Help requested! Risk Analyst interview tomorrow in London!

Hello,

Sorry for the long post. But I need some help urgently.

I’m applying for a new position within a major CIB (let’s call it Bank ABC) in London and my interview is tomorrow. The job is risk analyst position in their loan department. I have done my research and I have a good idea of the questions which might come up during the interview as they have been asked from previous candidates.

Please give feedback (negative or positive! ). Especially the answer to question #2 could use some improvement, I think!

Question 1: How do you think the recent current crisis might impact the credit/counterparty risk of our bank? What kind of customer segment could be impacted the most? How would you go about models would you build in order to anticipated the impact?

I would start of with an analysis using stress testing and scenario analysis using the following steps:

1. divide the loan portolio into different customer segments, such as mortgage loans, corporate bonds etc.
1. divide the customer segments into sub-segments based on the customer credit score/rating.
1. Use top-down macroeconomic analysis to decide the main drivers to be tested. Eg. GBP growth, unemployment rate, interest rate levels (LIBOR) etc.
1. Run the different segments/sub-segments through different scenarios/stress tests to obtain estimates of Prob.of Default, LGD and Expected Loss at Default.
1. Collect the data and synthesis the finding into one analysis showing the PD, LGD and EAD for the entire loan portfolio.

As I don’t believe that there is only one failproof model to estimate/model credit and cpty risk, I would then also use Historical scenarios. I would run the current loan portfolio through the past historical crisis, such as 1987 and 2008 in order to see how the portfolio would have survived.

Question2 :

Imagine you are introduced a credit risk estimation model which preditcts the Prob. of default of a customer, say a someone who has a mortgage. How would you go about analysing whether the model works? What kind of statistical tools would you use?

I would start of my calculating a “base rate” figure which I would compare to the results provided by the credit risk estimation model. Base rate would be calculated in the following manner:

1. Choose approx. 1000 mortgage borrowers who has already paid off their loan or defaulted on their obligations and who share the same credit score as the customer.
1. Trusting the Law of Large Numbers, calculate the default probability of the sample.
1. Compare the probability to the probability calculated using the credit risk estimation model
I would be very thankful if you could give some feedback on my responses.

Question 2:

Depends on how I am implementing the score. Here are some high level topics to brush up on:

1. Run Champion / Challenger model strategy on a subset of live deals. Set the benchmarks to measure performance (6 month DQ for example). Decide on the statistical test appropiate to test the difference (I’d use bayesian statistics on a binary outcome of DQ or not, but that is because I’m becoming a fan of expressing probabilities based on models)
2. Confusion Matrix where we calculate sensitivity and specificity on out of sample data
3. KS Test on out of sample data
4. ROC curve on out of sample data (This may be one of the more common)

I think talking any of the above 4 would work. ROC curve may be the easiest to learn. Champion / Challenger is the only true experimental method that isolates the variable of the new model change on all aspects of the business (defaults, do you win the deal to begin with, etc).

You need to think in terms of rank ordering risk. For each change in score, you should expect the odd ratios to be mostly monotonically increasing or decreasing.

Comparing the average default rate vs. the predicted default rate is not very useful IMO