Proposed Solution to COTOR Challenge, Round 2

1
Proposed Solution to COTOR Challenge, Round 2

• Jonathan Evans, FCAS, MAAA

2
Centered Ogive Sample Density
3
Log-Log Space Graph
4
Leap of Faith Log-Log Linear Regression Above
10,000
5
Pareto Tail Estimates Based on Regression
6
Some Key Points

• Regression assumptions are only slightly
  satisfied.
• Confidence interval is for alpha statistic and
  layer pure premium random variable (not a
  statistic).
• Similar regressions (even non-linear) on sample
  densities (or even distribution) have been
  published before.
• Challenge raises deep questions about Bayesian
  philosophy and estimation with extremely high
  intrinsic uncertainty.