Exercise 1: Credit Data Preparation Prompt Engineering
Objective:
Learn to craft effective prompts for comprehensive credit data preparation and feature engineering.
Background:
Credit Analytics Data Scientists are responsible for preparing and transforming data for credit risk analysis. A key challenge is developing robust data preparation pipelines that handle complex financial data sources.
Exercise:
1. Scenario:
You need to prepare and transform data from multiple sources for a credit risk modeling project that will predict small business loan defaults.
2. Basic Prompt Example:
How should I prepare data for a credit risk model?3. Prompt Improvement Activity:
- Identify the limitations of the basic prompt
- Add specific details about the data sources
- Include context about the modeling objectives
- Request structured data preparation methodology
- Ask for feature engineering recommendations
4. Advanced Prompt Template:
I am a Credit Analytics Data Scientist at a [size] financial institution preparing data for a small business loan default prediction model with these characteristics:
Data sources:
- Internal loan performance data: [size, format, time period]
- Financial statement data: [balance sheets, income statements, cash flow statements]
- Payment history data: [delinquencies, payment patterns]
- Credit bureau data: [business and personal credit reports]
- Macroeconomic data: [indicators, sources, granularity]
- Alternative data: [any non-traditional data sources]
Data challenges:
- Missing values: [extent and patterns of missingness]
- Data quality issues: [inconsistencies, errors, outliers]
- Class imbalance: [default rate in historical data]
- Temporal considerations: [point-in-time vs. current data]
- Regulatory constraints: [data usage limitations, privacy requirements]
Modeling objectives:
- Target variable: 90+ day delinquency within 12 months
- Model type: [logistic regression, machine learning, etc.]
- Interpretability requirements: [need for transparent features]
- Performance goals: [discrimination power, calibration]
- Implementation constraints: [production environment limitations]
Please help me develop a comprehensive data preparation and feature engineering plan by:
1. Outlining a structured data preparation approach that includes:
- Data exploration methodology
- Data quality assessment framework
- Missing value handling strategy
- Outlier detection and treatment
- Data integration approach
- Temporal alignment considerations
- Train/test/validation splitting strategy
2. For each data source, recommend:
- Key variables to extract
- Necessary transformations
- Quality checks to perform
- Derived features to create
- Common pitfalls to avoid
- Documentation requirements
3. Suggesting feature engineering techniques for:
- Financial statement data (ratios, trends, volatility measures)
- Payment behavior (delinquency patterns, utilization metrics)
- Credit bureau information (score transformations, attribute extraction)
- Macroeconomic factors (relevant indicators, transformations, lags)
- Alternative data sources (meaningful signals, transformations)
- Interaction terms and non-linear transformations
4. Recommending approaches for:
- Feature selection methodology
- Multicollinearity assessment
- Handling class imbalance
- Point-in-time feature creation
- Feature stability testing
- Data pipeline documentation
- Reproducibility and version control
Format your response as a comprehensive data preparation and feature engineering plan that balances statistical rigor with practical implementation considerations.5. Evaluation Criteria:
- Does the prompt clearly describe the data sources and their characteristics?
- Does it provide context about data challenges and modeling objectives?
- Does it request a structured data preparation approach?
- Does it ask for specific feature engineering techniques for each data source?
- Does it consider practical implementation considerations?
6. Practice Activity:
Create your own advanced prompt for credit data preparation related to:
- A retail credit card fraud detection model
- A commercial real estate loan loss forecasting model
- A consumer loan prepayment prediction model
Exercise 2: Credit Model Development Prompt Engineering
Objective:
Develop skills to craft prompts that help design and implement effective credit risk models using advanced analytics.
Background:
Credit Analytics Data Scientists must develop models that accurately predict credit risk. A key challenge is selecting and implementing appropriate modeling techniques that balance predictive power with interpretability and regulatory compliance.
Exercise:
1. Scenario:
You need to develop a machine learning model to predict early warning signs of credit deterioration in a commercial loan portfolio.
2. Basic Prompt Example:
What machine learning model should I use for early warning signs?3. Prompt Improvement Activity:
- Identify the limitations of the basic prompt
- Add specific details about the commercial loan portfolio
- Include context about early warning objectives
- Request structured model development methodology
- Ask for model evaluation and implementation recommendations
4. Advanced Prompt Template:
I am a Credit Analytics Data Scientist at a [size] financial institution developing a machine learning model to predict early warning signs of credit deterioration in our commercial loan portfolio with these characteristics:
Portfolio details:
- Loan types: [term loans, lines of credit, etc.]
- Industry segments: [distribution across sectors]
- Size range: $[X] to $[Y] in exposure
- Current portfolio performance: [delinquency rates, charge-offs]
- Available historical data: [time period, default events]
- Current early warning approach: [manual reviews, simple rules]
Early warning objectives:
- Prediction horizon: 3-6 months before severe deterioration
- Warning signs to detect: [covenant breaches, payment issues, financial deterioration]
- Desired sensitivity/specificity balance
- Action triggers: [watch list placement, relationship manager notification]
- False positive tolerance: [resource constraints for follow-up]
- Explainability requirements: [need to justify to relationship managers]
Technical constraints:
- Data refresh frequency: [daily, weekly, monthly]
- Implementation environment: [technology stack, processing limitations]
- Integration requirements: [existing systems, workflows]
- Monitoring capabilities: [drift detection, performance tracking]
- Regulatory considerations: [model risk management, fair lending]
Please help me develop a comprehensive credit early warning model by:
1. Recommending appropriate modeling approaches:
- Comparison of suitable algorithms (tree-based, neural networks, etc.)
