accident/report/calendar year

Before you can build a triangle, you have to decide which 'year' a claim belongs to — and there is more than one sensible answer. Did a 50,000 claim count for 2023 because the crash happened then, for 2024 because that is when it was reported, or for 2025 because that is when the cheque cleared? Each answer groups claims differently, and each grouping answers a different question. Choosing how to slice time is the first decision in any reserving exercise.

Precisely: an accident year groups claims by when the loss event occurred — the most common basis for reserving, because it matches losses to the period whose risk caused them. A report year groups them by when the claim was first reported, which removes pure-IBNR uncertainty (every claim in a report year is, by definition, already known) and suits claims-made coverage. A calendar year captures everything that moved during a 12-month window regardless of when the accident happened — it is what hits the income statement and is the basis for things like calendar-year paid losses. A related cut, policy year, groups by the policy that was in force. The same claim sits in different cells depending on which basis you choose; for example, a 2022 accident reported in 2023 and paid in 2024 belongs to accident year 2022, report year 2023, and shows payment activity in calendar year 2024.

This matters because mixing bases silently corrupts a triangle. Accident-year triangles develop because of both late reporting and late payment; report-year triangles develop only because of payment/settlement (the reporting lag is already absorbed). A frequent and costly mistake is reading a development pattern off one basis and applying it to data organised on another. A calendar-year diagonal cuts across many accident years at once, which is why a single bad calendar year (a spike in inflation or a court ruling) shows up as a diagonal effect, not a row effect.