Every entry from the homepage's Development section, plus everything that came before it — sorted by date.
A top-down approach to reserving across multiple business lines: first aggregate several triangles, calculate the reserve for the combined dataset, then allocate that reserve back to each individual business line. Among other benefits, this can reduce the impact of extreme values, provide a clearer understanding of uncertainty, and improve operational efficiency through a coherent portfolio-level analysis. Expected summer 2027.
Operational time — a shift in a portfolio's reporting or settlement speed over time — can distort development patterns if left untreated. We're developing an algorithm to detect whether a dataset exhibits operational time and quantify its magnitude, so it can be corrected for when the effect is significant. Expected November 2026.
Our original method produces a constant, taken as the average of the most recent selected number of mean cost per claim or payment values. Our new method instead fits a linear regression across those same recent values, producing a trend line rather than a flat constant. The two methods are closely related to local constant and local linear smoothing, respectively, but are formulated in a more intuitive way. Optimal smoothing method is picked via validation.
Built on the earlier DCL/TEDCL validation work with a structured framework for reserve uncertainty: the FIMI-residual, decomposed into model, parameter, and bias error, with variance modelled against reserve size via a Gamma-type mean-variance relationship — validated on a real financial-loss dataset and built toward operationalizing directly in GRID.
Added the ability to produce meaningful reserve analysis even when data is limited or noisy, by averaging a chosen number of recent periods into a constant mean cost per claim or payment figure.
Added seasonal effect calculation at quarterly granularity, for both counts and payments, using historical data or manually entered periods — with a choice of how many recent periods to use when computing missing year parameters.
Added Bornhuetter-Ferguson as a modelling option, applicable to counts and/or payments — counts as exposure, payments as either exposure or inflation.
Added the ability for every model (DCL or TEDCL, with or without calendar effects) to produce two sets of forward factors — incurred and paid — plus the ability to input your own forward factors and check which model comes closest to them.