Back testing

[Sponge]

what is back testing actually doing?

Is it simply running your model after the evens have occurred and assesisnbjow closely your model represents the real world scenario that it is testing?

 

[Simon James]

Yep, that's the idea. For example you know what your historical experience (eg equity markets/inflation etc) has been over the last year, so you can put these into your model and see if your model outcome is anywhere close to where you actually are.

It gives you confidence that your model has some credibility (at least for the particular scenario that you have experienced, not necessarily for what might happen in future!)

 

[Sponge]

Simon, I am still confused about what the input is. In an insurance (life) system, say my initial input are still historical data say at YE17 my cut off is June 2017 at that point I might or might not include estimates for the last 6 months in the year and then I get my say estimated mortality rates at YE17.

Is back testing looking at the impact of more up to date data (lets say now available at YE18) having developed between my cut off date of June 2017 to December 2017, putting the developed data into the same model and seeing what the impact of the revised YE17 estimate is compared to the actual mortality rates at YE17.