/ Bayesian Statistics

Case Study: Bayesian Modelling of Loss Curves

EXECUTIVE SUMMARY

Using Bayesian modelling techniques, we helped a general insurer improve estimates of claims reserves despite high levels of uncertainty and changes in reserving philosophy.

OUTLINE

An insurer with large commercial liability exposures was unhappy with standard actuarial techniques for estimating their reserves, wanting help using more advanced statistical approaches to the problem.

SOLUTION

Through the use of Bayesian hierarchical modelling, we fit a growth-curve model based on the cumulative claims amounts across each accounting period. Our generative model allowed us to account for complications such as changes in reserving philosophy and changes is business mix.

BUSINESS OUTCOMES

  • Visual exploration of the policy data and concentrations of risk
  • Better understanding of uncertainties in estimates
  • Fuller understanding of implications of reserving changes
  • More efficient capital usage due to increased confidence of model outputs

KEY MODEL FEATURES

  • Business knowledge naturally incorporated into the model
  • Model accounts for uncertainty in a quanitfyable manner
  • Generative modelling approach enables straightforward iteration
  • Posterior distributions provide intuitive outputs

IMPLICATIONS

  • Pooling of data naturally results in more sensible outputs
  • Reduced need for manual data adjustments enhanced auditability

KEY CLIENT BENEFITS

  • Increased confidence in model outputs
  • Explicit statement of assumptions improved awareness
  • Better understanding of output uncertainty improved decision making
  • Knowledge of the approach allowed its use in other business problems

CONTACT DETAILS

Mick Cooney
M: +353 (0)87 819 5992
E: mcooney@agrippadataconsulting.com

Michael Crawford
M: +353 (0)87 996 7437
E: mcrawford@agrippadataconsulting.com

This post was a collaboration between

Mick Cooney, Michael Crawford

  • Mick Cooney

    Mick Cooney

    Mick is a theoretical physicist with a MSc in high performance computing and a PhD in quantitative finance. He runs a couple of data science meetups, is an expert in the use of R and is a Bayesian.

    More posts by Mick Cooney.

    Mick Cooney
  • Michael Crawford

    Michael Crawford

    Michael trained as an actuary but has always worked in IT. He is responsible for providing, project management, data-driven insights, machine learning and AI advice to our financial services clients.

    More posts by Michael Crawford.

    Michael Crawford
Case Study: Bayesian Modelling of Loss Curves

Subscribe to Agrippa Data Consulting