1. Course wrap up
You did it! You've now mastered some basics of credit risk modeling in Python. Let's look at how far you've come since starting the course.
2. Your journey...so far
You started by preparing the data for modeling the probability of default, which is a crucial first step. Prepared data allows simple models to perform well, and you learn more about the data while preparing it.
You then developed, scored and understood the full breadth of using logistic regression models as well as gradient boosted trees. Gradient boosted trees are very popular right now, and having experience using them will prove useful.
After that, you analyzed the performance of the models, understood their financial impact, and used the results to formulate an overall strategy.
3. Risk modeling techniques
In this course, you used a specific type of model and framework. The model is what is known as a discrete time hazard model. This means all of the training data and the predictions are set for a specific point in time.
The framework used is known as a structural model framework meaning that the training data was used to explain the probability of default.
Some alternatives include a through the cycle model, where time is continuous, and a reduced-form model framework where a statistical approach is used for the probability of default based on distributions like the Poisson distribution.
4. Choosing models
There are many machine learning models available, so why did we use only a logistic regression and gradient boosted tree? Two primary reasons are these models are simple, and their performance for predicting the probability of default is good enough.
It's important to understand that many financial sectors prefer these simple models for statistical inference and interpretability. A model like a deep neural network can be seen as a black box. If it denies someone a loan and they ask you to explain why you can be legally obligated to explain why they were denied.
5. Tips from me to you
I have experience working with real-world data and applying machine learning in many different domains. With that, I have two tips to leave with you as you continue your data science journey.
First, focus on the data. I know it's not the most glamorous part of data science, but data engineering is very valuable. You learn about the business by preparing and understanding the data more than you do modeling. You also create value for the business because data is often considered a valuable asset.
Second, model complexity is a two-edged sword. Complex models may perform really well, but their complexity makes them difficult to understand. If business users cannot understand the model, they are unlikely to use it.
6. Thank you!
Thank you for taking this course! I hope you learned a lot about machine learning and credit risk. I'm proud of you for sticking with it and completing the class. Make sure to save your certificate from this class as a testament to your achievement!