Swiss Re Sigma data and falling volatility

In which we back test our previous method, almost accidentally p-value hack, and then run a test on how unusual the last 5 years have been.

In our last two posts, we analyzed the Swiss Re nat cat Sigma data [1]. We built a simple model to validate a claim that a $300bn year is a 1 in 10 year event (which didn’t look like an unreasonable claim), and we looked at how the volatility in annual natural cat claims had changed over time (it had actually reduced).
​
We said last time that we’d do one final post on this data set before moving on, firstly checking that our volatility modelling approach was sound. And secondly, looking at how unusual the last 5 years of losses have been, and whether this is more consistent with noise, or more consistent with some sort of underlying change.

Since I've been rinsing the Swiss Re data, I thought I'd find a nice photo of Switzerland, and stumbled across this great one of the Weisshorn. ​@samferrara

Read More