- Pros and cons of each approach for this specific use case
- Ensemble methods to consider
- Handling class imbalance in the modeling approach
- Appropriate threshold selection methodology
- Recommended approach with rationale
2. Outlining a model development methodology:
- Feature importance and selection approach
- Hyperparameter tuning strategy
- Cross-validation framework
- Performance metric selection and justification
- Model stability assessment
- Interpretability techniques
- Bias detection and mitigation
3. For the recommended approach, provide:
- Detailed implementation steps
- Code structure recommendations
- Common implementation pitfalls
- Computational efficiency considerations
- Testing and quality assurance approach
- Documentation requirements
4. Suggesting an implementation and deployment strategy:
- Model scoring process design
- Alert generation and prioritization
- Integration with existing workflows
- User interface considerations
- Feedback loop mechanisms
- Monitoring and maintenance approach
- Periodic retraining framework
Format your response as a comprehensive early warning model development plan that balances predictive power with practical implementation considerations and regulatory requirements.5. Evaluation Criteria:
- Does the prompt clearly describe the commercial loan portfolio and early warning objectives?
- Does it provide context about technical constraints?
- Does it request comparison of appropriate modeling approaches?
- Does it ask for detailed implementation steps and deployment strategy?
- Does it consider regulatory requirements and practical constraints?
6. Practice Activity:
Create your own advanced prompt for credit model development related to:
- A behavioral scoring model for credit line management
- A loss forecasting model for stress testing
- A customer response model for credit marketing campaigns
Exercise 3: Credit Portfolio Analytics Prompt Engineering
Objective:
Learn to craft prompts that help generate actionable insights from credit portfolio data for risk management and strategy.
Background:
Credit Analytics Data Scientists must analyze portfolio data to identify trends, opportunities, and risks. A key challenge is translating complex analytical findings into actionable business insights.
Exercise:
1. Scenario:
You need to develop a comprehensive credit portfolio analysis to identify concentration risks, performance trends, and optimization opportunities.
2. Basic Prompt Example:
How should I analyze our credit portfolio?3. Prompt Improvement Activity:
- Identify the limitations of the basic prompt
- Add specific details about the credit portfolio
- Include context about analysis objectives
- Request structured analytical approaches
- Ask for visualization and communication recommendations
4. Advanced Prompt Template:
I am a Credit Analytics Data Scientist at a [size] financial institution conducting a comprehensive analysis of our credit portfolio with these characteristics:
Portfolio composition:
- Loan types: [consumer, commercial, mortgage, etc.]
- Total exposure: $[X] billion
- Number of accounts: approximately [Y]
- Geographic distribution: [regional concentration]
- Industry sectors: [for commercial portfolio]
- Risk rating distribution: [current risk profile]
- Performance metrics: [delinquency rates, charge-offs by segment]
Analysis objectives:
- Identify concentration risks (geographic, industry, borrower)
- Detect emerging performance trends and early warning signals
- Assess risk-adjusted returns across segments
- Identify optimization opportunities (pricing, exposure limits)
- Support strategic planning and risk appetite setting
- Enhance regulatory reporting and stress testing capabilities
Available data:
- Account-level performance data: [time period, granularity]
- Borrower information: [demographics, financials]
- Collateral details: [for secured lending]
- Pricing and profitability data: [margins, fee income]
- Macroeconomic indicators: [relevant to portfolio performance]
- Risk parameters: [PD, LGD, EAD estimates]
Please help me develop a comprehensive credit portfolio analysis framework by:
1. Recommending key analytical approaches for:
- Concentration risk assessment (single name, sector, geographic)
- Vintage analysis and cohort performance tracking
- Migration analysis and transition matrices
- Correlation and contagion risk assessment
- Risk-adjusted return analysis
- Stress testing and scenario analysis
- Comparative benchmarking
2. For each analytical approach, provide:
- Detailed methodology and calculation steps
- Required data elements and preparation
- Statistical techniques to employ
- Interpretation guidelines
- Common pitfalls and limitations
- Business implications and action triggers
3. Suggesting effective visualization techniques for:
- Executive-level portfolio overview
- Detailed risk manager dashboards
- Trend analysis and forecasting
- Concentration visualization
- Performance attribution
- Scenario comparison
- Drill-down capabilities
4. Recommending approaches for:
- Translating analytical findings into business recommendations
- Prioritizing identified risks and opportunities
- Developing actionable insights for different stakeholders
- Creating effective reporting frameworks
- Implementing ongoing monitoring mechanisms
- Measuring impact of portfolio actions
Format your response as a comprehensive credit portfolio analysis framework that balances analytical rigor with practical business application.5. Evaluation Criteria:
- Does the prompt clearly describe the credit portfolio and analysis objectives?
- Does it provide context about available data?
- Does it request specific analytical approaches with detailed methodologies?
- Does it ask for effective visualization techniques?
- Does it consider practical business application of analytical findings?
6. Practice Activity:
Create your own advanced prompt for credit portfolio analytics related to:
- A CECL/IFRS 9 provision forecasting analysis
- A credit strategy optimization for a specific product line
- A competitive analysis of market credit risk and pricing