Perfectly Awesome

Tweedie - insurance premium estimation

Example code on GitHub, R v3.4.3
Background
  • you don't know your product costs until they're sold.
  • count models
  • gamma model (claim severity)
  • claim frequency = claim count/exposure
  • claim severity = claim cost/claim count
  • loss cost = claim frequency * claim severity
    • Tweedle Notes
  • used as distribution for GLMs
  • with cluster at x=0
  • use case: any industry with mix(zeros, non-negative continuous data)
  • features: mean, variance = mu * mean^p
  • mu = dispersion, p = index parameter
  • not defined for p=0->1, Y>0
  • continuous for p=1->2, with mass at y=0
  • continuous for p>2, for Y>0
  • p=1 --> Poisson; p=2 --> gamma, p=3 --> invGaussian