Streamlining Lending Decisions with a Probability of Default Model based on Transactional Data
A Problem for QuickCash
Imagine yourself as a lender working for QuickCash, a business lending company specialized in swift loan application procedures. During the day, you are sitting at your desk, surrounded by a pile of loan applications. As a responsible lender, you understand the importance of making informed decisions that balance risk and reward.
The dilemma is clear: you can either accept the loan request without a thorough risk analysis and hope for the best, or decline the application and potentially missing out on a credible borrower and the opportunity to make a profit. It's a tough call, especially for a lender that keeps the promise of a fast loan accepting procedure.
Faced with these difficult choices, you begin to ponder a solution. How can you make more informed lending decisions? How can you accurately assess the creditworthiness of borrowers without compromising the efficiency of your operations? In search of a solution that would help you the best, you think of a solution that can provide a definitive answer of either accepting or declining the loan application.
The credit scoring tool, RiskNavigator, fills this void by implementing a PD model which enables the lender to make a credit acceptance decision based on a binary outcome ‘Accept’ or ‘Decline’. This powerful tool provides a systematic and data-driven approach, allowing lenders like QuickCash to assess borrower credibility accurately and efficiently, while streamlining the decision-making process.
How a Probability of Default Model Assesses the Credit Worthiness of a Client
A PD (Probability of Default) model is a tool used by lenders to estimate the likelihood that a borrower will fail to repay a loan. It helps assess the creditworthiness or the risk associated with lending money to an individual or a business.
Think of it this way: When you apply for a loan, the lender wants to determine if you are likely to repay the loan or if there is a higher chance that you may default, meaning you won't be able to pay it back. The PD model uses historical data and various factors such as your credit history, income, and other relevant information to make an educated guess about the probability of you defaulting on the loan.
Prominent banks have already implemented robust Probability of Default (PD) models that perform similar tasks in their loan application processes. These models typically extend beyond the binary outcome of solely accepting or declining a loan. They are designed to provide multiple outcomes, enabling banks to diversify their offerings with varying interest rates based on the borrower's risk profile. For these models to recognize patterns for creditable and non-creditable loan applicants a PD model it requires data.
This poses an immediate challenge for QuickCash. Given their focus on relatively small loans and a swift application procedure, the applicants are not required to submit extensive documentation that proves their credibility. Additionally, analyzing these documents would introduce a longer application process and increase the costs. However, there exists a valuable and efficient source of data that can be leveraged — the power of real-life transactions. By using PSD2 regulation, the RiskQuest Navigator now enables the retrieval and transformation of applicant transaction data into relevant predictors for the PD model.
Now, one might wonder: how can we effectively train this PD model?
Training a Probability of Default Model for Lending Decisions
To make the transaction data usable, it must first pass through the RiskNavigator pipeline. In the pipeline, transactions will be classified so that they can later be aggregated by their corresponding label. Simple aggregations could be performed like, for example, taking the sum or count of a certain expense category. Additional predictors could be the average amount paid to debt-collectors, variability in the time-window between rent payments , or hand-crafted predictors, such as supplier or customer concentration.
All predictors are then divided into buckets, allowing for easy interpretability in the risk assessment process and outcome. To train the PD model, the RiskNavigator employs common modelling practices from the domain of credit risk. Simple models are preferred over more complex models due to their explainability, and established reputation.
Notably, the company sector (SBI) proves to be a strong predictor, with sectors like 'Information & Communication' associated with low default risk, and 'Wholesale and retail trade: automotive repair' associated with higher risk. Other significant predictors include free cash flow, supplier concentration, and customer concentration. The final model can be summarized into a scorecard (as shown below), where the predictors are assigned risk points based on their corresponding bucket. Finally, the score is the sum of all points based on the input value of the observed predictors.
How QuickCash uses the Probability of Default Model in their Loan Application Procedure?
The PD model, integrated into RiskNavigator, offers lenders invaluable benefits in their workflow. It enables quick and effective assessment of applicants' eligibility for credit, allowing lenders to promptly decide whether to accept or decline a loan.
One of these lenders, QuickCash, are impatiently awaiting the result of the model, and are very curious how they can use it to score their applicants. A typical scorecard of a credit risk model with the earlier stated predictors from above could look as follows:
QuickCash has to determine their risk appetite by means of a score cutoff, for which they settle at 550. On a given day, a client applies for a loan with the following data:
SBI-code: Construction (‘F’) = 100 points
Free Cash Flow: 350000 = 230 points
Customer concentration: 0.5 = 120 points
Supplier concentration: 0.3 = 150 points
Total: 600 points
Hooray! We can provide a loan to this applicant. Through trial & error QuickCash can determine whether the cutoff value of 550 matches their risk appetite. (footnote: these scores are purely for illustration purposes and cannot be directly compared to a framework such as FICO).
Months later, business is booming and manually inspection scorecards has becoming intensive manual labour for the employees at QuickCash (who are known for their swift procedures). Since the RiskNavigator can be connected easily through API-connection, the lenders at QuickCash can now easily request PD-scores, compare them against their cutoff, and automatically accept or decline incoming applications.
With RiskNavigator and the power of transaction-based PD models, lenders like QuickCash are revolutionizing the lending landscape. By leveraging real-time transaction data and implementing statistical modelling techniques, lenders can now make more informed and accurate credit acceptance decisions. This streamlined process not only enhances efficiency but also ensures responsible lending practices. As the financial industry continues to evolve, embracing advanced technologies and data-driven solutions will be key to staying ahead in the competitive lending market.
If you want to learn how the RiskNavigator can revolutionize your business, contact Hans Heintz [email protected] or Tom Corten [email protected]