Hands-on Machine Learning for Credit Risk Analytics - Masterclass in Python & SAS
In house (i.e., face-to-face at your company location, or online)
Anytime

Details:

• Topics you don't want to miss:
  • Demystify machine learning techniques and learn how to successfully apply them for credit risk prediction using real data
    
  • "Boost" your PD models
  • Learn how to program decision trees, random forests, neural networks (and more) for default and PD prediction from scratch
  • See how efficient feature selection is implemented
  • Compare and interpret various train/test/validation-split strategies specific for credit risk
  • Create your own cross-validation strategies
  • Apply various under-/over-/synthetic-sampling strategies
  • Learn how to compare model outputs: stability, discrimination (ROC and CAP) and calibration 
  • Forecasting PDs: TTC and PIT
  • LGDs: discount rates and selection bias
  • Prepayment models for mortgage and corporate loans
  • CECL and IFRS 9 models for multi-period risks
• Speakers: Professor Daniel Roesch and Professor Harald Scheule
• Content: in this course, we will develop hands-on credit risk solutions. All examples will be implemented on site using real credit data with over 622,000 observations with 15,000 default and loss observations, over 250 teaching slides and approximately 1,500 lines of code. Here are some examples (click on links)
  • Example 1: Validation using the area under the ROC curve
  • Example 2: Decision Trees
  • Example 3: Predicting lifetime expected losses for CECL and IFRS 9
• Please bring your laptops. All executions in parallel on multiple screens using Python and SAS. Python is open source, SAS will be provided in the cloud. Wifi access is available
• Outcomes:  after this course you will have a good understanding of current challenges in the credit risk industry, merits and pitfalls of various methods, have successfully built your own models from real world credit data
• A confirmation of 12 hours of continuing professional development will be provided upon completion of the Masterclass

Day 1:

Time	Topic
9.00	Data Pre-processing for Credit Scoring Credit scoring techniques for corporate & retail data Outlier detection Standardisation & categorisation Default definitions Non-monotone relations between defaults and risk factors Weight-of-evidence, categorisation, splines Multivariate non-monotone relationship
10.30am	Morning tea
11.00	Probabilities of Default (PD) Discrete time models (Logit & Probit) Comprehensive modeling including vintage, age and time effects Maximum likelihood estimation Model stability, discrimination & calibration Through-the-cycle & point-in-time Predicting Financial Crises
12.30pm	Lunch
1.30pm	Machine Learning Concepts Statistical and Machine Learning Thinking in Machine Learning Loss Functions Train-Test-Validation-Split Standardisation and Scaling Bias-Variance-Tradeoff Oversampling and Undersampling Cross-Validation Hyper-Parameters and Tuning Supervised and Unsupervised Learning Classification and Regression vs. Dimensionality Reduction
3.00	Afternoon tea
3.30pm	Machine Learning Validation Validation framework Backtesting PDs: Discrimination, Calibration, Stability Confusion matrix, ROC curves, AUROC & accuracy ratio Brier scores Binomial test Jefreys prior test Hosmer-Lemeshow Calibration curve and reliability diagram Tests on Model Stability Portfolio dependence Traffic lights Qualitative validation
5.00	Networking with fingerfood and drinks

Day 2:

Time	Topic
9.00	Unsupervised Machine Learning for PDs Bayesian regression techniques and classifiers K Means Clustering Hierarchical Clustering Principal Component Analysis
10.30am	Morning tea
11.00	Supervised Machine Learning for PDs Neural Networks and Deep Learning K Nearest Neighbours Bagging and Boosting Decision Trees Random Forests (Extreme) Gradient Boosting Support Vector Machines
12.30pm	Lunch
1.30pm	Loss Given Default (LGD) and Exposure at default (EAD) Computing observed LGD LGD discount rates Marginal modeling (linear, fractional logit & beta regressions) PD-LGD models Conversion measures Credit lines & flexible repayment schedules Backtesting LGD and EAD
3.00	Afternoon tea
3.30pm	Current expected credit losses (CECL) for US GAAP and IFRS 9 Loan loss provisioning and Basel capital 12-month expected loss vs. lifetime expected loss Rating class formation and roll rate analysis Macroeconomic forecasts Multi-period PD forecasts based on the macroeconomy and lifecycle Prediction of lifetime expected losses Significant increase in credit risk (SICR)
5.00	End of day

Terms and conditions:

• Participant numbers are flexible (past classes were offered to 10-200 participants)
• Costs are between $100-1,000 per participant
• Face-to-face delivery is on two consecutive days
• Online delivery is via WebEex and may be in weekly blocks
• Content changes are possible
• Language SAS, Python or both
• All codes and teaching materials are included
• Classes are hands on most participants bring laptops
  

Registration:

RETURN TO TRAINING EVENTS