s inflation-adjusted wage measure fell to $10.76 an hour, down by 2 cents YoY. You can read about our deep-dive into US inflation measures and their effects on Fed policy here.

In FinTech financings, SoFi is reportedly looking for a billion-dollar credit line as it seeks to offer more financial products online. SoFi is expanding into checking accounts with SoFi Money and this credit line could be a precursor to a potential IPO.

Kroll has rated Prosper’s latest $500 Mn securitization A+ on the senior tranche. This is only the 2nd time that Prosper has received an A+ rating. Progressa, a Canadian paycheck lender, has raised $84 Mn to expand operations.

In this week’s newsletter we continue our series on valuations of whole-loans and dive into How to Forecast Future Cashflows. You can read our previous piece on the Top 5 Questions to Ask to Know if Your Valuation Makes Sense here.

Valuation Framework

The price of a loan is the present value of the loan’s cashflows (after accounting for losses and prepayments), discounted by an appropriate discount rate:

Once we know the value of each loan in the portfolio, we can then calculate portfolio value by summing up individual loan values.

In this newsletter, we’ll break down the mechanics of forecasting cashflows – the “numerator” – at a loan and at a portfolio level. In a later newsletter, we’ll share how to determine the appropriate spread and discount rate.

Forecasting Cashflows using a Credit Model

There are multiple methods to forecast cashflows. The best technique depends on the quality, consistency, and history of the data. Segmenting portfolios by cohorts based on vintages, credit scores, term, age and other attributes is a very popular approach.

If there is sufficient history of loan-level data as well as predictive attributes, PeerIQ will apply a loan-level credit model. A loan-level model better captures each loan’s idiosyncratic behavior rather than smoothing away differences in larger cohorts.

Cashflow Modeling with Credit Models

PeerIQ’s credit model is a machine learning-based, ‘sparse’ transition model. What is a credit model? The inputs of a credit model are loan attributes and borrower factors such as:

  • Originator of the Loan
  • Term of the Loan
  • Interest Rate on the Loan
  • Original Principal of the Loan
  • Age of the Loan (Months on Book)
  • Loan Grade
  • Prior Loan Status
  • Borrower’s Credit Score at Origination

The model output is a prediction of the next month’s loan status.

The sparse transition model assumes 7 possible loan statuses which describe the performance of the loan:

  1. Current (or Status C)
  2. 1 Month Delinquent (or Status 1)
  3. 2 Month Delinquent (or Status 2)
  4. 3 Month Delinquent (or Status 3)
  5. 4 Month Delinquent (or Status 4)
  6. Default (or Status D)
  7. Paid Off (or Status P)

States 1 thru 5 above are transient states. A loan status of current means the loan is continuing to perform. Loans occupy these transition states until they finally arrive at a terminal “absorbing” state such as Default, Paid Off. Loans can transition from one state to another under a specific set of rules.

For example, a loan can only transition to “2-month delinquent” next month if the state of the loan in the previous month is “1-month delinquent”. However, as depicted in the chart below, a loan can move to the Current state (“C”) for up to 4 months after it has been delinquent:

Exhibit 1: Transition states

Source: PeerIQ

The output of our model is a matrix that displays the “transition rates” – probabilities that a loan moves from state to another given its current state. Although the modelling may seem complex, the transition matrix output is very intuitive.

The table below indicates that that a loan that is two months of book that current has a 97% probability of staying (“transitioning to” the subsequent state of) current. However, a loan that is 1-month delinquent has a 79.9% chance of becoming 2-months delinquent.

Once a loan arrives at a terminal absorbing state (“Defaulted” or “Paid Off”) there is no future possible transition as there are no further cashflows. Therefore, the probability of a loan remaining defaulted is 100%.

Exhibit 2: Transition Rates/Probabilities for Month on Book 2

Source: PeerIQ

Transition matrices are very useful. A portfolio manager can use these transition probabilities to perform Monte Carlo simulations on their P&L. A risk manager can check to see if there is deterioration in the probabilities of loans curing from one month to the next. An analyst can compare the model-estimated transition probabilities to empirical probabilities to test for leading trends.

Model Calibration

We estimate transition probabilities by fitting the model to historical data.

PeerIQ derives expected transition probabilities for each loan using our machine learning approach on public MPL datasets.

The charts below show cumulative losses and prepayments on public MPL loan data that is used as a training set for the PeerIQ model.

 Exhibit 3: Cumulative Loss Profile of Personal Loans by Vintage

Source: PeerIQ

Exhibit 4: Cumulative Prepayment Profile of Personal Loans by Vintage

Source: PeerIQ

Bringing it All Together – Testing for Model Fit

Below is a sample prediction of Expected Default Rates for a loan as compared to realized defaults. As you can see, the projected default curve is very close to historically realized defaults indicating strong out-of-sample fit:

Exhibit 5: Historical vs Predicted Default Rates

Source: PeerIQ

Next week, we will look at the methodology we use to determine the discount rates. Reach out to learn how PeerIQ can help you with valuations.

 

PeerIQ Mentions: 

 

Conferences:

 

Industry Update:

Lighter Fare: