Excellent presentation by Ailish. She explained the various measures very clearly with detailed examples. Usually, no one likes to explain such concepts so clearly with examples but these so called 'experts' talk round and round using jargons. However, it is amusing to watch her discomfort when answering questions!! In fact, the questions were very easy to answer compared to what she explained.
To the question in what can cause an override at 46:10. A typical override can be if the relationship manager gets a call from the customer that they are considering their financial options due to divorce. The model cannot capture that a joint obligation is considering divorce which increases the risk of defaulting. The relationship manager can then override and give the obligation a higher PD.
53:14 I dont agree. A PD model is a classification model and you need to validate the model against overfitting. You need an out of sample and out of time data to validate that the model works across different time periods. Of course, if you have a low default portfolio, then it would nearly be impossible to do so if you don’t have enough defaults to model on.
Excellent presentation by Ailish. She explained the various measures very clearly with detailed examples. Usually, no one likes to explain such concepts so clearly with examples but these so called 'experts' talk round and round using jargons. However, it is amusing to watch her discomfort when answering questions!! In fact, the questions were very easy to answer compared to what she explained.
To the question in what can cause an override at 46:10. A typical override can be if the relationship manager gets a call from the customer that they are considering their financial options due to divorce. The model cannot capture that a joint obligation is considering divorce which increases the risk of defaulting. The relationship manager can then override and give the obligation a higher PD.
Very useful!
53:14 I dont agree. A PD model is a classification model and you need to validate the model against overfitting. You need an out of sample and out of time data to validate that the model works across different time periods. Of course, if you have a low default portfolio, then it would nearly be impossible to do so if you don’t have enough defaults to model on.