Chaudhury - review of key issues in op risk cap modeling

[southeuro]

writing on external data issues, chaudhury mentions reporting bias. specifically he says, "while the compilation of truncation point used by an external data vendor is knows (so far so good), truncation level of loss size above which loss events reach the public domain is not knows" (huh?!).

Further he says "the higher this unknown reporting threshold, then the greater the extent of the under-reporting phenomenon, the reported losses will appear more skewed towards higher severities than the true losses are."

Does anyone make sense out of this gibberish? Thanks for clarifying

 

[Roshan Ramdas]

writing on external data issues, chaudhury mentions reporting bias. specifically he says, "while the compilation of truncation point used by an external data vendor is knows (so far so good), truncation level of loss size above which loss events reach the public domain is not knows" (huh?!).

Further he says "the higher this unknown reporting threshold, then the greater the extent of the under-reporting phenomenon, the reported losses will appear more skewed towards higher severities than the true losses are."

Does anyone make sense out of this gibberish? Thanks for clarifying

Hi,
Per my understanding, losses that get reported in the public domain (external loss data) tends to be biased towards high magnitude events that generally receive a lot of publicity.
Basing your operational risk regulatory capital charges predominantly on external loss data can thereby result in the overstatement of required capital.
Thank you

 

[southeuro]

Yes! Now it makes more sense. Here's what I learned since I posted that question: external data needs to be scaling to offset certain biases they intrinsically have (namely, scale --big firms have big loss, small/small--, truncating --no capture of data below a certain threshold--, and data capture --correlation between size and reporting-- biases). Thanks!