The Actuary and IBNR Techniques: A Machine Learning Approach

Significant time is spent adjusting triangles and selecting parameters when calculating IBNR reserves. However, these choices are often made using subjective “expert judgement” with limited to no scientific basis and no measure of predictive accuracy. In this session, we will demonstrate how machine learning techniques can be used to provide an automated, scientific, and rigorous basis on which to choose parameters that maximize predictive accuracy and stability of reserves. We demonstrate this framework on several real-world triangles and show that it improves predictive accuracy while ensuring reserves are adequate.

  • Date:Tuesday, September 14
  • Time:10:00 AM - 11:15 AM EDT
  • Session Type:Concurrent Session
  • Session Code:AR-4
  • Learning Objective 1:Understand how common machine learning techniques can be used for selecting optimal reserving parameters, and how these parameters impact results.
  • Learning Objective 2:Understand how different scoring metrics impact the selection of optimal parameters, and which metrics suit which situation.
  • Learning Objective 3:Understand possible future applications, enhancements, and practical use cases to take the methods further.
  • Level of Knowledge:Level 2: General knowledge of the subject
  • Moderator:Chris Holt