New convergence reports proposed

[johcarter]

The reports outlined the attached excel specification are designed to answer the question: ‘How much sampling error is there in my annual loss estimates?’ for any sample based loss model output. ORD_convergence_tables_v5.xlsx

The two new reports proposed are;

1. Convergence in Annual Loss Table (C ALT)
2. Convergence in Exceedance Probability Table (C EPT)

• Only relevant for sample-based output
• Can be produced for any summary level and financial perspective
• Used to guide the choice of number of samples for similar runs

The tables contain the following information:

• confidence intervals and standard errors for the annual loss estimates
• recalculated estimates using increasing subsets of the total number of samples to help visualise convergence of the estimates
• attribution of overall standard error to primary uncertainty (events/hazard), and secondary uncertainty /vulnerability i.e. sampling variability given the event. This helps to understand the relative importance of running additional samples for robustness of a given metric.

References OasisLMF project documentation:

• Approach to improving convergence proposal convergence_proposals_shared_v2.pdf
• Research paper ‘Quantifying the sources of simulation uncertainty in natural catastrophe models’ (Kaczmarska, Jewson, Bellone)
• Mathematical methodology anova_technique_methodology_v1.pdf

[benhayes21]

postponed until proved to be useful