Edition 4 | Week 2 of Reserving | Actual vs. Expected | Alternative Lens

The Decision Question

In edition 3 we concluded that the scatter signals when a question should be asked but does not answer it.

This edition addresses one of those unanswered questions: how did each accident year evolve?

This visual gives a view of ultimate estimates by accident year, showing the initial view, the current view, and the movement between them.

Direction of Revision by Accident Year	For each accident year, we immediately see: · Strengthening (current above initial) · Release (current below initial) · Stability (little to no movement)
Magnitude of Revision (Level or %)	Depending on which version: Top chart → absolute magnitudeBottom chart → proportional magnitude This makes visible: · Which accident years were materially revised · Whether small years were volatile in percentage terms · Whether recent years behave differently from older ones It moves from “are we calibrated?” to: “Which cohorts caused us to change our mind?”
Stability vs Instability	The intermediate markers show: · Smooth evolution (gradual adjustment) · Volatility (zig-zag behaviour) · One-off shocks
Concentration of Risk	You can now answer: • Is instability concentrated in recent accident years? • Do older years stabilise quickly? • Is revision systematic across cohorts? That was impossible in the scatter.

It shows revision behaviour by cohort — not emergence mechanics

This dumbbell chart can be harder to explain:

This view introduces a time-within-time structure (accident year and valuation year), which increases cognitive load. It is diagnostic, not reassuring.
It’s about behaviour, not outcome. The scatter answered: Did we get it broadly right? This answers: How did we change our mind? That’s a different, more technical question.

This dumbbell chart can be easier to misread:

Movement can be mistaken for volatility, even if it was minor calibration refinements.
Large levels dominate perception (top chart).
Readers May Infer causality – but the chart shows revision, not error.

Use this when….

Don’t use this when…

You want to understand how each accident year evolved from its initial estimate to the current view.
The question has shifted from “Are we broadly calibrated?” to “Which cohorts caused us to change our mind?”
You need to assess direction and magnitude of revision by accident year.
You want to evaluate stability versus volatility in the reserving process.
The audience is comfortable discussing cohort-level behaviour rather than portfolio-level reassurance.
You are supporting reserving committee, peer review, or governance discussions.

The goal is simply to provide high-level assurance to management or the Board.
The key question is overall calibration, not behavioural evolution.
The message is that there has been no material change (this would overcomplicate it).
The audience is not familiar with the distinction between accident year and valuation year.
You need to explain why revisions occurred (this chart shows movement, not drivers).
You are trying to communicate emergence timing within the development cycle.