I had a great time this week at Risk.Net’s Cyber Risk NA conference this week. I moderated a panel on Modeling Cyber Risk with Jack Jones (EVP RiskLens), Ashish Dev (Principal Economist at the Federal Reserve), Manan Rawal (Head of US Model Risk Mgmt, HSBC USA), and Sidhartha Dash (Research Director, Chartis Research).

We only had 45 minutes and ran out of time before we could get to all the topics I had on my list, so I wanted to included some notes here of things we covered:

- I opened with a scenario where I asked the panelists if they were presenting to the board would it be more honest to disclose the following top risks: 1) IOT, GDPR, and Spectre/Meltdown or 2) Our Top Risk is that we aren’t modeling cyber risk well enough. Most everyone chose option 2 :-)
- We talked about whether there was a right way to model
- Poisson, Negative Binomial, Log Normal
- Frequentist vs Bayesian

- Which model for scenarios makes more sense: BASEL II categories or CIA Triad?
- Level of abstraction required for modeling
- Event funnel: Event of interest vs incident vs loss event
- Top Down vs. Bottoms Up

- What are key variables necessary to model cyber risk (everyone agreed that some measure of frequency of loss and impact/magnitude are necessary)

Things we wanted to get to but ran out of time:

- What is necessary to get modeling approved and validated by Model Risk Management
- Should you purchase an external model or build your own?
- Can we use our Cyber Models for stress testing/ CTE calculations?
- Do we combine cyber scenarios with other operational risk scenarios?
- One audience question that we ran out of time for was “How was the FAIR approach different than LDA & AMA and how does it address their weaknesses (Frequency and severity correlation)”
- This was a good question but to be fair, FAIR wasn’t designed to be a stress testing model. However, many of the inputs used for FAIR are also used for LDA and AMA.

- There were lots of other audience questions about the use of FAIR which is always encouraging!