Understanding Net Interest Margin and Lending Terms

Net Interest Margin (NIM)

• Banks make money by charging interest on loans and paying interest on deposits. The difference between these rates is the NIM.
• Credit risk is a key factor in lending profitability. This is because borrowers may default on their loans, reducing the interest revenue generated by lending.
• Loan interest rates should be aligned with default risk and operating costs to ensure net gains and minimize losses.

Lending Terms

• Current: Borrower is up-to-date with loan payments.
• Delinquent: Borrower is behind on loan payments and may incur late fees and penalty interest. This is an early stage of default, typically measured in days (30, 60, 90, 180). Higher delinquency days increase the likelihood of default, triggering collection actions or suspension of the credit account.
• In Default: Borrower fails to repay the debt as agreed. This can be triggered by missing a certain number of payments or being delinquent for a specific period. Banks may treat 180 days of delinquency as being in default. Banks have to write off the debt as a loss.

Characteristics of Retail Lending

• Large number of product loans, with small values.
• Automation is required to be cost-effective due to high operating costs.
• Examples of retail lending products include: home mortgages, auto loans and leasing, personal loans, and credit cards.

Determinants of Lending Profitability

• Interest rate charged on loans: Higher rates generate more revenue.
• Default rate of the loan portfolio: Lower default rates minimize losses.
• To minimize losses due to defaults:
  • Assess potential borrowers' likelihood to default.
  • Lend to individuals who are better credit risks.
  • Limit the amount of credit provided.
  • Ensure the interest rate charged matches the average risk of default in the portfolio.

Credit Assessment

• Aim to estimate whether an applicant will successfully repay their loan.
• Based on
  • Applicant characteristics
  • Credit bureau information
  • Application information from other applicants
  • Repayment behavior of other applicants
• Types:
  • Judgemental: Based on the experience of the lender/loan officer and knowledge of the customer. Uses the "5Cs" of character, capital, collateral, capacity, and condition. Manual and not practical for high-volume lending.
  • Statistical: Based on multivariate correlations between inputs and risk of default. Can be automated.

Scorecards

• Credit scorecards are a key approach to retail lending credit assessment.
• Scorecards use historical lending data to develop models that estimate the probability of default based on a customer's profile.
• Scorecard Mechanics:
  • Points are assigned to each attribute and summed to get a credit score.
  • Points are based on predictive strength and correlation between characteristics.
  • Scores are compared to a threshold set by the institution. If the score meets or exceeds the threshold, credit is approved. Otherwise, it is denied.
• Characteristics of Scorecards:
  • Decision support tool for assessing credit risk.
  • Consistent with how credit analysts would evaluate credit.
  • Predictive variables represent major information categories.
  • Usually between 8 and 15 variables.
  • Score reflects the likelihood of an applicant being a good or bad credit risk.

Credit Bureaus

• Companies that keep records on consumers and businesses, such as their credit use and abuse.
• Data comes from creditors like banks.
• Banks can also get data about customers' loan details from other institutions.

Statistical Machine Learning Predictive Models: Scorecard Model Parameters

• Loss Event Definition: Defines what constitutes a loss event including:
  • 30/60/90 delinquent days
  • Debt that needs to be charged off or written off due to default
  • Borrower bankruptcy
  • Claims over $1000
  • Fraud over $500
  • Negative Net Present Value (NPV)
  • Less than x% of debt owed collected.
• Variable Availability and Selection: Determines which variables to use in the model, including demographic, behavioral data, and bureau score information.
• Credit Scores: Are obtained from credit bureaus or third parties that use bureau data to generate scores. Examples include FICO scores, which range from 300 to 850. Also includes raw bureau data like number of credit checks, total amount of credit, and delinquency history.

Scorecard Usage

• Low Scores:
  • Higher mortgage down payment
  • Higher interest rates
  • Requesting further documentation
  • Lower beginning credit limit
• High Scores:
  • Preferential rates (lower interest rates)
  • Higher credit limits
  • Premium cards

Drivers for Using Credit Scorecards

• Generate and test multiple scorecards based on different input parameters or customer segments.
• Easy to understand and use.
• Automation, processing efficiency, and repeatability.
• Regulatory concerns: Fairness and transparency - provide reasons for declining credit.

Scorecard Development Strategies

• Buy a generic or pooled data scorecard: Not specific to an institution's customers and products.
• Outsource to a credit risk vendor: Might not fully integrate business logic or be fully understood.
• Develop in-house: Requires data, internal expertise, and software tools.

Data Exploration Lab

• Histogram: .hist() function, used to visualize the distribution of numerical variables.
• Boxplot: .boxplot() function, visual representation of outliers for numerical variables. Identifies valid and invalid outliers.
• Countplot: .countplot() function, bar charts used to visualize the distribution of categorical variables.
• Kernel density estimation (kde): .kdeplot() function, used to visualize the distribution of events and non-events.
• Correlation heatmaps: .corr() and .heatmap() function, used to understand which variables may overlap and be redundant.
• Z-score: Measures how many standard deviations an observation is away from the mean. Outliers have a z-score of more than 3.

Data Merging

• Left joins can create duplicate records, which can inflate the number of bad/good events and skew results.
• One approach to avoid duplicates is to append observations as columns in one row, but this can lead to many empty columns.
• Feature selection and engineering can address this by selectively adding additional columns, such as averages or medians.

### Common Data Problems

• Dirty/Noisy: Negative age, for example.
• Inconsistent Data: Zero may mean actual zero or a missing value.
• Incomplete Data: Missing income data.
• Data Integration and Data Merging Problems: Currency differences.
• Overlapping Data Fields: Salary vs. Income

Missing Data

• Completely at random: The reason for missing data is unrelated to other data.
• At random: The reason for missing data can be inferred from other data.
• Not at random: There is an underlying reason for missing data that is structurally related.

Options for Dealing with Missing Values

• Keep: The variable that is missing could contain important information. Encode missing values.
• Discard: The variable may not be important.

Dive into the concepts of Net Interest Margin (NIM) and learn how various lending terms relate to borrower payments. This quiz explores the implications of credit risk, delinquency, and default on banking profitability. Test your knowledge on these crucial financial principles now!